MISMANAGED

TRADE?
 KENNETH FLAMM

MISMANAGED
TRADE?
Strategic Policy
and the
Semiconductor Industry

BROOKINGS INSTITUTION PRESS

Washington, D.C.
 Copyright © 1996 by
THE BROOKINGS INSTITUTION

1775 Massachusetts Avenue, N.W., Washington, D.C. 20036

All rights reserved

Library of Congress Cataloging-in-Publication data:

Flamm, Kenneth, 1951—

Mismanaged trade : strategic policy and the semiconductor industry /
Kenneth Flamm.
p. cm.
Includes bibliographical references and index.
ISBN 0-8157-2846-8 (cloth: alk. paper) — ISBN 0-8157-2847-6
(pbk.: alk. paper)
1. Semiconductor industry—Government policy—United States.
2. Semiconductor industry—Government policy—Japan. 3. United
States—Foreign economic relations—Japan. 4. Japan—Foreign
economic relations—United States. I. Title.
HD9696.S43U466 1996
382'.4562138152'0973—dc20 94-21324
CIP

987654321

The paper used in this publication meets the minimum requirements of the
American National Standard for Information Sciences—Permanence of Paper for
Printed Library Materials, ANSI Z39.48-1984.

Set in Times Roman

Composition by Harlowe Typography, Inc.

Cottage City, Maryland

Printed by R.R. Donnelley and Sons Co.

Harrisonburg, Virginia
THE BROOKINGS INSTITUTION

The Brookings Institution is an independent organization devoted to nonpartisan research,

education, and publication in economics, government, foreign policy, and the social sciences
generally. Its principal purposes are to aid in the development of sound public policies and
to promote public understanding of issues of national importance.
The Institution was founded on December 8,1927, to merge the activities of the Institute
for Government Research, founded in 1916, the Institute of Economics, founded in 1922,
and the Robert Brookings Graduate School of Economics and Government, founded in
1924.
The Board of Trustees is responsible for the general administration of the Institution,
while the immediate direction of the policies, program, and staff is vested in the President,
assisted by an advisory committee of the officers and staff. The by-laws of the Institution
state: “It is the function of the Trustees to make possible the conduct of scientific research,
and publication, under the most favorable conditions, and to safeguard the independence
of the research staff in the pursuit of their studies and in the publication of the results of
such studies. It is not a part of their function to determine, control, or influence the conduct
of particular investigations or the conclusions reached.”
The President bears final responsibility for the decision to publish a manuscript as a
Brookings book. In reaching his judgment on the competence, accuracy, and objectivity of
each study, the President is advised by the director of the appropriate research program
and weighs the views of a panel of expert outside readers who report to him in confidence
on the quality of the work. Publication of a work signifies that it is deemed a competent
treatment worthy of public consideration but does not imply endorsement of conclusions
or recommendations.
The Institution maintains its position of neutrality on issues of public policy in order to
safeguard the intellectual freedom of the staff. Hence interpretations or conclusions in
Brookings publications should be understood to be solely those of the authors and should
not be attributed to the Institution, to its trustees, officers, or other staff members, or to
the organizations that support its research.

Board of Trustees

James A. Johnson Vartan Gregorian Constance Berry Newman

Chairman Bernadine P. Healy Maconda Brown O’Connor
Samuel Heilman Samuel Pisar
Leonard Abramson Warren Heilman Rozanne L. Ridgway
Michael H. Armacost Robert A. Helman Judith S. Rodin
Ronald J. Arnault Thomas W. Jones Warren B. Rudman
Elizabeth E. Bailey Breene M. Kerr Michael P. Schulhof
Alan M. Dachs Thomas G. Labrecque Robert H. Smith
Kenneth W. Dam Donald F. McHenry Vincent J. Trosino
D. Ronald Daniel Jessica Tuchman Mathews John D. Zeglis
Stephen Friedman David O. Maxwell Ezra K. Zilkha

Honorary Trustees

Vincent M. Barnett, Jr. Walter Y. Elisha Donald S. Perkins

Rex J. Bates Robert F. Erburu J. Woodward Redmond
Barton M. Biggs Robert D. Haas Charles W. Robinson
Louis W. Cabot Teresa Heinz James D. Robinson III
Edward W. Carter Andrew Heiskell David Rockefeller, Jr.
Frank T. Cary Roy M. Huffington Howard D. Samuel
A. W. Clausen Vernon E. Jordan, Jr. B. Francis Saul II
John L. Clendenin Nannerl O. Keohane Ralph S. Saul
William T. Coleman, Jr. James T. Lynn Henry B. Schacht
Lloyd N. Cutler William McC. Martin, Jr. Robert Brookings Smith
Bruce B. Dayton Robert S. McNamara Morris Tanenbaum
Douglas Dillon Mary Patterson McPherson John C. Whitehead
Charles W. Duncan, Jr. Arjay Miller James D. Wolfensohn
.
 For
Dianne
 Foreword

Since 1959 the United States and Japan have repeatedly negotiated with
one another to resolve disputes involving trade in semiconductors. During
these four decades the American and Japanese chip industries have been
reluctant pioneers, cutting their way through a forest of issues created by
the unique conditions of high-technology trade and investment. The new
directions explored by the two countries’ policymakers evolved into the
1986 U.S.-Japan Semiconductor Trade Arrangement. Today, as that ex¬
periment and its 1991 successor agreement are about to expire, the
United States and Japan seem poised to again change course.
In this study Kenneth Flamm analyzes the public policy issues raised
during these years of friction over semiconductor trade. Motivated by
strategic economic and national security rationales, both the United
States and Japan pursued policies designed to advance their industrial
objectives. Policy conflicts occurred over trade, technology, development,
investment, and antitrust enforcement as Japan’s microelectronics indus¬
try caught up with its American competitors and pulled ahead in some
areas. U.S. short-range tactics sometimes undercut long-term goals. The
author concludes that over the long term improved trade rules and
strengthened enforcement mechanisms must be developed to ensure open
and competitive international markets for semiconductors. Effective so¬
lutions for trade in semiconductors will undoubtedly be applicable to
other industries and ultimately may serve as the model for further
strengthening the more general multilateral trading system codified in
the General Agreement on Tariffs and Trade.
The author is grateful to Dan Hutcheson, F. M. Scherer, and Michael
Smith for the detailed comments they made on the first draft of this study.
Extensive comments received from Daryl Hatano, Thomas Howell, Matt
IX
X FOREWORD

Rohde, John Steinbruner, Makoto Sumita, and Alan Wolff also greatly
improved this book, as did comments from Bill Finan, Gary Hufbauer,
Peter Reiss, Gary Saxonhouse, Carl Shapiro, Jack Triplett, and Philip
Webre on articles based on this research. Executives in U.S., European,
and Japanese electronics companies were generous with their time and
extraordinarily helpful when they spoke with the author as he went about
researching this book, as were many government officials in the United
States, Japan, and the European Community. The author is also grateful
to Japan’s Ministry of International Trade and Industry and Ministry of
Finance for their financial and logistical support of short stays at the
Tokyo research institutes associated with those ministries and their help
in arranging interviews with Japanese government and industry officials.
The author thanks the Rockefeller Brothers Fund and the John D.
and Catherine T. MacArthur Foundation for their financial support of
this study.
First-rate research assistance was provided by Yuko Iida Frost and
Kaori Nakajima. Michael Treadway and James Schneider edited the man¬
uscript. Melanie Allen, Ian Campbell, Cynthia Iglesias, and Andrew
Solomon verified its factual content. Kris McDevitt and Ann Ziegler
provided staff assistance. Ellen Garshick proofread the pages and Julia
Petrakis compiled the index.
The views expressed in this book are those of the author and should
not be ascribed to the persons or organizations whose assistance is ac¬
knowledged or to the trustees, officers, or other staff members of the
Brookings Institution.

MICHAEL H. ARMACOST
President
May 1996
Washington, D.C
 Contents

1. Introduction 1
Challenges for Public Policy 1
Technology and Industrial Structure in Semiconductors 7
International Competition in Semiconductors 18
Strategic Policy in the United States 27

2. New Competition: The Japanese Ascent in

Semiconductors 39
First Steps 40
Early Exports in Electronics 49
The Seal of Approval 52
Protecting the Japanese Market 55
The Integrated Circuit Arrives in Japan 59
A Secret Truce in the “Patent War” 68
Battles in the LSI Market 70
First Steps toward Liberalization and “Dumping” 73
The “Calculator War” 75
Further Liberalization and Crisis 77
Into the Japanese Market 88
NTT Arrives on the VLSI Scene 90
The VLSI Project 94
The Continuing Role of Government 113
Dependence in Silicon 119
Summary 124

3. The Genesis of an American Trade Policy in

Semiconductors, 1959-84 127
A Threat to National Security, 1959 128
Private versus Public Policy:
Television Exports in the 1960s 132
Competition in the 1970s 136

xi
xn CONTENTS

“Below Cost” Dumping 141

Mounting Frictions: “Quality Dumping” 144
Organizing a Response 147
Competition, Collusion, or Predation?
The 64K. DRAM Wars 148
Sectoral Negotiations 153

4. The Semiconductor Trade Arrangement and

Its Aftermath 159
A Thumbnail Historical Sketch 159
Evolution of the Semiconductor Trade Regime 162
Initial Implementation of the Arrangement,
August 1986 to November 1987 175
The Privatization of Restraints,
December 1987 to Mid-1989 201
“Coordination Structures” and “High Price Stability,”
1989 to 1990 212
U.S. Memories to the Rescue 216
Aftermath: The Second Semiconductor Trade
Arrangement 223

5. Effects of the Semiconductor Trade Arrangement

on Semiconductor Markets 227
Impacts on Semiconductor Supply and Demand 231
Impacts of Regional Price Differentials on Regional
Welfare 272
DRAMs versus EPROMs 278
Import Promotion and the STA 279
Conclusions 292
Appendix 5A: The Economics of Contract Pricing 294
Appendix 5B: Construction of Estimated FMVs 301

6. Dumping in DRAMs 305

The Economic Rationality of Below-Marginal-Cost
Pricing 307
Cost Structures and Dumping in the Semiconductor
Industry 311
Modeling the Semiconductor Product Life Cycle 313
Model Solution 334
Conclusions 358
Appendix 6A: A General Solution to the Problem of Optimal
Capacity and Production Choice 359
CONTENTS Xlll

Appendix 6B: Detailed Solution with Specific Demand and

Learning Curve Assumptions 365

7. Strategic Issues 372

Conceptions of “Strategic” 372
Semiconductor Dependency and Strategic Trade Policy 382
Anecdotal Evidence on Private Collusion 396
Empirical Tests for Competition 398
The Costs of Facing a Cartel 405
More Complex Stories about Cartels 415
Conclusion 417
Appendix 7A: Solution of the Model with Strategic
Behavior 418

8. Conclusion: Mismanaged Trade? 425

Why Semiconductors? 427
Strategic Policy in Semiconductors 429
Effects of the Semiconductor Trade Arrangement 433
Where Do We Stand? 437
Pricing and Dumping 442
Where Do We Go from Here? 449
The Global Semiconductor Conference 454
From Here to There 455
Mismanaged Trade? 457

Index 461

Tables
1-1. Share of Worldwide R&D Expenditures and Employment for
U.S.-Based Companies and Their Majority-Owned Foreign
Affiliates, by Parent’s Industry, 1989 14
1-2. Capital Costs of a Typical Semiconductor Wafer Fabrication
Facility, Mid-1970s-Early 1990s 16
1-3. Domestic R&D Expenditures and Employment by U.S.-Based
Companies and Their Majority-Owned Foreign Affiliates as
Shares of Worldwide Totals by Parent’s Industry, 1989 19
1-4. World Market Shares of Merchant and Captive Semiconductor
Manufacturers, by Region Where Headquartered, Selected
Years, 1981-93 20
1-5. U.S. Government Shares of Total R&D Funding,
by Industry, 1989 28
1-6. U.S. Industrial Consumption of Semiconductors,
by Consuming Sector, Selected Years, 1963-87 35
xiv CONTENTS

1-7. Sources of R&D Funds in the U.S. Semiconductor Industry,

Selected Periods, 1958-76 36
1- 8. Federal Contracts as Share of Total U.S. Semiconductor
Shipments, Selected Years, 1963-87 37
2- 1. Transistor Output of U.S. and Japanese Companies during
Transition to Overseas Production, 1957-68 47
2-2. FILP Loans to Japanese Machinery and Electronics Industries,
1956-84 64
2-3. Sales by Japanese Semiconductor Companies to Affiliated and
Nonaffiliated Companies, 1966-71 76
2-4. Japanese Measures to Liberalize Trade and Investment in the
Electronics Industry, 1962-76 88
2-5. Japanese Expenditures on Integrated Circuit R&D,
by Funding Source, 1963-79 97
2-6. Shares of Total VLSI Research Association Project Personnel,
by Lab Group, 1976-86 108
2-7. MITI Expenditures on R&D Programs Related to
Microelectronics, Fiscal Years 1983-92 114
2-8. Joint Patent Applications with NTT, by Number of
Participating Companies, 1980-86 117
2- 9. Government and NTT Semiconductor-Related R&D Funding
in Relation to Private Japanese Semiconductor R&D Funding,
1988 118
3- 1. Factory Prices of Finished Integrated Circuits in Europe,
Japan, and the United States, 1976 139
5-1. Market Concentration in DRAMs, 1979-89 256
5-2. Market Concentration in EPROMs, 1980-89 257
5-3. Shares of U.S. and Japanese EPROM Manufacturers in the
Japanese Market, 1988-92 265
5-4. Estimated Welfare Effects of Removing Japanese Border
Controls on DRAMs, 1987-88 274
5- 5. Foreign Share of the Japanese Semiconductor Market Due to
Changes in Foreign Penetration and Composition, Selected
Periods, 1986-94 286
5-A1. Distribution of DRAM Contracts to Start (Lead) and Length,
1984-89 296
5-B1. Foreign Market Value (FMV) Projected Cost Estimates for
Japanese DRAM Manufacturers, by Company, Selected
Quarters, 1986-89 302
6- 1. Regression Results from Analysis of Firm-Level Learning
Curves 340
6-2. Regression Results from Analysis of 1M DRAM Demand 344
6-3. Assumed Values for Empirical Parameters 346
6-4. Solution of the Optimal Control Problem, Nonstrategic
Baseline Simulation 348
CONTENTS XV

6- 5. Semiconductor Industry Characteristics under Alternative

Symmetric Equilibrium Conditions 350
7- 1. Solution of the Optimal Control Problem, Strategic Simulation 410
7-2. Simulated Welfare Effects of Cartelization on the DRAM
Market 413

Figures
1-1. Year-to-Year Changes in DRAM Prices, 1972-89 10
1-2. U.S. Semiconductor and Electronic Component Company
R&D as Share of Sales and Japanese Company IC R&D as
Share of Semiconductor Sales, 1976-94 12
1-3. U.S. Company Semiconductor Plant and Equipment
Investment and Japanese Company IC Plant and Equipment
Investment as Share of Semiconductor Sales, 1976-94 15
1-4. World Market Shares of Merchant Semiconductor Companies,
by Region, 1970-95 22
1-5. U.S. Share of Combined U.S. and Japanese Semiconductor
Industrial Employment, Value Added, and Sales, 1968-93 23
1- 6. Federal Government Share of U.S. Electronic Components
R&D Expenditures, 1972-92 29
2- 1. Output of Transistors and Diodes by Japanese Manufacturers,
September 1954-March 1956 46
2-2. Volume of U.S. Transistor Production, by Type, 1955-65 50
2-3. Value of U.S. Transistor Production, by Type, 1955-65 51
2-4. Import Shares of Japanese Semiconductor Consumption,
1966-94 62
2-5. Japanese Tax Incentives and Subsidized Lending for
Electronics Development, and Investment in Electronics by
Major Companies, 1961-71 65
2-6. Japanese R&D Subsidies for Electronics Development, and
R&D in Electronics by Major Companies, 1961-71 67
2-7. Japanese Consumption of Integrated Circuits and Import
Market Share, 1968-74 78
2-8. Major Results of the Japanese VLSI Project, 1975-90 101
2-9. Expenditures on VLSI Project and Subsequent Cooperative
Research, by Category, 1976-86 103
2-10. Linkages within VLSI Industrial System 104
2-11. Public and Private Funding of VLSI R&D Association,
1976-86 112
4-1. Errors in MITI Forecasts of DRAM Production, Quarterly,
1986-88, and Semiannually, 1987-94 198
4-2. Errors in MITI Forecasts of Japanese EPROM Production,
Quarterly, 1986-88, and Semiannually, 1987-94 199
4-3. Year-to-Year Changes in Fisher Ideal Price Indexes for DRAMs
and EPROMs, 1972-89 210
xvi CONTENTS

4- 4. Output of 1M DRAMs by Japanese and Non-Japanese

Manufacturers, 1989 213
5- 1. Prices of 64K DRAMs in Contract, Distributor, and Spot
Markets, 1983-85 235
5-2. Year-to-Year Changes in DRAM Prices and Quantities
Produced, 1972-89 238
5-3. Year-to-Year Changes in EPROM Prices and Quantities
Produced,1972-89 239
5-4. Prices for Four Generations of DRAMs, Pi Rule, 1974-89 241
5-5. Prices for Three Generations of DRAMs, Bi Rule, 1987-95 242
5-6. Quarterly Dataquest Monday Contract Prices and Average
Selling Prices for 256K DRAMs, 1987-92 245
5-7. Quarterly Dataquest Monday Contract Prices and Average
Selling Prices for 1M DRAMs, 1987-92 246
5-8. Monthly Contract Prices for 256K DRAMs in the United
States and Japan, 1985-92 247
5- 9. Monthly Contract Prices for 1M DRAMs in the United States
and Japan, 1987-92 248
5-10. Quarterly Japanese and World Contract Prices for 64K
DRAMs, 1985-89 249
5-11. Monthly Spot Prices for 256K DRAMs in the United States
and Japan, 1985-91 250
5-12. Monthly Spot Prices for 1M DRAMs in the United States and
Japan, 1986-92 251
5-13. Quarterly Fisher Ideal Price Indexes for Large-User Purchases
of DRAMs in the United States and Japan, 1986-89 253
5-14. Quarterly Wholesale Prices for 128K and 256K EPROMs in
the United States and Japan, Selected Years, 1985-92 255
5-15. Hirschman-Herfindahl Indexes and Cumulative Output for
Five Generations of DRAMs 258
5-16. Hirschman-Herfindahl Indexes and Cumulative Output for
Six Generations of EPROMs 259
5-17. Yield Plot for IBM’s 64K DRAM 261
5-18. Japanese Output of DRAMs and EPROMs, 1986-95 263
5-19. Japanese Exports of DRAMs and EPROMs as a Share of
Total Output, 1986-95 264
5-20. Average Foreign Market Value (FMV) across Companies for
256K DRAMs, 1986-91 268
5-21. Average Foreign Market Value (FMV) across Companies for
1M DRAMs, 1986-91 270
5-22. Prices and Estimated Foreign Market Value (FMV) for 512K
and 1M EPROMs, 1987-90 271
5-23. Alternative Measures of the Foreign Share of the Japanese
Semiconductor Market, 1986-95 282
6- 1. Relationship between Semiconductor Price and Cost,
Assuming “Small” Output-Sensitive Variable Costs 327
CONTENTS Xvii

6-2. Empirical Approximation of the Learning Curve in

Semiconductor Manufacture 330
6-3. Time Profile of 1M DRAM Costs and Prices in Simulated
Nonstrategic Equilibrium with Gamma = 1 353
6-4. Time Profile of Semiconductor Costs and Prices in Simulated
Nonstrategic Equilibrium with Gamma = 0 354
6-5. Historical Prices for 1M DRAMs Compared with Simulated
Nonstrategic Equilibrium Time Profiles 355
6- 6. Time Profile of Semiconductor Costs and Prices in Simulation
with Very High Initial Yields 357
7- 1. “Food Chain” Concept in Semiconductor Industry 375
7-2. Theoretical Case for Domestic Production of Semiconductors 384
7-3. Mitsubishi Competitive Marginal Revenue versus Cost of
Manufacture (COM) Reported to Commerce Department,
1989-91 405
7-4. NEC Competitive Marginal Revenue versus Cost of
Manufacture (COM) Reported to Commerce Department,
1989-91 406
7-5. Historical Prices for 1M DRAMs Compared with Simulated
Strategic Equilibrium Time Profiles 412
I

'
MISMANAGED
TRADE?
 CHAPTER ONE

Introduction

In July 1991 the U.S. government formally concluded negotiation of a

new Semiconductor Trade Arrangement (STA) with Japan, and set out a
framework for trade and investment in microelectronics between the two
nations for the next five years. That agreement, replacing an earlier,
pathbreaking 1986 pact, continued a controversial experiment in trade
policy with the potential to alter the direction of the international trading
system for high-technology goods. In 1996, as renewal of the 1991 agree¬
ment again loomed large, there were few signs that a consensus had yet
emerged as to the lessons of the past decade’s experience with semicon¬
ductor trade agreements. The rules of the game in international trade
and investment continue to avoid very real challenges posed by both
underlying economic fundamentals and real-world national policies in
high-technology industries. This book is about how and why the semi¬
conductor industry turned into the front line in a battle between national
trade and industrial policies, and about the profound issues raised by this
conflict.

Challenges for Public Policy

The 1991 semiconductor negotiations received surprisingly little notice
as they were being concluded, probably because they addressed what was
thought to be a troublesome little “sectoral” dispute between U.S. and
Japanese companies. The U.S. semiconductor industry’s global output,
after all, was then only a tenth the size of the U.S. motor vehicle market.

1
2 INTRODUCTION

U.S. firms’ sales of semiconductor chips in 1990 were $19 billion (out of
a worldwide industry total of $50 billion), compared with $200 billion in
American motor vehicle shipments. And aside from the semiconductor
industry itself, the only segments of U.S. industry to take an active
interest in semiconductor trade policy were the computer and commu¬
nications equipment sectors, the chief users of the products most affected
by the negotiations.
Despite the lack of notice, the discussions carried out behind closed
doors in government offices near the White House were anything but a
minor sideshow. As in other high-technology sectors, where global com¬
petition had mainly pitted American and Japanese companies against
one another, bilateral agreements between these two nations were his¬
torically the model for expanded arrangements as other countries’ market
presence grows.
By 1996, stakes that were already big had grown larger. Global semi¬
conductor sales in 1995 had soared to $150 billion, now more than 20
percent of world vehicle sales. Forecasts predicted that by 2000, global
chip sales would amount to 40 percent of global motor vehicle sales.1
Increasingly, able new competitors in East Asia—particularly Korea and
Taiwan—had emerged as major new players in the global chip industry,
pushing past European companies in their share of the global market in
1994. As negotiators from Washington and Tokyo convened their trade
talks, corporate and political leaders in Seoul, Taipei, and the capitals of
Europe made little effort to disguise their intense interest in the pro¬
ceedings. Today’s bilateral precedents were clearly likely to shape to¬
morrow’s multilateral system.
Precedent-setting features of the semiconductor pacts have already
been mimicked in other sectoral trade negotiations. The semiconductor
agreements’ market share target for foreign sales in the Japanese chip
market, widely hailed as a step toward “managed trade” by both sup¬
porters and opponents, was acknowledged by American government of¬
ficials to be the inspiration for a 1992 bilateral agreement setting quan¬
titative targets for increased sales of U.S.-made auto parts in the
Japanese market.2 More recently, apparent success in achieving market

1. See Dataquest, “Dataquest Predicts Worldwide Semiconductor Market to Double

by 2000,” press release, San Jose, Calif., October 2, 1995; and Department of Commerce,
U.S. Global Trade Oudook, 1995 (October 1995).
2. “[President George] Bush’s aides ... are philosophically opposed to ‘managed
trade’—the setting of quantifiable goals and market shares for each nation’s industries. But
INTRODUCTION 3

share targets in semiconductors led the Clinton administration’s trade

policymakers to focus on quantitative indicators of access to Japanese
markets in their approach to a bilateral economic framework.* * 3 Japanese
officials seemed just as determined to resist any further use of such
targets or indicators.
Even more important, trade frictions in semiconductors were just the
beginning of what promises to be a massive tangle of trade disputes in
other high-technology sectors. In computers, telecommunications, sat¬
ellites, pharmaceuticals, advanced materials, and aircraft, continuing in¬
dustrial skirmishes have been taking place. Although the particulars vary,
the central issues are almost always the same as those at the heart of the
semiconductor negotiations.
There are two important issues. How can government subsidies and
support for advanced technology—for decades an accepted and econom¬
ically beneficial practice in the United States, Japan, and virtually all
other industrialized countries—coexist with open international economic
competition in global high-technology industries? The postwar trading
system aspired to an ideal of free competition among firms from all
nations operating in a single, open global market, with market outcomes
determined only by the efficiency and effectiveness of individual com¬
panies. How can this vision be reconciled with the realities of a world in
which national governments make large investments in new technologies,
which may totally alter the industrial landscape, and inevitably favor
those firms with the easiest access to the innovations created with these
subsidies?4

they also realize that, as recent experience shows, that is the only way to deal with Ja¬
pan. . . . ‘We knew we had to pattern this agreement after the semiconductor agreement,’
said an official in the Commerce Department. ‘But we could not appear to be agreeing to
specific numbers.’
The Japanese know that, too, and, eager to make Mr. Bush’s trip a success, the Gov¬
ernment here spent the last few weeks cajoling and arm-twisting to force Japanese auto¬
makers to once again revise plans to increase their imports. Because the announcements
came from the companies, not the Government, the United States was able to argue that
such ‘unilateral’ proposals from Japanese companies did not amount to managed trade.”
David E. Sanger, “Bush in Japan; A Trade Mission Ends in Tension as the ‘Big Eight’ of
Autos Meet,” New York Times, January 10, 1992, p. Al.
3. Bob Davis, “Economy: Clinton Team to Suggest Import Goals for Japan as Trade
Talks Approach,” Wall Street Journal, May 20, 1993, pp. A2, All.
4. The 1993 GATT Agreement skirted the margins of this issue by explicitly permitting
government subsidies of up to 75 percent of basic research conducted by industry and up
to 50 percent of applied R&D costs through development of the first prototype. “World
4 INTRODUCTION

Second, how must the world trading system be adapted to deal with
the presence—or absence—of accepted norms for competitive behavior
among high-technology firms, and for relations with their home govern¬
ments? How much cooperation or collusion among nominally competitive
firms, possibly mediated or coordinated by government, can be tolerated
without making the idea of a competitive marketplace meaningless? In¬
vestments in high technology, if successful, create (at least temporarily)
some element of monopoly power based on a firm’s proprietary control
over unique technological assets. What will be the ground rules for com¬
petition in global markets that must, virtually by definition, stray far from
the textbook vision of perfect competition?
How these issues are settled will shape the world trading system for
decades to come. Given its central importance, one might presume that
thinking through the outlines of a coherent policy would be an overriding
concern of U.S. trade policymaking. But this has not always been the
case. The semiconductor industry has been simmering—or perhaps stew¬
ing—on the front burner of trade policy for over three decades. Instead
of writing down a recipe, however, rapidly changing shifts of American
trade policymakers have simply tossed additional ingredients into the pot,
often ignoring the outcomes of their predecessors’ experiments. Instead
of drawing operational guidance from a long-term strategy and vision,
American policy has largely been devised in terms of short-run tactics to
fix the irritant of the moment on an industry-by-industry, case-by-case
basis.
This study analyzes how thirty years of fierce international economic
competition, and high-stakes political intervention, came to lock the
American and Japanese chip industries onto their current course as re¬
luctant pioneers of a new high-technology world order. For it is no acci¬
dent that semiconductors have historically been at the cutting edge of
trade friction in high-technology industries. There are three reasons why
this has been so. First, semiconductors are an extreme example of a
product undergoing rapid and sustained technical innovation, showcasing
the distinctive features of high-technology industries that make them
both economically important, and a challenge to the international trading
system. Second, from the start, semiconductor companies fixed their gaze
on a global market, producing a sharp and sustained international com-

Trade Talks: The Uruguay Round’s Key Results,” Wall Street Journal, December 15, 1993,
p. A6.
INTRODUCTION 5

petition. More rapidly than in other high-technology sectors, a broad

array of companies from around the world—with various forms of sup¬
port from their national governments—have entered into a mortal eco¬
nomic combat on this common, contested terrain. Setting the vague rules
of the game for this battle inevitably involves nations, companies, and
the very institutional fabric of the world trading system. Finally, govern¬
ments and companies have often made decisions taking into explicit ac¬
count the predicted reactions of others outside their own national, in¬
dustrial, or firm boundaries: policies have been designed with strategic
objectives in mind.
In short, semiconductors have been at the center stage of trade policy
debate because they embody all the features of a high-technology indus¬
try most likely to test the limits of the existing trade regime. The rapid
international diffusion of semiconductor technology created national
companies from all corners of the industrialized world capable of com¬
peting in a single, global arena. Strategic concerns have linked the actions
of companies and governments to one another in a way that clearly rebuts
the usual presumption that the actions of individual, atomistic firms,
disciplined only by the impersonal forces of a competitive market, are
likely to determine industrial outcomes.
The bulk of this book analyzes how Japanese companies were able
to successfully challenge American companies in semiconductors, the
role that governments in the United States and Japan played in fostering
their respective industries and regulating trade conflict when it periodi¬
cally erupted, and the economic consequences of those government pol¬
icies. The study covers a broad range of topics, and uses a variety of
analytical tools. The chapters vary widely in their appeal to different
sorts of readers.
Chapters 2 through 4 are historical and descriptive, covering the de¬
velopment of the Japanese semiconductor industry through the mid-
1980s, the history of U.S.-Japan trade friction and trade policy over this
period, and the negotiation and operation of the Semiconductor Trade
Arrangement. In contrast with many earlier discussions of the develop¬
ment of the Japanese semiconductor industry, the account in this book
highlights the recurring tendencies toward rivalry and competition among
domestic chip producers. Japanese government policy basically focused
on sheltering or assisting national firms against competitive threats from
abroad. Successful cooperation among Japanese firms, and between gov¬
ernment and industry, generally built on the common perception of a
6 INTRODUCTION

clear external threat (for example, a technological challenge, or U.S.

trade actions). As the government’s direct leverage over companies di¬
minished with increasing economic liberalization in the 1970s and 1980s,
the appearance of such an external threat became even more important
in asserting government leadership over industry and organizing coop¬
erative action within industry. The dilemma this has created for U.S.
policy is one important subject discussed in this study. There is a recur¬
ring tension in U.S.-Japan trade policy between pursuit of a difficult,
long-term struggle against anticompetitive patterns of behavior within
Japanese government and industry that affect U.S. economic interests,
and a quest for short-term success in narrower industrial trade objectives
using policy tools that ultimately may distract from, rather than reinforce,
the long-term objectives.
The next chapters are more technical and analytical. Chapter 5 pro¬
vides a detailed empirical study of the economic impact of the STA.
Chapter 6 develops a model of semiconductor production that incorpo¬
rates key features of the industry—large, up-front R&D costs and sig¬
nificant learning curves in manufacturing—and through simulations,
studies the relationship between price and cost over the life cycle of a
memory chip under different assumptions about industrial structure. This
relation between price and cost is at the heart of discussions of “dump¬
ing” of chips as an international trade problem. Chapter 7 explores the
issue of strategic behavior—by governments or firms—and its implica¬
tions for the national interest and policy. A concluding chapter summa¬
rizes the lessons of this study, takes a look at where U.S. policies are
going, and proposes new policy initiatives in two major areas. The first
is some revision of the rules of the game in international trade and
investment that makes them more consistent with the special character¬
istics of high-technology products such as semiconductors. The second is
a reformulation of our economic strategy with Japan that better supports
our long-term goal of an open and competitive international marketplace,
and is more deliberate in creating policies—and incentives—that rein¬
force movement by Japanese government and industry toward that goal.
The balance of this chapter builds some foundations. I next introduce
the basics of technology and industrial structure in semiconductors. The
next section presents a thumbnail sketch of the history of international
competition in the industry. As a prelude to a study that is mainly about
interaction with Japan, the last part of the chapter summarizes the evo¬
lution of the U.S. industry before Japan came on the scene and the role
INTRODUCTION 7

of the U.S. government in the development of semiconductor technology

through the mid-1980s, when the current framework for competition was
negotiated.

Technology and Industrial Structure in Semiconductors

In semiconductors, the ubiquitous silicon chips at the electronic heart

of all manner of high-technology goods, an unprecedented number of
the thorny economic issues that spawn trade conflicts in high-technology
industries intersect. The chip industry’s rate of technological progress
places it at the dizzying forefront among high-tech industries, surpassing
even the well-documented case of computers. (Memory chip prices have
historically dropped at a rate of almost 35 percent a year, compared with
a quality-adjusted price decline of roughly 20 percent in mainframe com¬
puters.)5 With the high rate of innovation have come very short product
lives, as newer generations of chips rapidly render older vintages obso¬
lete. The relative size of investments in research and development marks
the semiconductor business as one of the most technology-intensive of
industries, again surpassing even computers, and is a source of significant
economies of scale.
Other issues, less exclusively identified with high technology, further
complicate matters. Semiconductor production is extraordinarily capital
intensive, creating another potential source of economies of scale. Learn¬
ing economies—the way that unit costs in some businesses fall with
cumulative production experience—although well established in some
other high-tech fields, like aircraft production (as well as in less technol¬
ogy-intensive areas, such as shipbuilding), and unimportant in others,
are particularly critical to successful competition in semiconductors.
Breakneck innovation, heavy investments in R&D and capital (and as¬
sociated economies of scale), and learning economies make the semicon¬
ductor industry a virtual laboratory for high-tech trade friction. In a
single arena, all the complications for the textbook model of perfectly
competitive markets that are commonly found in one or another high-
technology industry (but rarely are all simultaneously present in a single
industry) have been combined.

5. See figure 1-1 and Jack E. Triplett, “Price and Technological Change in a Capital
Good: A Survey of Research on Computers,” in Dale W. Jorgenson and Ralph Landau,
eds., Technology and Capital Formation (MIT Press, 1989), pp. 127-214.
8 INTRODUCTION

Modern semiconductor electronics was originally developed during

and after World War II. The first such solid-state components were diodes
and rectifiers, used where previously certain types of vacuum tubes had
been employed. In 1947 the transistor was invented at the Bell Telephone
Laboratories, and proved to be the seed of a technological revolution.
The transistor, with properties of both amplifier and switch, made it
possible to replace hot, bulky, power-hungry, and fragile vacuum tubes
with fast, cool, and small semiconductor components. From the start it
was realized that these solid-state devices could revolutionize the just-
emerging technology of the computer. The U.S. military invested heavily
in both technologies, and the crucial link between computers and semi¬
conductors continues to be the reason governments pay so much attention
to the semiconductor industry.
Transistors and diodes are discrete devices, however, with just one
circuit element per semiconductor device. The microelectronics revolu¬
tion was to begin in earnest in 1959, when the integrated circuit (IC)—a
semiconductor component with many devices, entire electronic circuits,
constructed on a single silicon chip—was invented.
The primary force behind rapid technological improvement in semi¬
conductors has been continuous advance in semiconductor manufacturing
processes. Improvements in fabrication technology have steadily reduced
the size of electronic circuit elements, making it possible to pack ever
greater numbers of miniaturized circuit elements on a single chip, and
stimulated the development of novel types of physical structures imple¬
menting standard electronic functions.
Memory chips are the largest single segment of the semiconductor
market, estimated as 31 percent of global sales in 1995.6 The dominant
product (75 percent of memory sales in 1995) was the dynamic random
access memory (DRAM), which by itself accounted for 23 percent of
world semiconductor sales in 1995. The DRAM was to play a leading role
in the unfolding U.S.-Japanese rivalry from the late 1970s to the present.
DRAMs have in many respects been the bellwether product of the
semiconductor industry, since they have historically been the first product
manufactured with each new vintage of processing technology.7 By nature

6. Based on estimates found in Dataquest, “Worldwide Semiconductor Market Grew

40 percent in 1995,” press release, San Jose, Calif., January 8, 1996; and Dataquest,
“Worldwide Memory Market Is on Pace to Double by 1999; DRAM Key Driver,” press
release, November 6, 1995.
7. Although this is somewhat less true today. Increasingly, performance in micropro-
INTRODUCTION 9

they are also well suited as a measure of advance in semiconductor

manufacturing technology. The abstract logical design of a DRAM is
simplicity itself, the endless repetition of a simple cell, and easily mas¬
tered; the details of its implementation and manufacture are everything.
The principal and overwhelmingly important characteristic of a DRAM
from the point of view of its consumers is its bit capacity, the amount of
information it can hold. The impact of technical improvement is typically
measured in cost per bit.
The first widely used commercial DRAM was the IK memory (1
kilobit, or K, means 1,024 bits of information) introduced in 1970 by
American semiconductor companies. New-generation DRAM chips
(each with four times the capacity of the previous generation) have been
introduced approximately every three years since the mid-1970s.

Innovation: The “Pi” Rule

Despite the more sophisticated processing required for higher density

ICs, technological progress in semiconductor manufacturing is widely
believed to have historically held the fabrication cost per area of silicon
processed approximately constant. (Or equivalently, because the size of
a mature chip product has been about constant, the manufacturing cost
per chip as mass production has peaked has held roughly constant for
successive generations of memory chips.) As a consequence, the cost per
electronic function has fallen by three-quarters with the arrival of each
new generation of DRAM with its fourfold greater capacity, implying a
compound annual rate of decline of 37 percent in the cost of a bit of
storage. This stylized, long-run relationship has even acquired a colorful
label—the pi rule—since every generation of DRAM, it was claimed,
approached a price of approximately pi (the mathematical constant
3.14159) dollars as production peaked.* * * * * 8
This, it turns out, was approximately true. Figure 1-1 shows two mea¬
sures of quality-adjusted cost per DRAM and confirms that this extraor-

cessors has been limited by clock speed and the distances between transistors on a chip.
High-performance computer processor chips are now often crammed as full of circuit
elements as the densest memory chips, and the manufacturing technology used by the
makers of the most advanced microprocessors is as demanding as that used in leading-edge
memories.
8. See M. P. Lepselter and S. M. Sze, “DRAM Pricing Trends—The -it Rule,” IEEE
Circuits and Devices Magazine (January 1985), p. 53.
10 INTRODUCTION

Figure 1-1. Year-to-Year Changes in DRAM Prices, 1972-89

Rate of change (percent)

Source: Author’s calculations. See chapter 5.

dinary decline in price was more than a fanciful phrase.9 One DRAM
price index in common use within the industry, the aggregate price per
bit of memory sold, fell at a compound annual rate of 31 percent from
1977 to 1988. Over the entire 1974-89 period, the compound rate of
quarterly decline in average cost per bit was 10.6 percent, equivalent to
an annual rate of 36.1 percent. Another, technically superior index of
DRAM price is a so-called Fisher Ideal price index. It more accurately
measures the effects of price declines and shows a drop closer to 34
percent a year from 1977 to 1988.10 A trend line fit to an annual Fisher
Ideal price index, whose changes are depicted in figure 1-1, over the
1971-89 period, shows a long-term decline of about 42 percent a year.

9. Consulting firm Dataquest’s quarterly estimates of average selling prices and quan¬
tities sold for all types of DRAMs were used to construct these measures.
10. Average price per bit is calculated by dividing aggregate sales revenues by total bits
sold for all generations of chip. Even if prices remained absolutely constant for each type
of chip from one period to the next, the average price per bit would change as a consequence
of shifts in demand from one type of chip to another. In essence, price per bit is a weighted
average of chip prices, where the weights vary across any two periods being compared.
The Fisher Ideal index, in contrast, uses identical weights in the two periods compared,
thus avoiding confusion of the effects of price changes with shifts in usage from one
generation to another. See chapter 5 for further details.
INTRODUCTION 11

Other types of memory chips show only slightly less rapid declines in
price. One study found a Fisher Ideal price index for all memory chips
declining at an annual rate of almost 33 percent from 1977 to 1988.”
Interestingly, the long-term decline in DRAM prices is even more
rapid if the period of extraordinary rise of chip prices after 1986, when
the STA was signed, is excluded. Figure 1-1 shows how unusual the
annual rates of change in DRAM prices after 1986 were: for the first time
in the history of the industry, a substantial price increase was registered
in 1988, and declines far smaller than historical norms were recorded in
1987 and 1989. How the STA might have been related to this slowing of
the historical rate of price decline in DRAMs is an important subject
explored in detail in this study.
A corollary to the rapid improvement in manufacturing technology
that made these dramatic price declines possible has been an equally
torrid pace for the introduction of new products. Even in the late 1950s,
for example, more than half of all new types of transistors introduced
were obsolete within two years; roughly the same life span applied to
chips used in computer systems in the mid-1970s.12 At IBM, memory
products historically had about a one-year lifetime.13 Nor does innovation
appear to have slowed—one analysis shows new generations of DRAMs
in the late 1980s being introduced at roughly the same rapid pace ob¬
served in the 1970s.14

Technological Intensity

Rapid technological change, of course, is the intended consequence

of investment in research and development, and relative levels of re¬
sources plowed into R&D are the usual criteria used to define a high-

11. See Ellen R. Dulberger, “Sources of Price Decline in Computer Processors: Se¬
lected Electronic Components,” in Murray F. Foss, Marilyn E. Manser, and Allan H.
Young, eds., Price Measurements and Their Uses (University of Chicago Press and National
Bureau of Economic Research, 1993), table 3.6.
12. See John E. Tilton, International Diffusion of Technology .The Case of Semicon¬
ductors (Brookings, 1971), p. 83; and Douglas W. Webbink, Economic Report on the
Semiconductor Industry: A Survey of Structure, Conduct and Performance (Federal Trade
Commission, 1977), p. 131.
13. William E. Harding, “Semiconductor Manufacturing in IBM, 1957 to the Present:
A Perspective,” IBM Journal of Research and Development, vol. 25 (September 1981),
p. 653.
14. M. Therese Flaherty and Kathryn S. H. Huang, “The Myth of the Shortening
Product Life Cycle,” Harvard Business School, May 1988.
12 INTRODUCTION

Figure 1-2. U.S. Semiconductor and Electronic Component Company

R&D as Share of Sales and Japanese Company IC R&D as Share of
Semiconductor Sales, 1976-94

Percent

Source: MITI IC R&D as percent of semiconductor sales was estimated by taking the IC R&D-to-sales ratio from
the MITI survey and adjusting to reflect aggregate value of semiconductor production versus IC production from
MITI industrial production statistics. Semiconductor Industry Association, SIA 1995 Annual Databook: Review of
Global and U.S. Semiconductor Competitive Trends, 1978-1994 (San Jose, Calif.: Technicon Analytic Research,
1995), p. 40; Thomas R. Howell and others. The Microelectronics Race: The Impact of Government Policy on
International Competition (Boulder, Colo.: Westview, 1988), pp. 218-19; National Science Foundation, Research and
Development in Industry: 1982, NSF 84-325 (Arlington, Va., 1984), 1988, NSF 90-319 (1990), 1991, NSF 94-325 (1994)
and unpublished MITI surveys of twelve Japanese IC companies.
a. Members of the Semiconductor Industry Association (merchant producers only).
b. Twelve Japanese semiconductor companies surveyed by the Ministry of International Trade and Industry (MITI).
c. Total funds for R&D as a share of net domestic sales by U.S. electronic component companies.
d. Company funds for R&D as a share of net domestic sales by U.S. electronic component companies.

technology industry. In the late 1980s American semiconductor compa¬

nies funded R&D expenditures amounting to 11 to 13 percent of their
global sales; available statistics suggest that Japanese companies invested
equivalent percentages of sales in R&D over the past decade and a half.
The size of the technological ante in this high-stakes global game has also
clearly been rising. Figure 1-2 shows that, for both American and Japa¬
nese companies, R&D as a portion of sales roughly doubled from 1976
to 1986. Official National Science Foundation data, which include semi¬
conductors in the broader industry category “electronic components” (in
INTRODUCTION 13

which semiconductors accounted for almost 50 percent of electronic com¬

ponent value added, and a somewhat smaller share of sales in 1989), show
total, or company-funded, domestic R&D as a portion of domestic elec¬
tronic component sales rose over time, just like semiconductors. But
although the technology intensity of semiconductors and the average for
all electronic components were roughly equal in 1976, by the early 1970s
semiconductors absorbed nearly double the R&D funds per dollar of
sales averaged in all components.
Bearing in mind that “electronic components” includes considerably
more than semiconductors, we can compare these figures with those for
other high-technology industries on a global basis (since high-technology
industries typically sell a substantial amount of output in foreign markets,
it is essential to include research performed and sales booked overseas
by American-based companies in presenting an accurate comparison).15
Table 1-1, based on data reported by U.S. multinational companies for
1989, shows some useful comparisons of electronic components with
other high-technology businesses.
One common definition of technological intensity is an R&D-to-sales
ratio exceeding the average for all manufacturing industry. By this defi¬
nition electronic components rank near the top in a list of technology¬
intensive industries, trailing only drugs and computers. Because semi¬
conductors are considerably more research-intensive than electronic com¬
ponents as a whole, it seems likely that if more disaggregated data on
R&D share were available, semiconductors would be at the very top of
this list. Electronic components—and presumably semiconductors—rank
considerably lower when R&D employment as a share of global employ¬
ment is the criterion, reflecting the fact that semiconductor manufacturing
is relatively labor intensive. Unlike other high-technology businesses that
produce systems—such as computers—U.S. chip companies have not for
the most part found it profitable to separate design and manufacture into
separate activities in their quest for competitive advantage.16

15. The National Science Foundation’s data on industrial R&D spending show domestic
R&D as a share of domestic sales, which is a misleading basis for cross-industry compari¬
sons when substantial research, or significant sales, take place within overseas affiliates and
the share of overseas affiliates varies substantially from industry to industry.
16. Although there are important examples of “fabless” chip companies, which do no
manufacturing and instead have their designs produced in outside fabrication facilities under
contract, these firms account for a relatively small share of industry output. The most
prominent examples are microprocessor and computer peripheral systems design specialists
such as Cyrix, Chips and Technologies, Weitek, and Cirrus Logic, and might arguably be
14 INTRODUCTION

Table 1-1. Share of Worldwide R&D Expenditures and Employment for

U.S.-Based Companies and Their Majority-Owned Foreign Affiliates,
by Parent’s Industry, 1989
Percent

R&D as R&D employment

proportion of sales as share of total
Industry to nonaffiliates employment

All industries 2.4 3.0

Petroleum 0.5 n.a.

Manufacturing 4.1 4.6
Chemicals and allied products 4.4 6.2
Industrial chemicals 3.6 6.2
Drugs 8.5 8.5
Other chemical products 2.9 5.6
Machinery, except electrical 6.3 6.9
Computers and office equipment 9.6 10.3
Electric and electronic equipment 6.1 5.3
Audio, video, and communications 7.7 5.9
Electronic components 7.8 5.8
Transportation equipment 6.2 6.3
Motor vehicles and equipment n.a. 2.1
Other n.a. 11.6
Other manufacturing 2.4 3.0
Instruments and related products 7.8 7.1
Wholesale trade 0.3 1.5
Services 0.8 0.7
Business services 1.9 1.2
Computer and data processing services 5.0 8.2
Source: Author’s calculations based on Bureau of Economic Analysis, U.S. Direct Investment Abroad, 1989
Benchmark Survey, Final Results (Department of Commerce, 1992), tables 2K1, 2P1, 2Q1, 2Q4, 2R1, 3E8, 3F23a,
3G10, 3G11, 315. Data are classified by industry of parent and reflect operations by all parents of nonbank foreign
affiliates and by majority-owned foreign affiliates only of these same parents,
n.a. Not available.

Scale Economies

The research-intensive nature of chip production has had significant

consequences for the semiconductor industry. Perhaps the most notice¬
able has been important economies of scale in the production of chips.
To a first approximation, whether one produces 10,000 chips or 10 mil¬
lion, the costs of developing the necessary technologies are roughly con-

designated as specialized computer peripheral producers (where the entire system just
happens to be incorporated into a single chip or set of chips), rather than producers of
standard semiconductor components. The blurring of lines between systems producers and
semiconductor component manufacturers is an important topic addressed later.
INTRODUCTION 15
Figure 1-3. U.S. Company Semiconductor Plant and Equipment
Investment and Japanese Company IC Plant and Equipment Investment
as Share of Semiconductor Sales, 1976-94a

Percent

Source: See figure 1-2.

a. Japanese IC investment-to-semiconductor sales share estimated using MITI production data as in figure 1-2.
b. Gross investment less depreciation and retirements.

stant. Thus the larger the sales base, the lower the average unit cost of
production.
These economies of scale in the use of the outcomes of R&D have
created a constant pressure for manufacturers to seek out the widest
possible market for their products to maximize the return on their rela¬
tively fixed R&D investments. As a consequence, the industry has been
unrelentingly international in focus right from the start. Major producers
quickly turned to foreign markets in their quest for a larger sales base.
U.S. companies surged into Europe in the late 1960s and, after overcom¬
ing numerous obstacles, into Japan in the 1970s and 1980s.
Another important source of scale economies in semiconductor pro¬
duction has been the capital-intensive nature of the chip manufacturing
process. Both American and Japanese companies have typically plowed
from 10 to 20 percent of their sales into new plant and equipment invest¬
ment, year after year. Figure 1-3 shows that in the early 1980s, one of the
16 INTRODUCTION

Table 1-2. Capital Costs of a Typical Semiconductor Wafer Fabrication

Facility, Mid-197Os-Early 1990s
Mid-1970s Mid-1980s Early 1990s

Total cost of wafer fabrication

facility (millions of dollars)3 30 100 300

Depreciation share of total wafer

fabrication cost (percent) 15 49 61
Source: Integrated Circuit Engineering, cited in Reese M. Reynolds and Don R. Strom, “CEM: Process Latitude
In a Bottle,” Semiconductor International (October 1989), p. 123.
a. Assumes 4,000 wafer starts per week.

keys to the Japanese industry’s rapid penetration of global markets was

a share of sales invested in new plant that was substantially higher than
that of U.S. firms.
Although investment as a share of company sales has been relatively
stable for both American and Japanese semiconductor companies in re¬
cent years, the cost of a single full-scale manufacturing facility has been
growing rapidly. Packing the maximum amount of circuitry onto a state-
of-the-art chip requires increasingly expensive manufacturing equipment
and facilities. The capital costs of a fabrication line for leading-edge chips
rose from about 15 percent of the total fabrication cost in the mid-1970s,
to about half of cost by the mid-1980s (table 1-2), and was projected
to pass 60 percent of total cost in the early 1990s.17 Today, the costs
of building a leading-edge mass production facility for chips exceeds
$1 billion.
Much of this equipment was highly specialized, with little or no scrap
value outside of the semiconductor business; and because of the rapid
pace of technological change, it had a short economic life. (Note the
increasing share of investment absorbed by retirements and depreciation
in table 1-2.) Investments in semiconductor manufacturing facilities,
therefore, were often difficult to liquidate for more than a fraction of
their acquisition cost. These investments thus took on the character of a
sunk cost. The increasing magnitude of such sunk costs has made entry
and exit from the industry more expensive and difficult. Moreover, since

17. See Reese M. Reynolds and Don R. Strom, “CEM: Process Latitude in a Bottle,”
Semiconductor International (October 1989), pp. 122-29. These figures refer to a state-of-
the-art facility for manufacturing a high-volume, mass-produced product. Facilities invest¬
ments relative to other costs would be substantially lower for a smaller facility used in
producing smaller volumes of more specialized products. Unit manufacturing cost would
also be a smaller share of product price for such more specialized, noncommodity products.
INTRODUCTION 17

it typically takes a year or more to get a new plant up and running, a

notoriously cyclical market adds another element of risk to such lengthy
investment projects.18

Learning Economies

Given that processing costs are the major component of manufacturing

cost, improving yields—the number of good chips that can be extracted
from some area of processed silicon wafers—is the key to being profita¬
ble, especially in commoditylike semiconductors such as DRAMs, where
proprietary design details count for little. Improved yields come from
better control over the manufacturing process, which in turn comes with
experience and learning by doing, as well as reductions in feature size,
which permit more parts to be crammed onto a given surface area.
Learning economies—decreases in unit cost that come from experi¬
ence—are regarded as critical: typically, industry sources have reported
that every doubling in cumulative output brings with it a 28 percent
decline in unit costs.19
By definition all high-tech industries share the characteristic of re¬
search intensity with semiconductors. Available data, however, portray
semiconductors as an extreme case in terms of both the importance of
investments in R&D and the pace and speed of innovation produced by
that investment in technology. Only a subset of high-tech industries dis¬
plays the high capital intensity seen in semiconductors, and still another,
perhaps smaller subset, exhibits learning economies as important as those
observed in chip making. In short, the chip industry wraps up in a single
package virtually every industrial attribute likely to spawn challenges to
rules for international trade and investment based on textbook premises
about competitive markets. The likelihood that the potential for conflict
would deliver actual trouble was increased by a history of rapid diffusion
of semiconductor technology across national boundaries, creating an in¬
tense international rivalry at a very early stage.

18. The world record for bringing a new chip plant on line appears to be held by NMB
Semiconductor, a defunct Japanese company, which claimed that it took only nine months
to go from initial groundbreaking on a new fabrication facility to initial production of 256K
DRAMs in 1985. See Larry Waller, “DRAM Users and Makers: Shotgun Marriages Kick
In,” Electronics (November 1988), pp. 29-30.
19. Robert N. Noyce, “Microelectronics,” Scientific American, September 1977, p. 67.
18 INTRODUCTION

International Competition in Semiconductors

Solid-state electronics was largely developed in the United States, but

overseas competitors soon came on the scene. Spurred by the generous
terms on which American firms—particularly AT&T’s Bell Telephone
Laboratories and RCA—made the technology available to others, Eu¬
ropean and Japanese firms had achieved a significant presence by the late
1950s. By the 1960s Japanese solid-state radios and televisions were dis¬
placing a significant share of the American products sold in the U.S.
market, and by the 1970s, Japanese IC-making skills honed on consumer
electronics were being directed toward penetration of industrial elec¬
tronic markets dominated by U.S. companies. The rapidity with which
an enormous American technological advantage was whittled down to
approximate parity was striking—and undoubtedly was an important
reason semiconductors came to prominence as the prime battlefield for
high-technology trade frictions.

The Global Spread of Semiconductor Manufacturing

Table 1-3 illustrates how, despite the international dispersion of tech¬

nological competence, R&D in the U.S. semiconductor industry, as in
other high-technology industries, has for the most part stuck close to
home. (Published statistics suggest that semiconductor producers account
for the vast bulk of sales by U.S. multinational electronic component
firms, and that measures of the latter therefore are a good proxy for
activities of the former.)20 Nine percent of R&D funding by U.S. multi-

20. Bureau of Economic Analysis, U.S. Direct Investment Abroad: 1989 Benchmark
Survey, Final Results (U.S. Department of Commerce, October 1992). Observe first that
firms primarily producing semiconductors almost certainly account for the overwhelming
share of output of multinational electronic component producers. Output of U.S. electronic
component factories was about $60 billion in 1989; domestic and export sales by R&D-
performing U.S. electronic component companies were about $53 billion. U.S. semicon¬
ductor sales accounted for 43 percent of electronic component output; sales of multinational
electronic component home offices amounted to 42 percent of sales of R&D-performing
electronic component companies. Since virtually all significant semiconductor producers
operate foreign affiliates, one may infer that statistics on the populations of U.S. multina¬
tional electronic component firms and statistics on U.S. semiconductor companies are
largely describing the same group.
Since parent-firm sales by U.S. electronic component makers operating as multination¬
als (in the Commerce Department benchmark survey) amounted to well under half of
domestic sales by all R&D-performing U.S. electronic component makers (in the NSF
INTRODUCTION 19

Table 1-3. Domestic R&D Expenditures and Employment by U.S.-

Based Companies and Their Majority-Owned Foreign Affiliates as
Shares of Worldwide Totals, by Parent’s Industry, 1989
U.S. R&D U.S. R&D
expenditure as share employment as
of worldwide share of worldwide
Industry R&D expenditure R&D employment
All industries 91.2 87.1
Petroleum 90.5 n.a.
Manufacturing 91.2 86.7
Chemicals and allied products 86.7 81.6
Industrial chemicals 89.7 83.7
Drugs 85.4 81.0
Other chemical products 89.4 83.3
Machinery, except electrical 86.8 84.2
Computers and office equipment 86.6 84.3
Electric and electronic equipment 95.8 92.2
Audio, video, and communications 97.8 96.1
Electronic components 90.9 88.2
Transportation equipment 94.1 88.2
Motors vehicles and equipment n.a. 52.7
Other n.a. 96.5
Other manufacturing 91.6 89.2
Instruments and related products 94.1 91.1
Wholesale trade 94.2 88.8
Services 86.8 88.6
Business services 93.7 95.0
Computer and data processing services 94.7 95.7
Source: See table 1-1.
n.a. Not available.

national companies and their foreign affiliates classified as primarily pro¬

ducing electronic components, and 12 percent of R&D employment, was
funneled into facilities located outside the United States in 1989. This is
similar to the pattern observed in other high-technology industries and
contrasts markedly with the distribution of sales. In 1989 majority-owned
foreign affiliates of U.S. electronic component companies accounted for
32 percent of worldwide sales to unaffiliated customers.

survey) in 1989, there were clearly a large number of American electronic component
companies selling almost exclusively to the domestic market, including the large U.S.
defense market. About half of electronic component output is active (vacuum tubes and
semiconductors) and passive (resistors and capacitors) components. The balance is coils
and transformers, connectors, printed circuit boards, miscellaneous parts, modules, and
circuit assemblies.
20 INTRODUCTION

Table 1-4. World Market Shares of Merchant and Captive

Semiconductor Manufacturers, by Region Where Headquartered,
Selected Years, 1981-93
Percent

Manufacturers 1981 1987 1990 1993“

North America
Merchant 46.9 33.1 30.3 32.4
Captive 17.0 12.1 9.4 7.0
IBM 9.4 7.4 6.5 4.5
Europe 7.4 9.6 10.9 8.7
Japan 26.6 42.1 44.2 42.6
Rest of world 2.1 3.1 5.1 9.4

Memorandum:
IBM share of North American
captive production 55.2 60.9 68.9 65.5
IBM purchase/world market n.a. 1.5 n.a. 0.9
IBM share of world production 9.4 7.4 6.5 4.5
Source: Integrated Circuit Engineering, Status 1982, Mid-Term 1988, Status 1991, Status 1994: A Report on the
Integrated Circuit Industry (Scottsdale, Ariz.), pp. 1-10.
n.a. Not available,
a. Estimated.

Local production of components is widespread in the U.S. electronics

industry for both political and economic reasons. Historically, for exam¬
ple, IBM had a well-known policy of sourcing most of the components
used in its European and Japanese computer production from facilities
in those regions.
However, lines around the “U.S. semiconductor industry” must
be drawn carefully. Semiconductors are produced not only by firms
labeled as semiconductor companies because that is their primary
activity, but also by firms primarily engaged in producing other
products, particularly electronic systems such as computers, using
semiconductors as inputs. Indeed, some important producers of
semiconductors sell virtually none of their output on the open market
and instead consume it all; these are called captive producers.
Producers who sell a significant amount of output externally are
referred to as merchant producers.
All the major captive producers are located in North America. Indeed,
one firm—IBM—historically accounted for about two-thirds of world¬
wide captive production. Table 1-4 shows one set of estimates of IBM’s
output relative to worldwide merchant chip production for selected years
INTRODUCTION 21

in the 1980s and 1990s.21 In 1991 an estimated captive IC production of

$4 billion ranked it as the largest American IC manufacturer and the
third largest worldwide after Japan’s Nippon Electric Corporation (NEC)
and Toshiba.22 In that same year IBM was estimated to account for about
6 percent of world semiconductor production, and purchased another 1
percent of world semiconductor output from outside suppliers. Within
two years, however, IBM had slipped to third largest U.S. IC producer
and sixth largest in the world.23 Because most U.S. semiconductor in¬
dustry data generally excluded IBM’s production (in this book, I will
note specifically when IBM’s captive production is included), much of
the empirical data fueling policy debates omitted this major American
player. In fact, as is apparent in table 1-4, the historical trend toward
reduced U.S. market share would have been even more pronounced if
IBM’s operations had been included.
The spread of competition for U.S. chip makers, in their global mar¬
ketplace, is typically illustrated by figure 1-4, which shows a precipitous
decline in the share of U.S.-based chip companies (excluding IBM) in
world semiconductor sales, or some other measure of output, in the
1980s. Since the U.S. decline is largely the mirror image of a Japanese
ascent, it is useful to focus on even more detailed comparisons of the
U.S. and Japanese industries. Figure 1-5 shows that American companies
accounted for about 40 percent of global sales by U.S.- and Japan-based
firms in 1989, compared with almost 70 percent in 1970 when the con¬
sulting firm Dataquest first began producing these estimates. Since 1990
the U.S industry has regained global market share for the first time in
fifteen years and is now at rough parity with its Japanese rivals in sales.
The same loaf may be sliced a bit differently if one compares activity
by American and Japanese factories, that is, discriminating according to

21. The decline in the share of American captive production in world output during
the 1980s reflects several factors: systems companies shutting down high-cost internal chip
fabrication lines, rapid growth in merchant sales by companies such as Intel, and shifts in
strategy leading captive producers to sell their chips on the open market as well. Until the
late 1980s, when it began to sell its semiconductor products in significant volume, AT&T
was the second largest captive producer in the United States; today it is classified as a
merchant producer. Since 1991 IBM has aggressively begun to seek sales of its semicon¬
ductor output on the open market. See IBM, Annual Report, 1991 (Armonk, N.Y., 1991).
22. See Integrated Circuit Engineering, Status 1992: A Report on the Integrated Circuit
Industry (Scottsdale, Ariz., 1992), pp. 2-14 to 2-17, 2-32, 3-6.
23. Integrated Circuit Engineering, Status 1994: A Report on the Integrated Circuit
Industry (Scottsdale, Ariz., 1994), pp. 1-14, 3-6.
22 INTRODUCTION

Figure 1-4. World Market Shares of Merchant Semiconductor

Companies, by Region, 1970-95

Percent

Sources: Dataquest unpublished historical estimates; Industry and Trade Strategies, The U.S. Electronics Industry
Complex, report to the U.S. Congress, Office of Technology Analysis (Berkeley, Calif., October 1988), table 2;
Dataquest, "Worldwide Semiconductor Market Grew 40 Percent in 1995,” press release (January 1996); and "World¬
wide Semiconductor Market Grew by 28 percent in 1994,” press release (January 1995).

regions where the output is produced rather than the corporate nation¬
ality of the producer.24 U.S. census data on factory shipments include
Japanese firms’ production in the United States, which, although still
modest, increased significantly in the late 1980s.
More interestingly, the data show that the share of semiconductor
value added within U.S. factories, despite a decrease through 1989, still
greatly exceeded the share of U.S. factory sales. Since 1989 the U.S.
share of value added has risen sharply and now exceeds that of Japanese
factories, but U.S. factory shipments remained stagnant at under 40
percent of the binational total. Materials consumption in the United
States, which probably more or less tracks wafer fabrication activity,
declined even more steeply than the U.S. share of shipments. What these
figures seem to show is that Japanese output is mainly concentrated in

24. I use official U.S. and Japanese manufacturing census data to construct these series.
Note that the U.S. concept of value added corresponds to the Japanese concept of gross
value added. (Japan also reports “plain” value added net of depreciation.)
INTRODUCTION 23
Figure 1-5. U.S. Share of Combined U.S. and Japanese Semiconductor
Industrial Employment, Value Added, and Sales, 1968-93

Percent

Sources: See figure 1-4; MITI, Census of Manufacturers, various years; and Bureau of the Census, Annual Survey
of Manufactures, 1993.

manufacturing-intensive “commodity” products with a relatively small

differential between price and materials costs, whereas U.S. output has
increasingly emphasized relatively design-intensive products, where the
spread between price and materials cost is much larger. In some cases,
so-called fabless U.S. semiconductor companies entered the market over
this period, shipping products designed internally but wholly manufac¬
tured outside, under contract, often in Japanese fabrication facilities.
Thus although the data show a sharp decline in U.S. semiconductor
manufacturing over the course of the 1980s, the U.S. semiconductor
industry, particularly in the area of design, remained very much alive,
even during the darkest days of the late 1980s.
American and Japanese firms today dominate global competition in
semiconductors, with other Asian companies (mainly from Korea, but
also in Taiwan) rapidly becoming significant players. For that reason, this
book is mainly concerned with U.S.-Japanese competition in the indus¬
try. But it is worth considering for a moment the experience of the Eu-
24 INTRODUCTION

ropean companies, which were technically ahead of Japanese firms in the

1950s, and at rough parity in the 1960s, but which today trail far behind
their Japanese competitors.25

The European Experience

The development of the European industry illustrates well how the

organization and structure of the semiconductor industries of Japan and
Europe have evolved quite differently from the way they did in the United
States. In the United States, established electronic component produc¬
ers—for the most part producers of vacuum tubes affiliated with electri¬
cal equipment manufacturers—became victims of technological change
and fell by the wayside as entrepreneurial start-up companies were
formed to push the development of new products.26 As a consequence,
by the mid-1960s, the American semiconductor industry was dominated
by merchant producers, young companies that had specialized in the
production of chips, then sold these chips at arm’s length to an entirely
different set of firms, the electronic equipment producers.
The development of this distinctive semiconductor industry structure
in the United States was linked to a number of factors. On the demand
side, much was owed, first, to the willingness of the military, the largest
consumer of leading-edge components, to buy very expensive products
from brand-new firms who offered the ultimate in performance in lieu of
an established track record, and second, to the rise of a highly compet¬
itive commercial computer industry, which was also willing to buy the
most advanced component technology from whomever offered it for
sale.27 Other factors at work included the high degree of labor mobility
within American industry, which made it easy for engineers to leave
established firms and start new ones if an existing company was slow to
commercialize new developments; the ready availability of venture capi-

25. See, for example, Tilton, International Diffusion of Technology, p. 27, who shows
Japanese firms trailing slightly behind most European firms in imitating major semicon¬
ductor innovations of the 1950s, but ahead of the Europeans (although still behind the
United States) in the 1960s.
26. The standard source for this history is Tilton, International Diffusion of Technology.
27. See Kenneth Flamm, Targeting the Computer: Government Support and Interna¬
tional Competition (Brookings, 1987), chap. 4, and Flamm, Creating the Computer: Gov¬
ernment, Industry, and High Technology (Brookings, 1988), pp. 13-19. In the 1950s and
early 1960s, military purchases also accounted for the bulk of sales of the most technolog¬
ically advanced computers.
INTRODUCTION 25

tal to fund such new spin-off companies; huge federal investments in

R&D in the underlying technology base from which companies drew to
develop their commercial products; and a first-class educational and sci¬
entific university infrastructure (likewise built with large doses of federal
support and disposed to cooperate with industry as a consequence of the
conditions tied to that funding).
The development of the semiconductor industry was quite different in
Europe.28 A more traditional, even hidebound, industrial structure pre¬
vailed: there was limited employee mobility among firms; scarce venture
capital to fuel start-ups; little (until much later) government disposition
to plow huge amounts via public procurement or R&D subsidy into
leading-edge electronics or computers; and an academic sector with few
links to, or interest in, industrial matters. Established electrical equip¬
ment manufacturers were the primary force driving investment in semi¬
conductor electronics as they sought to produce cheaper components for
use in their product lines. For the most part, semiconductors were de¬
veloped and produced within existing, vertically integrated electrical
equipment companies. Roughly the same (if vastly more successful) pat¬
tern of internalization of semiconductor production within larger systems
producers emerged in Japan and is explored in the next chapter.
Driven largely by demand for use in consumer and industrial products,
European production of discrete semiconductors (transistors and diodes)
grew nicely in the 1950s. However, when the development of the inte¬
grated circuit first used in significant quantities in computers, particularly
military computers, touched off an explosion in commercial demand for
computers, European electrical equipment producers largely missed the
shift in the market. As the hot new boxes produced by the American
computer industry took over the European market in the early 1960s (the
European computer industry lagged well behind), local electronics man¬
ufacturers began to realize that much of the American success was based
on advanced semiconductors. When European countries embarked on
crash programs to develop national computer industries in the mid- to
late 1960s, the programs often contained some support for ICs.
But these national development programs for both computers and
semiconductors were generally failures. Why they failed is an interesting

28. An extensive discussion of the comparative development of the European semicon¬

ductor industry is found in Franco Malerba, The Semiconductor Business: The Economics
of Rapid Growth and Decline (University of Wisconsin Press, 1985).
26 INTRODUCTION

and complex question that cannot be explored here.29 In brief, the basic
European strategy from the late 1960s on was to protect national markets
with high tariff walls, then select “national champion” firms who were
given favored treatment within the protected national market (generally
receiving both direct subsidies and preferences in government procure¬
ment).30 The major reasons for failure in the case of semiconductors were
twofold: first, their sheltering from competition in the open market often
meant that the European firms felt lessened pressure to stay technologi¬
cally abreast in a rapidly changing marketplace; second, misguided and
failed policies in the computer sector obstructed the development of a
dynamic upstream market for chips used in computers like the one that
was driving the IC industry in the United States.
Indeed, attempts to protect the European semiconductor and com¬
puter industries from imports created a vicious circle of sorts. High tariffs
and high costs for imported semiconductors meant higher prices—and
diminished sales—for European computer systems makers in both na¬
tional and global markets. Diminished computer sales meant a smaller
demand for locally produced semiconductors to be used in those com¬
puter systems. A weaker national semiconductor industry meant greater
political pressure for protection, and so on. This apparent contradiction
between protecting a chip industry and fostering a competitive, chip¬
using computer industry downstream is not unique to Europe, of course;
it became a major source of division within the U.S. electronics industry
in the late 1980s, after the Semiconductor Trade Arrangement was
signed.
Because European chip manufacturers are largely vertically integrated
divisions within electronics systems companies, this inherent policy con-

29. For a detailed analysis in computers, see Flamm, Targeting the Computer, chap. 5,
and Creating the Computer, chap. 5.
30. See Giovanni Dosi, Industrial Adjustment and Policy, vol. 2: Technical Change and
Survival: Europe’s Semiconductor Industry (Sussex, U.K.: University of Sussex, 1981), pp.
26-41; and Malerba, Semiconductor Business, 1985, pp. 129-31, 188-200, for concise dis¬
cussions of the pattern of government funding for European semiconductor companies in
the late 1960s and 1970s. Although the United States, Japan, and Canada virtually elimi¬
nated all tariffs on semiconductors in 1985 on a most-favored-nation basis, the European
Community continued to maintain a steep 14 percent duty on integrated circuits. In late
1995 the European Union agreed to slash semiconductor duties beginning in 1996. Tariffs
greater than 7 percent were to be cut to 7 percent and those less than 7 percent were to be
abolished. See “Europe to Cut Duties on Semiconductors,” Electronic Engineering Times
(December 18, 1995), p. 20.
INTRODUCTION 27

flict was internalized within the European electronics firms. The solution
chosen was to protect the domestic chip market, but to permit free
investment within Europe by foreign producers. In that way, access to
semiconductor technology developed abroad could be maintained and
some degree of protection granted to the domestic industry, yet enough
domestic competition preserved over the long run to ensure that prices
eventually approached costs. Only a mildly negative effect on the com¬
puter industry would be felt, to the extent that foreign producers of the
latest, most proprietary technology might be able to discriminate on price
in the European market, behind the shelter of tariff and other barriers
impeding chip imports. But for mature products, with multiple sources
of supply, competition within the European market would eventually
drive prices down.
The strategy contrasted markedly with that adopted in Japan, which
built formidable walls around its domestic chip market, blocking both
trade and investment, and strictly regulated the terms under which for¬
eign technology could be imported. Another vital difference in Japanese
policy seems to have been that Japan did not focus its promotional efforts
on a single national champion, but instead chose to actively foster com¬
petition at home within the ranks of its sheltered domestic producers
(while simultaneously promoting cooperation in R&D). As will be seen,
this policy of maintaining competition within the domestic market was
successfully defended by government bureaucrats against pressure from
Japanese politicians to organize a national semiconductor champion. In
both Japan and Europe—and increasingly in the United States—the
driving force behind national support programs in semiconductors was
the view that this sector was somehow strategic.

Strategic Policy in the United States

Finally, exploration of the role of U.S. policy in global competition in
semiconductors must start by explicitly recognizing the enormously im¬
portant role of government in developing the technology in the United
States, a role that remains significant today. Table 1-5 clearly shows the
important role that federal funding continues to play in electronic com¬
ponent research. In 1989 almost 20 percent of U.S. R&D performed by
multinational electronic component makers was funded by sources out¬
side the company itself (the vast bulk of this, funding may be safely
28 INTRODUCTION

Table 1-5. U.S. Government Shares of Total R&D Funding,

by Industry, 1989
Percent
U.S. federal and other
noncompany funds for R&D as
Industry share of global R&D performed“

All industries 27.3

Petroleum 0.9
Manufacturing 29.3
Chemicals and allied products 1.6
Industrial chemicals 3.4
Drugs 0.5
Other chemical products 0.7
Machinery, except electrical 8.4
Computers and office equipment 8.7
Electric and electronic equipment 42.7
Audio, video, and communications 56.8
Electronic components 18.6
Transportation equipment 51.1
Motor vehicles and equipment n.a.
Other n.a.
Other manufacturing 28.3
Instruments and related products 39.3
Wholesale trade 19.8
Services 5.0
Business services 0.7
Computer and data processing services 0.7
Source: See table 1-1.
n.a. Not available.
a. These funds are overwhelmingly federal funds.

assumed to have come from the American government).31 Although

smaller than that for some other high-tech industries (transportation
equipment, which is dominated by aerospace R&D; communications and
“other” electronic equipment; and instrumentation) with significant de¬
fense links, this figure is vastly greater than the federal share in other
commercial high-tech businesses such as computer hardware, drugs,
chemicals, and computer services.

31. The government share is considerably smaller (about 10 percent; see figure 1-6) if
all R&D-performing electronic component companies (that is, those with no foreign affil¬
iates as well as those with them) are included. Virtually all significant U.S. semiconductor
makers have foreign affiliates, however, and the multinational segment of the electronic
components industry provides the more relevant comparison.
INTRODUCTION 29

Figure 1-6. Federal Government Share of U.S. Electronic Components

R&D Expenditures, 1972-92a

Percent

Source: Author's calculations based on National Science Foundation, Selected Data on Research and Development
in Industry: 1992 (Arlington, Va., 1994), tables SD-3, SD-4, SD-6, SD-7, SD-8, SD-9; National Science Foundation,
Research and Development in Industry: 1988 (1990), tables B-4, B-7, B-11, B-18, B-19.
a. Mid-1970s data are not available.

Furthermore, the trend has clearly been downward over time, so that
semiconductors today benefit less from federal R&D funding than in the
past. Figure 1-6 shows the decline of the government share in funding of
the U.S. electronic component industry’s domestic R&D from levels
approaching 40 percent in the mid-1970s. The significance of these figures
is that the semiconductor industry—in the United States, as well as
overseas—has long been, and continues to be the focus of intense gov¬
ernment involvement, with the power to shape competitive outcomes.
A brief consideration of the history of that involvement suggests that
until fairly recently the motivation for U.S. government support in semi¬
conductors was strategic only in the least economically interesting sense
of the word, that is, driven primarily by national security concerns. Gov¬
ernment’s initial involvement with semiconductor technology had oc¬
curred during World War II as an effort to improve the reliability of
30 INTRODUCTION

silicon diodes used in radar. The military sponsored a huge research

program investigating the fundamental properties of germanium and sil¬
icon that involved thirty to forty U.S. research labs.32
In the immediate postwar period the principal organization undertak¬
ing research into semiconductor electronics was the Bell Telephone Lab¬
oratories, which had resumed a prewar research effort into solid-state,
semiconductor amplifiers. By the end of 1947, that effort had yielded the
invention of the first crude transistor, a feat that later won Bell physicists
William Shockley, Walter Brattain, and John Bardeen a Nobel Prize.
Although Bell was not to receive a postwar semiconductor R&D contract
from the military until after its invention of the transistor, defense users
quickly grasped the transistor’s potential value. It became clear that
transistors, with their greater reliability, smaller size and weight, lower
power consumption, and lower manufacturing cost, could potentially re¬
place many types of vacuum tube amplifiers. Devices constructed from
semiconductor materials could also perform most other electronic func¬
tions, including those of various types of diodes, resistors, and capacitors.
The U.S. Army’s Signal Corps Engineering Laboratory at Fort Mon¬
mouth, New Jersey, quickly became the focal point for Defense Depart¬
ment efforts to apply the new technology for military ends.33 Initial small-

32. See Ernest Braun and Stuart Macdonald, Revolution in Miniature: The History and
Impact of Semiconductor Electronics, 2d ed. (Cambridge University Press), 1982, pp. 226-
30. Some of the experiments undertaken at Purdue University as part of this effort were
so similar to the later experiments undertaken at Bell Labs that resulted in the invention
of the transistor that some have speculated that the transistor might have been invented
years earlier if the Purdue researchers had been searching for amplification effects. See
Richard R. Nelson, “The Link Between Science and Invention: the Case of the Transistor,”
in National Bureau of Economic Research, The Rate and Direction of Inventive Activity:
Economic and Social Factors (Princeton University Press, 1962), pp. 549-86; and Thomas
J. Misa, “Military Needs, Commercial Realities, and the Development of the Transistor,
1948-1958,” in Merrit Roe Smith, ed., Military Enterprise and Technological Change: Per¬
spectives on the American Experience (MIT Press, 1985), pp. 256-57.
33. After the war the Signal Corps also sponsored development of a photolithographic
process for creating resistors and capacitors on ceramic substrates at the National Bureau
of Standards and the Centralab division of the Globe Union Corporation for use in min¬
iaturized proximity fuses. The basic photolithographic techniques later used in the manu¬
facture of integrated circuits were derived from this technology. Jack S. Kilby, coinventor
of the integrated circuit, had worked on this technology before moving on to Texas Instru¬
ments. See Kilby, “Invention of the Integrated Circuit,” IEEE Transactions on Electron
Devices, vol. ED-23 (July 1976), pp. 648-54.
The National Bureau of Standards and the Signal Corps also worked together on the
research programs that led to the development of printed circuit boards and wave soldering
techniques, which were to revolutionize electronics manufacture. As late as 1961, half of
INTRODUCTION 31

scale Signal Corps funding of transistor research at Bell in 1950 had risen
to 20 percent of total funding by 1952, and 50 percent of transistor work
by 1953, and stayed at that level through 1955.34 About 25 percent of Bell
Labs’ semiconductor research budget over the period 1949-58 was funded
by defense contracts, and all of the early production of Western Electric,
the Bell System’s manufacturing affiliate, went to military shipments.35
After 1955 the Signal Corps began to fund semiconductor research
and fundamental development at companies other than the Bell Labs
(adding RCA and Pacific Semiconductor to the list in 1955), and in 1956
it doubled the size of the R&D effort to an average of $1 million a year.
Even greater funding went into engineering development (“fundamental
development” left off at the prototype stage, while “engineering devel¬
opment” created efficient and economic mass production technology).
Army transistor engineering development contracts over the 1952-64 pe¬
riod amounted to $50 million, averaging $4 million annually.36 The en¬
gineering development effort also jumped sharply in 1956—in that year
alone, $15 million in contracts was appropriated, with funds flowing to
virtually every semiconductor company in the United States.37
The occasion for the sharp increase in R&D funding in 1956 was a
breakthrough in transistor manufacturing technology. In late 1955, influ¬
enced by an ongoing antitrust suit brought by the U.S. Justice Depart¬
ment, Bell Labs had permitted the transfer of its “diffusion process”
technology to other companies. This technology, invented at Bell, with a

the value of U.S. shipments of printed circuit boards went to military users. See S. F.
Danko, “Printed Circuits and Microelectronics,” Proceedings of the Institute of Radio and
Electronic Engineers, vol. 50 (May 1962), pp. 937-38, 941; and Misa, “Military Needs,”
pp. 263-64.
34. Misa, “Military Needs,” p. 273.
35. Indeed, the first practical application of the transistor in laboratory equipment was
undertaken by Bell for the navy in 1949. The first use of transistors in a computerlike digital
circuit was in a gating matrix built at Bell Labs that same year as part of a “simulated
warfare” computer. The first transistorized computer built in the United States and the
early development of the power transistor were funded at Bell by the air force. C. A.
Warren, B. McMillan, and B. D. Holbrook, “Military Systems Engineering and Research,”
in M. D. Fagen, edA History of Engineering and Science in the Bell System: National
Service in War and Peace (1925-1975) (Murray Hill: Bell Telephone Laboratories, 1978),
pp. 617-48; and W. S. Brown, B. D. Holbrook, and M. D. Mcllroy, “Computer Science,”
in S. Millman, ed., A History of Engineering and Science in the Bell System: Communica¬
tions Sciences (1925-1980) (Indianapolis: AT&T Bell Laboratories, 1984), pp. 351-98. See
also Richard C. Levin, “The Semiconductor Industry,” in Richard R. Nelson, ed., Govern¬
ment and Technical Progress: A Cross-Industry Analysis (Pergamon, 1982), p. 26.
36. Misa, “Military Needs,” pp. 275-76.
37. Misa, “Military Needs,” p. 282.
32 INTRODUCTION

large share of the cost borne by the Signal Corps, made possible a great
improvement in performance (in particular, the ability to deal with high-
frequency signals) over the first crude point-contact transistors, and later
generations of junction transistors.38 With the rapid adoption of produc¬
tion technology for diffused base transistors by the U.S. industry, the
Signal Corps had reason to spread its research largesse much more
widely, as it pursued the development of high-performance transistors.
In its quest for new high-performance products, the military was partic¬
ularly inclined to take a chance on new firms: in 1959, for example, new
firms (those with no background in the older vacuum tube business)
accounted for 69 percent of military sales, compared with a 63 percent
share of all semiconductor sales.39
The military influence was pervasive. In 1958 and 1959, for example,
the federal government was directly funding about 25 and 23 percent,
respectively, of the R&D (excluding engineering development contracts)
undertaken within the U.S. semiconductor industry.40 If university and
federal laboratory work were factored in, along with engineering devel¬
opment funds, and indirect R&D funding embedded in contracts to
procure new devices at premium prices, that figure would have been
much higher: a congressional committee report estimated that the federal

38. See Misa, “Military Needs,” p. 281; and Tilton, International Diffusion of Tech¬
nology, p. 76. The early point-contact transistors were manufactured by attaching two
metal wires to the surface of a piece of germanium semiconductor. These devices were very
sensitive to environmental factors, had high noise levels, worked only at low frequencies,
and could be used only in low-power applications. Point-contact transistors were superseded
by so-called junction transistors in the early 1950s, which worked by creating junctions
between regions of semiconductor materials treated with impurities. The two principal
methods of creating junction transistors involved adding the impurity while growing a single
silicon crystal (grown junction transistors) or melting silicon materials already treated with
impurities into one another (alloy junction transistors). Although they improved the con¬
trollability of semiconductor characteristics, junction transistors still operated only at low
frequencies. The diffusion process worked by exposing a semiconductor substrate to a
carefully controlled atmosphere of heated, gaseous impurities, which diffused into the
surface of the substrate. Very precise control of the composition and thickness of the
impurities in the silicon could thus be achieved, and the operating frequencies of the new
transistors so produced jumped by two orders of magnitude.
39. Tilton, International Diffusion of Technology, p. 91.
40. See Tilton, International Diffusion of Technology, pp. 93-94. Tilton stresses that
formal R&D contracts were concentrated in older, established electronics firms, but it is
clear that substantial government funding flowed into research at newer companies. New
entrant Texas Instruments, for example, had about half of its $30 million in R&D spending
for 1959—-which compares with $70 million shown for the entire U.S. semiconductor in¬
dustry in 1959 by Tilton’s sources—contributed by the government. “Business Week Re¬
ports on: Semiconductors,” Business Week, March 26, 1960, pp. 92-108.
INTRODUCTION 33

government paid for 85 percent of U.S. electronics R&D in 1959.41 Two

of the largest recipients of U.S. government support in the early days of
the industry—Texas Instruments and Motorola—were later to become
the biggest producers in the U.S. semiconductor industry.42
The U.S. military played a particularly overwhelming role in driving
the development and production of high-performance transistors in the
late 1950s. At Western Electric, for example, all the diffusion-process
transistors manufactured through 1958 (roughly 13 percent of the com¬
pany’s cumulative output of all transistor types) went to military appli¬
cations, as did 54 percent of alloy junction, 30 percent of grown junction,
and 53 percent of point-contact transistors produced through that year.43
In addition to R&D funding and procurement, the military provided
funds to individual firms to build semiconductor capacity far in excess of
what was then required. By 1953, pilot transistor production lines at
Western Electric, General Electric, Raytheon, RCA, and Sylvania were
funded.44 Between 1952 and 1959, roughly $36 million was provided to
individual firms to build transistor factories that operated at only a frac¬
tion of their rated capacity.45 To use that excess capacity, firms were eager
to find new markets for their products.
Foremost among the products in which the new, high-performance
transistors were being applied by both military and commercial users
were electronic computers. Computers were among the earliest defense
systems into which transistors were inserted, and transistor developers
such as Bell Labs and later Philco received military contracts to build
computers from the high-performance components they had developed.46

41. Coordination of Information on Current Federal Research and Developmen t Projects

in the Field of Electronics, prepared for the Subcommittee on Reorganization and Inter¬
national Organizations of the Senate Committee on Government Operations, 87 Cong.
1 sess. (GPO, 1961), p. 130.
42. John G. Linvill and C. Lester Hogan, “Intellectual and Economic Fuel for the
Electronics Revolution,” Science, March 18, 1977, pp. 1107-14.
43. Misa, “Military Needs,” table 3.
44. Tilton, International Diffusion of Technology, p. 92; and Linvill and Hogan, “In¬
tellectual and Economic Fuel,” pp. 1107-08.
45. Braun and Macdonald, Revolution in Miniature, p. 81, report that the industry in
1955 had the capacity to produce 15 million transistors a year, whereas actual output was
3.6 million. Of the $36 million in funding for production capacity, $11 million was provided
for alloy junction transistors in 1952, another $15 million for diffused base transistors in
1957, and $10 million for integrated circuits in 1959. Linvill and Hogan, “Intellectual and
Economic Fuel,” pp. 1108-09.
46. On the intimate link between semiconductor and computer development see
34 INTRODUCTION

Commercial computers soon followed military computers as the primary

market for American semiconductors. Table 1-6 shows how computer
manufacturers have continued to be the single largest category of com¬
mercial users of semiconductors in the United States from 1960 to the
present.47
The U.S. military was thus a critically important early market for
semiconductors, dominating sales through the 1950s and well into the
1960s before finally stabilizing around 10 percent of the overall U.S.
market in the late 1970s and 1980s. Yet the next major development in
the semiconductor industry, the integrated circuit, was even more depen¬
dent on military customers for its rapid growth in sales than the transistor
had been.
Before the announcement of the IC in 1959, the military had funded
efforts aimed at developing miniaturized electronics with functional char¬
acteristics similar to those of ICs, but the companies that developed the
IC avoided military funding in order to ensure that the technologies
developed remained privately owned. Nonetheless, the initial stimulus to
creating the IC was the announced intention of the armed services to
provide a significant market for components with the appropriate char¬
acteristics.48
Even more than it had in transistors, the military dominated the market
for ICs. Whereas the federal government’s share of the market for discrete
semiconductors peaked briefly at almost 50 percent in 1960, its share of the
IC market remained above half of total sales until 1967. For close to a
decade, the market for ICs was dominated by military systems.49
The first major production application of the IC was in a general
purpose, stored-program computer running the Minuteman II guided

Flamm, Creating the Computer, pp. 15-19, 92-94,119-20, 122-23, and Targeting the Com¬
puter, pp. 49-51. See also Misa, “Military Needs,” p. 282, note 59.
47. Note that much of the output of “communications equipment” historically went to
military customers. Also, the rapid increase of semiconductor consumption within the
business services sector in the 1980s probably reflected the explosive growth in demand for
standardized microprocessor memory chips integrated into computer systems by value-add
resellers and computer retailers and probably should be considered a subset of demand for
chips used in computers.
48. See Norman J. Asher and Leland D. Strom, “The Role of the Department of
Defense in the Development of Integrated Circuits,” IDA paper P-1271, Arlington, Va.:
Institute for Defense Analyses, pp. 1-7.
49. In 1962, for example, all ICs produced were used in defense systems. That share
fell to 85 percent in 1964, 53 percent by 1966, and 37 percent by 1968. See Tilton, Inter¬
national Diffusion of Technology, table 4-8.
INTRODUCTION 35

Table 1-6. U.S. Industrial Consumption of Semiconductors,

by Consuming Sector, Selected Years, 1963-87
Percent of total

Sector 1963 1967 1972 1977 1982 1987

Military ordnance 3.0 4.5 1.2 1.6 2.1 1.1
Yarn mills 0.0 0.0 0.4 0.2 0.2 0.0
Misc. plastic products 0.2 0.0 0.0 0.1 0.2 0.2
Power hand tools 0.0 0.0 0.0 0.1 0.1 0.1
Computers and office machinery 23.3 14.4 17.7 17.6 15.8 12.4

Electrical industry equipment 6.0 3.5 2.4 1.9 2.2 0.1

Household appliances 0.1 0.0 0.0 0.0 0.1 0.3
Electric lighting 2.8 0.0 0.2 2.4 0.5 0.5
Radio, TV, audio 15.2 15.9 14.1 12.5 6.4 4.5
Communications equipment 25.7 26.9 20.0 17.5 11.3 3.4

Other electric components 7.2 17.1 7.3 6.8 3.0 5.3

Misc. electric machinery 0.3 0.2 0.9 2.1 2.6 1.7
Motor vehicles 3.3 1.8 2.3 3.4 2.7 6.1
Aircraft 4.1 0.2 8.5 3.6 7.2 1.4
Other transport equipment 0.0 0.2 0.1 0.0 0.1 0.0

Scientific and measuring instruments 2.1 1.8 1.5 2.4 2.1 10.3
Optical and photo 0.3 0.5 4.9 8.3 13.7 8.8
Misc. manufacturing 0.0 0.4 0.5 0.2 0.3 1.0
Transportation and warehousing 0.9 0.8 0.7 0.3 0.3 0.4
Communications services 0.0 1.4 1.9 4.0 7.5 9.1

Personal services 4.3 8.0 7.5 7.4 9.1 10.1

Business services 0.0 0.9 8.0 7.6 12.8 23.1
Health, education, nonprofit 0.0 0.4 0.0 0.0 0.0 0.0
State and local enterprises 0.0 0.0 0.0 0.0 0.0 0.0
Other 1.1 1.2 0.0 0.0 0.0 0.1
Sources: Author's calculations using data from Bureau of Economic Analysis, Input-Output Structure of the U.S.
Economy: 1963, vol. 1 (Department of Commerce, 1969) and 1967 (1974); The Detailed Input-Output Structure of
the U.S. Economy, 1972 (1979) and 1977 (1984); The 1982 Benchmark Input-Output Accounts of the United States
(1991); and Benchmark Input-Output Accounts of the United States, 1987 (1994).

missile.50 Awarded in 1960, the contract for this D37 computer—the first
built using ICs—alone accounted for about one-fifth of industry sales in
1965.51 ICs were not shipped in commercial computers until 1965, four
years after the first D37 rolled off the production line.
From 1959 through the late 1960s, military and space contracts pow¬
ered a surge in American IC production. The guidance computer for the

50. Texas Instruments engineers put 94 percent of the computer’s electronics on a set
of thirteen custom-designed integrated circuits. See Richard C. Platzek and Jack S. Kilby,
“Custom-Integrated Circuits—A Military Computer Application,” Proceedings of the IFIP
Congress 65, vol. 2 (Washington: Spartan Books, 1965), pp. 425-26.
51. Asher and Strom, “Role of the Department of Defense,” p. 21.
36 INTRODUCTION

Table 1-7. Sources of R&D Funds in the U.S. Semiconductor Industry,

Selected Periods, 1958-76
Millions of dollars unless otherwise specified

Federal funding
Federal Company as percent
Period funding funding Total of total

1958-69 495 569 1,064 46.5

1970-76 207 1,144 1,351 15.3
1958-76 702 1,713 2,415 29.1
1958-74 930 1,200 2,130 43.7
Sources: For 1958-69, 1970-76, and 1958-76 see Industry and Trade Administration, A Report on the U.S.
Semiconductor Industry (Department of Commerce, 1979), p. 8. For 1958-74 see John G. Linvill and C. Lester
Hogan, "Intellectual and Economic Fuel for the Electronics Revolution,” Science. March 18, 1977, pp. 1107-08.

National Aeronautics and Space Administration’s Apollo spacecraft gave

a potent kick to Fairchild’s IC profits, much as the D37 had helped boost
Texas Instrument’s rising star. Other important contracts went to Moto¬
rola, Westinghouse, Signetics, and RCA.52 Premium prices paid on these
military contracts fueled the rapid application of the technology to com¬
mercial products.
Industry participants have estimated that between 1958 and early 1970s
the federal government directly or indirectly funded between 40 and 45
percent of all industrial semiconductor R&D (table 1-7). This share was
to fall sharply in the 1970s as the technology matured, as technically
advanced components moved into a booming commercial mainstream,
and as the government’s share of the overall market declined. Nonethe¬
less, in key areas of semiconductor technology—for example, devices
using exotic substrate materials, computer-aided chip design and manu¬
facturing process modeling, rapid turnaround production technology, and
innovative reduced instruction set computer (RISC) chips—military sup¬
port continued to prime the pump of semiconductor electronics in the
1970s and 1980s.53
By the late 1980s, government funding of semiconductor-related re¬
search held at a stable level of $450 million to a little over $500 million.54
The government share of the overall chip market, after dipping in the

52. See Asher and Strom, “Role of the Department of Defense.”

53. For references to these programs, see Flamm, Targeting the Computer, pp. 70-72;
and Flamm, Creating the Computer, pp. 18-19.
54. See Federal Interagency Staff Working Group, The Semiconductor Industry (Wash¬
ington: National Science “Foundation, November 1987), p. 31; and Congressional Budget
Office, The Benefits and Risks of Federal Funding for SEMATECH (September 1987),
p. 60.
INTRODUCTION 37

Table 1-8. Federal Contracts as Share of Total U.S. Semiconductor

Shipments, Selected Years, 1963-87
Year Percent Year Percent

1963 35.5 1973 5.8

1967 27.6 1983 4.1
1971 12.7 1987 10.7
Source: Bureau of the Census, Shipments to Federal Government Agencies, 1963, MA-175 (Department of Com¬
merce, 1965), 1971 (1973), 1973 (1975), 1983 (1985), 1987 (1989); 1987 is the most recent available report.

early 1980s, climbed back into the 10 percent range in the late 1980s
(table 1-8). However, this number—based on direct sales of chips to
government and to primary and subcontractors building government sys¬
tems—significantly understates government’s total (direct and indirect
purchases) share of the semiconductor market. A calculation using an
input-output table to estimate the semiconductor content of commercial
equipment procured by the government suggests that in the mid-1980s,
purchases by defense agencies alone—direct and indirect—accounted for
over one-quarter of U.S. semiconductor shipments.55
A significant change, however, had occurred in the relationship be¬
tween the military and the semiconductor industry. Whereas in the 1960s
defense systems had ridden at the leading edge of semiconductor tech¬
nology, by the late 1970s they trailed far behind the standard set in a
booming commercial market.56 Perceiving a threat, the military proposed
a program designed to push the technological frontier in semiconductors
and close the gap between military and commercial electronics. The huge
and costly effort, known as VHSIC, the Very High Speed Integrated
Circuit program, accomplished neither objective, but in the process it set
the stage for an important shift in the articulation of objectives for gov¬
ernment investments in semiconductor technology.

55. Computer runs from the Defense Economic Impact Modeling System (DEIMS)
input-output model estimated that defense spending accounted for $274 million in direct
and $3,791 million in indirect consumption of semiconductors in 1986, about 26 percent of
U.S. shipments in that year. See Directorate for Information, Operations, and Reports,
Department of Defense, Projected Defense Purchases, Detail by Industry and State, Cal¬
endar Year 1986 through 1991 (1987), p. 51, and Calendar Year 1991-1997, p. 58. The DEIMS
model shows $66 million in direct and $3,447 million in indirect sales in 1985 (about 21
percent of U.S. shipments) and $206 million and $3,906 million in 1991 (just under 20
percent of U.S. shipments).
56. See John A. Alic and others, Beyond Spinoff: Military and Commercial Technolo¬
gies in a Changing World (Harvard Business School Press, 1992), p. 269, for a perceptive
analysis of this shift.
38 INTRODUCTION

For the first time, the potential impact of an R&D program in im¬
proving the competitiveness of the American industry, although not a
primary goal of Defense Department planners, played an important role
in building political support for the program. The strategic objectives of
U.S. policies affecting semiconductor technology had begun to swing
away from the purely military considerations of the cold war and toward
economic conceptions of a strategic industry or policy. This swing was to
pick up momentum rapidly throughout the 1980s, breaking through an
important political and ideological barrier in 1987 with the formation of
the Sematech (semiconductor manufacturing technology) consortium, a
large-scale R&D effort with explicitly commercial objectives.
Conceived in a moment of crisis for the semiconductor industry, during
the initial months of the Semiconductor Trade Arrangement, Sematech
marked an extraordinary shift in policy for an American administration
for which government intervention was ideological anathema. Although
framed in terms of national security, the Sematech initiative was clearly
intended primarily to improve the health of a broad base of U.S. com¬
mercial semiconductor manufacturers (many of whom also supplied the
Department of Defense). Together with the market share targets accom¬
panying the STA, Sematech constituted a truly radical change in Amer¬
ican policy toward its high-technology industries.
What had brought all this about was the extraordinary development
of the Japanese semiconductor industry in the previous decade. For the
first time, an American postwar high-technology success story—the sem¬
iconductor industry—stood at the brink of decisive competitive defeat at
the hands of a foreign competitor. How this came to be (for the most
part, within the space of a single decade), and how policy responses
articulated in terms of strategic arguments were to become a staple of
American economic debate, is—it shall be argued—one of the most
important episodes in the economic history of the postwar international
system.
 CHAPTER TWO

New Competition:
The Japanese Ascent
in Semiconductors

Japanese companies have worked for over forty years to master lead¬
ing edge semiconductor technology. Through the mid-1970s these enter¬
prises lagged several steps behind the pioneering American firms that
had founded the industry. But in less than a decade, by the mid-1980s,
the mantle of overall leadership in semiconductor manufacturing (al¬
though not in design) had clearly passed to Japanese producers, and this
development was accompanied by profound shifts in the marketplace.
The pattern of Japanese investments in semiconductor technology in
many respects resembled that observed in computers.1 As they did in
computers, government laboratories took the first steps in semiconductor
R&D in the early 1950s (although unlike in computers it was small Jap¬
anese firms who rushed the new components to market, embedded in
innovative consumer electronics products). During the 1960s high trade
barriers were put in place, shielding the domestic market. As the private
sector took on more of the burden of technology development, the gov¬
ernment labs shifted toward more long-term projects. As in computers,
a large amount of government aid in the mid-1970s (following the infusion

1. For more details see Kenneth Flamm, Creating the Computer: Government, Industry,
and High Technology (Brookings, 1988), chap. 6; and Targeting the Computer: Government
Support and International Competition (Brookings, 1987), chap. 5.

39
40 THE JAPANESE ASCENT IN SEMICONDUCTORS

of significant resources into the computer industry) cushioned the shock

of imminent liberalization of the domestic market for Japanese produc¬
ers. By the early 1980s Japanese companies had arrived at the technolog¬
ical frontier in key products and were making their presence felt in global
markets.
Changes in Japanese policy in semiconductors often seem to have
trailed closely behind those in the computer industry. Careful scrutiny of
the historical record suggests a compelling explanation: the computer
industry has historically been the primary focus of government policy
toward the electronics industry in Japan, and assistance to domestic semi¬
conductor producers was primarily designed to further long-range objec¬
tives in computers.

First Steps

The earliest development of semiconductor technology in Japan took

place within two Japanese research institutes, one affiliated with Nippon
Telegraph and Telephone Corporation (NTT, the quasi-public telecom¬
munications monopoly that had exclusive control over Japan’s national
telephone network through the early 1980s), and the other with the Min¬
istry of International Trade and Industry (MITI). The Electrical Com¬
munications Department had been spun off from the Ministry of Posts’
Electrotechnical Laboratory in 1948, attached to what was to become
NTT, and renamed the Electrical Communications Laboratory (ECL);
the Electrotechnical Laboratory (ETL) proper was reorganized within
MITI in its current form in 1952.2 These two public research institutes
(in ECL’s case, quasi-public) initially led, and continue to exert a major
role in, the development of semiconductor technology in Japan.3 The first
transistors built in Japan were fabricated within ECL and ETL.4 The exit
of key researchers from these laboratories and their move to the private

2. See Electrotechnical Laboratory, Guide to ETL, 1983-1984 (Ibaraki, Japan, 1983).

3. Until 1985 NTT was a public corporation. In that year it was reorganized as a private
corporation in which the Japanese government owned one-third of the shares; the remaining
two-thirds of the company’s stock was sold off to the Japanese public over a ten-year period.
4. There is some dispute over who was first. ETL conducted a point-contact transistor
experiment in 1951, but ECL apparently produced the first properly functioning device.
See Makoto Watanabe, “Electrical Communications Laboratories: Recent LSI Activities,”
Japan Telecommunications Review (January 1979), pp. 3-8; and Yasuzo Nakagawa,
Semiconductor Development in Japan (Tokyo: Diamond Publishing, 1985), pp. 22-31 (in
Japanese).
THE JAPANESE ASCENT IN SEMICONDUCTORS 41

sector, where they helped organize company development programs in

semiconductors, played an important role in the early development of the
Japanese industry.5
Over the years 1949-51, both government research institutes began
research into transistors, and they soon constructed working devices du¬
plicating the Bell Telephone Laboratories’ pioneering 1947 discovery. By
1953 three large Japanese electrical conglomerates—Hitachi, Toshiba,
and Nippon Electric Corporation (NEC)—had begun research on tran¬
sistors.6 Ironically, these companies—the largest established players in
the Japanese electrical machinery sector—did not rush to commercialize
the new technology. Despite the very generous terms on which the Bell
Laboratories made know-how about the new technology available to all
comers, no Japanese companies were among the first licensees of the
transistor.7

5. For example, two prominent semiconductor technologists left NTT to join Sanyo and
Nippon Electric Corporation in 1957; a senior semiconductor specialist left MITI’s ETL to
join Sony Corporation in 1961. Nearly thirty semiconductor researchers left the NTT labs
in 1962. See Nakagawa, Semiconductor Development in Japan, pp. 282-84.
A historically significant research program in semiconductors was also launched at
Tohoku University about this time; Tohoku to this day remains a major academic resource
for the Japanese semiconductor industry.
6. Nakagawa, Semiconductor Development in Japan, p. 286.
7. Under the auspices of a 1949 Joint Services contract to develop transistors for defense
applications, a five-day symposium was convened in September 1951 by Bell Labs and
Western Electric, the manufacturing arm of American Telephone and Telegraph Company
(AT&T), to disseminate information to the military and their contractors, including indus¬
trial and university representatives. Representatives of twenty-two American universities
and eighty-six other American companies attended. Two individuals affiliated with foreign
firms, serving in their governments’ technical liaison offices to U.S. government agencies,
attended this meeting: one came from International Standard Electric, in England, and the
other from the French subsidiary of ITT Corporation.
A second symposium, explicitly designed to teach “the art and science of transistor
technology, theory, and practice as it was known” to commercial licensees was held in April
1952. Representatives of a grand total of thirty-seven firms are known to have either
attended the 1952 Bell Labs technology transfer symposium or signed licensing agreements
with the Bell Labs by the end of 1952. Eleven of these firms were foreign (twelve if one
counts ITT as a foreign-based firm—ITT was U.S.-owned and headquartered but sold its
products in foreign markets). Five were British (Automatic Telephone and Electric, British
Thomson-Houston, General Electric, Pye, and English Electric), four were German (Sie¬
mens and Halske, Telefunken Gesellschaft, Felton and Guilleaume Carlfswerk, and Brush,
an American affiliate of Intermetall), one was Dutch (N.V. Philips), and one Swedish (L.
M. Ericsson). In addition, representatives of three European subsidiaries of ITT attended
the eight-day April 1952 symposium: Laboratoire Centrale de Telecommunications
(France), Standard Telephones and Cables (Britain), and Siiddeutsche Apparatfabrik (Ger¬
many). See F. M. Smits, ed., A History of Engineering and Science in the Bell System, vol.
42 THE JAPANESE ASCENT IN SEMICONDUCTORS

Instead, it was small Japanese companies, not the giants, who took
the lead with the new technology. The first company to produce a working
transistor was Kobe Kogyo, a small machinery producer,8 whose tech¬
nology chief had come to the United States in 1951 and been introduced
to the new device during visits to RCA and the Bell Labs. A bootleg
research project was started on his return. By the spring of 1952 an
experimental transistor had been built and the needed technology ac¬
quired and licensed from RCA. Hitachi and Toshiba signed licensing
agreements with RCA in that same year. Because of cross-licensing ar¬
rangements between Bell and RCA then in effect, a separate license
from Bell was not required in order to use the basic Bell patents.9 Indeed,
use of patented transistor concepts was initially cheaper via the RCA
channel: the Bell Labs required a $25,000 advance toward future royal¬
ties in exchange for patent rights, while RCA did not.10
RCA’s licensing of its electronic componentry in 1952 was followed by
the transfer of its monochrome television system to its Japanese partners
in 1953.11 RCA was willing to not only sell patent rights, but also actively
transfer know-how. To service its Japanese licensees RCA set up a Tokyo
engineering facility in 1954.12 In 1960 RCA expanded its presence to
include a Japanese research laboratory.13 In 1962 RCA sold rights to its

6: Electronics Technology (1925-75) (Murray Hill: AT&T Bell Laboratories, 1985) pp. 28-
29); Thomas J. Misa, “Military Needs, Commercial Realities, and the Development of the
Transistor, 1948-58,” in Merritt Roe Smith, ed.. Military Enterprise and Technological
Change: Perspectives on the American Experience (MIT Press, 1987), pp. 267-68. Foreign
licensing and seminar participation have been established from unpublished lists contained
in communications from D. S. Hochheiser, AT&T Archives, June 1990 and February 1991;
also John E. Tilton, International Diffusion of Technology: The Case of Semiconductors
(Brookings, 1971), pp. 102, 104, 106. Attendance records for the 1952 symposium may not
be absolutely complete; the only surviving historical record of participation is a series of
group pictures of attendees.
8. Its name had been changed from the Kawanishi Machine Manufacturing Company.
See Nakagawa, Semiconductor Development in Japan, p. 75.
9. See Tilton, International Diffusion of Technology, p. 77.
10. See William F. E. Long, “Price and Nonprice Practices Under the Uncertain Con¬
ditions of Rapidly Improving Technologies—A Case Study,” Ph.D. dissertation. Depart¬
ment of Economics, George Washington University, June 1967, pp. 24-25. However, pay¬
ment of the $25,000 to Bell entitled early licensees to attend the important April 1952
symposium in which detailed transistor know-how was transferred to them. See also “Busi¬
ness Week Reports On: Semiconductors,” Business Week, March 26, 1960, pp. 93-94.
11. See U.S. Tariff Commission, Television Receivers and Certain Parts Thereof, TC
publication No. 436 (1971), p. 13.
12. See Jack Baranson, The Japanese Challenge to U.S. Industry (Lexington Books,
1981), p. 40.
13. See “RCA Prepares to Open Far East Research Lab,” Electronics, July 24, 1960,
THE JAPANESE ASCENT IN SEMICONDUCTORS 43

color television technology to Japanese consumer electronics makers.14

Japanese electrical engineers later characterized access to RCA electron¬
ics technology as a “treasure mountain.”15
The year 1952 also marked the visit to the United States by Masaru
Ibuka, who with Akio Morita had in 1946 started a small Japanese elec¬
tronics company, Tokyo Telecommunications Engineering, later renamed
the Sony Corporation. Ibuka and Morita had been occupied with the
development of electronic circuitry for the Japanese military during the
war, and after the war’s end their government connections were to prove
useful in getting their fledgling company off the ground.16 During his
American visit, Ibuka learned of developments in transistor technology
and the terms of the Bell Labs licensing offer. In 1953 Morita was sent
to America to secure a license and needed technology from Bell, and he
returned with a provisional signed agreement and a mountain of technical
information.
In 1950, however, at a time of great scarcity of foreign exchange, MITI
had been given authority to approve or reject all contracts for technology
with foreign companies.17 MITI repeatedly used this authority over the
next twenty years, together with its control over approval of direct in¬
vestment in Japan by foreign companies and a licensing system for im¬
ports, to improve the terms on which Japanese electronics companies
could obtain access to foreign technology, and to force foreign companies
to transfer or license their technology to Japanese enterprises in exchange
for access to the Japanese market. In this case MITI did not approve the
licensing arrangement signed by Morita until 1954; it approved similar
agreements with Bell Labs for Toshiba, Hitachi, and Kobe Kogyo later

p. 11. The initial focus of its Japanese research effort was on basic studies of solid state
phenomena.
14. See U.S. Tariff Commission, Telephone Receivers and Certain Parts Thereof, p. 13.
Heavy and Chemical Industries News Agency, Foreign Investments in Japan, 2d ed. (Tokyo,
1967), chap. 5, shows virtually every major Japanese manufacturer of television receivers
and parts signing licensing and technical assistance contracts with RCA in the color tele¬
vision area in 1962.
15. Nakagawa, Semiconductor Development in Japan, p. 97.
16. Ibuka’s contacts were instrumental in initial sales of voltmeters and broadcast
equipment to Japan Broadcasting (NHK), Japan National Railways, and other government
agencies in the early years of the company. See Nick Lyons, The Sony Vision (Crown,
1976), p. 14.
17. See Merton J. Peck and Shuji Tamura, “Technology,” in Hugh Patrick and Henry
Rosovsky, eds., Asia’s New Giant: How the Japanese Economy Works (Brookings, 1976),
p. 535.
44 THE JAPANESE ASCENT IN SEMICONDUCTORS

in that year.18 MITI was later to use its control over technological tie-ups
to obstruct entry into the market by other Japanese firms.19
Many early concerns over patent rights were to become moot in 1956.
In that year AT&T signed a consent decree with the U.S. Department
of Justice that required all existing AT&T, Bell Labs, and Western Elec¬
tric transistor patents to be licensed free of charge to all interested do¬
mestic companies. Since Bell Labs had essentially dominated technology
development up to that point, this had the effect of placing the main¬
stream transistor technology of the day—junction transistors—into the
public domain.20
In other respects as well, U.S. semiconductor technologists were rel¬
atively open about permitting Japanese engineers unencumbered access
to their technology in the 1950s and 1960s. One individual posted at a
Japanese trading company’s New York office, for example, estimates that
he made arrangements for visits by approximately 3,000 Japanese elec¬
tronics engineers during a three-year assignment.21 Japanese researchers
also were given free access to American semiconductor research confer¬
ences, and came in large numbers.22 The methods some of these Japanese
visitors used to gather information may have contributed to a certain
degree of resentment visible in later years on the American side.23

18. See Nakagawa, Semiconductor Development in Japan, p. 285.

19. Mitsubishi, Sanyo, Fuji Electric, and Oki were reportedly hindered from entering
the transistor market by MITI’s controls over technology transfer. See NHK, Electronics-
Based Nation: Japan’s Autobiography, vol. 2 (Tokyo, 1991), p. 41 (in Japanese).
20. As Tilton, International Diffusion of Technology, pp. 76-77, points out, however,
continuing Bell Labs R&D on semiconductor technology, and the resulting patents, made
it imperative for semiconductor firms to work out licensing agreements covering new ad¬
vances with AT&T.
21. The favorite sites for visits were the Bell Labs, RCA, Western Electric, and General
Electric plants. NHK, Electronics-Based Nation, vol. 1.
22. A recent Japanese television documentary series featured the recollections of an
American engineer present at some of these meetings. At one large International Electron
Device Conference meeting in the 1960s, for example, the American researcher recalls that
about 20 percent of the attendees were Japanese. Often, he notes, all of the authors on
Japanese papers with multiple authors attended, apparently for the purpose of monitoring
and thoroughly documenting the presentations of other attendees. See NHK, Electronics-
Based Nation, vol. 2, pp. 347-53.
23. Some of the anecdotes publicly recalled on the Japanese documentary series would
be regarded by an American audience as reflecting poorly on their narrator. For example,
a top technologist at NEC described how, when he visited General Electric in 1960, GE
executives refused to disclose know-how for new products under development, on the
ground that GE’s existing technology contract with NEC only covered items currently in
production. Angered, the NEC technologist cultivated direct contacts with young GE
THE JAPANESE ASCENT IN SEMICONDUCTORS 45

Concentrating on acquiring existing know-how rather than creating

new technology, Japanese chip makers also faced a business environment
significantly different from that of American producers. Unlike firms in
the United States, where a massive military market sparked the growth
of the U.S. electronics industry, Japanese companies could not look to
government sales as their engine of growth. Commercial applications
instead had to spur development. NTT’s labs pointed the way when they
publicly exhibited a prototype transistor radio and a transistorized hear¬
ing aid in 1953. Although the Regency Corporation was first to actually
market a transistor radio, in the United States in 1954, Sony and Kobe
Kogyo quickly followed this lead. In Japan, Kobe Kogyo was the first to
develop a commercial transistor radio, which it announced in January
1954, but the reception of the crude set was poor, and the radio was not
a success.24 The transistor radio that Sony brought to market in late 1955
was of a markedly higher quality and set Sony on a successful growth
path that ultimately transformed the company into today’s multinational
giant.
The key to Sony’s success was its early mastery of the technology
needed to mass-produce transistors. By 1956 Sony transistor production
had topped 300,000 units, and a year later it had more than doubled, to
800,000 units. Sony’s closest competitor was Toshiba, which began mass
production of transistors in 1956. Kobe Kogyo did not begin mass pro¬
duction until 1957, and Hitachi and NEC did not build large-scale tran¬
sistor plants until 1958. When the large companies did enter the game,
though, they came in force, and their presence was quickly established.
An avalanche of transistors roared down (figure 2-1). In 1958 Toshiba
completed a single plant that was to produce 1.2 million transistors per
month.
By 1959 Toshiba was the largest Japanese semiconductor producer
(with 26 percent of the market), followed by Matsushita (16 percent),
Hitachi (15 percent), NEC (15 percent), Sony (11 percent), and Kobe

engineers involved in the project and so obtained the needed know-how. In another in¬
stance, a former Toshiba researcher described how, forbidden to take notes while visiting a
New York office, he would feign a need to go to the bathroom, where he would write
important points down. He did this so often that those with him thought he had a stomach
ailment. See NHK, Electronics-Based Nation, vol. 1, pp. 264-76, which contains material
from this series of NHK documentaries.
24. One of Kobe Kogyo’s engineers, Leo Esaki, moved from Kobe Kogyo to Sony in
1955, where he was to discover the tunnel diode, which earned him a Nobel Prize in 1973
and became Japan’s first significant technical contribution to semiconductor technology.
46 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-1. Output of Transistors and Diodes by Japanese

Manufacturers, September 1954-March 1956“

Transistors Diodes

80,000

50,000

40,000

20,000

Source: Twenty-Year History of the Electronics Industry (Tokyo: Dempa Publications, 1968), p. 110.
a. Manufacturers are Toshiba, Sony, Kobe Kogyo, Hitachi, and NEC. Mitsubishi, Fuji Tsushin, and Matsushita
were engaged in experimental research only.
b. Begins with sale of Sony transistor radios.
THE JAPANESE ASCENT IN SEMICONDUCTORS 47

Table 2-1. Transistor Output of U.S. and Japanese Companies during

Transition to Overseas Production, 1957-68
Millions of units unless otherwise specified

Output
Percent of Percent of
United Japanese transistors Japanese transistor
Year States Japan used in radios radios exported
1957 29 6 67 n.a.
1958 47 27 67 n.a.
1959 82 87 55 77
1960 128 140 48 70
1961 191 180 41 67
1962 240 232 34 76

1963 300 268 35 81

1964 407 416 33 69
1965 608 454 30 75
1966 856 617 26 86
1967 760 766 23 83
1968 883 939 20 90
Sources: For the United States, Electronic Industries Association, Electronic Market Data Book, 1977(Washington,
D.C., 1977), table 77, p. 115; for Japan, John E. Tilton, International Diffusion of Technology: The Case of Semicon¬
ductors (Brookings, 1971), pp. 156-57.
n.a. Not available.

Kogyo (5 percent). Despite Japan’s lag in the technology needed to

produce the most advanced products, imports into its highly protected
domestic market came to only 2 percent of consumption.25
The late 1950s marked a period of explosive growth in Japanese tran¬
sistor production. In 1957 Japanese unit output of transistors amounted
to about one-fifth of American production; by 1959 Japanese output had
passed U.S. levels. Until 1960 one-half to two-thirds of Japanese output
was incorporated into transistor radios, and through 1961 better than
two-thirds of these were exported (table 2-1). In those days, transistor
fabrication was a highly labor-intensive activity, and relatively low Japa¬
nese wage rates made it possible for Japanese companies to compete
successfully on price, particularly in relatively low-quality transistors for

25. These figures are reported in Tilton, International Diffusion of Technology, p. 144.
Japan’s lag in high-performance transistor technology for use in industrial applications
(such as communications equipment) was apparent in an internal battle that took place
within NEC, when it made a decision to support the construction of production facilities
for transistors. Research staff favored the use of indigenously developed transistor tech¬
nology, while manufacturing engineers advocated the use of more advanced American
transistor technology. A bitter conflict was ended in 1957, when NEC’s chairman chose to
go with the American technology. The episode is detailed in Nakagawa, Semiconductor
Development in Japan, pp. 91-93.
48 THE JAPANESE ASCENT IN SEMICONDUCTORS

use in consumer electronics, where cost, not performance, was critical.

In the late 1950s and early 1960s, for example, the consumer electronics
market was supplied almost exclusively with low-frequency germanium
transistors produced with relatively cheap alloy junction and grown junc¬
tion technologies, while the computer, defense, and industrial markets
were the destination for more advanced high-frequency transistors made
with sophisticated diffused base technologies.26 The most advanced high-
performance transistors were sold almost exclusively to the computer
industry.
The emphasis on mass production of low-end consumer products,
where cost, not quality, was paramount, was a successful strategy in the
1960s. Some in the United States dismissed any notion that Japanese
producers might pose a threat in the higher performance applications (in
computers and military systems) that drove the U.S. industry’s techno¬
logical advance. It would be incorrect, however, to conclude that the
early Japanese effort in semiconductors was exclusively oriented toward
consumer electronics. There was some attempt to use Sony point-contact
transistors when MITI’s Electrotechnical Laboratory constructed its first
transistorized computer in 1956, but there were problems with these
components, and imported products ultimately seem to have been used.27
A successor to this machine was built in 1957, which successfully made
use of more advanced junction transistors made in Japan by Hitachi, as
well as germanium diodes produced by NEC.28 The computer design was
transferred to Japanese companies and was the basis for the first com¬
mercial computers shipped by them.29 But the performance of the com¬
ponents used in these machines lagged well behind those used in the
United States, where newly developed surface barrier transistors were
setting new standards of speed and reliability.

26. See Long, “Price and Nonprice Practices,” pp. 30-36. For a description of the
different techniques, see chapter 1, note 39 of this volume.
27. Takahashi writes that an improved version of a Sony transistor was selected for the
computer (the ETL Mark III) design, but the official history of computer development at
the ETL notes that point-contact transistors imported from the United States were actually
used. See Osamu Ishii, “Research and Development on Information Processing Technology
at Electrotechnical Laboratory: A Historical Review,” vol. 45 (1981), p. 315 [contract
translation by IBRD]; and Shigeru Takahashi, “Early Transistor Computers in Japan,”
Annals of the History of Computing, vol. 8 (April 1986), p. 146.
28. Takahashi, “Early Transistor Computers in Japan,” p. 149; Ishii, “Research and
Development on Information Processing Technology,” p. 315.
29. See Flamm, Creating the Computer, pp. 175-76.
THE JAPANESE ASCENT IN SEMICONDUCTORS 49

Early Exports in Electronics

The mid- to late 1950s are an early breakpoint in the history of the
semiconductor industry. On the technical side, Bell Labs made public
new technologies for making transistors, while other techniques devel¬
oped outside the Bell Labs emerged in the commercial market.30 These
new methods soon displaced older technologies in higher performance
applications. Transistors based on a silicon semiconductor material,
which offered higher performance (although initially at a higher cost),
quickly began to replace germanium transistors in industrial applica¬
tions—such as computers—where better performance justified a pre¬
mium price. Most important, prototypes of the first integrated circuits
(ICs) were built in 1958-59, a development which was to transform the
electronics industry over the decade of the 1960s.
In the consumer market, Japanese firms made their first great push
into exports, triggering a cycle of reaction and response that in many
respects was repeated over the next three decades. The initial vehicle for
Japan’s export push was the transistor radio (table 2-1). In the early 1960s
the push into transistor radios was augmented by exports of transistorized
versions of other consumer electronics products, particularly televisions,
tape recorders, audio equipment, and, in the late 1960s, transistorized
calculators.31
The rapid Japanese expansion in consumer electronics and the sharply
increasing production of relatively inexpensive, lower performance tran¬
sistors used in those applications led to sweeping changes in the U.S.
electronics industry. Figures 2-2 and 2-3 show rapid changes in the com¬
position of American transistor output over the decade spanning Japa¬
nese entry into inexpensive transistors and the consumer electronics in

30. As part of its 1956 consent decree, AT&T had agreed to license all existing patents
royalty-free to any domestic company, and not to sell semiconductors in commercial markets
(that is, excluding defense and space applications).
31. On Japanese consumer electronics exports see “Japan Boosts TV Set Output,”
Electronics, February 26,1960, p. 48; “Japanese TV Sets Arriving This Week,” Electronics,
April 29, 1960, p. 32; “Sales and Imports of Radio-TV Rise,” Electronics, May 19, 1961,
p. 12; U.S. Department of Commerce, Bureau of Domestic Commerce, The U.S. Con¬
sumer Electronics Industry (1975), pp. 8-9, 11. On calculators see U.S. Department of
Commerce, Bureau of Domestic Commerce, The Impact of Electronics on the U.S. Cal¬
culator Industry, 1965 to 1974 (November 1975), pp. 13-15,44; and Badiul Alam Majumdar,
Innovations, Product Developments, and Technology Transfers: An Empirical Study of
Dynamic Competitive Advantage, The Case of Electronic Calculators (University Press of
America, 1982), chap. 5.
50 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-2. Volume of U.S. Transistor Production, by Type, 1955-65

Percent

100

80

60

40

20

0
1955 1960 1965

Consumer Industrial Industrial silicon

germanium
Sources: Electronic Industries Association, Electronic Industries Yearbook (Washington, 1959), p. 38; Electronic
Industries Association, “Factory Sales of Transistors,” monthly series, and “Sales of Transistors by Market,” series
of reports. Cited in William F. E. Long, “Price and Nonprice Practices under the Uncertain Conditions of Rapidly
Improving Technologies: A Case Study,” Ph D. dissertation, George Washington University, 1967, tables 1, 6, 10.

which they were incorporated. There was a sharp decline in the share of
transistors for consumer applications (mainly germanium for most of this
period) in U.S. output, in terms of both value and quantity; this decline
coincided with the Japanese push into consumer electronics exports that
began after 1957.
Relatively more of U.S. output shifted into higher performance tran¬
sistor types used in industrial applications, especially the recently intro¬
duced silicon transistors.32 Rapidly declining costs made the new silicon

32. Over this period the principal performance difference between transistors for the
consumer market and those for the industrial market was the ability to operate at high
frequencies. Transistors made using grown or alloy junction technology generally operated
at low frequencies, while transistors manufactured using newer diffused base techniques
operated at much higher frequencies. Virtually all consumer applications required only
THE JAPANESE ASCENT IN SEMICONDUCTORS 51
Figure 2-3. Value of U.S. Transistor Production, by Type, 1955-65
Percent

llllll Consumer Industrial Industrial silicon

germanium
Sources: See figure 2-2.

transistors more attractive for low-end applications over time, and some
resurgence by American producers in the market for consumer-use tran¬
sistors was evident by the mid-1960s. In a pattern that was to recur later,
the United States passed Japan in transistor output in the early 1960s as
demand shifted to more technically advanced silicon transistors. Japan
then raced to catch up in the new technology, ultimately surpassing the
United States again in total transistors produced in the late 1960s. This
phenomenon—of new technology providing the basis for improved com¬
petitive performance by American producers vis-a-vis foreign rivals man-

low-frequency operation, while industrial applications typically required high-frequency

operation. Higher performance transistors could typically be used in consumer applications
with less exacting operational requirements but would not normally be an economic choice
because of their greater cost. However, some rejected, “not-up-to-spec” high-performance
transistors were adequate for consumer applications, and such rejects were sold into the
consumer market. See Long, “Price and Nonprice Practices,” pp. 30-31.
52 THE JAPANESE ASCENT IN SEMICONDUCTORS

ufacturing older products at lower cost—was (as will be seen below)

repeated again in the 1960s and in the 1970s.
For the most part, the Japanese government’s role in the early devel¬
opment of the Japanese semiconductor industry was as a mediator of
contacts with markets outside Japan and as a gatekeeper to the Japanese
market. Although the government exercised significant regulatory powers
in managing technical and trade contacts with the outside world, and
government labs played an important part in Japanese technology devel¬
opment, it was largely the private sector that took the first initiatives in
the newly emerging solid state electronics technology, with an eye on
markets for consumer electronics products.
The Japanese government was to greatly increase its role in shaping
Japan’s electronics industry in the 1960s. It was computers, however, and
not consumer electronics, that led the government to take significant steps
to stimulate technological advance in the Japanese semiconductor indus¬
try. MITI’s attempts in the 1960s to help Japanese computer manufac¬
turers catch up with developments abroad were to be the backdrop
of a successful drive in the 1970s to the technological frontier in semi¬
conductors.

The Seal of Approval

A landmark event for the Japanese electronics industry was the pas¬
sage of the Law on Temporary Measures for Promoting the Electronics
Industry (the Electronics Industry Promotion Law) in 1957.33 (The basic

33. The Electronics Industry Promotion Law (Denshinho) was actually preceded by a
Law on Temporary Measures for the Promotion of Specified Manufacturing Industries in
1956, which, like the 1957 electronics law, provided for low-interest loans from the Japan
Development Bank (JDB) for investments in modernization and technology within selected
industrial sectors. Machine tools, auto parts, and other basic machinery were among the
nonelectronic sectors enjoying favored access to JDB capital. From 1956 through 1960
slightly less than 10.6 billion yen in loans was extended to these sectors. This compares
with 2.2 billion yen in loans to the consumer electronics sector granted over the seven years
from 1957 through 1963.
The Specified Manufacturing Industries Law was renewed in 1961 and again in 1966,
expiring in 1971. The Small Business Finance Corporation (SBFC) joined the JDB in
providing subsidized credit to the machinery industry, with loans extended by these two
institutions amounting to 53.8 billion yen over 1961-65, and 48.9 billion yen over 1966-70.
In 1964 the electronics law was extended for another seven years. Loans over this period
amounted to 12.1 billion yen, with the bulk of the funds going into electronic components.
In 1971 the electronics and machinery industry laws were combined into a single mea-
THE JAPANESE ASCENT IN SEMICONDUCTORS 53

structure was extended by successor laws: the Machinery and Electronics

Industry Law, 1971; the Machinery and Information Processing Industries
Law, 1978.34) Ironically, the 1957 law was primarily intended to promote
the consumer electrical and electronics products industries, and only after
its renewal in 1964 did the emphasis shift to computers and semiconduc¬
tors.35 Perhaps the law’s most important provision was its establishment
of an Electronics Industry Division within MITI, and an Electronics
Industry Council to coordinate policymaking with the Japanese electron¬
ics industry. This established MITI as the bureaucratic patron—in Japa¬
nese parlance, genkyoku—under whose jurisdiction the electronics in¬
dustry fell. As Komiya writes, a genkyoku typically strives for “its
industry to be orderly and organized and for there to be no disruptions
of any kind.”36 The contrast between this bureaucratic perspective and
the mercurial ups and downs characterizing the history of the highly
entrepreneurial U.S. electronics sector is noteworthy.
The Electronics Industry Promotion Law, in a pattern that was to be
repeated in later legislative frameworks for that sector’s promotion, es¬
tablished three classes of support efforts. For important technologies,
R&D subsidies were to be given. The second type of program provided
low-interest loans from the government-run Japan Development Bank

sure, the Law on Temporary Measures for the Promotion of Specified Electronics Industries
and Specified Machinery Industries (Kidenho). Over the seven years through 1977, the
electronics industry received approximately 20 percent of some 70.6 billion yen in loans
under the new law. In 1978 this measure was revised and extended as the Law on Temporary
Measures for the Promotion of Specified Machinery and Information Industries (Kijoho).
Loans from the SBFC and the JDB amounted to approximately 11 billion yen per year,
with approximately 80 percent going into electronics.
See Seiritsu Ogura and Naoyuki Yoshino, “The Tax System and the Fiscal Investment
and Loan Program,” in Ryutaro Komiya, Masahiro Okuno, and Kotaro Suzumura, eds.,
Industrial Policy of Japan (Academic Press, 1988), pp. 147-48.
34. The 1957 Denshinho law was extended in 1964, then followed by the 1971 Kidenho
law, which in turn was followed by the 1978 Kijoho (see the previous note.) Although the
Kijoho expired in 1985, another law, the Josokuho (established in 1960, amended to include
many new provisions for investment subsidies, and reenacted in 1986) provides legal au¬
thority for many of the sorts of subsidies to the computer and semiconductor industries
granted under the Kijoho. All recent JDB investments in information technology and
related industries, for example, have been made under the framework of the Josokuho.
See Japan Electronic Computer Corporation (JECC), Computer Notes (Tokyo, 1989), pp.
62-63 (in Japanese).
35. See Koji Shinjo, “The Computer Industry,” p. 342; and Ogura and Yoshino, “The
Tax System,” in Komiya and others, eds., Industrial Policy of Japan, p. 147.
36. See Komiya, “Introduction,” in Komiya and others, eds., Industrial Policy of Japan,
54 THE JAPANESE ASCENT IN SEMICONDUCTORS

(JDB), and later from its sister institutions, for initial production and
subsequent capacity expansion in specified products.37 The third type of
support provided additional JDB loans and special depreciation breaks
where “industrial rationalization”—that is, improvements in quality or
production technology—was desirable.38 The law also authorized selec¬
tive exemption from the antimonopoly law, allowing MITI to establish
research and production cartels.39
Although semiconductor-related R&D and production rationalization
were thus supported virtually from the start, the resources applied to
these ends were clearly quite small.40 All electronics-related R&D sub¬
sidies averaged a little over 230 million yen (just under $650,000) per
year during the seven-year period ending in fiscal 1963. Similarly, JDB
loans to the electronics industry accounted for an annual average of
324 million yen (about $900,000) over the same period.41 As previously
remarked, funding mainly went into consumer electronics—only a small
portion of these funds was applied to semiconductors.
In contrast, subsidized JDB loans to other machinery industries auth¬
orized by similar legislation amounted to over 2 billion yen per year over
1956-60, and almost 11 billion yen annually over 1961-65. Thus, the
“temporary measures” of 1957 initially led to only minimal new subsidies
for semiconductors. More important, perhaps, was the rigorous policy of

37. It is frequently argued that, in addition to the funds themselves, and the signal to
invest in the promoted sector provided to highly regulated private financial institutions
(which was quite significant in the capital-short Japanese economy of the 1960s, but less so
in the late 1970s and 1980s) or the direct subsidy element in loans at below-market rates,
these funds provided an implicit government guarantee for private sector loans to these
sectors.
38. This discussion follows Shinjo, “The Computer Industry,” p. 342.
39. MITI is reported to have used this law to block entry into semiconductors by some
Japanese latecomers, including Sanyo Electric. See Nakagawa, Semiconductor Development
in Japan, pp. 101-03.
The authority to sanction legal cartels in electronics was not often invoked. Over the
years 1957-70 some six legal “rationalization” cartels were established under the auspices
of the Electronics Industry Promotion Law: three over the period 1964-66 and three over
the period 1969-70. This compares with some 114 established in machinery under the
machinery promotion law over the same period. See Yutaka Kosai, “The Reconstruction
Period,” in Komiya and others, eds., Industrial Policy of Japan, p. 47.
40. Categories of products that were promoted, and the years over which the support
extended, are shown in Badiul A. Majumdar, “Industrial Policy in Action: The Case of the
Electronics Industry in Japan,” Columbia Journal of World Business, vol. 23 (Fall 1988),
pp. 27-29.
41. Majumdar, “Industrial Policy in Action,” p. 30-31.
THE JAPANESE ASCENT IN SEMICONDUCTORS 55

trade protection for the electronics industry formalized and justified by

passage of the Electronics Industry Promotion Law.42

Protecting the Japanese Market

Notwithstanding Japan’s early successes in exporting transistors and

consumer electronics to overseas markets, virtually all segments of the
internal Japanese electronics market were tightly walled off from foreign
competition by this time. The implements used to construct steep walls
around the Japanese market were forged during the immediate postwar
years, with the approval of the U.S. occupation authorities.
The two principal measures used to protect the domestic market, the
1949 Foreign Exchange Control Law (FECL) and the Foreign Investment
Law (FIL) of 1950, were parts of a complex of laws affecting foreign
exchange remittances and export-import trade.43 The FIL essentially re¬
quired case-by-case approval by MITI of all foreign investments in Japan
and was used to effectively block all foreign investment not deemed to
be in the national interest.44 The law also provided the legal rationale for
an extensive system requiring prior approval of technological agreements
between Japanese and foreign companies that generated foreign exchange
remittances. MITI exercised tight controls over licensing, royalty, and
technology transfer agreements, in order to influence the pricing and
composition of technology imports and the structure of high-technology
industries in Japan.45
The FECL and related measures had the effect of giving MITI con¬
siderable discretionary power over foreign trade transactions. By 1960

42. See Motoshige Itoh and Kazuharu Kiyono, “Foreign Trade and Direct Investment,”
in Komiya and others, eds., Industrial Policy of Japan, pp. 147, 160.
43. Other elements included the Export Trade Control Law (ETCL), the Import Trade
Control Law (ITCL), and the rules of the “Standard Method of Payments” governing
imports. On the operation of these measures, see Lawrence Krause and Sueo Sekiguchi,
“Japan and the World Economy,” in Patrick and Rosovsky, eds., Asia’s New Giant, pp.
411-27, 451-52; Mark Mason, American Multinationals and Japan: The Political Economy
of Japanese Capital Controls, 1899-1980 (Harvard University Press, 1992), pp. 159-61; and
Dennis J. Encarnation, Rivals Beyond Trade: America versus Japan in Global Competition
(Cornell University Press, 1992), pp. 46-50.
44. On the mechanics of how this system was used, see Mason, American Multinationals
and Japan, pp. 154-61; Encarnation, Rivals Beyond Trade, pp. 47-50.
45. See Terutomo Ozawa, Japan's Technological Challenge to the West, 1950-1974: Mo¬
tivation and Accomplishment (MIT Press, 1974), pp. 16-24, 52—66; and Peck and Tamura,
“Technology,” 544-58.
56 THE JAPANESE ASCENT IN SEMICONDUCTORS

the foreign exchange control system had evolved into a broad system of
MITI-administered import quotas. Mounting foreign criticism led to Ja¬
pan’s first steps toward trade liberalization at the end of the 1950s, when
257 commodities, covering 44 percent of all Japanese imports, were first
freed from licensing requirements.46 The 1960 program of Trade and
Exchange Liberalization Guidelines initially considered selected elec¬
tronics items for liberalization, but MITI later withdrew electronics goods
from the list of items to be liberalized.47
Tight control over foreign imports of and investment in semiconductors
was maintained throughout the 1950s and early 1960s. The limitations on
investment were particularly important in ICs, products that often re¬
quired considerable technical support from manufacturers. Without a
local subsidiary to provide that support to customers or distributors, an
effective sales effort was difficult, if not impossible.
In 1962, for example, Robert N. Noyce, then one of the key technical
figures at Fairchild Semiconductor and co-inventor (with Jack S. Kilby of
Texas Instruments) of the integrated circuit, visited Japan to arrange an
investment in a local production facility and discuss the sale of licenses
to use Fairchild’s key patents to Japanese firms.48 (Together, the TI and
Fairchild patents were the key intellectual properties at the center of the
emerging IC market.)
Noyce, and Fairchild, were completely outmaneuvered in their initial
foray into the Japanese market. MITI flatly refused to permit Fairchild
to invest in its own Japanese factory. Frustrated, Noyce then contacted
an NEC executive he had met in the United States at a defense semicon¬
ductor conference,49 complained about MITI’s reaction, and set about
negotiating with NEC over patent rights. Informed of the need for NEC
to secure approval of the terms from MITI, Fairchild was pressured to
reduce its royalties and fees; only later did the U.S. company discover

46. Krause and Sekiguchi, “Japan and the World Economy,” p. 414.
47. See “Japanese Put Off Freeing Electronics Imports,” Electronics, July 8, 1960,
p. 11. The initial, “temporary” delay of three years was to stretch into sixteen years;
quantitative restrictions were not completely removed from electronics imports until 1976.
48. This episode is described in Nakagawa, Semiconductor Development in Japan,
pp. 135-40.
49. Noyce apparently met Mr. Naganuma, the NEC executive, at a conference de¬
scribed in NHK, Electronics-Based Nation, vol. 2, pp. 253-56: the U.S. Navy-sponsored
Semiconductor Devices Research Conference, held in Boulder, Colorado. Although the
conference was closed to the general public, some foreign engineers including Naganuma
were allowed to attend.
THE JAPANESE ASCENT IN SEMICONDUCTORS 57

that NEC’s own president also chaired the MITI licensing approval ad¬
visory committee responsible for this decision!50
A deal was finally struck in 1963. Noyce negotiated an exclusive license
with NEC for patent rights in the Japanese market. The agreement so
completely transferred the Japanese rights to Fairchild’s IC technology
to NEC that it was later used to block attempts by Fairchild to set up a
Japanese sales office. NEC also charged other Japanese companies for
their use of covered technologies. It was a poor deal for Fairchild, but it
reflected the then-prevailing climate for foreign investors in Japan. That
same year, TI began its attempt to enter the Japanese market, and took
a much tougher bargaining stance. But TI was not to gain even a small
degree of access to the Japanese market until half a decade later, in 1968,
after an exhausting battle.
Through 1963 only seven foreign semiconductor producers had been
permitted to invest in the Japanese market, in all cases through joint
ventures in which they held a minority interest, and in all cases with
extensive transfer of technical know-how a part of the deal. Four of these
ventures were broad tie-ups in electrical equipment and electronics,
which also covered semiconductors. Over the period from 1952 through
1962 the Dutch company Philips NV had acquired a 35 percent interest
in Matsushita Electronics, the semiconductor-producing affiliate of Mat¬
sushita.51 From 1953 to 1960 ITT’s International Standard Electric sub¬
sidiary had increased its holdings in NEC to 17.7 percent. General Electric
(GE) had built up a 5.6 percent interest in Toshiba from 1953 to 1959.52
Three new ventures were exclusively concerned with semiconductors.
From 1957 to 1961 International Rectifier had established a 39 percent
stake in Nippon International Rectifier. In 1961 Raytheon Company had
invested in New Japan Radio, a semiconductor-producing joint venture
with Japan Radio (Raytheon’s share had increased to 33 percent by 1963).
In 1962 Mitsubishi TRW (with a 20 percent stake held by TRW Incor¬
porated, and another 20 percent held by Pacific Semiconductor) and

50. See Mason, American Multinationals and Japan, pp. 196, 325. Mason’s account is
based on interviews with Noyce and then-executive vice president of Fairchild, Richard
Hodgson.
51. This equity was finally sold back to Matsushita in 1993. See Douglas R. Sease,
“Abreast of the Market: With Confusing Results for 1st Quarter, It Isn’t Surprising the
Market Is Stalled,” Wall Street Journal, May 3, 1993, p. Cl.
52. See Heavy and Chemical Industry News Agency, Foreign Investment in Japan,
pp. 375, 377.
58 THE JAPANESE ASCENT IN SEMICONDUCTORS

Komatsu Hoffman (46 percent owned by silicon rectifier maker Hoffman

Electronics) had been formed to produce specialized semiconductors.53
A critical moment in the Japanese semiconductor industry’s relations
with the outside world came in 1963, when Texas Instruments—co-owner
with Fairchild of the intellectual birthrights to the IC—decided to apply
for permission under the FIL to set up a wholly owned Japanese manu¬
facturing subsidiary. TI had earlier approached Sony with a proposal to
set up a manufacturing joint venture, but negotiations had collapsed in
1959 when Sony insisted on exclusive rights to sell the output.54 TI had
also reportedly contacted NEC in 1960, when top executives were visiting
Japan, about a joint venture in semiconductors. The proposal had been
rejected outright.55
At roughly this same time, International Business Machines Corpo¬
ration (IBM), using its vast patent portfolio and the promise of technol¬
ogy transfer as a bargaining chip, had managed—after years of negotia¬
tions and acceptance of various restrictions on its activities—the highly
unusual feat of negotiating with MITI the establishment of a wholly
owned computer manufacturing facility in Japan in I960.56 TI’s interest
in establishing a Japanese facility was reportedly motivated at least
in part by the prospect of supplying IBM (with which TI had excel¬
lent relations elsewhere in the world) and its new Japanese facility with
components.57
After numerous presentations in Tokyo in 1963, TI filed a formal
application in January 1964.58 Greeted with outright hostility by both

53. See Heavy and Chemical Industry News Agency, Foreign Investments in Japan,
chap. 5. See also Organization for Economic Cooperation and Development (OECD),
Electronic Components, Gaps in Technology (Paris, 1968), p. 186.
54. Mason, American Multinationals in Japan, p. 176.
55. Nakagawa, Semiconductor Development in Japan, pp. 125-27. TI then reportedly
joined other U.S. semiconductor producers in petitioning for restrictions on Japanese
transistor imports, in what appears to have been the first trade policy initiative taken by
the U.S. chip industry (discussed in chapter 3).
56. More details and an analysis of this episode may be found in Flamm, Creating the
Computer, pp. 180-82; Mason, American Multinationals in Japan, pp. 187-91; and Marie
Anchordoguy, Computers Inc.: Japan’s Challenge to IBM (Harvard University Press, 1989),
pp. 22-24.
57. Nakagawa, Semiconductor Development in Japan, p. 154.
58. The most complete account of this affair from the American side may be found in
Mason, American Multinationals in Japan, pp. 176-87. Mason was given access to the TI
corporate archives in preparing his research.
THE JAPANESE ASCENT IN SEMICONDUCTORS 59

MITI and the Japanese press, the application was pigeonholed—neither

approved nor disapproved—while MITI stalled for time.59

The Integrated Circuit Arrives in Japan

MITI’s interest in electronics was rapidly increasing in the early 1960s.
The market for electronic computers was booming, and it was becoming
clear that this was destined to be a major industry.60 After IBM had been
permitted to set up an affiliate manufacturing computers for the Japanese
market in late 1960, MITI actively urged three Japanese computer makers
to join a cooperative R&D program—the FONTAC project—to develop
important computer technologies. This project, begun in 1962, was the
first cooperative research project established among competing Japanese
electronics manufacturers. Although FONTAC was a relatively small-
scale project (it was largely finished by 1964 and received government
funding of 338 million yen, or $930,000),61 it was the prototype for later
government-industry collaboration on joint R&D programs.
The introduction by IBM of an innovative new family of computers—
the System 360 product line—in 1964 shook up both the mushrooming
computer industry and Japanese policy. IBM injected a whole new gen¬
eration of advanced semiconductor components—integrated circuits—
into the commercial mainstream, and this and other innovations rendered

59. It is interesting to note that in Mason’s account from the U.S. perspective (Amer¬
ican Multinationals in Japan, p. 180) the Japanese industry is portrayed as pressuring MITI
to resist TI’s demands. In Nakagawa’s version (Semiconductor Development in Japan, pp.
154-66), told from a Japanese perspective, MITI is credited with having a long-range vision
of a competitive threat that the domestic industry did not yet see, and organizing an industry
all too prone to cutting individual deals on the side with TI. According to Nakagawa,
Hitachi, Mitsubishi, and Toshiba all held secret talks with TI. According to Mason
(p. 323), Matsushita attempted to cut its own deal through Philips, which had an equity
share in (as noted above) and extensive cross-licensing arrangements with Matsushita
Electronics.
60. See Flamm, Creating the Computer, chap. 6.
61. The project spent a total of 1.126 billion yen, of which 787 million yen was provided
by the participating companies (NEC, Fujitsu, and Oki). The Computer Technology Re¬
search Association in which these companies joined together as members was not actually
dissolved until 1973. Thirty Year History of Mining and Manufacturing Industry Technology
Research Associations (Tokyo: Japan Industry Technology Promotion Association, 1991),
p. 70. See also Shinjo, “The Computer Industry,” p. 343.
60 THE JAPANESE ASCENT IN SEMICONDUCTORS

the Japanese FONTAC effort obsolete virtually overnight.62 TI’s contin¬

uing political fight to secure the right to establish a Japanese subsidiary
also spotlighted the importance of ICs for the nascent computer industry.
Although the Japanese were already experimenting with ICs by this
time, development had proceeded slowly. The first crude, experimental
IC built in Japan was constructed inside MITI’s Electrotechnical Labo¬
ratory in 1960. The private sector lagged well behind: NEC developed its
first experimental IC in 1962.63 There was little demand in Japan. Lacking
the huge defense effort in solid state communications and computer com¬
ponents that was driving technology development in the United States,
IC production in Japan was not to expand greatly until the solid state
calculator came to market in the late 1960s.64
The first major push to actually produce ICs came after IBM’s 1964
announcement of its System 360 series. Shocked into action, MITI
launched several initiatives. Some $80,000 (29 million yen) in grants for
R&D on ICs for use in computers was hurriedly made available to Jap¬
anese semiconductor makers in 1965.65 Even more significant, a ten-year,
6-billion-yen program of subsidized loans from the JDB to Japanese chip
producers was announced.66 And in April 1966 an outright ban on all foreign
ICs with more than thirty-four circuit elements was put into place.67
Even as TI’s application for a wholly owned subsidiary sat on hold,
MITI signaled that it was willing to permit some degree of foreign par¬
ticipation in the most sophisticated niches of the Japanese semiconductor

62. Actually, the IBM computers did not use true “monolithic” (that is, etched from a
single piece of silicon) ICs. Discrete semiconductor components were instead bonded to a
ceramic substrate to form so-called hybrid ICs. Although competitors were to soon intro¬
duce computers using true monolithic ICs, the IBM System 360 line nonetheless marked a
significant step forward in the sophistication of components used in commercial computers.
63. See Martin Fransman, The Market and Beyond: Cooperation and Competition in
Information Technology Development in the Japanese System (Cambridge University Press,
1992); and Charles Cohen, “Japanese Components Men Stress Integrated Circuits and
Diodes,” Electronics, May 4,1962, pp. 20-21. Mitsubishi brashly announced the availability
of a complete line of ICs in 1961 but was unable to produce a functioning complete circuit
until 1964. See Nakagawa, Semiconductor Development in Japan, pp. 130-35; NHK, Elec¬
tronics-Based Nation, vol. 3, pp. 83-93.
64. However, Mitsubishi’s early IC effort was driven by contracts from the Japan Self
Defense Agency to downsize communications equipment. NHK, Electronics-Based Nation,
vol. 3, p. 92.
65. Yasuo Tarui, “Japan Seeks Its Own Route to Improved IC Techniques,” Electronics,
December 13, 1965, pp. 90-98. The funds were made available to the “big six” of the
Japanese computer industry: Fujitsu, Hitachi, Toshiba, NEC, Oki, and Mitsubishi.
66. Nakagawa, Semiconductor Development in Japan, p. 158.
67. “Onward and Upward,” Electronics, October 3, 1966, p. 257.
THE JAPANESE ASCENT IN SEMICONDUCTORS 61

market, as long as it was effectively controlled by domestic interests. In

September 1964 American semiconductor specialist Bernard Jacobs
formed Kyodo Electronics Laboratories, in a joint venture with five
smaller Japanese electronic component companies.68 Kyodo quickly be¬
came the only significant purely merchant IC producer in the Japanese
market—by 1966 it was second only to NEC in sales in the infant Japa¬
nese IC market.69
Although Japanese chip makers had invested in experimental IC de¬
velopment in the early 1960s, mass production was not to start until 1966,
a full half decade behind the American industry.70 The first ICs coming
off these new production lines were for the most part used in the latest
computer models manufactured by the computer divisions of the same large
companies.71 But in 1966 imports dominated the emerging market, account¬
ing for almost two-thirds of Japanese IC consumption (figure 2-4).72
More powerful means were needed to counteract the mounting com¬
petitive threat from rapid technical progress in American computers. In
1966 MITI launched a new program of large-scale national R&D proj¬
ects: among the first three projects announced was an effort to develop
a “Very High Speed Computer System” (VHSCS).73 The five largest
Japanese computer producers—Fujitsu, Hitachi, Nippon Electric, To¬
shiba, and Oki, which also were among the largest semiconductor makers
in Japan—were teamed in an ETL-directed cooperative project to build
a large time-sharing computer system to compete with the IBM System
360 Model 67.

68. The five Japanese companies were Toko, Nippon Chemical Condensor, Koden Elec¬
tronics, Pioneer Electronics, and Alps Electric. See Tarui, “Japan Seeks Its Own Route,”
p. 91. Jacobs owned 25 percent of the venture’s equity and received additional compensation
for technical assistance. See Heavy and Chemical Industries News Agency, Foreign Invest¬
ments in Japan, 2d ed., pp. 48, 372.
69. See “Onward and Upward,” pp. 257-58.
70. Nakagawa, Semiconductor Development in Japan, p. 280. By late 1966 only NEC,
Kyodo Electronic Laboratories, and Fujitsu were producing ICs (in small quantities);
Hitachi and Toshiba had just completed manufacturing facilities. “Onward and Upward,”
pp. 257-58.
71. “Onward and Upward,” pp. 257-58.
72. In figure 2-4, IC exports from the United States by captive producers such as IBM
and the U.S. affiliates of Japanese companies (growing rapidly since the 1980s) are counted
as imports, while production by Japanese affiliates of U.S. companies, such as TI, IBM,
and Motorola, are counted as domestic output.
73. For more detailed references to MITI projects in computer technology development
and their outcomes over this period, see Flamm, Creating the Computer, pp. 184-92.
62 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-4. Import Shares of Japanese Semiconductor Consumption,

1966-94

Percent

Sources: Pre-1970 measures of IC production from Machinery Promotion Association Economic Research Institute,
Japan-US Semiconductor Industry Research Survey (Tokyo, 1980), p. 114. Post-1970 production data from Research
and Statistics Department. MIT1, Yearbook of Machinery Statistics, various years. Import data from Japan Tariff
Association, Japan Exports and Imports, Commodity by Country (Tokyo, various years). Import share of consumption
is defined as value of imports divided by apparent consumption (Value of production plus value of imports less value
of exports). Hybrids were not a separate category in trade data until 1989. Hybrids are excluded in production data
only.
a. April 1966: ban on foreign ICs with more than 34 circuit elements.
b. September 1970: Liberalization of ICs with less than 100 elements.
c. Late 1971: MITI crackdown on LSI IC import licenses.
d. 1973: removal of licensing of ICs with less than 200 elements.
e. December 25, 1974: liberalization of IC imports with more than 200 elements.
f. 1975: liberalization of computer ICs.
g. U.S.-Japan Semiconductor Trade Arrangement.

As part of this project, Fujitsu, Hitachi, and NEC shared a 400-

million-yen ($1.1 million) grant to develop next-generation integrated
circuits: so-called LSI (large-scale integration) ICs, containing up to
10,000 transistors on a single chip.74 This was the largest infusion by
MITI of R&D funding into semiconductors yet, and it achieved impor¬
tant results.75 In addition to fueling progress on LSI within all three

74. Yasuo Tarui, “ICs in Japan—a CloseupElectronics, May 13, 1968, p. 106. Al¬
though there is no exact definition, medium-scale integration (MSI) generally is taken to
mean ICs with up to 1,000 circuit elements, while very large scale integration (VLSI) devices
have up to 100,000 circuit elements.
75. MITI’s ETL, which directed the project, developed a metal oxide semiconductor
THE JAPANESE ASCENT IN SEMICONDUCTORS 63

companies, the funding helped Hitachi develop high-performance logic

chips and NEC develop IC memory chips (Japan’s first). Chips utilizing
semiconductor technology developed in this project soon showed up in¬
side computer systems shipped by Hitachi, NEC, and Fujitsu.76
The increased interest in electronics is clearly visible in the fragmentary
statistics that are available on MITI subsidies to investment and R&D in
the electronics industry in the 1960s. The Electronics Industry Promotion
Law had been renewed for another seven years in 1964 and was later com¬
bined with the Machinery Industries Promotion Law in 1971 (and again
extended in 1978).77 As before, the three major means of subsidy under
these revised measures were R&D subsidies, special depreciation tax
breaks, and subsidized loans from the public development banks.
Table 2-2 shows the volume of subsidized loans from public financial
institutions extended under these successive laws in the relevant multiyear
periods.78 In years prior to 1964, subsidized loans to the electronics in¬
dustry averaged one-seventh to one-thirtieth of levels going to the ma¬
chinery industry annually. Furthermore, little of this funding went to
areas other than consumer electronics. After 1964, electronics funding
jumped to about one-sixth of machinery industry lending (with the major
share going into electronic components), and after 1971 to one-quarter
(and going mainly to IC producers). After 1978, annual lending in elec¬
tronics was quadruple the levels going to machinery companies. Elec¬
tronics in general, and semiconductors in particular, were clearly being
pushed closer and closer to the front of MITI’s industrial policy queue
in the late 1960s and 1970s.

(MOS) device structure known as the diffusion self-aligned (DSA) circuit, which producers
used extensively in following years. The first Japanese IC memory, a 144-bit static random
access memory (RAM) chip, was developed by ETL and manufactured by NEC in 1968.
An electron beam exposure system was developed in cooperation with JEOL, now the top
supplier of electron beam production equipment to the semiconductor industry. This system
was the very first such machine developed by JEOL; the second and third systems built
were shipped to Sanyo and Toshiba. A step-and-repeat photolithographic system using laser
alignment was built by Nikon for ETL as part of this project. This design was later used in
the very first commercial wafer stepper system built by Nikon, which was to become a top
supplier of such machines to IC makers a decade later. Author’s interview with Yasuo Tarui,
director of the MITI VLSI project, March 18, 1992; Osamu Ishii, “Research and Devel¬
opment”; and Machinery Promotion Association Economic Research Institute, Japan-U.S.
Semiconductor Research Survey (Tokyo, 1980), p. 30.
76. See Flamm, Creating the Computer, pp. 189-90, notes 43-44.
77. See note 34.
78. These statistics capture only lending to industry under the Electronics Industry
Promotion Law and exclude funds borrowed under the auspices of other measures.
 s: ^ §
o
2 §
s: sc in
d ■se.
«0 r^ <u
<N g I'
cl .5 3 ^
to -*2 E
« 2 3 -«
3 2
U CO
o 55
53

u.
3 os
o >s
^ ?
■o 3
= T3
, ^ vd
O v. SC
§ 2 •
go —
a . cn SO
J5 ^ a-
. \6 d m 2 o O'
s c in
(N
•—
w 5—
£ S 2
i^I
' ^ ca
Table 2-2. FILP Loans to Japanese Machinery and Electronics Industries, 1956-84

“ CD “
i g ■§
&
* Q§.£

f. Lending was mainly to auto parts manufacturers (within machinery) and integrated circuit producers (within electronics).
P^.g.
«U E
cs
cn (N , «
: 5 : : (N r-
r- :Q
- 3 .
cS't*
«- C O
y

E a !
2 ^1
M S3 !
2 a: •
CL O

I31 cn o vo o 00 iJ
I c d i-H ^-H d (N 00 :

^ a
a> «-
§-e;
d. A major proportion of this lending went to electronic components manufacturers.

^ <
— c

h On (N
e. Japan Development Bank lending specifically budgeted for electronics.

(N VO E I1
<N <n d d JS S'
e9 vo
■O ^
c. Period saw a rapid increase in loans to auto parts manufacturers.

c ^ >%
CS flj a> TJ
> o
E.S. _ t-
_C 3
<l> jr
o
00
</5
o
CO -E
X LT a. o
c2
H 22
'f
£P « oo (N
D oo •-s s<u
■o 2 3
£ s
*
• d
°°.
ov (N
.c
H "! <u -Sr "E
> o ^

rrs sa
<U
VC
: Cl
« 2 «
‘o
0)
O ^
3 “^ u.- £3
o
Cl
is
sf 1
S® •E c<u <*-o
15 $ X) -,-3 O

3 £ 3
Q. «2
— CO
(#i
U.
o 0-
CO O § £
C. T33
o

-C vo f— ^
*E
vj QJ '— o
d m
in
oo VO in C -o ^ 3 a.
C/5 fc-
*5 m CO CO
i= s
i< S » ^
"c <U O c
3 S o -J5
C
<u £^ 9- <u
^ C
^
<u
>5 cE jj

a -C O -o 1) 1> /) Ci.w 3 ^
O
VO
cn IX)
VO vo
o
r-
O
r-
GO
NO
o r-
r-
^ ^ C
00
■^3 i
l i
i ii i i
i 1 8^-:^ §
3 >,CC f- J
•2 vO i-H vo '-t OV i-H 00
l! iX) IX) vo vo vO s VO r- r- o .VJ O'
co
CQ On
Ov On Ov
’“H
Ov On Ov
t-H
Ov On
i-H
Ov " 2S
CL
d.
CL
^
THE JAPANESE ASCENT IN SEMICONDUCTORS 65

Figure 2-5. Japanese Tax Incentives and Subsidized Lending for

Electronics Development, and Investment in Electronics by Major
Companies, 1961-71

Billions of yen

Sources: Electronic Industry Almanac, 1975 (Tokyo: Dempa Publications, 1974), p. 176; and Machinery Promotion
Association Economic Research Institute, Japan-US Semiconductor Industry Research Survey, pp. 98, 119.

These subsidized funds, furthermore, were significant in relation to

resources the private sector was investing in semiconductors in the 1960s
and 1970s. Although not all these funds were going into semiconductors,
it is useful to compare the total flows with IC facilities investment expen¬
ditures undertaken by major Japanese semiconductor firms. As table
2-2 shows, subsidized electronics loans exceeded a quarter of IC facilities
investments in the late 1960s, and stayed above 10 percent of IC plant
investments in the 1970s, before dropping under 4 percent in the 1980s.
The importance of subsidized loans and tax breaks can be examined
in greater detail for the decade of the 1960s and early 1970s. Figure 2-5
compares the volume of special accelerated depreciation given to the
electronics industry from 1961 through 1971 with major companies’ in¬
vestments in IC plant and equipment in those years. Because the mar¬
ginal corporate tax rate was quite high throughout this period (roughly
66 THE JAPANESE ASCENT IN SEMICONDUCTORS

55 percent),79 the tax benefits would also have been quite large. However,
this special depreciation was spread over all qualifying electronics-related
investments (including the computer industry), not just ICs—unfortu¬
nately, further disaggregation is not possible.
Figure 2-5 also compares JDB loans in electronics with investment in
IC facilities. Again, subsidized loans are relatively large in relation to
investments. Like the special tax breaks, the JDB loans also went to
electronics firms making products other than ICs, but we do know that
the largest share of the funding after 1964 did go to IC producers. Also,
since the JDB was not the only public lender to the industry under the
auspices of the Electronics Industry Promotion Law, other relevant sub¬
sidized loans are not being captured in this figure.
In contrast to depreciation benefits and loan subsidies, MITI R&D
subsidies to electronics under the Electronics Industry Promotion Law
were much less important. Figure 2-6 compares all electronics R&D
subsidies with IC-related R&D spending by major Japanese producers.
If it is recalled that significant R&D subsidies in semiconductors began
only after 1965, it is clear that these monies could not have been partic¬
ularly large in relation to spending levels within the industry. However,
these R&D subsidies capture only MITI monies spent under the Elec¬
tronics Industry Promotion Law—they do not cover large-scale national
R&D projects, internal funding of the ETL, or funding provided under
other statutes. Nonetheless, it is probably reasonable to suggest that for
semiconductors in the 1960s, in the aggregate, tax breaks and subsidized
loans, and not R&D subsidies, were the most significant instrument for
MITI’s promotion of the semiconductor industry.
In fact, R&D subsidies in semiconductors seem to have been focused
on relatively narrow themes, and principally on supporting national pol¬
icy in computers. But while the big ticket semiconductor R&D projects
were in digital ICs and were clearly designed to strengthen the Japanese
computer industry, now at the center of MITI’s electronics strategy, some
resources went into ICs oriented toward other applications. Analog cir¬
cuits for use in communications and consumer electronics were funded
at Fujitsu, Toshiba, Hitachi, Shiba Electric, Nippon Columbia, NEC,
Matsushita, and Toyo Communication Equipment in the latter part of
the 1960s.80 A MITI-funded consortium of five television manufacturers,

79. See Ogura and Yoshino. “Tax System and the Fiscal Investment and Loan Pro¬
gram,” p. 126.
80. Tarui, "ICs in Japan," p. 103.
THE JAPANESE ASCENT IN SEMICONDUCTORS 67

Figure 2-6. Japanese R&D Subsidies for Electronics Development, and

R&D in Electronics by Major Companies, 1961-71

Billions of yen

Sources: See figure 2-5.

seven component firms, four universities, and two Osaka-area research

institutes began a successful joint R&D effort on ICs for use in television
sets in 1966.81 A substantial ($172,000, or 62-million-yen) project at Hay-
akawa Electric (later renamed Sharp) pioneered the miniaturization of
electronic calculators in a design using LSI ICs around this time.82 Mit¬
subishi, the largest supplier of ICs to Japanese calculator manufacturers
at the time, used the MITI funding it received for its work on the Sharp

81. The joint research group included Osaka Onkyo, Hayakawa Electric (later renamed
Sharp), Matsushita Electric, Mitsubishi Electric, Sanyo Electric, Elna-Fox Electronics,
Hoshi Electric, Kobe Industrial, Matsuo Electric, Murata Manufacturing, Nichicon Capac¬
itor, Sodensha, and Osaka University. See Eizi Sugata and Tashihiko Namekawa, “Inte¬
grated Circuits for Television Receivers,” IEEE Spectrum, May 1969, p. 74.
82. “Japan: LSI Calculator,” Electronics, November 11, 1968, p. 307.
68 THE JAPANESE ASCENT IN SEMICONDUCTORS

calculator chips for its LSI research.83 This push into calculators, indeed,
was integral to the next major twist in the saga of Japanese semiconductor
development.

A Secret Truce in the “Patent War”

Just as the Japanese electronics industry was faced with new and po¬
tentially devastating changes in its global competitive environment, Texas
Instruments’ stalled 1964 application to enter the Japanese market flared
up into a hot dispute. Tensions had escalated in mid-1965 when TI’s
Japanese investment applications were published, some five years after
their initial filing.84 All Japanese semiconductor producers opted to op¬
pose the TI application, and the “patent war” began in earnest. But
whereas most Japanese chip makers were apparently willing to make
some sort of deal, NEC—armed with its exclusive Japanese rights to the
Fairchild IC patents—opted for an aggressive legal counterattack in
Japan.
By 1966, however, Japanese firms had begun to eye the United States
as a market for a new generation of consumer electronic products con¬
taining ICs. In the fall of 1966, after beginning shipments of a new radio
containing ICs, Sony was forced to discontinue sales by the threat of
patent infringement suits from TI.85 In that same year, Hayakawa Electric
(Sharp) began to mass-produce calculators.86 The product appeared to
have enormous export potential but would be vulnerable to legal attack

83. Mitsubishi decided, however, not to commercialize these experimental chips, and
Sharp began to talk to foreign suppliers. The linkups with foreign suppliers of LSI
chip sets for calculators are described below. “Japan: LSI Calculator,” p. 308; and NHK,
Electronics-Based Nation, vol. 3, p. 233.
84. Full patent rights were not granted to TI until 1989 (although some limited patent
rights were approved in 1977). Andy Zipser, “Texas Instruments Gets Japanese Patent;
Analysts See Sizable Addition to Revenue,” Wall Street Journal, November 22, 1989,
p. 1A.
85. Mason, American Multinationals in Japan, pp. 183-84.
86. Sharp’s entry into calculators was closely linked to MITI policy. In the early 1960s
Sharp had actually developed a prototype transistorized computer system but, on advice
from MITI, decided not to enter the business. MITI instead had suggested that Sharp make
smaller products using digital logic designs that could be mass-produced. By 1964 Sharp
had developed its first transistorized calculator. Commercialization of desktop calculators
was encouraged with subsidized loans in 1964-65; further investments in production facil¬
ities for calculators using ICs were made eligible for subsidized loans and special deprecia¬
tion benefits from 1966 to 1970. See NHK, Electronics-Based Nation, vol. 3, pp. 129-39;
and Majumdar, “Industrial Policy in Action,” pp. 28-29.
THE JAPANESE ASCENT IN SEMICONDUCTORS 69

in the United States, where TI’s IC patents were recognized. These fears
were realized when TI warned Sharp about patent violations in 1967.
MITI, alarmed by the threat to exports, finally responded to TI in
August 1966 on its 1964 application for a Japanese affiliate. MITI offered
to permit an equal partnership with a Japanese firm, provided that TI
also agreed to license its patents to Japanese companies at reasonable
rates and accept MITI restrictions on its output in the first three years
of production.87 TI turned down the offer. MITI nonetheless reportedly
began to work on persuading NEC to drop its adamant opposition to the
TI patents, and the agency secretly began to search for a possible Japa¬
nese joint venture partner for TI.88
Mitsubishi, supplier of the ICs used in the Sharp calculator (and the
largest producer of calculator ICs in Japan), was extremely concerned,
and the company sent a team to Texas to seek a solution to the Sharp-TI
patent conflict.89 Mitsubishi was particularly interested in a solution in¬
volving some manufacturing tie-up with TI because of its own weakness
in leading edge semiconductor technology compared with other Japanese
semiconductor producers.90 Mitsubishi made some new and interesting
proposals to TI and offered advice on negotiations with MITI, fully
expecting that it would emerge with a new relationship with TI.91
At this point, however, MITI intervened. Rather than simply continue
the exchange with Mitsubishi, TI’s negotiating team in Tokyo was asked
to speak with other Japanese companies. Sony soon came forward with
a proposal for a temporary 50-50 joint venture under TI control, to be
converted into a wholly owned affiliate at a later date, in exchange for
giving Japanese firms the license to use the TI patent portfolio. MITI
quickly agreed to go along with the Sony proposal. Mitsubishi was
shocked to learn it had been unceremoniously dumped; from the TI
perspective, it was all rather mysterious.92

87. NHK, Electronics-Based Nation, vol. 3, p. 229; Mason, American Multinationals

in Japan, p. 182; and Nakagawa, Semiconductor Development in Japan, p. 159.
88. Nakagawa, Semiconductor Development in Japan, pp. 164-66.
89. Nakagawa, Semiconductor Development in Japan, p. 226, writes that Mitsubishi
had 70 percent of the Japanese market for bipolar (the most important pre-MOS LSI
technology) ICs used in calculators in Japan. The visit, at the end of September 1967, was
“unannounced and unexpected and uninvited,” according to one TI executive. Mason,
American Multinationals in Japan, p. 184.
90. Nakagawa, Semiconductor Development in Japan, pp. 161-63.
91. See Nakagawa, Semiconductor Development in Japan, p. 163.
92. TI, in fact, never did figure out why Sony stepped forward and took the initiative.
Mason, American Multinationals in Japan, pp. 184-85; 322-23', note 169.
70 THE JAPANESE ASCENT IN SEMICONDUCTORS

In fact, MITI had orchestrated the Tokyo negotiations behind the

scenes. MITI had approached Akio Morita of Sony and asked him to
start a joint venture with TI for the sake of the future of the Japanese
electronics industry.93 In exchange for permission to enter Japan, TI ne¬
gotiated a 3.5 percent royalty rate for the right to use its patents. This
rate would apply to all Japanese semiconductor makers other than NEC:
under continuing pressure from MITI to settle its dispute with TI, NEC
finally agreed to a considerably reduced payment for use of the TI
patents.94
This negotiated truce to the “patent war” was concluded in late 1967
and announced in early 1968. On the surface, the public settlement ap¬
peared to reflect TI’s complete capitulation to MITI, since it had agreed
to a 50-50 joint venture with Sony, indicated that it would consult with
MITI on its production levels, and licensed its crucial patents to its
Japanese rivals—elements all present in the 1966 MITI offer rejected by
TI. In reality, however, in what may have been the first use of the tactic
in a semiconductor trade negotiation, two “secret side letters” guaran¬
teed TI its principal objective, namely, the right to set up a wholly owned
Japanese affiliate. One of the letters, from the Ministry of Finance, stated
that the interministerial committee it chaired would give “favorable con¬
sideration” to a TI purchase of Sony’s equity share after three years; in
the other letter Sony for its part agreed to sell its stake at that time.95

Battles in the LSI Market

The three-part program adopted by MITI in 1965-66—an outright

ban on imports of advanced ICs, a hefty program of subsidized loans and
tax breaks for capacity expansion, and a significant increase in semi-
conductor-related R&D subsidies—gave Japanese producers a substan-

93. Morita is quoted to this effect in Nakagawa, Semiconductor Development in Japan,

p. 164. Sony was apparently tapped for a variety of reasons: MITI’s lack of confidence in
Mitsubishi’s semiconductor technology capabilities, TI’s history of confidence-building con¬
tacts with Sony, and a top MITI official’s personal relationship with Sony’s Morita (both
came from the city of Nagoya).
94. See Nakagawa, Semiconductor Development in Japan, pp. 165-66; Mason, Ameri¬
can Multinationals in Japan, pp. 185-86; 323, note 170. Actually, since the patents were not
formally granted until much later, these royalties represented “advance payments.”
95. Mason, American Multinationals in Japan, p. 186.
THE JAPANESE ASCENT IN SEMICONDUCTORS 71

tial boost in the mid-1960s.96 By 1969 imports accounted for less than a
quarter of apparent Japanese IC consumption (see figure 2-4). But as it
had before, rapid technological change was to again threaten these gains
in domestic makers’ market share.
Calculator sales had expanded rapidly in the late 1960s, and continued
development of this product led to the next major crisis for the struggling
Japanese semiconductor industry. In the late 1960s the technological focus
was on replacing the relatively large number of chips in current models
with a much small number of more highly integrated LSI chips. In the
United States the production technology for LSI had developed rapidly,
and American semiconductor manufacturers were sweeping world mar¬
kets with low-cost LSI products.
Japanese semiconductor producers initially had a great deal of diffi¬
culty mastering LSI technology; they were also strapped financially by
the cost of investing in the whole new generation of production facilities
required to produce the more advanced chips. An advanced fabrication
technology, MOS (metal oxide semiconductor) technology, was also be¬
coming important in the manufacture of LSI chips, but the production
process for MOS devices was very demanding. Japanese chip makers
lagged well behind American companies in mastering it.97 The costs of
using patents owned by American chip makers put Japanese producers
at a further competitive disadvantage, forced to pay about 10 percent of
sales as royalties to their American competitors.98

96. My account in this section for the most part closely follows those of Nakagawa,
Semiconductor Development in Japan, pp. 190-205, and NHK, Electronics-Based Nation,
vol. 3, pp. 264-316, which are only available in Japanese.
A radically different Japanese view of the strength and behavior of Japanese semicon¬
ductor companies at the time of the “calculator wars” of the late 1960s and early 1970s has
also appeared in the English-language literature. Note, in particular, the view expressed
by Takuo Sugano (University of Tokyo) and Michiyuki Uenohara (NEC): “the Japanese
coauthors assert that Japanese companies were ready sooner than the Americans to under¬
take large-scale commercial production of MOS ICs. NEC foresaw a potential mass market
for the desk-top calculator, and, in cooperation with Hayakawa (the predecessor of Sharp),
developed calculators using MOS ICs. They completed a commercial model in 1966, and
the success of this venture helped to establish the practicality of the MOS IC.” Michiyuki
Uenohara, Takuo Sugano, John G. Linvill, and Franklin B. Weinstein, “Background,” in
Daniel I. Okimoto and others, eds., Competitive Edge: The Semiconductor Industry in the
U.S. and Japan (Stanford University Press, 1984), p. 15.
97. See, for example, NHK, Electronics-Based Nation, vol. 3, p. 311; U.S. Department
of Commerce, Global Market Survey: Electronic Components (1974), pp. 5-6, 81-82;
Franco Malerba, The Semiconductor Business: The Economics of Rapid Growth and De¬
cline (Frances Pinter, 1985), pp. 100-02, 151-52.
98. “IC Makers Shaken at the Dawn of IC Liberalization: Request for 7 Billion Yen
72 THE JAPANESE ASCENT IN SEMICONDUCTORS

Japanese calculator makers, who appreciated the advantages of highly

integrated MOS LSI chips for their products, had embarked on ambitious
projects to redesign their calculators using the new technology. Although
the domestic chip industry was receiving subsidies from MITI to develop
the technology, progress was slow, and local manufacturers were unable
to supply the chips needed by local calculator manufacturers. Calculator
producers, led by Sharp, then turned to American IC suppliers, who—
after initial reluctance—soon developed a thriving business providing
MOS LSI calculator chips to the Japanese calculator industry." In 1969
Sharp shipped the first LSI calculator, and a new cycle of rapid technical
innovation hit the marketplace.100
Booming Japanese calculator production provided American compa¬
nies with a rapidly growing market where—faced with an absence of
domestic competition in MOS LSI—they were finally permitted to sell
their products in volume. North American Rockwell (later renamed
Rockwell International) had forged a supply relationship with Sharp,
Canon struck a deal with Texas Instruments, and Ricoh turned to AMI.101
In the early 1970s the rapidly growing consumer electronics market
exploded, and foreign-made semiconductor sales in Japan soared. Fig¬
ure 2-4 documents this surge in imports. By 1971 over 35 percent of

Subsidy and the Rush to Establish Their Own Technology for Mass Production,“ Nihon
Keizai Shimbun, April 18, 1973.
99. A famous story often told in Japan relates how the chief executive of Sharp visited
the United States in 1968 to work out an arrangement for an American chip company to
supply MOS LSI chips needed in its new calculator designs. He visited eleven manufactur¬
ers, including Fairchild, TI, Motorola, AMI, National Semiconductor, RCA, Philco, and
Sylvania. All rejected his proposal, because the volumes he required were too high for
their current capacity, which was largely tied up with defense production. Finally, just as
he was leaving the United States, executives at North American Rockwell had him paged
at the Los Angeles airport, to tell him they had reconsidered their initial decision to reject
his request and would work with him. The details of this story may be found in Nakagawa,
Semiconductor Development in Japan, p. 192; NHK, Electronics-Based Nation, vol. 3,
p. 264. The Rockwell-Sharp collaboration was to continue well into the early 1970s. See
“How to Cut a Pocket Calculator in Half,” Electronics, February 1, 1971, p. 104. Other
Japanese calculator makers quickly followed Sharp in striking deals with American chip
suppliers. Nakagawa, Semiconductor Development in Japan, p. 192.
100. Machinery Promotion Association Economic Research Institute, Japan-U.S. Sem¬
iconductor Industry Research Survey, p. 30.
101. See Nakagawa, Semiconductor Development in Japan, p. 195; NHK, Electronics-
Based Nation, vol. 3, pp. 264-316.
THE JAPANESE ASCENT IN SEMICONDUCTORS 73

Japanese IC consumption was imported. American chip makers then held

three-quarters of the mushrooming Japanese MOS LSI market.102

First Steps toward Liberalization and “Dumping”

Calculator development continued to proceed apace, taking advantage
of rapid improvements in chip manufacturing technology. By 1970 one-
chip calculators had been introduced. One fruitful outcome of the col¬
laboration between Japanese calculator makers and American chip sup¬
pliers was the development of the first single-chip microcomputer, in
1971.103 But even as competitive pressure on Japanese firms intensified at
the high end of the market, in calculator LSI, a new attack emerged at
the low end.
Intense foreign pressure on Japan to liberalize its trade regime had
mounted in the late 1960s and early 1970s. In September 1970 Japan
removed older, less advanced ICs (chips with fewer than 100 circuit ele¬
ments) from the “negative” list of products requiring prior approval for
importation. Despite hefty tariffs (12 percent), imports came pouring in
at prices more than 20 percent below previous levels in the Japanese
market.104 The competition was intensified by the severe recession the
semiconductor industry experienced in 1970-71.
The “low price offensive”—American chip makers hawking these
older logic chips at bargain prices—subjected their Japanese competitors
to withering fire, at both the top (in calculator LSI chip sets) and the
bottom (standard logic) ends of the market simultaneously.105 This led to
what were perhaps the first charges of “dumping” in U.S.-Japan IC trade.

102. Arthur Erikson and Charles Cohen, “Japanese Electronics Firms Search Out New
Markets to Pierce Economic Fog,” Electronics, November 22, 1971, p. 132.
103. See NHK, Electronics-Based Nation, vol. 3, pp. 318-21; and Robert N. Noyce and
Marcian E. Hoff Jr., “A History of Microprocessor Development at Intel,” IEEE Micro,
vol. 1 (February 1981), pp. 9-13.
104. “Cheap IC Selling by U.S. Firms Posing Problem; Dumping Mooted,” Japan
Economic Journal, September 29, 1970, as excerpted in Semiconductor Industry Associa¬
tion (SIA), Japanese Protection and Promotion of the Semiconductor Industry: Japanese
Laws, Government and Industry Documents, and Press Reports Relating to Japan’s Pro¬
motion of Its Semiconductor Industry, 1967-85 (Washington: Dewey, Ballantine, Bushby,
Palmer and Wood, 1985).
105. Similar devastation in the market for standard logic occurred in Europe, where
this was the period of the so-called logic wars. A price war broke out among American
chip makers, inflicting a severe blow to European producers struggling to stay abreast of
their American competition. See Malerba, The Semiconductor Business, pp. 110-19.
74 THE JAPANESE ASCENT IN SEMICONDUCTORS

In the latter part of 1970 MITI and the Ministry of Finance launched
informal investigations of possible American dumping of ICs in Japan.
The intent was apparently to put pressure on American producers to
mute unwanted price competition with Japanese manufacturers; officials
suggested to reporters that a “dumping charge may rise to the surface if
the cheap sale of ICs continues further.”106 The tough talk apparently had
little effect on the hardheaded Americans; within months a continuing
“low price offensive” by the American producers had forced Japanese
producers to cut back production of these chips by 20 percent.107 By early
1972 MITI was even floating the idea of a rationalization cartel covering
logic ICs with fewer than 100 elements, but this apparently went no¬
where.108
Meanwhile, even at the protected high end of the market—for highly
integrated calculator LSI chips—product-starved Japanese calculator
manufacturers braved MITI’s wrath and red tape and fought for the right
to import needed inputs. The issue, quite simply, was survival. U.S. IC
manufacturers had crammed all the electronics for a calculator onto a
single chip, and these new products had revolutionized the economics of
the calculator business. Advances in ICs and reductions in the number
of components used, not the wages of workers assembling hundreds of
discrete components, became the key to competitive success. New Amer¬
ican players, including some U.S. chip makers, even reentered the elec¬
tronic calculator business, basing their designs on the revolutionary new
chips now hitting the market, and substantially reduced the share of the
American market captured by calculator imports.109
Chip prices dropped precipitously in the U.S. market, under the pres¬
sure of aggressive competition. The fierce competition was also exported
across the Pacific. From spring 1970 to early 1971, the prices of imported
calculator LSI chips dropped by half.110

106. “Cheap IC Selling.”

107. “Tension Mounts from Planned U.S. Entry into IC Industry,” Japan Economic
Journal, April 27, 1971 (excerpted in SIA, Japanese Protection).
108. Nihon Kogyo Shimbun, January 8, 1972 (excerpted in SIA, Japanese Protection).
109. U.S. Department of Commerce, The Impact of Electronics, pp. 13-18.
110. A contemporary account noted that there was little Japanese chip producers could
do about it, because many of the Japanese IC producers themselves purchased LSI chips
and other specialized products in their capacity as manufacturers of computer and com¬
munications equipment, and because Japanese LSI chips were distinctly inferior to Amer¬
ican ones. “Thus, stoppage of cheap U.S. imports by any means may invite opposition
from Japanese calculator firms.” “Prices of LSI Circuits Reportedly Reduced to Half
THE JAPANESE ASCENT IN SEMICONDUCTORS 75

The “Calculator War”

Under sustained assault by American competitors, Japanese IC mak¬

ers counterattacked. In the late 1960s and early 1970s the major produc¬
ers all built new MOS LSI production lines.* * 111 Still, the new technology
was difficult to master—costs were high and yields low. Japanese calcu¬
lator makers were inclined to maintain their new relationships with
American suppliers. Alarmed, the semiconductor producers branded the
calculator manufacturers as unpatriotic; Sharp’s chief executive came
under particularly strong attack as “a traitor who would waste Japan’s
precious foreign currency.”112 Sharp was further criticized for taking MITI
R&D money while buying foreign chips.113
MITI at that point made an example of Sharp by refusing to issue it
necessary import licenses. Other calculator and office equipment makers
joined Sharp in protesting, arguing that “we are buying parts from the
U.S. because they cannot be made at home; we cannot do anything if
we are not allowed to import.”114 For Sharp the lesson was clear: a
necessary minimum amount had to be purchased domestically.
Even this was insufficient to mollify angry domestic semiconductor
manufacturers, who faced a continuing crisis. Japanese chip makers con¬
tinued to lag at least one or two years behind the state of the art in a
very demanding new technology, and even as losses mounted, they were
pressed for the financial resources to build new generations of production
facilities.115 The ravages of increased competition from imports were
reflected in a rising import share (see figure 2-4) and a decline in the

by U.S. Makers,” Japan Economic Journal, January 5, 1971 (excerpted in SIA, Japanese
Protection).
111. NHK, Electronics-Based Nation, p. 311; U.S. Department of Commerce, Global
Market Survey, Electronic Components, p. 82. Large investments in upgrading plants to
produce CMOS (complementary metal oxide silicon) ICs were undertaken by most Japanese
producers in 1972 and 1973. “IC Makers Shaken.”
112. NHK, Electronics-Based Nation, vol. 3, p. 311 (author’s trans.).
113. On the other hand, Mitsubishi, which was working with Sharp to manufacture ICs
for a MITI-funded calculator project, actually declined to move to commercial production.
See note 83.
114. Nakagawa, Semiconductor Development in Japan, pp. 196-97 (author’s trans.).
115. For example, note the evaluations of the semiconductor makers’ problems found
in Nakagawa, Semiconductor Development in Japan, pp. 198-200, and U.S. Department
of Commerce, Global Market Survey, Electronic Components, pp. 81-82. The Nihon Keizai
Shimbun commented in 1973 that “it is estimated to take at least 2 to 3 years” for Japanese
companies to come up with the technology needed to compete with U.S. IC producers.
“IC Makers Shaken.”
76 THE JAPANESE ASCENT IN SEMICONDUCTORS

Table 2-3. Sales by Japanese Semiconductor Companies to Affiliated

and Nonaffdiated Companies, 1966-71
Millions of yen unless otherwise specified

1971
Sales 1966 1967 1968 1969 1970 (estimate)

Inside
company 230 984 4,860 14,576 24,780 28,223
Outside
company 380 2,120 5,847 11,008 16,172 20,951
Total 610 3,104 10,707 25,584 40,952 49,174

Inside
sales (percent) 37.7 31.7 45.4 57.0 60.5 57.4
Outside
sales (percent) 62.3 68.3 54.6 43.0 39.5 42.6
Source: Machinery Promotion Association Economic Research Institution, Japan-U.S. Semiconductor Industry
Research Survey, p. 124.

share of Japanese companies’ chip output sold externally (that is, to

customers other than in-house equipment-producing divisions). Table
2-3 shows that the external share of Japanese chip producers’ sales
dropped from 68 percent in 1967 to about 40 percent in 1970.
MITI ultimately came down firmly on the side of the Japanese chip
producers (who were parts of the same companies producing computers,
the centerpiece of MITI electronics policy) and severely chopped back
licenses for imports of American calculator chip sets. In late 1971, even
as Japanese calculator production headed toward a new annual peak of
1.3 million units, the head of the MITI electronics and electrical machin¬
ery division made clear the ministry’s determination to choke off imports:
“If we’d approved all the import applications, there would have been
enough kits this year to make 4 million calculators.”116
A typical experience was that of Fairchild, which in 1972 was about to
sign a $500 million contract with a Japanese company, only to have MITI
pull the plug on the deal.117 Even successful import contracts faced a
daunting gauntlet of bureaucratic obstacles. Top executives at Busicom,
the Japanese manufacturer that commissioned Intel Corporation to de¬
sign and build the first single-chip microprocessor for use in a calculator
design, had to make dozens of daily trips to MITI, the Bank of Japan,
and the Ministry of Finance in order to secure permission to pay Intel
for its development cost, and had to plead with administrators to let the

116. Erikson and Cohen, “Japanese Electronics Firms,” pp. 125, 132.
117. NHK, Electronics-Based, Nation, vol. 3, p. 18.
THE JAPANESE ASCENT IN SEMICONDUCTORS 77

company import a product deemed vital to its future. When the first
prototype of the Intel design arrived in Tokyo, it took over a month for
it to clear customs at Haneda airport.118
Less typical was the experience of Texas Instruments Japan, which in
1971, consummating its secret 1968 bargain with MITI and Sony, was
transformed into the only wholly owned affiliate of an American semi¬
conductor maker inside the Japanese market. In that same year TI Japan
began shipping a one-chip calculator IC that sold like hotcakes—at its
peak, 300,000 units per month.119 Although ostensibly perched comfort¬
ably behind the protective walls thrown up around the Japanese market,
TI continued to fight a silent guerrilla war against the bureaucracy. In
1972, for example, a top Japanese executive quit Mitsubishi and was
recruited by TI. Shortly before he was to be hired, he was approached
by MITI and warned that, if he joined TI, MITI “would act accord¬
ingly.”120 Heeding these warnings, he and others in his group joined
another Japanese company.

Further Liberalization and Crisis

The MITI clampdown on calculator LSI imports was successful in
providing domestic IC makers some degree of shelter from the competi¬
tive storm. The year 1972 marked a strong recovery in the semiconductor
business, but by any measure it produced absolutely incredible results in
the Japanese semiconductor market. Figure 2-7 presents the internal
market share estimates of a Japanese semiconductor company and shows
that the Japanese import market share in ICs plunged by almost
40 percent from the previous year, to 22 percent of total consumption.
(See also figure 2-4, which portrays identical changes.) There are con¬
flicting explanations for why this occurred. One obvious one is that, as
documented above, MITI had openly launched a highly successful ad¬
ministrative offensive against imported ICs by late 1971, applying pres¬
sure to both exporters and their potential Japanese customers.
Another explanation, offered by an NEC semiconductor executive in
a book that was widely read in Japan, is that the U.S. suppliers simply
failed their Japanese customers. According to this account, defective

118. See NHK, Electronics-Based Nation, vol. 4, pp. 131-54.

119. NHK, Electronics-Based Nation, vol. 3, p. 325.
120. Nakagawa, Semiconductor Development in Japan, p. 229.
78 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-7. Japanese Consumption of Integrated Circuits and Import

Market Share, 1968-74

Billions of yen

Source: Contemporary internal estimate by a Japanese semiconductor manufacturer.

calculators returned to Japanese manufacturers around 1971 were found

to have defective U.S. chips. Because U.S. IC makers refused to take
their defective product back on the grounds that their contract did not
require them to do so, the NEC executive alleged, Japanese customers
learned that “it was much safer to do business with Japanese makers,
who have better warranty and service.”121

121. Author’s trans. The NEC executive went on to blame the quality problem on
cheap offshore assembly of these ICs in Southeast Asia: too much emphasis on cost, too
little attention to quality control. Interestingly, no Japanese semiconductor user is quoted
as supporting this version of events. Indeed, one user interviewed by Nakagawa describes
the lesson taught over this period as being that a certain minimum amount had to be
purchased from domestic chip makers, but notes that even this left Japanese semiconductor
producers unhappy. Nakagawa, Semiconductor Development in Japan, pp. 197, 203-04.
But NEC had its own quality problems with LSI, according to other accounts in this
same volume. After copying Rockwell chips as a last resort, NEC finally managed to
develop its own ten-chip LSI calculator kit, but was unable to sell it. When a Japanese
calculator producer finally tried to use it, the NEC LSI chips were found to be defective.
Ibid., pp. 199-200.
THE JAPANESE ASCENT IN SEMICONDUCTORS 79

Yet another explanation was offered by a former NEC executive in a

1991 interview with the author.122 According to this account, U.S. pro¬
ducers simply failed to deliver the volumes Japanese customers de¬
manded, and created a perception that American customers were being
given first priority in a tight market. Product-starved Japanese users
then “begged” Japanese IC makers to start producing needed chips, and
ever since then Japanese calculator makers have relied on domestic
producers.123
Admirers of the famous Japanese film Rashomon might be tempted to
suggest, as did director Akira Kurosawa in that work, that there might
be an aspect of truth in all three accounts. The only relatively certain
facts are the mute percentages depicted in figure 2-7, and those numbers
show foreign IC makers regaining a lofty 36 percent share of the Japanese
market by 1974; this suggests that U.S. quality or delivery problems could
not have been so severe as to permanently depress American companies’
sales in Japan. Although it is not logically impossible for U.S. quality
and delivery problems to have coincided with a MITI crackdown on
calculator LSI chips in late 1971 and 1972, MITI officials, at least, are on
contemporary record as opining that imports would have soared absent
import restrictions.
But, then, why was import market share allowed to resume its upward
course in 1973 and 1974? One important factor may have been that, early
in 1973, the U.S. government substantially increased political pressure
on Japan to continue liberalizing IC imports. In February of that year
the U.S. Special Representative for Trade threatened to bring a case for
adjudication by the General Agreement on Tariffs and Trade (GATT),
in which it seemed inevitable that Japan would lose.124 In March 1973 the

122. Author’s interview with former NEC executive, March 1991.

123. But a third NEC executive offered a rather different perspective on imported
computer ICs in 1974. Speaking at a panel discussion, NEC’s executive director remarked
that “in the past, Japanese minicomputer makers have relied on American ICs, and MITI
gave administrative guidance, putting these things on the negative list ... if ICs for
microcomputers were not placed on the negative list, capital investment would not have
been possible by Japanese makers. This meant that since MITI put up the negative list and
gave administrative guidance, it was possible for us for the first time to stand on our own
feet.” “Plunging Into IC Liberalization,” Nikkan Kogyo, December 12, 1974 (excerpted in
SIA, Japanese Protection and Promotion).
124. “U.S. Presses Japan with Appeal to GATT for Liberalization of Electronic Com¬
puters and IC; Heading toward Majority Vote; Notification Given at Time of Special
Representative Eberle’s Visit to Japan,” Sankei Shimbun, March 1,1973 (excerpted in SIA,
Japanese Protection). The United States had threatened to file a.complaint with the GATT
80 THE JAPANESE ASCENT IN SEMICONDUCTORS

U.S. government made IC imports a major issue at a meeting of the

Organization for Economic Cooperation and Development (OECD) in
Paris.125
In short, Japanese policies in computers and semiconductors were
being subjected to increasingly intense scrutiny from abroad. The outside
world continued to put pressure on the Japanese government to further
liberalize its economy, and throughout the early 1970s it debated how to
cope with these foreign demands.126 In computers and in its sister sector,
integrated circuits, the debate was particularly acrimonious. IBM had
announced a new generation of computer systems using advanced LSI
components (the System 370 series) in 1970, and once again the Japanese
computer industry perceived itself under fierce assault. The continuing
pressure to liberalize the domestic market, coupled with accelerating
technological competition, combined to create an atmosphere of acute
crisis for the Japanese electronics industry.
MITI had taken the initiative in computers in 1971. Warning that
foreign machines would sweep through the Japanese market, wipe out
domestic computer makers, and leave “knowledge-intensive industries
. . . seized by foreign capital,” and predicting that the “center of Japan’s
nerve system will be controlled by foreign capital,” MITI successfully
argued for delay in full liberalization of foreign investment in computers
until 1976, and postponement of removal of restrictions on computer
imports until 1975.127 In exchange for this protection, at MITI’s insistence

if the liberalization were delayed, and U.S. officials publicly complained before the OECD.
“Liberalization of Electronic Computers and ICs Generally Decided; MITI to Hasten
Implementation Measures,” Asahi Shimbun, March 1, 1973; “Industrial Circles’ Plan Out
of Question—Liberalization of Electronic Computers, Eberle Expresses Strong Dissatis¬
faction,” Sankei Shimbun, March 27,1973; and “Import Liberalization during 1975; Cabinet
Decision on Electronic Computers Expected Today,” Nihon Keizai Shimbun, June 15, 1973
(all excerpted in SIA, Japanese Protection).
125. “Industrial Circles’ Plan Out of Question—Liberalization of Electronic Com¬
puter; Eberle Expresses Strong Dissatisfaction,” Sankei Shimbun, March 27, 1973 (ex¬
cerpted in SIA, Japanese Protection).
126. Japan had accepted Article XI of the GATT in 1963 and Article VIII of the Articles
of Agreement of the International Monetary Fund (IMF) in 1964, and had joined the
OECD in 1964. A three-year phased program calling for removal of some quantitative
restrictions and partial capital liberalization had been introduced in 1968, and a further
phased liberalization of trade and foreign investment was announced in 1971. See Krause
and Sekiguchi, “Japan and the World Economy,” pp. 425-30; and Itoh and Kiyono, “For¬
eign Trade and Direct Investment,” pp. 164-66.
127. “Contents of MITI Views on Electronic Computer Industry,” Asahi Shimbun, July
13, 1971 (excerpted in SIA, Japanese Protection).
THE JAPANESE ASCENT IN SEMICONDUCTORS 81

the six major Japanese computer manufacturers were combined into

three groups of two companies each, with each group to develop com¬
plementary models of advanced computer systems. This cartelization of
computer production was to be supported by a generous MITI subsidy
to cooperative research efforts in each of the three groups. The R&D
subsidy program, dubbed the “3.5 generation” program, targeted the
development of commercial products over the 1972-76 period.128
The three groupings of firms showed varying degrees of willingness to
work together. Fujitsu and Hitachi, organized into the “Ultra Advanced
Computer Development Technology Association,” received a 27-billion-
yen government subsidy over the fiscal 1972-76 period and formed a joint
venture to produce IBM-compatible computer peripherals and terminals
(Nippon Peripherals). That venture remained active until 1986.129 Semi¬
conductor R&D was one of the activities funded by this subsidy, but
actual research topics were assigned by the association, then carried out
(except in their joint venture in peripherals) within each firm separately.
The avenues and extent of technical information sharing between com¬
panies are little discussed but could not have been great—after a joint
product announcement in 1974, the two companies were soon locked in
head-to-head competition, with each announcing models that targeted
markets ostensibly assigned to the other.
The Mitsubishi-Oki tie-up was a substantially smaller-scale affair. The
two companies, formed into the “Ultra Advanced Computer Technology
Research Association ” received 6.4 billion yen in subsidies over fiscal
1972-76.130 Both companies had dropped out of the mainframe computer
business by the mid-1970s.
Perhaps the tightest of the three groupings was that formed by NEC
and Toshiba. Their New Computer Series Technology Research Associ¬
ation was given a government subsidy of 21.6 billion yen from 1972 to
fiscal 1976.131 NEC and Toshiba actually formed a joint venture company,

128. The name alluded to the view that IBM’s System 370 was only half a generation
more advanced than its so-called third-generation System 360. For further details, see
Flamm, Creating the Computer, pp. 193-95; Targeting the Computer, pp. 135-36.
129. The Fujitsu-FIitachi research association spent a total of 88 billion yen over the
1972-76 period, of which 61 billion yen was contributed by the companies. In addition,
another 67 billion yen (entirely contributed by the companies) was spent over the 1975-81
period. See Thirty-Year History, p. 77; Flamm, Creating the Computer, pp. 194-95.
130. The total budget for the research association was 22.2 billion yen over fiscal 1972-
76, plus 3.4 billion yen over an additional period extending through fiscal 1981. A govern¬
ment subsidy was only granted during the first period. Thirty-Year History, p. 79.
131. The research association’s total budget was 60.5 billion yen from 1972 through
82 THE JAPANESE ASCENT IN SEMICONDUCTORS

NEC Toshiba Information Systems (NTIS, which became a member of

their research association), to develop and market a line of computers.
R&D into ICs was one of the areas in which the research association was
active. Although the R&D was assigned to research groups established
within individual member firms, at least some research topics were in
part jointly investigated.132
This aid to computer firms was designed as short- and medium-term
assistance to help them cope with liberalization, and as a stimulus to
greater interfirm cooperation.133 At the same time, a new round of
resources was injected into longer term R&D, under the direction of
MITI’s Electrotechnical Laboratory. Another large-scale national R&D
project in computers, the “Pattern Information Processing System”
(PIPS), was launched in 1971. R&D on MOS LSI design and manufac¬
turing technology, including the design and manufacture of a sixteen-bit
single-chip microprocessor (demonstrated in 1977),134 IK fast and 4K
slow static RAM chips, and a 16K DRAM, were carried out as part of
the first portion (1971-75) of the PIPS program.135 About 3.5 billion yen
was expended in the area of “devices and materials” by the ETL and its
private contractors as part of the program over 1971-75.136 Almost 90

fiscal year 1976; 19.5 billion yen was spent over an “additional development” period from
1977 through fiscal 1981. Thirty-Year History, p. 78.
132. Thirty-Year History, p. 78.
133. A call by MITI Minister Yasuhiro Nakasone for “promotion of further tieup than
the present three groups” in early 1973, for example, was explained to reporters by MITI
officials to mean that these firms “cannot stand competition with the huge manufacturers
of the United States, unless they deepen mutual tieup in commonization of the software
and in taking shares in production and joint development of peripheral apparatus, and
unless they either unified into a home-production manufacturer like Britain’s ICL in the
future or are concentrated at least into a group for specific purposes (Oki-Mitsubishi) and
an alignment of electronic computers proper (remaining four companies). Explaining the
Minister’s statement of the 7th to the press, a MITI leader said, ‘For the present, this does
not necessarily mean unification of industry circles, but the emphasis of the administrative
guidance lies in promoting tieup among the three groups.’ ” “Cooperation in Concentration
Sought—Statement by MITI Minister,” Asahi Shimbun, March 8, 1973 (excerpted in SIA,
Japanese Protection).
134. NEC was the first Japanese company to announce a commercial sixteen-bit micro¬
processor, in 1978. Machinery Promotion Association Economic Research Institute,
“Japan-U.S. Semiconductor Industry Research Survey,” p. 30.
135. See Electrotechnical Laboratory, Pattern Information Processing System: National
Research and Development Program, 1978 (Tokyo, 1978), pp. 27-30; and Ishii, “Research
and Development on Information Processing Technology.”
136. These are my estimates derived from graphs found in ETL, Pattern Information
Processing System, p. 5. This compares with total spending of 17.4 billion yen over the years
1971-78.
THE JAPANESE ASCENT IN SEMICONDUCTORS 83

percent of the semiconductor funding went into contracts with Japanese

companies; NEC, Hitachi, and Toshiba were the contractors on the LSI
chips.137 NEC was later to become the first (in March 1977) and largest-
volume Japanese producer of the 16K DRAM; in 1980 Toshiba was to
market a commercial version of the sixteen-bit microprocessor it had
developed under contract for the PIPS project.138
Integrated circuits also benefited from policy changes set in motion
in 1971, by being exempted from immediate liberalization, although
the delay was less: 1974 for both liberalization of foreign investment
(100 percent owned subsidiaries automatically approved) and removal
of residual licensing restrictions on imports of ICs with more than
200 elements.139 Unlike in computers, however, no major plan to subsi¬
dize the technological development of ICs apart from activities within the
3.5 generation and PIPS programs was announced, reaffirming that pol¬
icy in semiconductors was viewed by MITI primarily as a component of
computer policy.
As the semiconductor manfacturers’ crisis deepened in 1971, however,
MITI began to consider more direct measures designed specifically to
help the industry cope with the effects of accelerating technological com¬
petition and the continuing effects of the partial liberalization of 1970.
In the fall of 1971 a plan to support the establishment of a joint IC design
venture among major domestic IC manufacturers was floated and then
abandoned.140 In late 1971 an industry group considered the formation
of a cartel among Japanese IC makers, to restrict production and allocate
products among potential producers. For a period in early 1972 MITI

137. ETL, Pattern Information Processing System, pp. 3, 5. The research themes (and
contractors) in devices and materials were semiconductor lasers (NEC, Toshiba), reversible
photosensitive materials (Sanyo, Fujitsu, Konishiroku), spatial modulation devices (Hita¬
chi, Matsushita, Hoya Glass), magnetic bubble domain devices (Hitachi, Hitachi Metal,
NEC, Tohoku Metal), and large-scale integrated circuits (Toshiba, Hitachi, NEC).
138. The chronology given in Machinery Promotion Association Economic Research
Institute, “Japan-U.S. Semiconductor Industry Research Survey,” p. 31, shows NEC as the
first company to sell a 16K DRAM, in March 1977. On Toshiba’s commercialization of its
PIPS microprocessor technology, see Hajime Iizuka, ’’Design and Implementation of a
Microprocessing Unit with a Flexible Architecture,” in T. Kitagawa, ed., Computer Science
and Technologies, 1982 (Tokyo and Amsterdam: Ohm and North-Holland, 1982), pp.
22-38.
139. Import quotas on ICs with 200 elements or fewer were removed in 1973. U.S.
Department of Commerce, Global Market Survey: Export Opportunities for Electronics
Industry Production and Test Equipment (GPO, 1974), p. 69.
140. “MITI Envisages Establishment of Joint Firm of Makers for Blueprinting ICs,”
Japan Economic Journal, August 3, 1971 (excerpted in SI A, Japanese Protection).
84 THE JAPANESE ASCENT IN SEMICONDUCTORS

apparently supported the cartel concept for more mature products.141 By

the spring of 1972, however, the idea was abandoned after encountering
bitter dissension within the industry over the fine points of the proposed
cartel’s organization.142
Continuing attempts to delay the impending liberalization were deci¬
sively defeated by mid-1973; restrictions on imports of ICs with fewer
than 200 elements were also removed in that year. IC imports subse¬
quently jumped sharply, almost back to their 1971 peak of more than a
third of Japanese consumption (see figures 2-4 and 2-7). MITI Minister
Yasuhiro Nakasone—the same Liberal Democratic Party (LDP) politi¬
cian who years later was to be chosen prime minister—floated the idea
of further consolidation of the three sets of computer groupings into a
single national computer firm, along lines followed in Europe (MITI
officials actually mentioned Britain’s ICL as a model).143 As before, how¬
ever, the large electronics companies resisted the idea of a forced merger
of their computer divisions into a single national champion.144
The political forces calling for consolidation of activities among com¬
panies in response to imminent liberalization widened their focus, to ICs,
shortly thereafter. Industry had called for a delay in liberalization and a
new 7-billion-yen subsidy for semiconductor technology development.145
Instead, a new MITI proposal for IC manufacturers to reorganize pro¬
duction, with each specializing in particular types of chips, was revealed
in mid-March 1973 (just before the next fiscal year was to begin).146 In
June, after the Japanese cabinet refused to further delay liberalization,

141. Nihon Kogyo Shimbun, January 8, 1972 (excerpted in SIA, Japanese Protection).
142. A news report on the proposal, considered within Japan’s Electronics Industry
Association, specifically notes that “trying to find out which maker is superior in technology
stands to be difficult owing to the swift pace of technology in this field. As such, [industry
quarters] felt that reaching a conclusion on such aspects was going to take some time.”
“Electronics Industry Plans Cartel for Types of Products Using IC,” Japan Economic
Journal, December 14, 1971 (excerpted in SIA, Japanese Protection). By March 1972 the
idea of a special Kidenho cartel had been abandoned, with firms disagreeing over how to
determine which producer would manufacture what types of ICs, and the four top compa¬
nies resisting the cartelization of more advanced chips. “Cartel Controlling IC Production
Types Will be Difficult to Implement for FY1972,” Nikkan Kogyo Shimbun, March 23, 1972.
143. ‘“Cooperation in Concentration Sought,’ Strengthening of Tieup among Three
Groups—MITI Policy,” Mainichi Shimbun, March 8, 1973 (excerpted in SIA, Japanese
Protection).
144. “Will Not Carry Out Another Reorganization—Electronic Computer Industry
Circles,“ Nihon Keizai Shimbun, March 8, 1973 (excerpted in SIA, Japanese Protection).
145. “IC Makers Shaken.”
146. “MITI Heading toward Reorganization of IC Enterprises; Adjustment of Produc-
THE JAPANESE ASCENT IN SEMICONDUCTORS 85

“shocked” IC producers were informed of MITI’s decision to begin re¬

organizing their ranks.147 The companies instead proposed increased sub¬
sidies and the enactment of an antidumping law to deal with anticipated
imports of low-priced U.S. chips.148
What finally emerged at the end of 1973 out of the blizzard of propo¬
sals and counterproposals from the LDP politicians, MITI, and the elec¬
tronics industry was a compromise—some degree of interfirm coopera¬
tion in R&D in ICs, in exchange for sharply increased subsidies. Even
earlier, NEC and Toshiba, which had been teamed by MITI in the New
Computer Series Technology Research Association, had yielded to the
pressure and agreed to develop computer memories jointly as part of
their common computer effort; bitter rivals Hitachi and Fujitsu, joined
by a shotgun marriage in another development group, had publicly com¬
mitted themselves to joint development of ICs to be used in their com¬
puter products (although they apparently did not follow through).149 As
its contribution to the 1973 compromise, MITI unveiled a two-year, 35-
billion-yen ($129 million) package of development subsidies for the IC
industry, organized into five areas. Three themes involved specialized
manufacturing technologies for LSI ICs used in computers: NMOS (neg¬
ative metal oxide semiconductor), CMOS (complementary MOS), and
silicon gate MOS LSI. The remaining two projects were to develop linear
ICs for use in industrial applications. Two of these projects were to be
organized as joint development efforts among several companies.150

tion Fields Planned for Six Exporter Firms in Preparation of Liberalization,” Nihon Keizai
Shimbun, March 15, 1973 (excerpted in SIA, Japanese Protection).
147. “IC Industry Circles Shocked; Caught Between U.S. Offensive and Liberaliza¬
tion,” Nihon Keizai Shimbun, June 15, 1973; and “IC Industry Reorganization Adjusting
Production Areas,” Nihon Keizai Shimbun, June 15, 1973, p. 6 (both in SIA, Japanese
Protection).
148. “IC Industry Circles Desire Establishment of Legislative Measures by Govern¬
ment—Prevention of Selling at Low Price,” Nihon Keizai Shimbun, June 16, 1973 (ex¬
cerpted in SIA, Japanese Protection).
149. “Three Electronic Computer Groups Strengthening Tie-Up, Planning on Joint
Development of ICs and Mutual Supply of Equipment,” Nihon Kogyo Shimbun, June 25,
1973 (excerpted in SIA, Japanese Protection). NEC and Toshiba apparently did do joint
R&D, whereas Hitachi and Fujitsu apparently did not. An official history of Japanese
technology research associations notes that in the NEC-Toshiba grouping, an R&D section
was established “in each member firm, which independently and jointly (partially) pursued
its assigned topic.” In the Ultra Advanced Computer Development Technology Research
Association (the Fujitsu-Hitachi grouping), in contrast, a research system was established
“within each company and engaged in a topic assigned to each.” Thirty-Year History, pp.
77-78.
150. “IC Industry Looks to the Government for Aid; Is It Quick Remedy or Only
86 THE JAPANESE ASCENT IN SEMICONDUCTORS

The structure of the “IC Liberalization Countermeasure” aid program

of 1973-74 was intended to stimulate the reorganization of the IC indus¬
try, but it had only limited success in encouraging the desired trend.151
Eight of Japan’s twelve domestic IC producers were involved in the pro¬
gram, but only five were finally organized into joint development efforts.
Two of these companies, Mitsubishi and Oki, had already been cooper¬
ating in the development of ICs for use in computers as part of the
computer liberalization subsidy program; the IC program extended this
cooperation into NMOS LSI technology development. Joint efforts ap¬
parently included creation of a “semiconductor consultative council” to
promote joint IC development and exchange technical information, and
joint development of an NMOS LSI microprocessor.152 Similarly, the sec¬
ond group, composed of Fujitsu, Sharp, and Kyodo Electronics Labo¬
ratories (Kyodo, still the only pure merchant Japanese producer, was a
supplier to both Fujitsu and Sharp) formed a “summit conference” and
a joint technology committee to promote joint development and infor¬
mation exchange.153

Narrow Escape?” Nihon Keizai Shimbun, November 5, 1973; “IC Subsidies to be Granted
to 8 Companies; Mitsubishi-Oki and Fujitsu-Sharp-Kyodo Electronic Groups Come to
Fore; Hitachi, Toshiba, and Japan Electric to Carry Out Unilateral Development,” Nihon
Kogyo Shimbun, November 29, 1973 (both excerpted in SI A, Japanese Protection). One of
the joint development efforts was the work on NMOS LSI, carried out by Mitsubishi and
Oki; the other was development of multipurpose linear ICs for use in industrial applications,
jointly undertaken by Fujitsu, Sharp, and Kyodo Denshi Gijutsu Lab. See Computer White
Paper, 1974 Edition, p. 22.
151. “MITI expects that groups of companies engaging in joint development will main¬
tain their cooperative structure in other fields henceforth without sticking to those kinds of
items which became the objects of a subsidy.” “IC Industry Circles to Make Every Possible
Effort for Development of Technology; Subsidy Amounting to 1.8 Billion Yen for Fiscal
1974 to be Given Shortly to Eight Companies, Including Hitachi and Toshiba,” Nikkan
Kogyo Shimbun, March 20, 1974. See also “IC Subsidies to be Granted,” Nihon Kogyo
Shimbun, November 29,1973; and “Joint Development for IC New Model Becomes Active;
Mitsubishi-Oki; Fujitsu-Sharp, etc.,” Nihon Keizai Shimbun, January 6,1974 (all excerpted
in SI A, Japanese Protection).
152. Mitsubishi and Oki were already cooperating on development of ICs for use in
computers as part of their participation in the 3.5 generation program. Joint work on their
IC promotion subsidy project—development of an NMOS microprocessor LSI—was to be
divided, with Mitsubishi developing the LSI chip itself and Oki the low-cost package. Nihon
Kogyo Shimbun, February 19, 1974 (excerpted in SI A, Japanese Protection).
153. Nihon Kogyo Shimbun, February 19,1974; and “IC Industry Circles to Make Every
Possible Effort for Development of Technology; Subsidy Amounting to 1.8 Billion Yen for
Fiscal 1974 to Be Given Shortly to Eight Companies, Including Hitachi and Toshiba,”
Nikkan Kogyo Shimbun, March 20, 1974 (both excerpted in SI A, Japanese Protection).
THE JAPANESE ASCENT IN SEMICONDUCTORS 87

On Christmas Day 1974, MITI finally delivered on its promise to

liberalize IC imports by the end of 1974. At almost the last possible
moment, ICs with more than 200 circuit elements (other than computer
ICs) were finally freed from licensing requirements.154 The Japanese were
far from confident of the ability of their producers to compete freely with
imports; it was generally believed that the domestic industry lagged a
year or two behind the foreign competition in technological terms.155 A
monthly publication of the Electronics Industries Association of Japan
pessimistically observed in March 1975 that, although Japanese producers
had caught up with the United States in production of MOS LSI for
electronic calculators, “the Japanese industry is thought to be two or
three years behind the U.S., even though it has been protected by Gov¬
ernment measures.”156
Japan had also largely liberalized foreign investment in ICs in 1974,
and firms were now for the most part free to set up wholly owned sub¬
sidiaries. The process of liberalization was finally completed in 1975,
when the last import restrictions on ICs used in computers were lifted.
Wholly owned foreign investment in computer chips was not permitted
until the very end of that year (table 2-4). Liberalization of capital in¬
vestment was not entirely unencumbered: hostile acquisitions of existing
firms, for example, were still barred, and advance notification—and prior
approval—of investments were still formally required.157 This was not to
change until the foreign investment law was revised in 1980. But by the
mid-1970s the formal doors to the Japanese market had finally swung
open for U.S. semiconductor companies. (It should be noted, however,
that though imports could now enter without a license, a stiff import duty
of 12 percent, as opposed to 6 percent in the United States, was levied
on ICs through the end of the decade.)

154. “Overall Import Liberalization Tomorrow, MITI Aid to Industry to Continue,”

Nihon Keizai Shimbun, December 24, 1974 (excerpted in SIA, Japanese Protection).
155. “Do We Advance toward a Reorganization of the IC Industry?” Nihon Keizai
Shimbun, December 24, 1974; and “Consideration of an Emergency Tariff,” Nikkan Kogyo
Shimbun, December 21, 1974 (both excerpted in SIA, Japanese Protection).
156. “Competition May Short-Circuit Japanese IC Industry,” Journal of the Electronics
Industry of Japan, March 1975 (excerpted in SIA, Japanese Protection).
157. Although Japan had made international commitments not to obstruct such invest¬
ment, it retained the formal internal legal power to block investment. Author’s interview
with Mr. Nishibe, MITI, and Mr. Sakamoto, Ministry of Finance, Tokyo, March 1992.
88 THE JAPANESE ASCENT IN SEMICONDUCTORS

Table 2-4. Japanese Measures to Liberalize Trade and Investment in the

Electronics Industry, 1962-76
Capital liberalization Technology
Import transfer
Products 50 percent 100 percent liberalization liberalization

Electronic computers
Major components 8/4/1974“ 12/11/1975“ 12/24/1975 7/1/1974
Peripheral device 7/1/1974 7/1/1974 7/1/1974 7/1/1974
Memory or terminal 8/4/1974 12/11/1975 12/24/1975 7/1/1974
Other partsb 8/4/1974 12/11/1975 2/1/1972 7/1/1974
Other components 8/4/1974 12/11/1975 12/24/1975 7/1/1974
c
Software 12/1/1974 4/1/1976 7/1/1974
Integrated circuits
Fewer than 100 8/4/1971“ 12/1/1974“ 9/1/1970 6/1/1968
Fewer than 200 8/4/1971 12/1/1974 4/19/1973 6/1/1968
More than 200 8/4/1971 12/1/1974 12/25/1974 6/1/1968
Source: Kenneth Flamm, Targeting the Computer: Government Support and International Competition (Brookings,
1987), p. 254; and MITI Machinery and Information Industry Bureau, Current Condition of the Electronics Industry
(Tokyo, September 1985), p. 8 (in Japanese).
a. Including integrated circuits for computer.
b. Input device, output device, communications control, and so forth.
c. No quotas applied.

Into the Japanese Market

Even before it had become clear that political pressure was finally
about to succeed in pushing open the door into the Japanese market,
American semiconductor firms had begun to position themselves for the
big event. Fairchild (in 1969) and National Semiconductor (before 1972)
had attempted an end run around investment restrictions by setting up
wholly owned manufacturing affiliates in U.S.-occupied Okinawa, where
the investment regime was more liberal than on the main islands of Japan.
The companies were gambling that, when Okinawa reverted to Japan in
1972, they would be permitted to retain their wholly owned Okinawa
manufacturing affiliates (despite a legal maximum of 50 percent foreign
ownership in IC manufacturing affiliates after the initial round of liber¬
alization in 1971) and achieve unhindered access to the Japanese mar¬
ket. 158 The bet failed to pay off. National was forced to close its operation,
while Fairchild was compelled to enter into an ill-starred 50 percent joint
venture with TDK Electronics, which finally closed in 1977 after years of
problems.

158. This episode is recounted in Mason, American Multinationals in Japan, pp

222-23.
THE JAPANESE ASCENT IN SEMICONDUCTORS 89

Motorola, the second-largest global vendor of semiconductors (behind

Texas Instruments) in the early 1970s, also met with frustration in the
1970s and early 1980s. It had established a sales office in Tokyo in 1962
but found itself unable to market its products to Japanese customers
without a Japanese manufacturing facility.159 Reluctant to alter a company
policy of avoiding joint ventures in selling its high-technology products
in overseas markets, Motorola nonetheless had decided to take the plunge
into a 50 percent-owned Japanese joint venture in semiconductors in
1971, when Japanese policy was changed to permit such ventures. After
much work, a joint semiconductor manufacturing venture with Japanese
audio producer Alps was set up in 1973, but it collapsed in 1975 during
an economic downturn, leaving Motorola again without a local factory.
Gun-shy, Motorola went back to a pure sales effort until 1980, when it
decided to acquire Toko, a small domestic producer of electronic com¬
ponents.160 Advised by Japanese political and business circles to go slow
on outright acquisition, Motorola formed a 50 percent joint venture with
an agreement to acquire Toko’s equity within three years, in a deal
reminiscent of TI’s 1968 bargain with MITI and Sony. The wholly owned
subsidiary was formed after two years and renamed Nippon Motorola,
but sales growth was mostly disappointing. Despite twelve years of in¬
vestment efforts, Motorola still felt it was stuck outside the core of the
Japanese electronics market in the early 1980s.
Despite their frustrating track record in joint ventures, American
firms continued to make forays into the Japanese semiconductor market
throughout the 1970s. In 1971 AMI set up a marketing office. Advanced
Micro Devices set up a Japanese sales office in 1974, followed by Signetics
and ITT in 1975, Intel in 1976, and Zilog in 1977.161 The head-on con¬
frontation with the American IC makers was clearly building toward a
climax after 1975, when the last residual restrictions on trade and invest¬
ment disappeared.

159. This account of Motorola’s experiences is drawn from Mason, American Multi¬
nationals in Japan, pp. 220-31; National Research Council, U.S.-Japan Strategic Alliances
in the Semiconductor Industry: Technology, Transfer, Competition, and Public Policy (Wash¬
ington: National Academy Press, 1992), pp. 91-101; David B. Yoffie and John Coleman,
“Motorola and Japan (A),” Harvard Business School case 9-388-056 1987, rev. 1989.
160. Both Toko and Alps had lost assets in semiconductor manufacturing when Kyodo
Electronics Laboratories, the startup founded as a joint venture by American Bernard
Jacobs and seven Japanese companies in 1964, disappeared in the mid-1970s.
161. Japan Electronics Almanac 1983 (Tokyo: Dempa Publications, 1983), p. 202.
90 THE JAPANESE ASCENT IN SEMICONDUCTORS

The rising crescendo of concern over imminent liberalization jumped

to a higher level of tension in the mid-1970s, when a new “technological
shock” was revealed. In 1974 senior officials at NTT became aware of
two startling technical developments in the United States. An antitrust
case against IBM, launched by Telex in 1973, had resulted in public
disclosure of IBM’s plans to develop a so-called “Future System” com¬
puter in the early 1980s, using a 1-million-bit (1-megabit, or 1M) IC
memory chip, at a time when the state of the commercial art was a chip
holding only 1,000 bits.162 Furthermore, on a visit to AT&T’s Bell Labs,
the president of NTT had observed an experimental system to etch the
finely detailed masks used to make very large scale integration (VLSI)
chips with small enough elements to pack onto a megabit memory.
Shocked into action by how far ahead American semiconductor R&D
appeared to be, NTT decided in 1975 to launch a major technological
initiative designed to close the technology gap.163

NTT Arrives on the VLSI Scene

Although MITI was the formal bureaucratic patron of the Japanese
IC industry, NTT had, since its establishment in 1952, played an impor¬
tant (though considerably less visible) role in supporting the domestic
industry. Barred legally from manufacturing its own equipment (as
AT&T, its American counterpart, was not), the Japanese telephone com¬
pany—organized as a special, public corporation—had long been en¬
gaged in electronics R&D, transferring the resulting technology to and
developing products with its Japanese suppliers in the private sector.
Fueled by subsidized loans from Japan’s public postal savings bank sys¬
tem and the compulsory purchase of its bonds by new telephone subscrib¬
ers, NTT had enormous financial resources.

162. NHK, Electronics-Based Nation, vol. 4, p. 356.

163. Anchordoguy, Computers Inc., p. 139, drawing on published interviews with NTT
President Yonezawa; and Yoshio Nishi, “The Japanese Semiconductor Industry,” in Bjorn
Wellenius, Arnold Miller, and Carl J. Dahlman, eds.. Developing the Electronics Industry
(Washington: World Bank, 1993), pp. 123-30; and the unpublished slides from the oral
talk on which this work is based. See also VLSI Technology Research Association, VLSI
Technology Research Association: Retrospective of the Past Fifteen Years (Tokyo, 1991), pp.
80-81. Fransman, The Market and Beyond, p. 57, and “NTTPC and Computer Makers Plan
Super LSI to Curb IBM Advance,” Japan Economic Journal, April 15, 1975 (excerpted in
SI A, Japanese Protection), contain variants on this basic story.
THE JAPANESE ASCENT IN SEMICONDUCTORS 91

Until the early 1980s, when it became the subject of protracted trade
negotiations between the United States and Japan, NTT’s procurement
was effectively closed to non-Japanese suppliers.164 Indeed, even many
Japanese suppliers were frozen out of NTT contracts, which for the most
part went to a small group of favored companies, the so-called “Den-
Den” family of suppliers. Within this small group, an even smaller group
of four principal suppliers—NEC, Fujitsu, Hitachi, and Oki—received
the bulk of the NTT contracts. In 1968, for example, these four firms
accounted for 70 percent of NTT’s procurement.165 This share was to
decline over the next decade, but remained around 50 percent in the
early 1980s.166
NTT’s support for Japanese electronics manufacturers has been or¬
ganized quite differently from that provided by MITI. NTT’s large inter¬
nal research laboratories conducted basic research and transferred the
results of that research to selected suppliers. The transfer of technology
typically occurred through joint projects with suppliers to develop equip¬
ment using new technology, and through the exchange of technical per¬
sonnel between supplier companies and the NTT laboratories.167 Support
for R&D by its suppliers has not taken the form of contract R&D per¬
formed by these companies for NTT.168 Instead, funding for supplier
R&D has been structured as a part of development and procurement

164. Indeed, the president of NTT’s widely quoted response to some of these foreign
pressures was to declare that “the only thing we could consider buying overseas would be
[telephone] poles and mops.” “High Technology Gateway: Foreigners Demand a Piece of
NTT’s $3 Billion Market,” Business Week, August 9, 1982, pp. 40-44.
165. Anchordoguy, Computers Inc., p. 42, citing a Japanese study.
166. See U.S. International Trade Commission, Foreign Industrial Targeting and Its
Effects on U.S. Industries, Phase I: Japan (Washington, 1983), p. 152; and “High Tech¬
nology Gateway,” pp. 42, 44. Data in these sources show NEC, Fujitsu, Oki, and Hitachi
accounting for between 42 and 49 percent of NTT’s purchases over the years 1980-82.
Companies affiliated with NEC and the Sumitomo trading group (to which NEC belongs)
account for another 5 percent of NTT procurement, pushing the big four share above
50 percent. In 1981 NEC was the largest single supplier, accounting for about 20 percent
(25 percent with other Sumitomo group affiliates added on) of NTT purchases. Fujitsu
accounted for about 13 percent of purchases, Oki 7 to 8 percent, and Hitachi 6 to 7 percent.
167. Kozo Yamamura, “Joint Research and Antitrust: Japanese vs. American Strate¬
gies,” in Hugh Patrick, ed., Japan’s High Technology Industries: Lessons and Limitations
of Industrial Policy (University of Washington Press, 1986), p. 194; and Jon Sigurdson,
Industry and State Partnership in Japan: The Very Large Scale Integrated Circuits (VLSI)
Project (Lund, Sweden: Research Policy Institute, 1986), p. 100.
168. Published R&D survey statistics, for example, confirm that only relatively small
amounts could have been paid out to external organizations for contract R&D in the 1970s
and 1980s. See Flamm, Targeting the Computer, p. 139, note 19.
92 THE JAPANESE ASCENT IN SEMICONDUCTORS

contracts for new equipment.169 Patents resulting from these joint devel¬
opment projects have generally been jointly owned by NTT and the
company concerned, so that companies have been free to use technolo¬
gies developed with NTT assistance.170
NTT’s early role in Japanese transistor R&D (it boasts of having built
the first Japanese germanium transistor, in 1952) as well as the manner
in which significant know-how was transferred to the private sector has
already been noted. NTT largely focused its semiconductor research
through the mid-1960s on the conservative goal of improving transistors
used in transmission systems. In the late 1960s, however, NTT began
research on electronic switching. Its first studies of the use of ICs were
not undertaken until 1965, when a project to design a family of IC logic
chips for use in switching systems was begun.171
Within the large semiconductor makers, all of which were Den-Den
family suppliers, development—and procurement—of ICs used in NTT
equipment were of considerable importance. In 1968, for example, NEC
had three lines of advanced digital ICs under development: one for “a
government-sponsored, large scale integration project” (presumably the
LSI component of MITI’s Very High Speed Computer System national
R&D project), another for use in NTT’s electronic telephone exchanges,
and a third for use in its next-generation advanced computer systems.172
Five Japanese manufacturers—Hitachi, Toshiba, NEC, Mitsubishi, and
Fujitsu—worked on ICs for use in NTT’s “DEX 2” electronic switching
system, developed in 1969. Experience on this project is credited by
Japanese semiconductor producers with being especially important
in establishing rigorous quality and reliability standards for domestic
products.173
Through the 1960s, though, NTT put little emphasis on support for
the Japanese electronics industry beyond its relatively narrow interest in

169. Ira C. Magaziner and Thomas M. Hout, Japanese Industrial Policy (Berkeley:
Institute of International Studies, University of California, Berkeley, 1980), pp. 107-08;
and Sigurdson, Industry and State Partnership.
170. Sigurdson, Industry and State Partnership; and Yamamura, “Joint Research and
Antitrust.”
171. Watanabe, “Electrical Communications Laboratories,” p. 4.
172. See Tarui, “ICs in Japan,” p. 105. The NTT logic design, known as controlled
saturation logic (CSL), was used in its D-10 electronic switching system. See Watanabe,
“Electrical Communications Laboratories,” p. 4.
173. NHK, Electronics-Based Nation, vol. 4, p. 14.
THE JAPANESE ASCENT IN SEMICONDUCTORS 93

creating a domestic supplier base for its specialized telecommunications

equipment designs. For example, NTT resisted MITI pressure to use
domestic computers; in 1971, 109 of its 172 computer systems were sup¬
plied by foreign companies.174 In the late 1960s, however, NTT began to
take a wider view of its domain and started to fund data communications
R&D. Its first big jump into the field was the decision to design and build
a large time-sharing computer, the Dendenkosha Information-Processing
System (DIPS-1), with a development effort initiated in 1968. A measure
of NTT’s relatively narrow support for IC development up to that time
was the fact that the first DIPS models (built by Fujitsu, Hitachi, and
NEC) used ICs and other pieces of technology developed for MITI’s
VHSCS project, in which these same companies participated.175
As its Japanese electronics suppliers faced the trauma of market lib¬
eralization in the early 1970s, however, NTT began to adopt a substan¬
tially broader view of its responsibilities toward ensuring their financial
health. When MITI funded the 3.5 generation computer project in 1971
to fend off the IBM threat, NTT stepped in and agreed to coordinate its
plans for a follow-up to the DIPS system with the MITI programs, cre¬
ating an additional 5-billion-yen ($14 million) subsidy for development of
essentially the same computer technologies.176 NTT’s increasing concern
over the competitive fortunes of domestic electronics companies, now
forced to grapple with technologically advanced foreign rivals on a rela¬
tively open playing field, was an essential element of its decision to plunge
into VLSI research on a large scale.
The NTT VLSI program begun in 1975 was an unprecedented, large-
scale effort, with a budget of 20 billion yen ($67 billion) over three
years.177 NTT’s internal laboratories cooperated with those of Fujitsu,
Hitachi, and NEC, the suppliers of NTT’s DIPS computers, in developing
the basic technologies—materials, processes, circuits, and systems—
needed to manufacture computer chips. Production of a 64K DRAM

174. Anchordoguy, Computers Inc., p. 31.

175. See Flamm, Targeting the Computer, p. 129; and Anchordoguy, Computers Inc.,
pp. 54-56.
176. Anchordoguy, Computers Inc., pp. 121-22.
177. The 20-billion-yen budget is reported in U.S.-Japan Study Group, p. 30, and Koki
Inoue, “The VLSI Technology Research Association and the Development of the Semi¬
conductor Industry,” Kikai Keizai Kenkyu (Machinery Economics Research), no. 23 (Tokyo:
Machinery Promotion Association, 1992), p. 63. See also Anchordoguy, Computers Inc.,
p. 138.
94 THE JAPANESE ASCENT IN SEMICONDUCTORS

was chosen as the vehicle for creating and testing the new process tech¬
nologies.178 The program met its goals: by 1977 a 64K DRAM had been
produced and significant progress made in photolithographic, X-ray, and
electron beam systems used to manufacture integrated circuits.179 By that
time, however, the NTT effort was no longer the only such program, or
even the largest one, on the scene.

The VLSI Project

MITI was equally disturbed by the reports from the United States
that suggested that, once again, the Japanese computer industry was
about to find itself set back by another round of rapid technical advance
overseas. The challenge was worsened by the fact that, with liberalization
almost complete, there would no longer exist a sheltered Japanese market
in which domestic producers could at least temporarily escape the rigors
of competition at the leading edge.
As discussed earlier, a complex political dance was already under way
over the future of the Japanese computer industry. Some forces within
the LDP, the ruling political party, wished to force the computer makers
into a single national champion firm, along the lines already being fol¬
lowed by British and French government policies. The six largest Japa¬
nese computer makers—all of them (except Oki) actually the computer
divisions of large and powerful Japanese conglomerates—resisted this
idea vigorously. In 1971, however, with Kakuei Tanaka as MITI minister,
they had been forced to reconfigure themselves as three loosely allied
pairs of firms in exchange for financial assistance from MITI to cope with
liberalization. But, as we have seen, these alliances were imposed from
outside, and the strongest companies did everything they could to pre¬
serve their independence. In 1973 MITI responded by using a program
of subsidies to IC R&D to encourage closer relations among the firms.
Only the Mitsubishi-Oki pairing (between the two weakest companies in
the Japanese computer industry) had responded to this carrot by struc¬
turing a cooperative research project; the others had simply gone their

178. Watanabe, “Electrical Communications Laboratories,” p. 7.

179. Watanabe, “Electrical Communications Laboratories,” pp. 7-8; and “VLSI Mem¬
ory Chip of Extremely High Concentration Is Developed,” Japan Economic Journal, May
2, 1978, and “Responding to Criticism of the Closed Nature, Complete Opening of Super
LSI Patents of NTT,” Keizai Sangyo Shimbun, February 8, 1980 (both excerpted in SI A,
Japanese Protection).
THE JAPANESE ASCENT IN SEMICONDUCTORS 95

own separate ways with their shares of the IC research subsidies.180 Com¬
plicating matters further from MITI’s perspective was NTT’s increasing
interest in computers and computer-related ICs, which historically had
been MITI’s responsibility.
When news of the double-barreled IBM-AT&T technological threat
from America hit Japan in 1974-75, important voices within the Japanese
industrial system clamored for action. NTT advanced further into MITI
territory with the announcement of its 1975 VLSI program. The industry
begged for huge new subsidies but was opposed by fiscal conservatives
within the government. Powerful groups within the LDP repeated their
call for a single Japanese national computer champion.181
Whether MITI was serving as a buffer between the conflicting interests
of LDP politicians and the private sector, or the bureaucrats were “mere
puppets for important political interests,” or “the politicians were really
only ‘cheering squads’ (oendan) for the bureaucratic armies” will prob¬
ably forever remain a topic for vigorous debate among students of Japa¬
nese political economy.182 Whatever the reality, a three-sided negotiation
among MITI, the electronics companies, and LDP politicians took place.
A compromise, brokered by the former head of the Dietmen’s League
for Promotion of the Information Industry, was struck, giving the industry
a large subsidy for VLSI research.183 In exchange, the firms would recon-

180. Fujitsu, however, did join a cooperative effort with two electronics companies not
involved in computer production, Sharp and Kyodo Electronics Laboratories. Also, al¬
though not a formal part of the IC subsidy program, a certain amount of cooperation in
R&D appears to have gone on between Hitachi and Fujitsu through their Nippon Periph¬
erals joint venture, and between NEC and Toshiba through their NTIS joint venture
company.
181. The account given below draws heavily upon Masato Hashizume, “Tracing the
VLSI Research,” in VLSI Technology Research Association, VLSI Technology Research
Association: Retrospective, pp. 80-87. Hashizume, a former section chief from the elec¬
tronics industry division of MITI, was appointed head councilor of the VLSI Research
Association. See also NHK, Electronics-Based Nation, vol. 4, pp. 358-61.
182. These different views of the Japanese political system are marvelously exposited
in Chalmers Johnson, “MITI, MPT, and the Telecom Wars: How Japan Makes Policy
for High Technology,” in Chalmers Johnson, Laura D’Andrea Tyson, and John Zysman,
eds.. Politics and Productivity: The Real Story of Why Japan Works (Ballinger, 1989), pp.
197-200.
183. The little-known Dietmen’s League for Promotion of the Information Industry
had earlier consulted with MITI when the original 1973-74 IC R&D subsidies, and addi¬
tional assistance for the computer industry, had been negotiated. See “MITI Informally
Decides on Liberalization Countermeasures Expenses—Electronic Computers: 43 Billion
Yen for Three Years: Aid to Be Extended for Development of New Models: Total Amount
96 THE JAPANESE ASCENT IN SEMICONDUCTORS

figure themselves into just two private industry laboratories for the pur¬
poses of the subsidy and do a significant amount of R&D jointly in a
single common facility.
The research effort was organized around five companies: Fujitsu,
Hitachi, Toshiba, NEC, and Mitsubishi. (Both Toshiba and Oki dropped
out of mainframe computer manufacture in the mid-1970s, but only Oki,
technically and financially the weakest of these companies, was excluded
from the project.184) Technical personnel from MITI’s ETL were put in
charge of the common facility, dubbed the VLSI Joint Laboratory, which
gathered personnel from all five private sector participants and the ETL.
NTIS, the NEC-Toshiba joint venture formed in 1971, which already had
established a joint research laboratory system, became one of the two
participating private laboratories. A new entity known as Computer De¬
velopment Laboratories (CDL), with Fujitsu, Hitachi, and Mitsubishi as
participants, was created as the second industry laboratory.185 The entire
project was budgeted at 74 billion yen ($236 million), to be spent from
1976 to 1979; of that total MITI was to contribute 40 percent. The overall
objective of the program was to develop the technologies needed to
manufacture a 1M DRAM, or the equivalent, by 1985.186
Quite simply, this was by far the largest infusion of R&D subsidies
ever received by the Japanese semiconductor industry, in both absolute
and relative terms. The VLSI project accounted for almost 40 percent
(over 50 percent if the NTT projects are counted as well) of Japan’s

of Subsidies to Reach 77.1 Billion Yen,” Sankei Shimbun, August 14, 1973 (excerpted in
SI A, Japanese Protection).
184. Various explanations have been advanced as to why Oki got dropped and not
Mitsubishi. The most common explanation is that Oki was simply too technologically
inferior. See “Responding to Criticism of the Closed Nature, Complete Opening of Super
LSI Patents of NTT,” Keizai Sangyo, February 8, 1980 (excerpted in SIA, Japanese Protec¬
tion)-, and Anchordoguy, Computers Inc., pp. 140-41, quoting MITI sources. Other reports
suggest that OKI’s close technical ties with American computer maker Sperry Rand through
their Oki-Univac joint venture were a factor. See “Computer Industry Seen Revamped;
Joint VLSI Development Agreed,” Japan Economic Review, August 15, 1975 (excerpted in
SIA, Japanese Protection). Oki was reportedly given 64K DRAM technology by NTT to
compensate for its exclusion. Other firms outside the NTT VLSI project, including Toshiba,
Mitsubishi, and Matsushita, also reportedly lobbied for access to the NTT work. See
“Responding to Criticism.”
185. Mitsubishi was the weakest company in terms of its level of IC technology. Author’s
interview with Yasuo Tarui, March 1992. Its presence in CDL probably posed little threat,
then, to either Fujitsu or Hitachi. Also, Anchordoguy, Computers Inc., p. 140, cites a
popular Japanese history of the period as suggesting that Fujitsu saw this as a chance to
penetrate the Mitsubishi keiretsu grouping’s large computer market.
186. Nishi, “The Japanese Semiconductor Industry,” p. 5.
 to O 3

+ £
d f' r~ oo
rt¬ no i/t m
E S>< a
* &"• = “
c"> S O ■f'
g
^3
s: 2-I S i
§ -o
£ O >N C
<n
&* 31 £ 53 ™
'c: .E oo
Q t 1L no iri u-> r- oo o j= r-
3i cj —
“3 <&, m o ic oo Q. CQ
N to 't N
OS 3 b 2 8S ^

U a 5 * ■i ^

a §
I = f2 2
a y 5 .2 li «
-3 3) • CL
+ ^ d m oo d d rf o —< >/n ro
R b-
O -t H Ol
s t-H oo Tt d no d m ri io OO N oi H 6 o\ t:rl
<D © £ rf d d cn ro m d d
no a O ^ g = §.
On 3 8^5
£ c oc
^ ij
r9
•" Ss-S a
•kj f- NO r'l ro oo M O 1; h Oi 2-st i
^ C NO On rt O On
3 o o 3 oo
rf d
■t 6 o oi —5 d' (3 id VO d N Ol O
t-H {N} T-H r-H
o a " o g ^
?>
c C.'i'
& O *J
£ s c
ri ^
to
I?
3 y q r- r- r- s-S
OS .2
ts Q
£ fc *8 &o ts
o s
vd id id
'O VO vO vO 'vO id id

52 1/5
b CJ
!<
I
Q is S *o «
u j c a
3 a> O O
^3^00

QS Sp ^3 § liu
S£ C % O MU O

a Q s5wP
S o « E
K aS
00 C''; © iO
3
"3 g. ri h oo id *0Sujo
£ c oo o in ov
CN (N t—i
<n
r-
so

Q & os
H -C)>N
C w “I
« j tso
"3 P =
CD ^ P£ . 2
S 1
5 2 h E
3 T3 M
c/5 e W)
S? - £- = 6
H Tf f\] (N
Q £ B litf-B
°8 «,
't ’t On 00
CO 00 Os iO
ca E .2 V5 c
<u o o w 3
3 o; c/joOC1*-
W l r^i CO ,
O
U ^
£5 *; g
■Cl
3 3 I -■§=,■§
3 © ^
*T-«>
*»A o p poop oooo ooooo
8 ^
-2 BS dH
r--

~3 O W
^ rt- ^t- Ht (N CO IO ID o o o in in e x> n- , S

6 ta <n cn N IT) ON O h : g.2gs

■e § > g S'
« u £ ?! 3
o _ W C/5

til j£
< E-
5H 3 32
s oo.£
y s;
& ^ s t-H iO O oooo oooo ooooo J« |< i
o ^ 2 «r g
Sg
JiQ 3 oi it! o m in r- Ov O O in in <n in o oo iD E
rs -
o y o'
On ^
O'
NO IT)
r-
t? K (N n *o oo O' N OO Hrt^-00^
<3 ta (N N n in P -H 3-1
' O M k
3 O' M
c ^
J- - 00 ■ ' c
U O' O £ Urc «j
c — no o o. DO DO
(U <U
O CL . <u X) -O
in >> ' <S) /-S - c 00
■ yN „
. <U UJ E E On
^ « W —
Cd os:
Ov
r- 2S O DO 60 E
0) I EU £ S £
i: o
s CO in iO r- 00 Oi o tH (N m rl* in iO 00 Ov iO
iO VO VO vO vO VO VO r- r- r- r- r- r- r^
3 X)
Ov Ov Ov Ov Ov
£ Ci
t-H
ON
t-H
Oi
t-H
Oi
T-H
Oi
t-H
Oi
T-H
Oi
t-H
Ov
t-H
Ov
t-H
Ov
*-H t-H T-H
Ov
t-H t-H, t-H
Ov
T-H t-H
Ov
t-H c Cuj (2
98 THE JAPANESE ASCENT IN SEMICONDUCTORS

national IC R&D effort in the late 1970s (table 2-5).187 In addition to the
subsidies shown in table 2-5, in-kind contributions of personnel and
equipment from MITI’s own lab worth 500 million to 1 billion yen,
exemptions from tariffs on machinery and equipment worth about
550 million yen, and reductions in taxes on fixed capital further aug¬
mented government support for the VLSI project.188
There is a virtual consensus in Japan that the effort was an exceedingly
important boost to Japanese semiconductor technology and played a key
role in making Japanese firms leaders in semiconductor manufacturing in
the 1980s.189 Certainly, tangible progress in DRAM manufacturing, in
particular, followed in the wake of the VLSI projects. The first 16K

187. My figures suggest that the VLSI project was a smaller fraction of overall IC R&D
than does a widely cited calculation by Ryuhei Wakasugi, because I am using revised MITI
statistics that increased estimates of private IC R&D in 1979. See “Research and Devel¬
opment and Innovations in High Technology Industry: The Case of the Semiconductor
Industry,” Japanese Economic Studies, vol. 17 (Fall 1988), p. 15.
188. Inoue, “VLSI Technology Research Association,” p. 63.
189. Typical appreciations are the following: “the most successful among [technology
research cooperatives] has been the VLSI Technological Research Cooperative. The reason
that it has been assessed as a success is, first, that there have been achievements in regard
to technologies that are common throughout the semiconductor industry as a whole. . . .
Second, there has been an efficient interfirm transfer of the common technologies.” Wak¬
asugi, “Research and Development,” pp. 18-19.
“Consequently, the VLSI Program contributed greatly to the Japanese semiconductor
industry both directly, in terms of its research and development achievement, and indirectly
by building a new culture for R&D planning and collaboration.” Nishi, “The Japanese
Semiconductor Industry,” p. 126.
“Today there exists a consensus that the VLSI Project in Japan, established in 1976 as
an engineering research association, had an exceptional success in promoting technological
development. . . . The project resulted in firmly raising the level of VLSI manufacturing
technology of the five participating companies. . . . This is a contributing factor for the
increasing shares taken by these companies in the world market for memory circuits.
However, the VLSI project also had a profound effect on companies outside the group of
5 companies; today Matsushita has become a major actor in the field.
Furthermore, the project similarly raised the technical level of two major groups of
supporting companies ... not only have Japanese VLSI companies established themselves
among the world leaders but its crystal and VLSI equipment manufacturers have also
established themselves in the top league.” Sigurdson, Industry and State Partnership, pp.
62-63.
“In the middle of the 1980s, Japanese companies caught up with their U.S. counterparts,
and had stayed in a dominant position in the memory chip market represented by 4M
DRAM and 16M DRAM until the beginning of the 1990s. It is well known that the ‘VLSI
Research Association’ assisted by MITI contributed to the success of the Japanese semi¬
conductor industry during this period to a great extent.” Ryo Sakamoto, “Will the Japanese
Version of ‘Sematech’ Save the Japanese Semiconductor Industry?” Foresight (November
1995), pp. 94-95.
THE JAPANESE ASCENT IN SEMICONDUCTORS 99

DRAM, introduced by the U.S. producer Intel in 1976, kicked off four
generations of DRAMs, through the 1M chip introduced in the mid-
1980s, in which the focus for innovation was on perfecting manufacturing
technology with which to squeeze a well-understood physical design into
ever smaller spaces.190 Improvements in manufacturing technology were
precisely the focus of the VLSI projects, and they provided an entry
point for Japanese manufacturers into world semiconductor markets.
Beginning in 1978, after the completion of the first NTT VLSI project,
the first significant innovations in DRAM design and manufacturing from
Japanese companies (from both NTT and project participant NEC) were
publicly unveiled, in the form of the first prototype 64K DRAMs. The
success of the first NTT project paved the way for a second 20-billion-
yen ($95 million) three-year project, ending in 1980.191 In the next gen¬
eration of DRAM, the 256K chip (appearing in 1980), Japanese compa¬
nies (again NTT and NEC) were even to take the lead as the first world¬
wide to unveil the new product. These new 256K DRAM designs, in
fact, were directly linked to the results of the VLSI projects.192 From
that generation on, Japanese producers clearly dominated the introduc¬
tion of new product and process technologies in DRAM manufacture.193
As the first large-scale experiment in which rival companies’ engineers
actually worked together in the same laboratories, the MITI VLSI pro¬
gram’s apparent success also made it the model for even more intensive
experiments with joint research, in Japan and abroad. However, recent
revisionist assessments of the VLSI project in the West have argued for
two qualifications. First, in assessing the project’s technological accom¬
plishments, it has been suggested that the “Japanese did not appear to
have made any major breakthroughs. In most areas, [an American VLSI

190. See Betty Prince, Semiconductor Memories: A Handbook of Design, Manufacture

and Application, 2d ed. (John Wiley, 1991), pp. 211-12. The earlier development of the IK
and 4K DRAMs has been characterized as a period of innovation in the basic physical
design of memory cells used in a DRAM, rather than in the process technology used to
implement this design.
191. Inoue, “The VLSI Technology Research Association,” p. 63.
192. NEC’s design used technology developed jointly with Toshiba as part of the MITI
VLSI project. See Steve Galante, “Japanese Semiconductor Firms Hustle to Short-Circuit
Image as Imitators,” Asian Wall Street Journal, June 16,1980, p. 4. Toshiba, on the strength
of its burgeoning VLSI skills, was to join the other three firms in the “club” on yet a third
NTT VLSI project, started in 1981, which targeted the 1M DRAM. “NTT Has Accepted
Toshiba to Join VLSI Project,” Nikkei Sangyo Shimbun, October 4, 1982, p. 1.
193. See Prince, Semiconductor Memories, pp. 218-73, for a historical retrospective on
DRAM innovations that supports this assessment.
100 THE JAPANESE ASCENT IN SEMICONDUCTORS

expert interviewed] felt that the Japanese had simply extended their tech¬
nology in ways comparable to developments that had already occurred in
the United States.”194
Second, it has been argued that in the VLSI project, as in previous
cooperative research programs sponsored by the Japanese government,
there was “very little research cooperation between the participating
Japanese companies, at least insofar as this refers to the joint generation
and sharing of technological knowledge ”195 Both of these points merit
some brief consideration.

The Technical Impact of the VLSI Project

It is important to remember that the VLSI project was conceived of

as a “catch-up” program, designed to bring Japanese producers up to a
standard of competence already assumed to be well under development
in the West. It was not intended to produce “major breakthroughs” in
the fundamental technologies of the integrated circuit; it did, however,
do a remarkable job of extending and improving on technological con¬
cepts already developed or under development overseas.
Figure 2-8 shows twenty-two “major results” of the MITI-subsidized
VLSI project, as identified by the VLSI Technology Research Associa¬
tion’s recent history of the project, and where they fit into the overall
process of IC manufacturing. Twelve of the advances were in equipment
used in processing the silicon wafers on which ICs are etched, two in
testing the processed wafer, two in manufacture of the large silicon crys¬
tals from which the wafers going into the manufacturing process are
sliced, two in technologies used to design chips, and six in demonstrating
prototypes of specific types of ICs. Thus, sixteen of the twenty-two were
demonstrations or improvements in equipment, or design tools, used in
the manufacture of any VLSI chip, and only six the development of
working prototypes of specific chips.

194. Uenohara and others, “Background,” in Okimoto and others, Competitive Edge,
pp. 38-39, also cited prominently in Fransman, The Market and Beyond, p. 84. It should
be noted that this assessment was made before the VLSI effort had actually ended. As will
be documented below, although the government subsidy ended in 1979, the companies
involved continued in a “private sector edition” of the project, using private funds only,
from 1980 to 1986, to further develop and commercialize the results of the initial, subsidized
project. Neither Okimoto and others nor Fransman seem even to be aware that the coop¬
erative effort extended beyond 1979.
195. Fransman, The Market and Beyond, p. 58.
Figure 2-8. Major Results of the Japanese VLSI Project, 1975-90

Direct drawing

>
O'
o
O
102 THE JAPANESE ASCENT IN SEMICONDUCTORS

In the view of the former director of the VLSI Project’s Joint Research
Laboratory, the four most important achievements of the program were
advances in electron beam machines (used to indirectly etch patterns for,
and directly write, circuit features on silicon chips); optical “steppers”
(photolithographic machines used to imprint patterns on chips); the tech¬
nology used to grow single silicon crystals, then slice and polish the silicon
wafers used in chip making; and techniques used to characterize and
measure silicon crystals.196 Another project participant’s list of major
research areas shows six topics in semiconductor manufacturing pro¬
cesses, three in materials, three in design and testing, two in packaging,
and four in particular device structures.197
Consistent with all these lists is the central thesis of the most detailed
published study of the VLSI project: that the most visible, concrete
effects of the program were in creating technical information flows be¬
tween Japanese equipment and materials suppliers and IC producers.198
It was in the manufacturing infrastructure for the Japanese semiconductor
industry that the most important advances appear to have occurred.
This is evident to some extent in the way the VLSI project’s budget
was spent. Figure 2-9 breaks out expenditure by the VLSI Technology
Research Association over the four years of the subsidized MITI project,
as well as seven additional years of wholly privately funded cooperative
research that continued the project after 1979 (this point is elaborated
below). Expenditure on facilities, materials, and procurement contracts
averaged about 70 percent of the project’s cost over the years 1976-79.
Yasuo Tarui, the project’s leader, estimates that 30 to 40 percent and
perhaps even more of the project’s budget went into the development of
new equipment.199 Another study estimates that a quarter to a third of
the project’s funding was spent on the purchase of the most advanced
semiconductor manufacturing equipment available in the United
States.200

196. Author’s interview with Yasuo Tarui, March 1992.

197. Nishi, “The Japanese Semiconductor Industry,” table 9.2.
198. Sigurdson’s excellent Industry and State Partnership monograph appears to be the
only published English-language study focusing on the actual technical content and results
of the VLSI project. Fransman’s The Market and Beyond is almost entirely devoted to the
formal organizational aspects of Japanese cooperative R&D programs.
199. Author’s interview with Yasuo Tarui, March 1992.
200. Kiyonori Sakakibara, “Organization and Innovation: A Case Study in Japanese
Semiconductor Project,” MIT Sloan School of Management, December 1982, p. 14.
THE JAPANESE ASCENT IN SEMICONDUCTORS 103
Figure 2-9. Expenditures on VLSI Project and Subsequent Cooperative
Research, by Category, 1976-86
Percent

Facilities Computer | | Materials

Personnel Procurement mi °ther

Source: VLSI Technology Research Association, VLSI Technology Research Association: Retrospective, p. 66.

Ironically, the most direct evidence for the link between the VLSI
project and Japanese semiconductor technology shows up in the com¬
mercial activities of firms that were not even members of the VLSI proj¬
ect. These equipment and materials firms cooperated and subcontracted
with the formal project participants to produce advanced materials and
equipment that had not previously been available to the Japanese indus¬
try (figure 2-10).201 The importance of these linkages is visible in the
subsequent success of Japanese equipment and materials suppliers in
global markets with products utilizing their VLSI project experience.

201. This figure, from Sigurdson, Industry and State Partnership, p. 120, is also partly
reproduced in Fumio Kodama, Analyzing Japanese High Technologies: the Techno-Para¬
digm Shift (London: Pinter Publishers, 1991), p. 89.
104 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-10. Linkages within VLSI Industrial System

Machinery VLSI computer Laboratories Materials Equipment Sub¬

makers system makers suppliers suppliers suppliers

DAI- USHIO
RICOH
NIPPON DENKI
VLSI
PROJECT
JOINT
LABS - TOPPAN

TOSHIBA
LITHO
T. MACH. TOSHIBA I.
T. SEIKI
TOKUDA CANON -1
TOK.DK. LITHO
ASIA E. II. KYOCERA

NIKON

FUJITSU LITHO
NGK-SPARK
III.
DISCO
SHINKAWA
ULVAC
CRYSTAL JEOL
NIHON SHINKU
TOKYO ELECTRIC
ANELVA
HITACHI SHIN ETSU ANDO
HITACHI WAFER
HANDOTAI
KOKUSAI-
DENKI
OSAKA
TESTING
TITANIUM

TOKYO
OHKA
ETL

NTT TAKEDA
LABS RIKEN

NHK
SHARP KDD KOKUSAI
SANYO DENKI

Source: Jon Sigurdson, Industry and State Partnership in Japan: The Very Large Scale Integrated Circuits (VLSI)
Project (Lund, Sweden: Research Policy Institute, 1986), p. 120.
THE JAPANESE ASCENT IN SEMICONDUCTORS 105

In wafer processing equipment, two key examples are those of Nikon

and Canon, which, marketing wafer processing systems initially devel¬
oped in collaboration with the VLSI project, went from minuscule mar¬
ket shares in 1981 to over half of the global market for optical lithography
equipment used in semiconductor manufacture by 1985.202 Another
equipment maker with a major presence in the global market, gained at
least in part through technology developed for the VLSI project, is To¬
shiba, which in 1978 began marketing an electron beam pattern generator
developed for the project.203 Dainippon Printing and Toppan Printing
(producers of masks used in etching features on chips), Kyocera (today
the world’s largest manufacturer of ceramic packages for ICs), NGK
Spark Plug (the second-largest manufacturer of ceramic IC packages),
and Ushio Denki (today the world’s largest supplier of high-intensity
lamps used in photo-etching machines for ICs) are other examples of
firms gaining a commanding presence on the world market after emerging
from the VLSI project experience. Other firms—such as JEOL in elec¬
tron beam equipment, Ulvac in ion beam implanters, Advantest (for¬
merly Takeda Riken) in IC testers, and Kokusai Electric in deposition
and etching equipment—underwent a similar transformation after their
experience with NTT’s VLSI project.204
In the late 1970s in Japan, chip manufacturing equipment, previously
purchased from U.S. vendors, was to an increasing extent displaced by
domestic equipment. In 1979 Japanese equipment manufacturers began
a push into foreign markets; within four years their global market share
had increased by two-thirds in wafer processing equipment (to 25 per¬
cent) and had tripled in test equipment (to 21 percent).205 This rapid

202. Presentation of John Poate prepared for a National Research Council seminar,
“Advanced Processing of Electronic Materials in the United States and Japan,” Washing¬
ton, June 4, 1986. In 1985 Nikon had 35 percent of the global optical lithography equipment
market, and Canon 17 percent. The leading American producer, GCA (Geophysics Cor¬
poration of America), had 30 percent of the market. In 1981 GCA had sold 73 percent of
wafer steppers shipped globally. See also Sigurdson, Industry and State Partnership, pp.
86-93; Jay S. Stowsky, “Weak Links, Strong Bonds: U.S.-Japanese Competition in Semi¬
conductor Production Equipment,” in Johnson and others, Politics and Productivity, p. 263;
and “VLSI Projection Aligner by Canon,” Electronic News, November 8, 1978, p. 28.
203. See John Hataye, “Toshiba Machine to Market E-Beam Pattern Generator,” Elec¬
tronic News, July 24, 1978, p. 42.
204. Sigurdson, Industry and State Partnership, pp. 93-105, 120; and Stowsky, “Weak
Links,” pp. 265, 268-69.
205. See John Hataye, “Japan Equip. Firms Eye Exports to U.S.,” Electronic News,
December 3, 1979, p. 20; and U.S. Department of Commerce, A Competitive Assessment
of the U.S. Semiconductor Manufacturing Equipment Industry (1985), p. 4.
106 THE JAPANESE ASCENT IN SEMICONDUCTORS

growth was to continue: by 1989 Japanese producers accounted for

44 percent of global wafer processing equipment sales (with shares in
individual market segments as high as 74 percent for optical steppers,
and 60 percent in diffusion furnaces).206
Similar histories are evident in semiconductor materials. All five of
the large Japanese silicon wafer manufacturers worked closely with the
VLSI project; advances in silicon wafer manufacturing technologies have
already been identified as one of the key accomplishments of the proj¬
ect.207 Between 1977 and 1984, Japanese manufacturers doubled their
share of world silicon wafer output, from 17 to 34 percent.208
In short, the circumstantial historical record suggests that the VLSI
projects of both MITI and NTT had direct links to significant improve¬
ments in semiconductor manufacturing technology in Japan, and that
much of this progress was reflected in a surge of Japanese equipment and
materials suppliers into the world marketplace. The IC manufacturers
who were the principals in these programs were to score equally impres¬
sive gains in global sales in the 1980s, largely on the basis of standardized
commodity products with cost structures dominated by manufacturing
(such as memory chips) rather than complex, design-intensive products
(such as microprocessors) where design and performance details were as
important as—if not more important than—production cost.

206. Department of Commerce, National Security Assessment of the U.S. Semiconduc¬

tor Wafer Processing Equipment Industry (April 1991), p. 26.
207. Author’s interview with Yasuo Tarui, March 1992. Sigurdson, Industry and State
Partnership, pp. 76-81, credits links forged between silicon wafer manufacturers and IC
producers during the VLSI project with the former’s rapid advance on the world market.
The information apparently flowed two ways. Sigurdson quotes an employee of Shin Etsu
Handotai (SEH), the largest wafer maker: “5-6 years back the IC makers in Japan were
behind (U.S.). (SEH) had good contacts with customers in the U.S. We could know the
secrets. Japanese IC makers were eager to know and could (provide) some secrets. Japanese
(companies) very happy with our information” (p. 79).
208. See Remo Pellin, “Semiconductor Market,” in Department of Energy, Proceedings
of the Flat-Plate Solar Array Project Workshop on Low-Cost Poly silicon for Terrestial Pho¬
tovoltaic Solar-Cell Applications, DOE/JPL 1012-122 (Pasadena: Jet Propulsion Laboratory,
1985), p. 407. Wafer production is measured in millions of square inches of silicon. If only
semiconductor grade wafers are counted (and lower quality wafers used to make solar cells
excluded), the global market share of Japanese wafer producers stood at more than
40 percent in 1984. See Strategies Unlimited, Silicon Wafer Industry Assessment (Mountain
View, Calif., 1986), p. 124.
THE JAPANESE ASCENT IN SEMICONDUCTORS 107

Technical Cooperation among Japanese Manufacturers

The VLSI project was the first case in which a MITI-supported re¬
search association set up its own, independent joint laboratory to under¬
take research, and the first to mix within a common space researchers
from the different participating companies. Its success provided the im¬
petus for later initiatives to establish joint industry R&D facilities.
However, it has rightly been pointed out that perhaps 15 percent of
the project’s total budget was spent within the joint R&D laboratory,
with the remainder expended within the two company-run joint research
programs and the individual companies, under the direction of the VLSI
Technology Research Association.209 Actually, the share of the work car¬
ried out by the joint lab may have fallen below even this low number; as
table 2-6 suggests, personnel in the joint lab as a share of the project’s
total personnel dropped rapidly from 18 to 11 percent over the four-year
life of the project. Furthermore, the topics pursued within the joint lab
were specifically chosen to be “common” and “basic” technologies far
removed from actual commercial device development.210
Even within these constraints, organizational walls between different
research groups in the joint lab were erected to protect perceived pro¬
prietary interests: Hitachi, Fujitsu, and Toshiba, for example, were re¬
luctant to have their researchers work with other companies in the area
of lithography equipment because each of these companies had active
commercial efforts under way for these products. As a result, each of
these companies’ researchers were segregated within their own compet¬
ing lithography laboratory within the overall joint lab, joined by research¬
ers from NEC and Mitsubishi (who were not in direct competition) but
not the other companies.211 One breakdown of a group of patent appli¬
cations resulting from the project shows only 16 percent of these patents
involving researchers from more than one company.212 Given these facts,
one revisionist critique has argued that there was limited joint generation
and sharing of knowledge within the VLSI project.213

209. Fransman, The Market and Beyond, p. 80.

210. For an interesting analysis, see Soichiro Seki, “Intra-Industry R&D Consortium—
Lessons from Japanese Cases,” Brookings Institution, May 1992; on the choice of themes
for the joint laboratories, see Yasuo Tarui, IC no Hanashi (Tokyo: Nihan Hoso Shuppa
Kyokai, 1982), p. 144.
211. Fransman, The Market and Beyond, pp. 65-67.
212. Tarui, 1C no Hanashi, p. 149.
213. Fransman, The Market and Beyond, pp. 76-79.
 fO O 00 On

28.7
CN cn
ON nC
CN (N
NO oc o oc
CN r-

23.0
CN GO O ’—1
o r- o in oo tj- 00 CN
-H r-H ^H (N r- CN in

oo oo m

24.5
On r- oo

oo ri h O' o in o cn o o
NO m in

no
oo
I
^h cn o

23.3
NO

On oo no c o in © cn 00 cn
i—( r-H r-H CN r- CN in

22.1
r- CN On
^ (N) CO NO r- O c^ 00 CN CN
CN r—< r-H r-H NO cn
-O

Source: VLSI Technology Research Association, History of the VLSI Research Association (Tokyo 1991), pp. 39, 56, 62.
r- oo in o

20.3
in On no

in
CN
r-i <n in NO
NO
cn -h oo
<n
r-H

c
c
o in o NO 00 GO © CN cn

17.4
£ o o cn 'Tf 00 cn o r-H 00
<n r-H r-H r-H NO r—H r-H m m
$

a. CDL is Computer Development Laboratories; NTIS is NEC Toshiba Information Systems.

o' oo O O O ON o o
0

in in CN CN
«L N- N- N- N-
e
o
0

On O O O
o
o <N in
N-
in
N-
CN
"xf
CN
>2
^2

-c
O o o o
0

« 00
m
oo
cn
O r-
o

<u
fti

Co <n o o o O O O o
0

cn
r-H r-H r-H r-H
CN CN NO NO
L

£ .£

£
J9
o T3
T3
<D
©
s«. <U <D
<3
-s; O
O
o
CO cj
O O co
to
to
to 03
NO C3 CO
I JD<& C a
-o 03
<N cS C
—
NTIS

Q- CL
JD ■J P jo E
P
1 1
3 C
S
h
£ o
aj o U rG cn E
o p Q
Q o PJ o H o
I u
£ ‘o
UfclSu Z H Z u
THE JAPANESE ASCENT IN SEMICONDUCTORS 109

But a broader view of matters belies this simplistic conclusion. Joint

patents may be useful in tracing those formal joint R&D efforts that
produce patents, but there is no reason to believe they are a particularly
useful indicator of information sharing. Indeed, the brief sketch earlier
in this chapter of the early history of Japanese semiconductor develop¬
ment stressed how useful open conferences, conversations, and obser¬
vations on visits to labs were in obtaining initial access to technical
information on American semiconductor technology.
Furthermore, sharing of technical information between participants in
MITI-sponsored projects has always been Japanese policy and practice,
even before the VLSI project’s experiment with a large joint R&D facil¬
ity. The form of organization before the VLSI program was more of a
hub-and-spoke system, with either the ETL or a research association
contracting pieces of the program out to individual companies, then
taking pains to disseminate the results through meetings and conferences
among the participants. A characteristic feature of the system was that
research association members were required to share research results
with other members.214
The results of ETL’s internal research through the early 1960s were
generally disseminated through technical workshops in which companies
participated. With the advent of the large-scale project system in the
mid-1960s, however, and coordinated contract research undertaken by
private companies, the ETL began to take a more direct role in planning,
coordinating, and disseminating contracted private research for projects
in which it was involved.215 For example, Tarui, describing MITI’s first
grants for IC development in 1965, notes that “the research accomplished
in each project will be available to all companies.”
Producers are also being encouraged by the Microcircuit Technical Com¬
mittee of the Japan Electronic Industry Development Association. Chair¬
man of the committee is Noboru Takagi, a professor at the University of
Tokyo; Tsuneo Momota of MITI’s Electrotechnical Laboratory is the vice
chairman. Because of their affiliations, they can disseminate technical in¬
formation widely.216

Efforts to disseminate technical information among firms were also

clearly on the agenda of the VLSI project, even when joint R&D was

214. See Akira Goto and Ryuhei Wakasugi, “Technology Policy,” in Komiya and others,
Industrial Policy of Japan, pp. 198-99.
215. Ishii, “Research and Development on Information Processing Technology.”
216. Tarui, “Japan Seeks Its Own Route,” pp. 90-91.
110 THE JAPANESE ASCENT IN SEMICONDUCTORS

not involved. Martin Fransman, principal exponent of the revisionist

thesis, himself inadvertently makes this point when he quotes the reply
of one VLSI participant to his query about the flow of information be¬
tween the two private labs (the Fujitsu-Hitachi-Mitsubishi and NEC-
Toshiba groupings): “they had open conferences and produced thick
reports. But of course the reports did not contain all the relevant infor¬
mation about know-how.”217 It is a long leap indeed to conclude that not
sharing all technical information means that such sharing was not impor¬
tant for the companies involved.
Moreover, the formal walls constructed between research groups
within the joint VLSI laboratory were ultimately, participants have sug¬
gested, broken down. Parties, social occasions, and evening drinking
sessions were used to break down the formal barriers between areas, and
considerable information and ideas did ultimately flow back and forth.218
The work carried out in the joint laboratory seems to have been
productive. One study identifies major results of R&D carried out in this
lab in the areas of electron-beam (E-beam) writing systems, E-beam
resists, silicon wafer production, silicon crystal micro defect control, LSI
testing equipment, plasma etchers, E-beam mask testers, and lithography
equipment. The companies associated with developing and testing this
equipment are a veritable who’s who of today’s semiconductor equipment
and materials industry in Japan: in E-beam writing systems, Fujitsu,
NEC, Fuji Electronic Chemicals, Hitachi, Toshiba, Toshiba Engineering,
Akashi Seisaku, Asia Seisaku, and Nihon Business Automation; in E-
beam resists, Tor ay and Tokyo Oka Kogyo; in wafer production and micro
defect control, a consortium made up of Shin Etsu, Komatsu, and Osaka
Titanium; in LSI testers, Ando Electronics; in plasma etchers, NEC
Anelva; in E-beam mask testers, Hitachi and Hitachi Software Engi¬
neering; in lithography equipment, steppers from Canon and Nikon, an
ultraviolet projection aligner from Canon, and X-ray components from
Rigaku Electronics.219
The Canon and Nikon equipment projects were particularly significant
activities. In addition to receiving R&D subsidies for the next-generation

217. Fransman, The Market and Beyond, p. 81. Fransman also describes some research
actually conducted jointly by the two laboratory groupings, but he minimizes its impor¬
tance.
218. NHK, Electronics-Based Nation, vol. 4, p. 358; Sakakibara, “Organization and
Innovation,” pp. 21-22; Sigurdson, Industry and State Partnership, pp. 50—51; and author’s
interview with Yasuo Tarui, March 1992.
219. Inoue, “The VLSI Research Association,” p. 68.
THE JAPANESE ASCENT IN SEMICONDUCTORS 111

projection aligner, 20 to 25 percent of the R&D funding for a prototype

stepper—and informal assurances that the system would be purchased
by VLSI project members if it met published specifications—were re¬
portedly received from the VLSI joint lab. Nikon reportedly received
subsidies from the VLSI joint lab to cover its effort to reverse-engineer,
and improve upon, American company GCA’s leading-edge stepper,
along with informal assurances that project members would buy the new
system.220
Although the VLSI project was the first large effort to bring companies
together to conduct joint research, it had some precedents. MITTs efforts
to cartelize the computer industry in the early 1970s did bring company
researchers together in joint R&D projects, often under the auspices of
the research associations. Bitter rivals Fujitsu and Hitachi did join to¬
gether on R&D through their Nippon Peripherals joint venture, which
was a member of and received funding from the research association
administering that piece of the 3.5 generation subsidy.221 NEC Toshiba
Information Systems also set up a joint research system that predated
the VLSI project; one of its purposes was R&D into ICs. Joint research
by NEC and Toshiba through NTIS on the VLSI project is actually known
to have produced at least one important piece of technology, an electron
beam machine.222 Finally, the IC subsidies of 1973-74 were specifically
structured to create incentives for joint R&D among the participating
companies, although the results were at best mixed.
Perhaps the ultimate test of the utility of the coordinated structure of
the VLSI project for the companies involved, however, was the market.
By 1979 objections raised to the large scale of the subsidy by U.S. in¬
dustry and government had made it politically difficult to continue the
government-subsidized project, and the joint research lab was shut
down.223 The industry, however, continued a “private sector” edition of

220. Ross A. Young, Silicon Sumo: V.S.-Japan Competition and Industrial Policy in
the Semiconductor Equipment Industry (Austin: University of Texas IC2 Institute, 1994),
pp. 185-88. Propelled by these new systems, Canon and Nikon turned the tables on industry
leader GCA. From a 10 percent share of the world stepper market in 1980, they jumped to
80 percent by the end of the decade. U.S. stepper makers’ share dropped to 10 percent,
and GCA, which invented the stepper, went out of business in 1992. Young, p. 99.
221. In fact, Fransman, The Market and Beyond, p. 52, notes how important it was for
“cross-fertilization” of ideas for engineers from the two rivals to sit down and work together
on research in the joint venture.
222. Anchordoguy, Computers Inc., p. 144.
223. See “VLSI Research Group Is Going to Halt Activities at End of March,” Japan
112 THE JAPANESE ASCENT IN SEMICONDUCTORS

Figure 2-11. Public and Private Funding of VLSI R&D Association,

1976-86

Billions of yen

the VLSI project for another seven years, funded entirely out of its own
monies, to continue development and commercialization of semiconduc¬
tor manufacturing technologies first explored under the subsidized pro¬
gram.224 Figure 2-11 shows the scale of the original and the additional
funding (the latter amounted to about 60 billion yen, or 50 percent more
than the companies’ contribution to the original program).225 Companies
clearly must have found their joint research through CDL and NTIS, and

Economic Journal (March 1980, excerpted in SIA, Japanese Protection)-, and Sakakibara,
“Organization and Innovation,” pp. 12-14.
224. See “Super LSI Volume Production 3-Year Plan, 10 Billion Yen Invested in First
Year,” Nihon Keizai Shimbun, July 31, 1980 (excerpted in SIA, Japanese Protection)-, and
unpublished slides for presentation of Yoshio Nishi, “VLSI Technology Perspective,” World
Bank, China Electronics Seminar, October 22, 1987.
225. The VLSI project’s government funding was actually given in the form of a “con¬
ditioned loan” with an obligation to pay back the subsidy if profits were earned on the
technologies and products developed. From 1983 to 1987 the companies actually paid back
some 8.5 billion yen to the government (out of a MITI subsidy of 28.6 billion yen). Inoue,
“The VLSI Research Association,” p. 63.
THE JAPANESE ASCENT IN SEMICONDUCTORS 113

the coordination and information sharing structured through the VLSI

Technology Research Association, to have been productive, to continue
an effort on this scale. The CDL and NTIS labs continued to be active
in joint research in other areas as well.226
The utility of VLSI-style joint research clearly seems to have been
proven to both government and industry. Immediately after the VLSI
joint lab was disbanded, two similar collaborative R&D labs were estab¬
lished as part of new government-industry R&D initiatives.227 Joint re¬
search labs have since become a familiar element of technology policy in
Japan.

The Continuing Role of Government

Even if no particular merit is attributed to the joint and coordinated
structure of the VLSI projects, it is clear that the resources associated
with the MITI and NTT VLSI programs alone must have had an enor¬
mous impact on the Japanese semiconductor industry. Table 2-5 shows
how resources allocated by the Japanese government and NTT to R&D
into ICs went from minuscule amounts to well over 30 percent of total
Japanese investments by the late 1970s.
In the 1980s political pressures from abroad forced MITI to cut back
the scale of its subsidies to R&D investments in ICs. Table 2-7 shows the
considerably smaller scale of the major MITI technology subsidies of the
1980s, particularly in comparison with private sector investments in
R&D. As the Japanese industry grew larger and stronger, however, mak¬
ing its presence felt in global markets, this assistance became less critical.
The crisis of the 1970s had passed, and the Japanese semiconductor
producers, using their newly honed manufacturing skills, were to become
the largest global suppliers of commodity memory chips, a product where
manufacturing cost was virtually the only thing that mattered.
Instead, the biggest subsidies to the Japanese semiconductor industry
in the 1980s came from NTT, the Japanese telephone monopoly. NTT’s
first VLSI project in 1975 had differed from that of MITI in emphasis,

226. CDL and NTIS were participants in the MITI-sponsored Next Generation Basic
Computer Technology program, which carried out research in the areas of software, com¬
puter processing of language, and computer peripherals (but not ICs).
227. See Izuo Hayashi, Masahiro Hiano, and Yoshifumi Katayama, “Collaborative
Semiconductor Research in Japan,” Proceedings of the IEEE, vol. 77 (September 1989),
pp. 1431-32.
 ON CO 00 _
03 vO VO co
cd cd 03 °o o (N in 00 m
d d d (N o o o d o CO o

(N 00 _: 00 CO (N (N
■^t cd cd r- 3 rxj o cd o NO in
d d (N c d d o (N o

in 03 03 03 03 03 cd cd
•^f d p: co
On C © d d d d d d d
I

On

cd CO t—' r- 03 03 03 03 cd 03 cd
tt co oo
d d d d d d d d
£
'-•-A
o
03 00 NO CO •
03 r- (N o 03 03 03 (N ^ cd cd
d d d d O c d d
2
• »*y
c
o
•*■<» o cd
O 03 cd 03 03 T-H cd cd
<L>
*^» d d d d d d d
O c
o

ON Tt 03
03 oo m 03 03 03 03 cd 03 cd
d d d d d d d d d
^3
>-*
Q
*^««»
r- on 03 03 03 03 cd cd 03 03
ft: tj- r- iT)
»3 d d d d d d d d
s
a
v
®o
o
co m
<N 00
Tt 03 03 03 cd cd cd 03
ft, co
d d d d d d d
Q
°8
a;
s T3 ON
co
r- in
in Tf 03 03 cd 03 cd 03 cd 03
o <L>
d d d d d d d d
<0 *3
s^ 0)
a Cu

■3 “
c 'is
<u C. u
Cv <D C/5
OJ
C/5
£
<D
U |1 I C
o
Lq o o C/5 C
cc — Q. U g Wh
o § >, 3 <D <D -3
03
. C/5
<L> - 1/3 Cl td £ c/5 > -a “
C s o
I*-*- C/5
^o vs 5s s« <n
op ^
'O C y
o
■C c <u
cu
”
Cl
0)
^
^ -S
c o gc g - <u ■4—> <D
!- O
CL Z3 •5b w o 5/5 • W) u y T3 ~ 1 2 2
Qc/> 03 c jf <c
0) o g ’S c « « a Ed.-
^ <U o o '5 - > 0>3c^
S 79 >%.s £
1— ■2 g
^ 03-3
5/5 > _C
(u o
<=« P 03 c/5 W-
■§11
r--' <u c -c g *C "c O (Ui > ^ cd
i Ot * a..<£ o = o cd e -a 8 II
■g 2 I
-n «« (U
<N — 03 C/5 'o
u £ a cd C
03 O -C 00 s o i 8 C ^
_0> c o 00 CL U s cd
s -a o C
XJ 3 3 o
■lo GO o -gc j s Z
e2 CQ Z
 r- o <N co O 00 vC (—>
vO (N O o CO On in o
ri
O'
9-
o
c
E
cJ
'<•
i
o (N o cd — U 3
1-H CO O' ^
'd'
f' o S

—' c2 J=
O (N £ -u
cd co O (N oo ^d" O r-
Td- cf O r- CO CO ON On
d in
^d- S 2W
dll
O'
1
Cd 03 cd
co o <N oo o (N oo
o r- CO in ON _
d d d co . - o U)
O c
Vj *r
i C; O
*3 g

dg9
03 03 cd
co O (N 00 o VO
TJ- o CO in in
d d d
in
CO

03 cd cd
co o <N 00 o in
^d- o r- CO 00 t>
d d d 00 o vd I P SJ
00 CNl I ON
(N on
<—•
vo S'
Q- \£5
CO a. *

cd os cd
co O (N 00 o co
rj- o vo "d- 00
d d d C1-— sr
r-
<N 'C"5

t- ^
O ^
03 03 cd
co o
o
CN ^ 00 av
t— <V
. >s “>
d d d a5 ^
C 1^ o
VO
(N a.^ >NfO
CN
oo S 5
ON ^ O TO
- bS M
9?.!; 5
q ° o r- 3 N! ^ Q
03 03 cd cd cd cd cd oo o CO
*dt w
d d d d d d c
CO
<N s.i'5.-
3 co
kj «s C33
O On O '« 1)
03 03 cd cd cd cd cd
VO
o On O | "fN>s 1/93 3
d d d d d d d in
ON cd <L < fflco
f2
3< ?9§ .22
Q. > - t
a. Average annual funding levels for multiyear projects.

K S O w
o 3
G O >, O H
b. OEIC is optoelectronic integrated circuit project.

<U
in c g 0-
os cd U cd cd cd cd On o o. u =
d d G d d d d VO <N 3 C E 3
2 o n) «J
_o VO
o V -a E
in fli W t.
o £7 > w O
£
o u.u T3
•- a
2 T3
c
3 U ® t5
CL
-C
CJ Q £ “X5 ^ IU
t-M
cd tU 0)
G
<D
o>
C/5 G u PC
a q 3 00
e
5 a D o 0> *c
PC G
O "5
c -v > V
u g 4
C ^
O — NO 2
G cS g e ’P o cd cd OX) G T3 iz H •
O
£ s 2 3
^ £ u cj 03 cd PC U S
S3 S
CL Lh
at
u c/5 _ .
P “2 E T3 £ c E js •« cd CJ
’> T3 T c O O a, s_✓
G G .> c c 5 o C/5 Cd
r O G OX) c S> S l;
<D 5 3 p bG (j o U
G “ I
G
O
_ _
H-* d -g oo c u- OX) 5 u S5 m
.22 cd cd C “ c G
0)
£
C/<u5 £
<U ~ V 43
O ■6 <u I £. | «
T3 .£
£ <u
C/5
G G "3
« g • 3

<u
u
s- O
■*->
O <2
0
<u
v a -
o
s « o 2 «5 G
G
c2 -o G
G G
o <L> e E E H o a M—<
X 43
c/5cd <D
^ o jr c c
W PH
>
U
G
co O < <L) Q.
P
!« 1- •§ S
GO CQ Z s S aaS
116 THE JAPANESE ASCENT IN SEMICONDUCTORS

focusing more on devices and device structures, and less on the manu¬
facturing process (although, as described earlier, some important pieces
of manufacturing equipment—testers and etching equipment—came out
of the program).228 The scale of the overall program was substantially
smaller than MITI’s, however, since firms were not required to match
NTT’s investment in R&D.
In 1978 NTT began a second three-year VLSI program, this time
focusing on development of a 256K DRAM, and like the first program
budgeted at 20 billion yen.229 By early 1980 NTT had announced a 256K
DRAM prototype.230 Yet a third NTT VLSI program was begun in the
fall of 1981, this time targeting 1M DRAMs and computer-aided design
systems for use in designing future generations of ICs.231 In this third
project, membership was extended beyond the three core members—
Fujitsu, Hitachi, and NEC—to Toshiba. Technology transfer from and
procurement of 256K DRAMs by NTT gave the Japanese semiconductor
makers an early boost in 256K DRAM production in the mid-1980s.232
NTT spending on R&D involves a substantial component of joint
research with other companies. Table 2-8, based on data reported by
Kodama, shows that about 28 percent of patent applications jointly filed
by NTT and its suppliers over the 1980-86 period involved two or more
of NTT’s suppliers.
By the mid-1980s NTT, through its relationships with its suppliers, had
quietly become the largest source of quasi-public subsidies to the Japa¬
nese semiconductor industry. An AT&T semiconductor technologist has
estimated, on the basis of his visits to NTT in the mid-1980s, that about
22 percent of all NTT R&D staff were working on semiconductor tech¬
nology in 1985, suggesting an internal 1988 R&D expenditure of about

228. Author’s interview with Yasuo Tarui, March 1992; Sigurdson, Industry and State
Partnership, p. 58; and unpublished slide from Nishi presentation, 1987.
229. “Responding to Criticism of the Closed Nature, Complete Opening of Super LSI
Patents of NTT,” Nihon Keizai Sangyo, February 8, 1980 (excerpted in SIA, Japanese
Protection).
230. See note 190.
231. “NTT Has Accepted Toshiba to Join VLSI Project,” Nikkei Sangyo Shimbun,
October 4, 1982, p. 1.
232. In 1982 NTT began procuring 256K DRAMs in production quantities (hundreds
of thousands) from Hitachi, NEC, and Fujitsu, using design and manufacturing technology
transferred to these firms at no cost. See Jack Robertson, “Japan Cos. Get NTT’s 256K
Skills,” Electronic News, October 11, 1982, p. 8; and Nikkei Shimbun, August 26, 1982,
p. 8.
THE JAPANESE ASCENT IN SEMICONDUCTORS 117

Table 2-8. Joint Patent Applications with NTT, by Number of

Participating Companies, 1980-86
Percent unless otherwise specified

Number of manufacturers involved

(not including NTT itself)
Joint patent
Year applications One Two Three Four

1980 627 72.9 8.0 1.1 17.8

1981 669 63.7 8.8 3.0 24.5
1982 652 73.5 5.1 1.2 20.2
1983 655 74.4 7.5 3.8 14.4
1984 540 69.3 5.2 0.9 24.6
1985 375 72.0 4.8 4.0 19.2
1986 655 81.2 8.1 3.4 7.3
Source: Fumio Kodama, Analyzing Japanese High Technologies: The Techno-paradigm Shift (London: Pinter
Publishers, 1991), pp. 98-99. An error in the published data for 1980 has been corrected.

36 billion yen.233 Table 2-9 estimates Japanese subsidies to semiconductor

R&D in 1988 and shows that they amounted to about 18 percent of all
public and private spending on semiconductor-related R&D. This ratio
is remarkably similar to that resulting from an analogous calculation for
the United States in the late 1980s and early 1990s.234 Although MITI
spending was no longer dominant, it was still significant (about one-
eighth of all nonprivate spending).
NTT’s relationship with private industry was to undergo a significant
restructuring in the mid-1980s as a result of the enterprise’s privatization

233. David L. Carter, “Estimates of Japanese Government Semiconductor R&D

Spending, Analysis,” unpublished background material furnished to the National Advisory
Committee on Semiconductors, 1990. These numbers do not appear to include external
semiconductor research by suppliers funded through development contracts.
234. In 1987 U.S. government spending on semiconductor-related R&D amounted to
about 20 percent of the national total. Government spending was $454 million, whereas
private industry R&D spending was about $1.8 billion. The latter is estimated to have been
12.9 percent of U.S. company sales of $13.6 billion. The R&D-to-sales ratio is an SIA
estimate; U.S. company sales are an estimate from Integrated Circuit Engneering (ICE).
Government spending is estimated in Philip Webre, The Benefits and Risks of Federal
Funding for Sematech (Congressional Budget Office, 1987), p. 60.
In 1992 U.S. Defense Department semiconductor-related spending is estimated to have
been about 18 percent of the national total: DOD spending was about $700 million, and
private spending about $3.2 billion. The departments of Energy and Defense are estimated
to have appropriated $686 million for semiconductor-related R&D in fiscal 1992; see De¬
partment of Defense, DOD Key Technologies Plan (July 1992), p. 5-28.
See generally Semiconductor Industry Association, 1995 Semiconductor Industry As¬
sociation Databook: Review of Global and U.S. Semiconductor Competitive Trends, 1978-
1994 (San Jose, Calif.: 1995), p. 40.
118 THE JAPANESE ASCENT IN SEMICONDUCTORS

Table 2-9. Government and NTT Semiconductor-Related R&D Funding

in Relation to Private Japanese Semiconductor R&D Funding, 1988
Agency or program Billions of yen

MITI R&D programs related to microelectronic materials and devices

National R&D projects
High-speed computer for scientific and technological uses 2.78
New function elements 1.21
Superconducting materials and devices 1.06

Industrial research support

Light-activated materials 0.23
Basic Technology Research Promotion Center
Sortech 1.43
OEICa 1.00
Coherent optical communication 0.72
Amorphous magnetic materials and electronic devices 0.38

All MITI programs 8.81

NTT R&D on semiconductors 36.00

Ministry of Education funding 9.00

Science and Technology Agency funding 16.90

Total government and NTT semiconductor-related R&D funding 70.71

Japanese industry IC R&D funding 320.70

Government and NTT funding as a share of all semiconductor-

related R&D (percent) 18.07
Sources: Japan Electronics Almanac 1989, pp. 68, 76, 77; Electronic Industry Almanac 1989 (Tokyo: Dempa
Publications, 1989), p. 185; AT&T Bell Laboratories-Electronics Technology Planning Center, memorandum, Feb¬
ruary 1990, pp. 3-5; and MITI unpublished twelve-company survey,
a. OEIC is optoelectronic integrated circuit project.

in 1984. From the privatized NTT emerged a major new, publicly funded
institution investing in information technology. A public battle between
MITI and the Ministry of Posts and Telecommunications in 1984 over
control of Japanese investments in information technology research was
resolved with the establishment of a jointly administered Basic Technol¬
ogy Research Promotion Center in 1985.235 The huge annual dividends
from the government’s remaining one-third share of the equity of the
privatized NTT are funneled into this center, which in turn funds up to
70 percent of joint R&D ventures with private industry. From 1986 to
1990 close to a billion dollars (100 billion yen) of NTT-derived funds were

235. See Johnson, “MITI, MPT, and the Telecom Wars,” pp. 227-30; Teruyuki Inoue,
NTT: A Giant Facing Competition and Division in the Information Age, 4th ed. (Tokyo:
Otsuki Shoten, 1992), pp. 108-14 (in Japanese).
THE JAPANESE ASCENT IN SEMICONDUCTORS 119

plowed into this center. The single largest recipient of these subsidies is
the Advanced Telecommunications Research Institute, with a core re¬
search staff mostly made up of personnel sent from NTT. The large
private corporations that participate in the Institute’s projects and fund
the balance of its budget also send researchers to work on these research
efforts, who then serve as the conduit for technology transfer back into
their home organizations.236
MITI’s stance toward the Japanese semiconductor industry also
changed dramatically in the 1980s. The agency lifted formal controls on
semiconductor imports and foreign investment for the most part by 1975,
but some signs that administrative guidance and other barely visible
measures were used to protect domestic producers lingered on through
the remainder of that decade. As late as 1979, for example, American
chip producers still publicly described alleged incidents in which MITI
had forced Japanese customers to cancel orders with American compa¬
nies and switch to Japanese suppliers.237 And one could still see the old
attitudes at work on occasion, even in the 1980s. Perhaps the clearest
example was a MITI initiative, in the early and mid-1980s, to reduce
dependency on foreign supplies of the high-purity silicon used to produce
semiconductors.

Dependence in Silicon

In 1978, working through its so-called Sunshine Project, MITI set up

a special study group within the Japan Electrical Manufacturers Associ¬
ation to examine the low-cost manufacture of polycrystalline silicon
(poly), the raw material used to make the silicon wafers on which ICs are

236. Inoue, NTT: A Giant Facing Competition and Division, pp. 108-14.
237. “Dr. Hogan tells of a $7 million order Fairchild got recently from an unnamed
Japanese customer. ‘Our people in Japan gave a party; they were delighted. But one week
later the Japanese customer was on the phone literally crying. MITI, he told us, had forced
the order to go to Hitachi, even though his people felt we were providing a better product.’ ”
John Reason, “Japan’s Electronics Markets: A Pair of Views,” IEEE Spectrum, vol. 16
(June 1979), p. 52.
“Some reported experience implied Japanese government involvement. It was reported
that sales have been cancelled after application to MITI for an import license, and that this
was due to telephone calls received by prospective purchasers asking why they were im¬
porting when essentially comparable domestic products were available.” U.S. International
Trade Commission, Competitive Factors Influencing World Trade in Integrated Circuits
(November 1979), p. 59.
120 THE JAPANESE ASCENT IN SEMICONDUCTORS

etched.238 The study group examined technologies for producing the

higher purity poly used in manufacturing wafers used in semiconductor
production, as well as the lower grade poly used to make solar cells. The
basic problem for all Japanese producers was that silicon production was
highly electricity-intensive, and electricity in Japan was very expensive.
In October 1980 the New Energy Development Organization (NEDO)
was formed as part of MITI’s Sunshine Project, to promote technological
development. One of its first projects was to develop energy-efficient
technology for low-cost polycrystalline silicon for solar cells. The R&D
work was divided between the two largest Japanese poly producers,
which also happened to be the largest Japanese makers of silicon wafers
used in IC fabrication: Shin Etsu Handotai (SEH) and Osaka Titanium
Company (OTC). (These same two firms worked closely with the chip
producers involved in MITI’s VLSI project; see figure 2-10.) OTC was
to work on low-cost production of a precursor chemical, and SEH to
develop the process for making this material into polycrystalline silicon.239
Although the OTC-SEH work on low-cost poly production was explicitly
aimed at solar grade material, there was a clear intention to apply the
work to semiconductor grade material if possible.240 The very same year
that the NEDO silicon project was set up, another Japanese chemical
producer, Tokuyama Soda, also began research into poly production.241
The extraordinarily rapid rate of growth in the Japanese semiconduc¬
tor market was apparent to all by the early 1980s, and foreign silicon
producers responded by increasing their Japanese presence. Japanese
semiconductor production was growing rapidly, and with it imports of
poly. By mid-1982 it was clear that, for the first time in the history of the
Japanese semiconductor industry, poly imports were going to exceed do¬
mestic production. Some in Japan lamented the apparent inability of
Japanese firms to be competitive, given the high cost of electricity, and

238. See Toshio Noda, “Processes and Process Developments in Japan,” in Proceedings
of the Flat-Plate Solar Array Project Workshop on Low-Cost Polysilicon for Terrestrial
Photovoltaic Solar-Cell Applications, JPL publication 86-11 (Pasadena, Calif.: Jet Propul¬
sion Laboratory, February 1986), p. 213.
239. Noda, “Processes and Process Developments,” p. 214.
240. For example, to the question, “Do you think it’s possible to use this process to
produce electronic-grade silicon?” Noda of OTC replied, “My understanding is that every¬
one here has a strong interest to see our process applied to produce electronic-grade
material. We would like to see that happen, too.” Noda, “Processes and Process Devel¬
opments,” p. 231.
241. See “Tokuyama Soda’s Polysilicon Business Dept.: Offense with Polycrystalline
Silicon,” Nikkei Sangyo Shimbun, December 18, 1984, p. 6.
THE JAPANESE ASCENT IN SEMICONDUCTORS 121

went so far as to accuse foreign poly makers of dumping.242 By the end

of 1982 it was clear that imports of poly (roughly 80 percent of which
were supplied by the German producer Wacker) had doubled, and Jap¬
anese silicon industry leaders had begun to perceive “import depen¬
dence” as a serious problem.243
Various manifestations of that concern soon appeared. At the urging
of the Japanese government, OTC decided to more than double its poly
production capacity.244 Tokuyama Soda was reported in late 1982 to have
made a decision to begin producing poly by 1984.245 By early 1983 it had
been reported in the Japanese trade press that financial support from the
Japanese government, in the form of low-interest loans from the Japan
Development Bank, were being made available to support increased poly
production.246 In the spring of 1983 MITI had also formed an industry
study group to examine the silicon industry’s “problem” and recommend
action; in discussion in the Japanese trade press of the issues to be
considered by the study group, the problem of import dependence is
given top billing, as “problem number one” of the silicon industry.247
The overriding theme of the study group was the rising share of im¬
ports in Japanese poly consumption. Japan’s semiconductor devices and
materials industries were clearly concerned about dependence, as imports
climbed past half of consumption.248 By early 1984 the major recommen-

242. Raru Metaru Nyuzu [Rare Metal News\, July 1, 1982.

243. “Polycrystal Silicon—Imports Will Surpass Domestic Production Next Year,” Rare
Metal News, December 1, 1982.
244. See Nihon Keizai Shimbun, July 19, 1982; “Polycrystal Silicon—Imports Will
Surpass Domestic Production Next Year,” Rare Metal News, December 1,1982; E. Costogue
and R. Pellin, “Polycrystalline Silicon Material Availability and Market Pricing Outlook
Study for 1980 to 88: January 1983 Update,” DOE/JPL-1012-79A (Pasadena: Jet Propulsion
Laboratory, February 1983), p. 14. Pellin recalls that in late 1982 or early 1983, when he
was working as a consultant to the company, officials at OTC told him that “an understand¬
ing between MITI and the Osaka Titanium Manufacturing Company has been reached
under which Osaka will increase its plant capacity to 800 annual metric tons in 1985 and to
1800 metric tons by 1990” (p. 13); and Remo Pellin, personal communication to the author,
1989.
245. “Polycrystal Silicon—Imports Will Surpass Domestic Production Next Year.” This
was also reported in Costogue and Pellin, “Polycrystalline Silicon Material Availability,”
p. 14.
246. On the provision of JDB loans for new silicon capacity see Rare Metal News,
March 8, 1983; Nikkei Sangyo Shimbun, December 12, 1983; Rare Metal News, October
24, 1984.
247. See Rare Metal News, March 8, 1983.
248. Kinzoku Jihyo [Metal Industry Review] no. 1180 (April 25,1984), pp. 9-10, stresses
the fact that imports had reached the magic “50 percent” ratio as a cause for alarm.
122 THE JAPANESE ASCENT IN SEMICONDUCTORS

dations of the study group report were being implemented, before the
report was even finished, and a full fifteen months before the final report
was published.249 These recommendations included expansion of domes¬
tic production, encouragement of entry by new Japanese producers, ne¬
gotiation of long-term contracts with foreign suppliers, direct investment
in foreign producers, and construction of foreign plants to take advantage
of lower electricity costs. Most of these recommendations were soon
translated into action.
In August 1984 it was announced that poly producer Hi-Silicon (a
joint venture between OTC and Japan Silicon) would double its annual
poly capacity to 1,080 metric tons, and that Tokuyama would also expand
its capacity to 1,000 metric tons. Shortly thereafter it became known in
the Japanese trade press that both Tokuyama and Hi-Silicon were receiv¬
ing subsidized industrial technology promotion loans from the JDB in
connection with these expansions.250 Other Japanese silicon producers
also scheduled capacity expansions in 1984. These expansions continued
even after the silicon market began to turn down in early 1985.
Aside from Tokuyama Soda’s pre-study group entry there was a sub¬
stantial wave of new investment in the silicon business on the part of
Japanese steel companies after the release of this report. Nippon Kokkan
announced its intention to enter the poly business in July 1985.251 The
next month Kawasaki Steel and LSI Logic announced a joint venture to
manufacture wafers.252 Nittetsu Electronics (owned by Nippon Steel) was
willing to talk about its entry into the wafer business, and a technical
link to Hitachi, in the second half of 1985.253
In addition, there was a substantial increase in foreign acquisitions
and investment over this period. SEH and Mitsubishi Metals purchased
substantial equity shares in the U.S. silicon producer Hemlock in 1984,
while the silicon study group was still meeting. Nippon Kokkan pur¬
chased Great Western Silicon from GE in 1985, shortly after the release

249. MITI’s study group was named the High Purity Silicon Issues Study Group (Ko-
jundo Shirikon Mondai Kenkyukai). Its Research Report [Chosa Hokokusho] was published
by the Japan Society of Newer Metals Association, a Tokyo industry association, in March
1985.
250. Rare Metal News, October 24, 1984.
251. See Nihon Keizai Shimbun, July 17, 1985.
252. Avra Wing, “See Japan Challenge to U.S. Wafer Makers,” Electronic News, July
8, 1985, p. 14.
253. See “International Report: Japan,” Solid State Technology, vol. 28 (September
1985), p. 20; and Rare Metal News, January 8, 1986.
THE JAPANESE ASCENT IN SEMICONDUCTORS 123

of the group’s report. Kawasaki Steel purchased wafer manufacturer

Siltec in 1985, in part to supply wafers to its joint venture with LSI Logic.
Mitsubishi Metals purchased wafer manufacturer NBK in 1986. OTC
acquired American epitaxial wafer manufacturer U.S. Semiconductor in
its 1986 fiscal year. SEH began production at a U.S. wafer factory in
April 1984, and it finished a plant in England in 1985. NKK announced
construction of a plant in Oregon in 1986, but that plan was later sus¬
pended when the market turned down.
These responses had a considerable effect: the import share of do¬
mestic silicon consumption abruptly dropped to well under 50 percent
after 1985, then leveled off. Since the import ratio had been rising in
the early 1980s, when the yen had been weak against other currencies
(which made foreign products less price competitive than Japanese
products) and then dropped decisively precisely when the yen strength¬
ened (making foreign products more price competitive), this was all
the more remarkable.254
Considerable activity also continued on the R&D front. By 1984 pro¬
totype processing equipment was in operation, and NEDO announced
that the construction of a plant with annual capacity of 100 metric tons
and using its energy-efficient technology was to begin in 1990.255 Komatsu
Denshi was also reported to have received a contract from NEDO to
improve production processes for materials used in silicon production.256
NEDO continued through the end of the decade to sponsor work
aimed at reducing electricity requirements in poly production. Although
the details were not spelled out, NEDO announced in May 1989 the
development of a new production technology that reduced the cost of
solar grade poly by 50 percent. Interestingly, the attitude toward imports
in at least that part of MITI has not changed much. Prominent play was
given to the technology’s possible role in eliminating dependence on
imports of foreign solar grade poly: “Japanese companies have, until
now, been totally dependent on imports for procurement of high-grade
silicon materials for solar cell production. NEDO’s new technology is

254. Indeed, a private antitrust suit (ultimately settled out of court) was brought by
American silicon producer Union Carbide in 1988. It alleged that Japanese silicon producers
had colluded within the MITI-sanctioned study group to exclude U.S. producers from the
Japanese market. See Louise Kehoe, “Japanese Silicon Suppliers Named in Antitrust Suit,”
Financial Times, October 5, 1988; and “Carbide’s Poly Charges Echo DRAM Flap,” Elec¬
tronic News, October 17, 1988.
255. See Noda, “Processes and Process Developments,” p. 231.
256. See Denryoku Jiji Tsushin, January 30, 1985.
124 THE JAPANESE ASCENT IN SEMICONDUCTORS

expected to pave the way for realizing 100% domestic production of solar
batteries, according to industry experts.”257
Aside from occasional episodes like these, however, the most obvious
and visible trappings of protection had clearly been lifted in semiconduc¬
tors. Even complaints about some of the less transparent forms of gov¬
ernment intervention, such as administrative guidance, were to drop
sharply as Japan turned its attentions fully toward global markets. As a
new decade of exports and international competition began in the 1980s,
it was a healthy and self-confident Japanese industry that emerged on the
world scene. On the other side of the Pacific the American industry faced
crisis and self-doubt, as for the first time it confronted a determined,
well-funded, and highly capable foreign competitor.

Summary

The portrait painted in this chapter of the development of the Japanese

semiconductor industry differs in some important respects from the sim¬
plest “Japan, Inc.” stereotypes of Japanese industrial policy. A great deal
of latent, and sometimes intense, competition within the Japanese sem¬
iconductor industry was an important element in its development. Con¬
tinuing economic battles among Japanese chip companies were moder¬
ated from time to time by the Japanese government, which intervened in
the name of the collective industry interest, typically in response to an
external (foreign) threat.
The first such threat was competition from imports and investment by
foreign companies, both of which were stoutly resisted from behind steep
protective barriers around the Japanese market. The government blocked
all such inroads and used all the weapons at its disposal to force foreign
firms to transfer their technology to Japanese partners on the best pos¬
sible terms for the latter. In the early 1960s, and again in the late 1960s
and early 1970s, initially successful forays by foreign chip producers into
the Japanese market were blocked and parried by the government at the
behest of the domestic industry. This theme—of government providing
the means to organize a sometimes intensely competitive and quarrel¬
some domestic industry into a cohesive united front in response to foreign

257. “NEDO Develops Technology to Produce Silicon Domestically,” Nikkei Sangyo

Shimbun, May 16, 1989, p. 13.
THE JAPANESE ASCENT IN SEMICONDUCTORS 125

pressures or threats—is pursued along other dimensions in the next

chapter.
The basically hostile environment toward foreign investment and im¬
ports visible in the 1950s and 1960s was softened considerably, in response
to foreign pressure, in the 1970s. By the early 1980s most formal protec¬
tionist barriers had disappeared. However, a long tradition of informal
“guidance” supplied to industry by the Japanese bureaucracy, coupled
with significant regulatory and administrative controls in the economy,
continues to provide the means for considerable, informal government
influence over private industrial behavior. Such influence was sometimes
visible in policies toward semiconductor materials infrastructure in the
early to mid-1980s, when reduction of dependence on foreign imports
became the objective of an informal, unannounced policy.
Japan’s chip producers are mainly internal divisions of larger Japanese
industrial conglomerates, and in the 1950s consumer electronics was the
major interest of industrial policy in electronics. In the early 1960s the
spotlight shifted to computers, and for the next thirty years industrial
policy in electronics was mainly organized around a campaign to build a
world-class computer industry in Japan. In the mid-1960s it became clear
that a competitive semiconductor industry might be needed as a basic
element of that plan, and in the 1970s, amid the VLSI revolution in the
United States, developing such capabilities for Japan became an essential
element of any plan to build an indigenous, self-sufficient computer in¬
dustry. At that point the policies that later pushed Japan into the forefront
of leading edge semiconductor manufacturing technology were put into
place. By the mid-1980s Japanese firms had clearly succeeded in semi¬
conductor manufacture (although an equivalent lead in chip design and
computer systems continued to elude them).
The first major policy initiatives in semiconductors in the 1960s fo¬
cused on subsidies to investments in plant and equipment by manufac¬
turers—little money was spent on R&D. In the 1970s the emphasis
shifted away from capital investment into R&D, particularly into invest¬
ments in narrowly focused applied R&D on semiconductor manufactur¬
ing equipment and infrastructure. In the mid-1970s a new model of co¬
operative industrial R&D was introduced and refined in the VLSI
projects, and further honed in later initiatives of the 1980s. Bowing to
foreign complaints, MITI reduced its relative levels of support for semi¬
conductor R&D, but a considerable part of the slack was picked up by
126 THE JAPANESE ASCENT IN SEMICONDUCTORS

others, particularly NTT, which continues to invest heavily in joint R&D

with the major Japanese chip producers. Today, the overall contribution
of public and quasi-public (that is, NTT) resources to national investment
in semiconductor technology is roughly equivalent to the relative burden
borne by government in the United States.
 CHAPTER THREE

The Genesis of an
American Trade Policy in
Semiconductors, 1959-84

The domestic semiconductor industry’s appeals to the U.S. govern¬

ment for assistance began with the first probes by Japanese transistor
producers into the international market in the late 1950s. These pleas
have generally been phrased in terms of a defensive policy against some
foreign threat. Over the years, however, the description of the threat and
of the behavior to be neutralized or counteracted by policy interventions
has evolved. The 1986 Semiconductor Trade Arrangement, in a sense,
reflected the growing force and sophistication of these strategic argu¬
ments for semiconductor trade policy. This chapter reviews just how ac¬
counts of the strategic challenge posed by foreign producers have evolved
over time, how new institutional mechanisms for dealing with these ques¬
tions were created in the early 1980s, and how a sharp increase in trade
frictions set the stage for the radical experiment in American trade pol¬
icy set in motion in 1986. One important theme that emerges from a de¬
tailed historical examination of U.S. trade policy in semiconductors is
that policy has often appeared to have been the cause and not just the
effect of changes in the competitive conduct of Japanese semiconductor
producers.

127
128 THE GENESIS OF AN AMERICAN TRADE POLICY

A Threat to National Security, 1959

Charges of sales of imports at unfairly low prices date back virtually
to the beginnings of the modern semiconductor industry. As discussed in
the last chapter, Japanese producers in the late 1950s made large invest¬
ments in transistor production facilities. In those days there was a fairly
sharp dividing line between relatively inexpensive, low-quality, low-
frequency transistors used in consumer electronics products, such as
transistor radios, which these same Japanese manufacturers were then
beginning to manufacture in large volumes, and the relatively expensive,
high-quality, high-frequency transistors used in high-performance de¬
fense and computer applications.
Early methods of transistor production generally made use of rela¬
tively primitive (by today’s standards) labor-intensive manufacturing
techniques, and open access to the fundamentals of recently developed
transistor technology, coupled with relatively low wage rates, made Jap¬
anese producers very competitive in the production of low-end semicon¬
ductors for use in consumer electronics. Much of the development effort
in the American industry, funded by the Department of Defense, was
oriented toward increasing performance to better serve high-end military
and computer systems markets; markets for lower performance consumer
electronic components received relatively little attention.
The first great Japanese export blitz in transistors and consumer elec¬
tronics at the end of the 1950s was to lead to the first serious clash between
the Japanese and American electronics industries. In 1959 a surge of low-
priced Japanese transistor imports first hit American markets. Whereas
in 1958 Japan had shipped 11,000 units worth a total of $7,000 to the
United States, in the first nine months of 1959 it was 1.8 million units
worth $1.1 million.1 In September 1959, citing national security concerns,
an American trade association—the Electronics Industries Association
(EIA)—petitioned the Office of Civil and Defense Mobilization
(OCDM, the executive branch predecessor to the Office of Emergency
Preparedness) to impose quotas on Japanese transistor imports.2 The
episode was notable not only as the first shot in a continuing political

1. See “Business Week Reports on: Semiconductors,” Business Week, March 26, 1960,
p. 113.
2. See “Import Study Nears Showdown,” Electronics, November 6, 1959, pp. 32-33;
“Electronics in Japan,” Electronics, May 27, 1960, pp. 99-100; and “Washington Rejects
Transistor Import Quota,” Electronics, June 8, 1962, p. 7.
THE GENESIS OF AN AMERICAN TRADE POLICY 129

battle over trade policy waged in Washington between American and

Japanese electronics companies, but also because it revealed patterns of
behavior that were to be repeated continuously over the following three
decades.
On the American side, it was a divided industry that petitioned the
government for relief. Numerous American companies were by then mar¬
keting products manufactured by Japanese partners: for example, in tran¬
sistor radios, Motorola and General Electric (GE) served as marketers
for Toshiba outside Japan, Emerson for Japan’s Standard Radio, and
RCA Victor and Channel Master for Sanyo. Other American electronics
companies, such as IBM, Remington Rand, and Ampex, controlled Jap¬
anese subsidiaries and were concerned about jeopardizing their sales in
the Japanese market. RCA, in addition to collecting royalty checks from
its Japanese licensees, negotiated a licensing arrangement with Elitachi
in computers in 1961, in order to sell within the protected local market.
A similar arrangement was negotiated between NEC and Honeywell in
1962. Still other American electronics companies had significant owner¬
ship stakes in Japanese electronics producers: GE in Toshiba, Westing-
house in Mitsubishi, ITT in NEC, and North American Philips’ Dutch
parent in Matsushita. As a trade publication noted in mid-1960, “the
‘haves’ effectively muzzled EIA (both havenots and haves being mem¬
bers) from making a firm presentment to OCDM—and so the subject
[of import quotas] quietly languishes.”3
Within a few years these tensions had worsened, and the American
electronics industry had openly split. On one side were U.S. electronic
components firms, who favored protection against consumer electronics
imports from Japan—in which the bulk of inexpensive Japanese compo¬
nents entering the U.S. market were embedded. On the other side of the
chasm sat American consumer electronics companies, who were con¬
cerned about protectionist precedents jeopardizing their future access to
foreign electronics markets, as well as their supply lines to Japanese
manufacturers. By 1968, for example, the EIA was testifying on both
sides of trade policy issues, with its components division in favor of
protection against electronics imports, and its consumer electronics di¬
vision opposed (with the exception of Sylvania, which supported protec¬
tion, and RCA, which took no position).4 This conflict between compo-

3. “Electronics in Japan,” p. 99.

4. Motorola, which produced both equipment and components, belonged to the con-
130 THE GENESIS OF AN AMERICAN TRADE POLICY

nents users and producers was to become a persistent theme in later

episodes of trade friction.
The appeal to national security concerns was another call to be echoed
over the years. In 1960, however, as was already noted, there was a
relatively clear dividing line between the types of transistors used in
consumer electronics and those used in defense (industrial and computer
applications). The Defense Department was at that time pouring sub¬
stantial resources into the development of advanced semiconductor tech¬
nology for use in military applications, and even directly subsidizing the
industry’s investment in new capacity, so it was not surprising that the
department’s Electronics Production Resource Agency submitted a study
to OCDM concluding that adequate capacity existed for both present
and future military transistor demand.* * * 5 Narrowly interpreted, defense
needs did not provide a particularly compelling rationale for action.
Geopolitical considerations beyond a narrow view of military produc¬
tion requirements also weighed in against a policy of protection. In the
early 1960s the United States was concerned to limit Japanese trade with
the Soviet Union and cement military cooperation with Japan. Building
a strong, friendly bulwark against Soviet and Chinese expansion in Asia
was a primary objective in U.S. policy toward Japan, and political and
security links were given conscious priority over economic relations.6
Even then, however, there was considerable concern over Japan’s
overtly protectionist trade and investment policies, and the United States
and the other industrialized countries exerted steady political pressure
on Japan to open up its markets. Pressure from the International Mon¬
etary Fund led in 1960 to promulgation of a program of Trade and
Exchange Liberalization Guidelines, but the Ministry of International
Trade and Industry (MITI) quickly withdrew electronics goods from the
list of items to be liberalized.
One concern voiced on the American side at this time—which was to
persist to the present day—was that use of protectionist threats against
Japanese exports would set a bad precedent, undermining progress to¬
ward more open national markets that had been achieved by the inter-

sumer electronics division and opposed protection. Foreign Trade and Tariff Proposals,
Pt. 8, Hearings before the House Committee on Ways and Means, 90 Cong. 2 sess. (Gov¬
ernment Printing Office, 1968), p. 3486.
5. “Import Study Nears Showdown,” p. 32.
6. See Dennis J. Encarnation and Mark Mason, “Neither MITI nor America: The
Political Economy of Capital Liberalization in Japan,” International Organization, vol. 44
(Winter 1990), p. 37, for references on this point.
THE GENESIS OF AN AMERICAN TRADE POLICY 131

national economic system since 1945. The delicate balance between the
threats of protection and the open trading system they were intended to
secure (a system placed in jeopardy if threats are transformed into actions
on too wide a scale) fueled policy debates in 1959, as it does today.
On the Japanese side, the response to the campaign for protection in
Washington was to blaze some trails that would become equally well trod
over time. A Japanese delegation traveled to Washington in 1959, where
it stressed that activities in consumer electronics created no threat to
U.S. defense, and that Japanese companies had “no immediate plans to
go after the markets for highly specialized transistors.”7 The delegation
also made clear its willingness to impose voluntary quotas or other ne¬
gotiated arrangements. In the late spring of 1960, in response to contin¬
uing frictions, MITI proceeded to impose quotas and floor prices on
transistor radios exported to the United States.8 The system was contin¬
ued in later years.9
The issue of third-country exports (evasion of quotas by shipping
goods to other countries for reexport to the United States) arose early
on and was dealt with administratively by MITI. By 1961 large quantities
of Japanese transistors were being exported to Hong Kong (clearly by
the same firms whose own exports of transistor radios to the United
States were being limited), where they were then assembled into transis¬
tor radios and exported to the United States. With sales of inexpensive
Japanese transistor radios in the United States undercut by large-scale
imports of even less expensive radios made in Hong Kong, MITI sus¬
pended transistor exports to Hong Kong in May 1962. A system of quan¬
titative limits on transistor exports to Hong Kong was set up a couple of
months later.10

7. “Import Study Nears Showdown,” p. 33.

8. Quotas for 1960 were set to equal a 20 percent increase over actual exports in 1958
and 1959. Penalties, including cancellation of the quota, were set for firms violating the
floor prices or evading controls by exporting to the United States and Canada through third
countries. See “Japanese Put Off Freeing Electronics Imports,” Electronics, July 8, 1960,
p. 11.
9. See “Japan Extends Quotas for Transistor Radios,” Electronics, January 6, 1961, p.
9; and “Japan Eases License Rules, May Cut Transistor Prices,” Electronics, May 26, 1961,
p. 9. By 1971 Japan had voluntary export controls on transistor radio shipments to twenty-
nine different countries. See U.S. Tariff Commission, Trade Barriers, vol. 5: Part II:
Nontariff Barriers (1974), p. 255.
10. MITI suspended transistor exports to Hong Kong and Okinawa on the grounds that
they were exporting transistor radios to the United States at a price $6 lower than that for
equivalent Japanese exports. This episode almost set off a trade war between Hong Kong
132 THE GENESIS OF AN AMERICAN TRADE POLICY

Such export cartels had by then been used for some time in postwar
Japan, initially in textiles, where they dated back to the early 1950s. By
the early 1960s a complex system of legal and administrative measures
permitted—indeed, often encouraged (since moderation of “excessive
competition” was an avowed objective of industrial policy)—the forma¬
tion of both domestic and export cartels, regulating pricing, capacity
investments, or production levels. By the early 1960s, when similar mea¬
sures covering transistor radios and transistors were put into place, the
Japanese government estimated that roughly 30 percent by value of its
exports to the United States were affected by some sort of quantitative
control, on domestic or export price or quantity.* 11
Ultimately, the U.S. producers’ petition to restrict Japanese transistor
imports was rejected in 1962, after a good two and a half years of dis¬
cussion, on relatively narrow national security grounds. The American
semiconductor industry, after an alarming slowdown in 1960-61, was
growing rapidly again, American producers were getting the vast bulk of
the defense business, and capacity seemed adequate to meet any future
surges in demand.12 Although some U.S. producers continued to com¬
plain to Washington that the closed Japanese market was being used as
the base for an export push into the American market, the government’s
attitude was perceived by the Japanese to be that Japan’s concentration
on the consumer market permitted U.S. firms to concentrate their re¬
sources on defense needs, and therefore contributed to national security.
As one Japanese observer later put it, the Japanese were pleased that the
United States, as a technologically advanced nation, could afford such a
fair decision.13

Private versus Public Policy: Television Exports in the 1960s

Considerable ambiguity over the contours of official Japanese policy

was to develop in the early 1960s, because of an often blurry borderline

and Japan. The U.K. government threatened to suspend Japanese cotton cloth imports by
the Crown Colony unless the Japanese lifted their export restrictions on transistors. See
“Japanese Transistors Sought by Hong Kong,” Electronics, June 8, 1962, p. 8; and David
Rose, “Hong Kong’s Transistor Radio Exports Soar to 100,000 a Month,” Electronics,
September 28, 1962, p. 24.
11. See Eleanor Hadley, Antitrust in Japan (Princeton University Press, 1970), p. 387.
12. “Washington Rejects Transistor Import Quota,” p. 7.
13. Yasuzo Nakagawa, Semiconductor Development in Japan (Tokyo: Diamond Pub¬
lishing, 1985), pp. 125-27 (in Japanese).
THE GENESIS OF AN AMERICAN TRADE POLICY 133

between overt Japanese government mandates and ostensibly voluntary,

private actions undertaken by groups of firms responsive to government
“suggestions.” This phenomenon is well illustrated by the system of price
floors devised in 1963 for television receiver exports, the next major
irritation to afflict U.S.-Japanese trade relations in electronics.
Japanese production of black-and-white television sets had grown
quickly in the late 1950s and by the early 1960s accounted for roughly 60
percent of the value of Japanese consumer electronics output (radios by
that time had dropped to about 30 percent of Japan’s consumer electron¬
ics shipments).14 As the Japanese market approached saturation in 1960,
new strategies were required. The Japanese government attempted to
expand the domestic market by setting new standards for color TV and
financing new television broadcasting channels.15 Producers were also
encouraged to turn their attention to foreign markets: the U.S. color
standard was adopted in order to facilitate exports of transistorized sets
to the United States.16 In mid-1960 MITI announced publicly that Japa¬
nese electronics producers were rapidly expanding production facilities
for color TV sets primarily intended for export (the sets were thought to
be too expensive for domestic consumers, and domestic sales were to be
merely a sideline).17 Japanese TV receivers first began to arrive in the
United States in significant quantities in I960.18 By 1966 color TV pro¬
duction had risen to a half million sets, with half of that output going to
the export market.19
Facing mounting trade frictions, seven television-producing members
of the Japan Machinery Exporters Association agreed to a private “or¬
derly marketing” agreement, setting minimum export “check” prices for
TV sets shipped to the United States in the fall of 1963. In enforcing this
system, Japanese television manufacturers also attempted to limit com¬
petition with each other in the U.S. market by agreeing to the so-called
five-company rule, which required every exporter to limit its U.S. sales
to five exclusive customers.20 Under the Export-Import Trade Law of

14. See Electronic Industries Association of Japan, Facts and Figures on the Japanese
Electronics Industry (Tokyo, 1986), p. 34.
15. “Japan Moves to Cut Television Set Surplus,” Electronics, October 14, 1960, p. 11.
16. “Japan Launches Color TV,” Electronics, January 22, 1960, p. 22.
17. See “Japanese Push Color TV Production Plans,” Electronics, July 24, 1960, p. 11.
18. See “Japan Boosts TV Set Output,” Electronics, February 26, 1960, p. 48; and
“Japanese TV Sets Arriving This Week,” Electronics, April 29, 1960, p. 32.
19. Electronics Industries Association of Japan, Facts and Figures, p. 35.
20. See Kozo Yamamura and Jan Vandenberg, “Japan’s Rapid-Growth Policy on Trial:
134 THE GENESIS OF AN AMERICAN TRADE POLICY

1952, private price- or quantity-fixing agreements among firms were legal

as long as they were eventually registered with MITI, but a MITI tradi¬
tion of informal, unwritten “administrative guidance” to firms made even
this boundary indistinct. Some details of this agreement were publicly
disclosed prior to its implementation, and floor prices were in fact initially
set in consultation with MITI.21
As this case illustrates, behind-the-scenes efforts to “privatize” actions
taken to deter or defuse trade conflict have often been preferred over
more public, visible (and perhaps precedent-setting) official interventions
by the Japanese government. It is probably safe to argue that, from the
early 1960s on, the construction of formal or informal export cartels was
the standard response to trade frictions involving Japanese electronics.22
The formal cartel agreements regulating television exports continued
until 1973.23
Interestingly, absent continuing direct involvement by MITI in the
operation of the cartel, discipline among the Japanese exporters ulti¬
mately began to break down. The minimum check price system began to
disintegrate by the late 1960s (exporters undercut the minimum prices
with a variety of under-the-table kickback schemes to U.S. buyers).
The five-company rule, although strictly enforced, could be and was
circumvented.24

The Television Case,” in Kozo Yamamura and Gary R. Saxonhouse, eds., Law and Trade
Issues of the Japanese Economy: American and Japanese Perspectives (University of Wash¬
ington Press, 1986), pp. 259-63; and Brief of Appellants, Zenith Radio Corporation and
National Union Electric Corporation, case nos. 81-2331, 81-2332, 81-2333, United States
Court of Appeals, Third Circuit (Philadelphia: International Printing Company, 1983),
pp. 14-19.
21. Accounts at the time made much of the distinction between price floors and actual
pricing, since “the actual export prices cannot be set because of Japan’s antitrust laws.”
See “Japan Firms Setting Minimum TV Prices,” Electronics, August 2, 1963, p. 7. A fairly
complete discussion of the history of price-setting schemes applied to Japanese TV exports
may be found in Yamamura and Vandenberg, “Japan’s Rapid Growth Policy on Trial,”
pp. 238-70.
22. It is not clear, however, that these cartels always worked well. For example, it is
known that cheating on check prices on television export prices, in the form of under-the-
table rebates to importers, sometimes occurred.
23. In 1974 Zenith Radio Corporation launched a widely publicized antitrust suit
against Japanese television exporters.
24. Brief of Appellants, pp. 20-28; and Yamamura and Vandenberg, “Japan’s Rapid
Growth Policy on Trial,” pp. 262-63. Despite the rule, some U.S. customers managed to
purchase TVs from more than one of the Japanese exporters. A Japanese exporter could
also circumvent the rule by naming as one of the five American purchasers its own U.S.
subsidiary, which would then be free to sell to any U.S. customer. Kenneth G. Elzinga,
THE GENESIS OF AN AMERICAN TRADE POLICY 135

Roughly coinciding with the formation of the TV export cartel was the
development of a domestic cartel to fix prices in the Japanese domestic
color television market.25 In 1964, after the continuing depressed circum¬
stances of the TV industry had triggered a round of severe price cuts in
the domestic market, a web of discussion groups was formed to set retail
prices, rebates, retail and wholesale profit margins, and to discuss such
matters as demand forecasts and bottom prices. Production, inventory,
and shipment data were exchanged and market shares and output quotas
assigned.26 In 1970 six of the companies were found by the Japan Fair
Trade Commission to have broken the antimonopoly law, although they
were also found to have ceased the violations. But various of the groups
associated with this domestic cartel continued to meet after the export
cartel disbanded in 1973. One midlevel group continued to hold monthly
meetings until 1977, and the highest-level group (the Okura Group,
named after the hotel where the presidents of seven top consumer elec¬
tronic producers met to discuss these issues for a variety of electric
appliances) continued to meet through at least 1974.27
The link between the domestic TV cartel and the export cartel has
always been controversial. MITI was clearly involved in the design and
establishment of what was nominally a privately administered export
cartel arrangement and sent representatives to meetings at which actions
were taken to set export prices and foreign market shares. If only because

“The New International Economics Applied: Japanese Televisions and U.S. Consumers,”
Chicago-Kent Law Review, vol. 64, no. 941 (1988), p. 964; and Franklin M. Fisher, “Mat¬
sushita: Myth v. Analysis in the Economics of Predation,” Chicago-Kent Law Review, vol.
64, no. 941 (1988), p. 973.
25. In the 1950s Japanese producers had set up a cartel to regulate prices in black and
white television sets. As David Schwartzman has commented, Japanese TV manufacturers
and distributors created a cartel, the Home Electric Appliances Market Stabilization Coun¬
cil, under the aegis of the Electronic Industries Association of Japan in 1956. This group
set up a series of retail price agreements and enforcement procedures and attempted to
organize production cuts (not fully successfully). In 1957 the Japan Fair Trade Commission
decided that the enforcement procedures violated Japan’s antimonopoly law and ordered
the Market Stabilization Council to halt punitive actions against discounters. (The JFTC
did not, however, rule against price fixing per se or force the council to disband.) See
Schwartzman, The Japanese Television Cartel: A Study Based on Matsushita v. Zenith
(University of Michigan Press, 1993), pp. 77-80.
The considerable evidence unearthed in the 1970 and earlier investigations is also cited
in Yamamura and Vandenberg, “Japan’s Rapid Growth Policy on Trial,” pp. 253-57; Brief
of Appellants, pp. 35-38; and Elzinga, “The New International Economics Applied,”
p. 964.
26. Schwartzman, Japanese Television Cartel, pp. 81-89.
27. Schwartzman, Japanese Television Cartel, pp. 87-89.
136 THE GENESIS OF AN AMERICAN TRADE POLICY

of the 1970 JFTC decision touching on selected elements of the domestic

TV cartel, MITI could hardly have been unaware that there was a parallel
cartel structure in place in the shadows for the domestic market. The
issue is significant because U.S. television producers always asserted that
the relatively higher domestic prices created by the actions of an illegal—
but government-tolerated—domestic cartel made it economically feasible
for the export cartel to set the much lower foreign prices that did them
such serious damage. Allegations of a similar two-tiered pricing struc¬
ture—low export prices and high domestic prices in a formally or infor¬
mally protected domestic market—were to be an important factor in
U.S.-Japan semiconductor trade friction in the late 1970s.

Competition in the 1970s

Although the immediate competitive threat to the American chip

industry was dissipated by its breakneck technical advance, American
firms remained most unhappy about being shut out of the Japanese mar¬
ket. As Japanese electronic exports making use of ICs began to enter the
U.S. market in the late 1960s, Texas Instruments (TI)—which, barred
from establishing a presence in Japan, had refused to license its patents
to Japanese companies—successfully jimmied open the Japanese market
a notch by threatening to petition to exclude from the U.S. market
Japanese exports of electronic equipment using semiconductors that in¬
fringed on its patents (a detailed account of this episode may be found
in chapter 2). TI was finally permitted to establish a joint venture in
Japan, subject to output restrictions, which was then converted into a
wholly owned subsidiary in 1971. TI’s early presence was the exception
that proved the rule of barriers to entering the Japanese market, however.
By the early 1970s rapid innovation had made American chip produc¬
ers the unchallenged leaders worldwide. In Japan, as was noted in the
last chapter, Japanese chip makers were even complaining about U.S.
companies “dumping” low-priced chips in the Japanese market, and
MITI took active steps to slow the inflow of U.S. imports of ICs.
This was not to say that American chip production did not come under
competitive pressure from foreign factories. In fact, the late 1960s and
early 1970s marked a rapid transfer of large segments of the American
chip industry to offshore locations, as U.S. semiconductor companies
moved their relatively labor-intensive assembly and testing operations to
THE GENESIS OF AN AMERICAN TRADE POLICY 137

foreign subsidiaries or contractors in low-wage countries.28 A substantial

impact from foreign trade was being felt within the U.S. industry: in
1971, for example, former workers at Sprague Electric semiconductor
plants petitioned for trade adjustment assistance as U.S. imports of semi¬
conductors assembled overseas surged.29 In 1973 and 1974 workers at GE
transistor and diode factories filed for similar assistance in the face of
continuing increases in imports.30 By the late 1970s approximately 70 to
80 percent of U.S. companies’ semiconductor shipments were assembled
offshore.31 But these actions mainly reflected internal restructuring within
American multinational companies and created little in the way of gen¬
uine trade friction likely to stir up international political confrontations.
Under continuing foreign pressure, Japanese quantitative import re¬
strictions on semiconductors had been gradually phased out by 1976.
Foreign investment was liberalized in 1974, and most formal trade barriers
had vanished by 1980. Restrictive practices in procurement by Nippon
Telegraph and Telephone (NTT), the state-owned telecommunications
monopoly, as well as in standards, certification and quality requirements,
and membership in R&D associations continued to generate complaints
by foreign chip makers, however. With the market opening of the 1970s,
Japanese chip producers began to face serious competition and were
frequently forced to rely on direct intervention by MITI with their cus¬
tomers to fend off American IC imports. As the inevitability of liberali¬
zation of the Japanese market became clear, government subsidies to
semiconductor R&D were greatly increased in order to help Japanese
producers adjust to the oncoming new realities. NTT and MITI launched
highly successful cooperative industrial research programs in 1975 and
1976, respectively. The programs were focused on technologies to im-

28. See Kenneth Flamm, “Internationalization in the Semiconductor Industry,” in Jo¬

seph Grunwald and Kenneth Flamm, The Global Factory: Foreign Assembly in International
Trade (Brookings, 1985), pp. 68-85.
29. See U.S. Tariff Commission, Capacitors and Semiconductors: Former Workers of
the Sprague Electric Company Plants at—North Adams, Mass., Worcester, Mass., Hillsville,
Va., Lansing, N.C., Concord, N.H., Barre, Vt., Grafton, Wis., TC publication 395 (May
1971).
30. See U.S. Tariff Commission, Transistors and Diodes: Workers of the Buffalo, N.Y.,
Plant of General Electric Co., TC publication 588 (June 1973); and U.S. International
Trade Commission, Transistors and Diodes: Workers and Former Workers of the Syracuse,
N.Y., and Auburn, N.Y., Plants of General Electric Co., ITC Publication 715 (February
1975).
31. Flamm, “Internationalization in the Semiconductor Industry,” pp. 83-84.
138 THE GENESIS OF AN AMERICAN TRADE POLICY

prove mass production of high-volume chips used in the computer indus¬

try, particularly DRAMs.
By 1977 it was becoming clear in the United States that the ongoing
technology push in Japan was achieving important results and was being
accompanied by increased investments in capacity. An initial trickle of
imported Japanese DRAMs had begun to surface in the U.S. market. A
surge in exports of Japanese chips to the U.S. market, similar to those
already seen in consumer electronics and automobiles, seemed likely.
Determined to maintain a more effective voice in Washington, American
chip producers formed the Semiconductor Industry Association (SIA) as
their lobbying arm.32
In October 1978 Robert N. Noyce, chairman of Intel, a U.S. DRAM
producer, launched the first public salvo across the Japanese bow, charg¬
ing that Japan’s protection of its home market, together with government
subsidies, made it possible for Japanese producers to engage in “two-tier
pricing.” Citing the example of the television market, Noyce argued that
firms were free to charge high prices in the sheltered Japanese market,
then price exports “as low as they want, since they need only to cover
incremental variable costs.”33 Without actually claiming that this was
currently occurring in semiconductors, Noyce noted that “if this pattern
is repeated in the semiconductor market, the U.S. market would be
flooded with underpriced Japanese integrated circuits and LSI products,”
and U.S. producers would “have no choice but to cut production or go
bankrupt.”34
Noyce’s fears were not without factual support. There is some evi¬
dence, largely anecdotal, that chip prices for leading edge digital ICs
tended to be higher in Japan than in the United States through much of
the 1970s. As one prominent Japanese manufacturing expert put it:
“When production of IC’s began in Japan around 1970, IC’s made in
America always sold for 20% less than the Japanese product. This low
price was made possible by the scale merit produced by the huge com¬
puter industry and the defense industry. When the Japanese companies
managed to cut their costs through rationalization of their industry, the

32. After a transcontinental pilgrimage to Washington intended to inform U.S. Trade

Representative Robert Strauss that “the Japanese are coming,” semiconductor executives
were reportedly dismayed to get a “so what?” reaction from Strauss. Author’s discussion
with industry executives and former government officials, September 1992.
33. See Peter Moylan, “Noyce Rips Gov’t as Peril to U.S. Semicon Industry,” Electronic
News, October 9, 1978, pp. 51-52.
34. Moylan, “Noyce Rips Gov’t,” p. 52.
THE GENESIS OF AN AMERICAN TRADE POLICY 139

Table 3-1. Factory Prices of Finished Integrated Circuits in Europe,

Japan, and the United States, 1976
Dollars

Type of device West Europe United States Japan

Calculator LSI 2.75 2.45 1.86

MOS 3.20 3.00 3.20
Bipolar digital 0.85 0.75 1.36
Bipolar linear 1.10 1.00 0.70
Microprocessor (including memory
and support circuits) 150.00 95.00 150.00

Calculator and watch displays (4 mm)

LED 1.80 1.70 1.50
LCD 5.00 4.50 4.50
Clock displays (15 mm)
LED 5.50 5.00 5.25
LCD 8.50 8.00 8.00
Source: Mackintosh Consultants, Market Survey of Semiconductors, vol. 4: Applications and Markets (Study un¬
dertaken on behalf of the Ministry of Research and Technology of the Federal Republic of Germany, December
1976), pp. 125-27. Quoted average costs reflect differences in product specifications and volume requirements.

American companies always cut their price even further.”35 Table 3-1,
although it does not control for volume and product mix, shows that unit
costs for a variety of advanced ICs tended to be significantly higher in
Japan in the mid-1970s, while products typically used in consumer elec¬
tronics (linear ICs, display chips) actually tended to be priced below U.S.
levels.
The SIA’s activity was successful in prodding a Senate subcommittee,
in December 1978, to order the U.S. International Trade Commission
(ITC) to launch an informational investigation into the competitive po¬
sition of the U.S. semiconductor industry.36 By the time the commission’s
report was delivered, at the end of 1979, the forecast threat had mater¬
ialized. Amid a boom in U.S. electronics production, a major shortage
of 16K DRAMs had developed over the summer of 1979.37 Viewing these
shortages as an opportunity, three major Japanese producers—Hitachi,
Fujitsu, and NEC—had plunged into the U.S. market in force. By the

35. Hajime Karatsu, “Quality Control—The Japanese Approach,” paper presented at

a seminar, “Quality Control: Japan’s Key to High Productivity,” Washington, March 25,
1980, p. 9.
36. See Richard Wightman, “ITC Launches Probe of U.S. Semicon Position in Japan,
Europe,” Electronic News, December 18, 1978, p. 44.
37. See Steven Hershberger, “Users Book 16K RAMs Through 1980,” Electronic News,
July 16, 1979, p. 49; and “Schottky, RAM Bind Hits User Revenues,” Electronic News,
September 3, 1979, p. 1.
140 THE GENESIS OF AN AMERICAN TRADE POLICY

end of the year they had collectively achieved about a 40 percent share
of the U.S. market for 16K DRAMs.38
During the nine-month ITC investigation, the SIA’s evolving theory
of Japanese industrial practices was further elaborated. At an ITC hear¬
ing in San Francisco in May 1979, the SI A for the first time suggested
that the two-tier pricing scenario was actually occurring, notably in sales
of 16K DRAMs.39 Also, apparently for the first time, it was suggested
that, in addition to being an example of what would now be called a
strategic trade policy implemented by the Japanese state, Japanese poli¬
cies contained an explicitly predatory element. Noyce, arguing on behalf
of the SI A at the May hearing, articulated a strategic conception of
the dangers of dependency on foreign suppliers as follows: “Now one
might argue that U.S. consumers benefit from these bargain prices. But
we must realistically ask how long such bargain prices last. Middle East¬
ern oil was a bargain until we became dependent upon it. Similarly,
sooner or later the Japanese losses on high density memories will be
recouped and I submit that it is foolish to assume any long run benefit
to consumers.”40
What is vague in these statements is whether these strategic calcula¬
tions are viewed as being undertaken by the Japanese state, with Japanese
companies passively responding to changes in state policy, or whether the
companies are being accused as active parties to the strategic plan. One
might, for example, conceive of “state predation,” where state subsidies
induce firms to cut prices in order to stimulate exit by foreign rivals.41 At
the time, though, the prevailing conception was one of “Japan, Inc.,”

38. See Lloyd Schwartz, “Mostek Chief: Japan Threatens Industry,” Electronic News,
October 15, 1979, p. 72; and Henry Scott Stokes, “Japan Goal: Lead in Computers,” New
York Times, December 12, 1979, p. Dl. Buoyed by surging demand for computers, even
IBM was reportedly contacting Japanese chip suppliers in search of 16K DRAM supplies.
See John Hataye, “IBM Shopping in Japan for 16K Dynamic RAMs,” Electronic News,
December 24, 1979.
39. See Jim Leeke, “‘Practices Abroad Unfair,’ SIA Says at ITC Hearing,” Electronic
News, June 4, 1979, p. 106.
40. “Statement of Dr. Robert N. Noyce, Vice Chairman of the Board, Intel Corpora¬
tion, on Behalf of the Semiconductor Industry Association,” Hearings before the U.S.
International Trade Commission, San Francisco, May 30, 1979, pp. 21-22.
41. Willig, for example, draws a distinction between “strategic dumping,” which relies
on national policies to protect exporting companies’ home market, in order to gain cost
advantages and create monopoly power for the exporters in importing markets, and “pred¬
atory-pricing dumping,” which is a company strategy to obtain monopoly power in an
importing country’s market. See Robert D. Willig, “The Economic Effects of Antidumping
Policy,” Organization for Economic Cooperation and Development, Paris, 1992, pp. 7-8.
THE GENESIS OF AN AMERICAN TRADE POLICY 141

with firms and state joined together in a collective strategic plan. In the
colorful words of one top American executive, it was the “33 companies
in the SI A taking on the sovereign nation of Japan.”42

“Below Cost” Dumping

The era of two-tier pricing was relatively short-lived. The evidence

submitted in 1979 was quite scanty—the head of Mostek, a U.S. DRAM
producer, asserted a 20 to 30 percent differential between U.S. and
Japanese 16K DRAM prices in 1978 and presented first-quarter 1979 data
on five selected sales contracts by a U.S. company to Japanese customers,
showing prices significantly above the prevailing prices for equivalent
Japanese products sold in the United States.43 In any event, an attempt
to investigate the issue in greater depth would have soon run into the
complexities of market structure and distribution patterns for semicon¬
ductors in the United States and Japan, which make price-to-price com¬
parisons between the two markets rather tricky.
Briefly put, roughly 70 to 80 percent of sales to U.S.-based semicon¬
ductor customers are transacted directly with chip manufacturers through
long-term contracts, with deals typically struck months ahead of delivery.
The balance is sold through distributors to smaller customers, and as spot
sales, often through an active “gray” (secondary) market of brokers,
distributors, and other arbitragers. Spot prices typically rise above con¬
tract pricing in tight markets and fall below contract prices when demand
is slack.44 Thus, to avoid comparing apples with oranges when searching
for price differentials, contract prices should properly be compared with

42. W. J. Sanders, president of Advanced Micro Devices at the time, quoted in Leeke,
‘“Practices Abroad Unfair,’” p. 106.
43. These data were contained in a confidential submission to the ITC dated August
17, 1979, but appear to have been presented later at a hearing of the congressional Joint
Economic Committee in October 1979. See ITC, Competitive Factors Influencing World
Trade in Integrated Circuits, publication 1013 (November 1979), pp. 70-71; and U.S.-
Japanese Trade Relations, Hearing before the Joint Economic Committee, 96 Cong. 1 sess.
(GPO, 1979), pp. 21, 26. Comparisons are further complicated by the fact that DRAMs
are sold on both long-term contracts and on a spot basis, for immediate delivery, and prices
can diverge substantially. It is unclear whether an appropriate spot-to-spot or contract-to-
contract comparison between the two markets was being made.
44. For detailed evidence on the structure of the U.S. semiconductor market, see
Kenneth Flamm, “Measurement of DRAM Prices: Technology and Market Structure,” in
Murray F. Foss, Marilyn E. Manser, and Allan H. Young, eds., Price Measurements and
Their Uses (University of Chicago Press, 1993).
142 THE GENESIS OF AN AMERICAN TRADE POLICY

other contract prices, not with spot prices, and vice versa. Dumping
complaints, historically, have not always drawn these distinctions.
Furthermore, the semiconductor market in Japan has a rather different
structure from that in the United States. For the most part, large chip
manufacturers in Japan sell directly only to sister electronic equipment
divisions of the parent corporation. The vast bulk of external sales to
large customers go through authorized sales agents, while smaller cus¬
tomers are served through secondary sales agents who order product
from the main sales agents. Even smaller quantities are sold on a spot
basis through retailers clustered in selected urban areas, such as Tokyo’s
Akihabara district.45 Prices quoted in Japanese trade sources typically
refer either to prices to large users through main sales agents or to spot
prices in Akihabara. The prices large users pay in Japan are roughly
comparable with U.S. contract prices, whereas U.S. spot prices are most
similar to Akihabara pricing.
The complexities of direct U.S.-Japan price comparisons never be¬
came a major issue, however, because charges of two-tier pricing had a
relatively short life. In the months after the San Francisco hearing, de¬
mand for 16K DRAMs surged, and as prices soared, complaints about
low-priced imports faded away. From mid-1979 until the present day,
charges that U.S. prices for Japanese chip imports were below Japanese
levels ceased to be an important irritant to trade relations.
Instead, complaints that the Japanese were selling below the cost of
production in both markets began to emerge.46 Texas Instruments, the

45. See U.S. Department of Commerce, International Trade Administration, “Japan—

Semiconductors/Nonvolatile Memory in Japan—ISA9106,” derived from Fuji Keizai Co.,
“The Semiconductors-Nonvolatile Memory Market in Japan,” Tokyo, June 1991; “Japan-
Semiconductors/Analog Devices in Japan—ISA9106,” derived from Fuji Keizai, “The
Semiconductors-Analog Devices Market in Japan,” Tokyo, June 1991; and “Japan—Semi¬
conductors/Logic Devices in Japan—ISA9106/’derived from Fuji Keizai, “The Semicon-
ductors-Logic Devices Market in Japan,” Tokyo, June 1991.
NEC and Mitsubishi were reported to sell 100 percent of their Japanese sales through
sales agents, while Hitachi, Fujitsu, and Toshiba sold 80 to 90 percent of their external
sales by this route. Smaller manufacturers such as Sharp were reported to sell as much as
20 to 30 percent of their shipments directly to users, and the remainder through sales
agents.
Primary sales agents are typically associated with a particular Japanese manufacturer.
From the standpoint of market access, Japanese sales agents rarely handle foreign products
that are very similar to, or compete with, products from the Japanese manufacturer with
which they are affiliated. Since most Japanese producers offer broad product lines, it is
unusual for a primary sales agent to sell imported semiconductors at all.
46. Testifying before a Senate committee in early 1980, Intel’s Noyce saw a political
THE GENESIS OF AN AMERICAN TRADE POLICY 143

only American producer manufacturing in Japan at the time, had at¬

tacked the SI A position in early 1980, asserting that prices received by
Japanese producers in their home market were actually lower than Amer¬
ican prices for the same Japanese chips; the SI A argued that the Japanese
were voluntarily restraining their exports in response to American in¬
dustry complaints and deliberately creating a soft market at home.47 Press
reports imply, however, that prices in a weakening U.S. market still felt
pressure from falling prices in the Japanese market. By mid-1980, for
example, Japanese DRAM producers were reported to be starting a
campaign of surveillance of shipments to their sales agents, to discourage
their resale at very low prices to customers in the U.S. market by gray
market traders (and further irritating American producers).48 By the
spring of 1981 Japanese producers NEC (which exported 60 percent of
its output to the United States) and Fujitsu had even suspended sales
into the U.S. spot market, to lessen the threat of an American anti¬
dumping action.49 Four of the major Japanese IC producers (NEC,
Hitachi, Toshiba, and Fujitsu) also announced plans to manufac¬
ture DRAMs at U.S. plants, in a complementary bid to reduce trade
frictions.50

intent in this development: “Indeed, Intel buys the 16K RAMs from Japan because we
have found that cheaper than to make them ourselves. Now, there is some artificial pricing
in that market. That’s what I’m suggesting.” Senator [Adlai] Stevenson: “Well, that’s what
I’m getting at. Is there? What do you mean by that?” Dr. Noyce: “Up until the San
Francisco hearings, we could buy 16K RAMs in the United States, from Japanese compa¬
nies at lower prices than we were selling the same product for in Japan.” Senator Stevenson:
“Is that artificial pricing or are they just more productive and efficient?” Dr. Noyce: “We
were meeting the market price in Japan. After the ITC hearings in San Francisco, U.S.
prices went up and Japanese prices went down. I think the prices had been artificial there,
but that is a very difficult thing to determine.” Trade and Technology, Part III, Hearings
before the Subcommittee on International Finance of the Senate Committee on Banking,
Housing, and Urban Affairs, 96 Cong. 2 sess. (GPO, 1980), p. 176.
47. “U.S. Semiconductor Firms Disagree on Import Strategy,” Denver Post, March 23,
1980, p. 40.
48. See “Exports of Japanese Semiconductors to US at Low Prices Conspicuous; Half
Price, Too, through Trading Firms; Manufacturers Strengthen Checking of Destinations,”
Nihon Keizai Shimbun, May 24, 1980, p. 6.
49. “NEC Suspends Shipments of RAM Chips to U.S. Market,” Japan Economic
Journal, March 3, 1981; and “Japanese Electronics Firms Delay Plans to Mass-Produce
64-K Chips,” Asian Wall Street Journal Weekly, April 6, 1981, p. 14.
50. See “Japan Firms Plan U.S. Production of Advanced Circuits,” Asian Wall Street
Journal Weekly, May 18, 1981, p. 17; Thomas J. Lueck, “NEC Plans $100 Million U.S.
Plant,” New York Times, June 27, 1981, p. Dl; and “Top Four Japanese IC Makers Expand
U.S. Operations,” Asian Wall Street Journal Weekly, July 13, 1981, p. 15.
144 THE GENESIS OF AN AMERICAN TRADE POLICY

Mounting Frictions: “Quality Dumping”

Japanese chip makers had in fact taken their first hesitant steps toward
setting up U.S. production facilities, amid intensifying political criticism,
several years earlier. Among the major Japanese companies, NEC had
been the pioneer in 1978, when it acquired Silicon Valley producer Elec¬
tronic Arrays.51 At least partly in explicit reaction to rising protectionist
sentiment, Fujitsu announced an investment in a San Diego manufactur¬
ing facility in 1979. In that same year Oki Electric opened negotiations
to have some of its products produced in the United States, Hitachi
unveiled plans to use a Dallas facility for semiconductor assembly, and
NEC began shipping U.S.-made chips from its California plant.52
By late 1979 it had become clear that U.S. imports of Japanese chips
had roughly doubled in value, compared with the previous year.53 Amid
an unforeseen boom in demand for chips used in computers, and man¬
ufacturing yield problems on American production lines for 16K
DRAMs, a shortage of DRAMs had developed in the United States.54
Japanese chip makers rushed into that vacuum. In a parallel develop¬
ment, many Japanese semiconductor equipment makers had begun their
first serious probes into the U.S. market. Key production tools developed
in cooperation with the NTT and MITI R&D programs in very large
scale integration—such as the Canon projection alignment system and
the Takeda Riken (later renamed Advantest) high-speed logic tester—
were first offered for sale in the U.S. market in 1979.55 The clearly evident
Japanese thrust into the U.S. semiconductor market provoked a rising
crescendo of political criticism within the United States.56

51. There had been at least two instances of Japanese investments in small American
semiconductor companies earlier in the 1970s: Toyo Electronics (which later changed its
name to Rohm) acquired Exar Integrated Systems in 1972, and Hattori Seiko acquired
Micropower Systems in 1977.
52. John Hataye, “Semicon Makers May Increase Production in U.S.,” Electronic
News, June 11, 1979, p. 24; and “Special Report: Japan is Here to Stay,” Business Week,
December 3, 1979, pp. 81-86.
53. See “Japan IC Exports to U.S. in 8 Months Double to $96M,” Electronic News,
November 19, 1979, p. 58.
54. “Special Report: Can Semiconductors Survive Big Business?” Business Week, De¬
cember 3, 1979, p. 85; and Cheryll A. Barron, “Microelectronics Survey: All That Is
Electronic Does Not Glitter,” Economist, March 1, 1980, pp. 3-4.
55. See John Hataye, “Japan Equip. Firms Eye Exports to U.S.,” Electronic News,
December 3,1979, p. 20. For a discussion of these products’ roots in Japan’s VLSI projects
see chapter 2.
56. For example, see Lloyd Schwartz, “Mostek Chief: Japan Threatens Industry,” Elec-
THE GENESIS OF AN AMERICAN TRADE POLICY 145

One of the oddest elements in the rising chorus of American com¬

plaints was the charge of “quality dumping” that emerged in late 1979,
when U.S. and Japanese DRAM prices were undeniably in approximate
parity.57 Reports from American chip users suggested that the Japanese
DRAMs they were buying had defect rates one-half to one-third those
experienced with comparable American products.58 U.S. chip makers for
the most part did not deny the bad news coming from their customers.
Instead, reasoning that higher levels of quality reflected increased expen¬
diture of resources on quality control,59 the American industry charged
that sales of higher quality Japanese products at the same price as Amer¬
ican products reflected a form of “dumping.” But U.S. companies also
increased their own investments in quality control programs.
The Japanese industry’s response to this complaint was to hold a
well-publicized seminar, “Quality Control: Japan’s Key to High Pro¬
ductivity,” in Washington in March 1980. The seminar’s presentations
made the argument that it was superior technology—in the form of
quality control techniques imported from American sources and ap¬
plied and improved upon in Japan—that was responsible for the Jap¬
anese quality edge. The seminar also worked to reinforce customer
perceptions of the American industry’s quality deficiencies, by spot¬
lighting a speaker from Hewlett-Packard Co. who made public data
documenting a large quality gap between American and Japanese
DRAMs purchased by that company.60

tronic News, October 15, 1979, p. 72; Steven Hershberger, “Mostek Chief: Limit Imports
from Japan,” Electronic News, November 19, 1979, p. 42; and Stokes, “Japan Goal: Lead
in Computers,” p. Dl.
57. See Stokes, “Japan Goal: Lead in Computers,” p. Dl.
58. “Can Semiconductors Survive Big Business?” p. 85.
59. Testifying before the U.S. International Trade Commission in 1979, SI A members
argued that “ ‘double’ testing techniques of the Japanese were expensive (‘It is an economic
issue, not a quality issue’) and that they considered this ‘better deal’ given to buyers of
Japanese circuits a type of market penetration technique.” U.S. International Trade Com¬
mission, Competitive Factors Influencing World Trade in Integrated Circuits, USITC publi¬
cation 1013 (November 1979), p. 24. The SIA notes that “To achieve this quality there is a
high dependence on the best available tools and automation. This adds to the capital cost,
but it may pay in the long run through higher yield and higher productivity. Japan, because
of its easier access and lower cost of capital has an inherent advantage.” The International
Microelectronics Challenge: The American Response by the Industry, the Universities, and
the Government (Cupertino, Calif., May 1981), pp. 27-28. See also “Japan Makes Them
Better,” Economist, April 26, 1980, p. 55.
60. See Ray Connally, “Japanese Make Quality-Control Pitch,” Electronics, April 10,
1980, p. 81; and “Japan Makes Them Better,” pp. 55-56.
146 THE GENESIS OF AN AMERICAN TRADE POLICY

Over the next several years American chip makers suffered through a
continuing series of unfavorable quality comparisons with the Japanese
by their customers. Although these assessments showed considerable
improvement relative to the Japanese competition, American chips were
still perceived to lag in quality through the early years of the 1980s.61 A
federal crackdown on fraudulent quality testing and certification practices
in Defense Department chip procurement over 1981-82 did little to help
this perception.62 The American semiconductor industry fought back by
touting its aggressive adoption of improved quality management practices
over this period, and questioning the existence of a quality gap with
Japan to the extent portrayed in customer reports.63 However, at least
some American chip producers, as they point to their current world-class
standards, frankly acknowledge the existence of a significant quality gap
in the early 1980s.64

61. See “U.S. RAMs Improve But Still Lag Japan’s,” Electronic Engineering Times,
November 10, 1980, p. 1; “Xerox Data Uphold HP Contention that Japanese RAMs are
Better,” Electronic Engineering Times, March 2, 1981, pp. 2, 16; Ray Connolly, “Copy
Japanese, U.S. Managers Urged,” Electronics, April 21, 1982, pp. 106-08; Jerry Lyman
and Alfred Rosenblatt, “The Drive for Quality and Reliability, Part 1,” pp. 125-28, and
“Part 3: Users Push for Quality,” pp. 141-43, Electronics, May 19, 1981; “HP, Motorola
Continue the Debate over Quality and its Management,” Electronic Engineering Times,
May 25, 1981, pp. 14-15; and Larry Waller, “Perception Lag Nags U.S. Chip Makers,”
Electronics, April 21, 1982, pp. 42-43.
62. See “A Crackdown on Chip Quality,” Business Week, January 11, 1982, p. 110; and
Marilyn Chase and Jim Drinkhall, “Grand Jury Bores in on the Faulty Testing of U.S.
Defense Gear,” Wall Street Journal, October 4, 1982, p. 1.
63. Waller, “Perception Lag Nags U.S. Chip Makers,” pp. 43-44.
64. Top Intel executive Craig Barrett, for example, was quoted in a 1991 Intel publi¬
cation as saying that, in the early 1980s, “we realized that while we had always been a
leader in technical solutions and product reliability, our product quality levels did not meet
world class standards. After benchmarking against the best manufacturers in the world,
we learned that we had a long way to go to match up.” Katie Woodruff, “Intel’s Journey
to Total Quality,” Intel Microcomputer Solutions (January-February 1991), p. 8.
It is worth contrasting this assessment with the public statements of Intel vice-president
Willard Kauffman in April 1982: “Kauffman, in fact, questions whether any such gap
actually existed to the extent it was perceived by Richard Anderson of Hewlett-Packard
Co.’s Computer Systems division. . . . According to Intel’s Kauffman, the Japanese still
excel in their ‘strategy to market quality. ... We have to sell our quality in the same way.
We have nothing to be ashamed of,’ he emphasizes.” Waller, “Perception Lag Nags U.S.
Chip Makers,” p. 44.
The customer critique of American chip producers’ quality practices in the early 1980s
is implicitly supported by U.S. semiconductor companies’ advertising and marketing liter¬
ature in the late 1980s, which stressed the impact of the adoption of rigorous statistical
process control procedures on defect rates. Advertising by National Semiconductor Cor¬
poration, for example, depicted a decline in defects from 7,800 parts per million (ppm) in
THE GENESIS OF AN AMERICAN TRADE POLICY 147

Organizing a Response

In the spring of 1981 the SIA unveiled its program to combat the
growing competitive threat from abroad. It identified three key areas for
action. One was international in focus: access to foreign markets. The
other two were domestic in nature: greater U.S. stimuli to capital for¬
mation, and increased investment in U.S. R&D and engineering educa¬
tion. Specific goals with respect to market access in Japan included a
reduction of the Japanese tariff on semiconductors from 10 percent to 4.2
percent by April 1982 (rather than in 1987, as scheduled by the Tokyo
Round agreement on multilateral tariff reductions under the General
Agreement on Tariffs and Trade), and equivalent treatment for U.S. and
domestic companies in Japan (national treatment) in “access to financing
at competitive rates, bureaucratic processing of subsidiary filings with
the government, and ability to recruit top Japanese engineering talent.”
For the first time also, the SIA called on the Japanese government to
create an “affirmative action program” to remedy the effects of past
discriminatory practices against foreign firms.65
Action quickly followed on several of these issues. In May 1981 the
U.S. government announced that it had negotiated the requested reduc¬
tion in Japanese tariff levels.66 Congress passed a 25 percent incremen¬
tal tax credit for R&D in 1981, after the SIA pressed its case for this
measure.67
The industry itself also took concerted action on the research and
education fronts. By late 1981 the SIA had formed the Semiconductor
Research Cooperative, a joint effort to funnel R&D funds into basic
research projects in universities.68 And at Stanford University, with joint
funding from both industry sponsors and the Defense Advanced Re-

1978 to 37 ppm in 1988. National’s vice president for quality and reliability was quoted as
saying that “We feel our ability to produce and supply high-quality, reliable products is
now [author’s emphasis] second-to-none in the industry.” National Semiconductor Corpo¬
ration, National Anthem (1989), p. 7. Similarly, Intel reported a decline in defects from
10,000 ppm in the early 1980s to 200 in 1989, with levels in the 10- to 50-ppm range for
mature products. Woodruff, “Intel’s Journey to Total Quality,” pp. 8-9.
65. Semiconductor Industry Association, The International Microelectronic Challenge,
pp. 22-23.
66. Clyde H. Farnsworth, “U.S. and Japan Plan Cuts in Semiconductor Tariffs,” New
York Times, May 12, 1981, p. Dl.
67. See Andrew Pollack, “Federal Aid Sought for Semiconductors,” New York Times,
May 19, 1981, p. Dl.
68. “Electronics Research Projects,” New York Times, December 27, 1981, p. Dl.
148 THE GENESIS OF AN AMERICAN TRADE POLICY

search Project Agency (DARPA), a new Center for Integrated Systems

was established to train graduate students in microelectronics and focus
research on relevant areas.69 The announcement by MITI in 1981 of its
ten-year project to develop a “Fifth Generation” computer prompted
calls within both the U.S. computer industry and the U.S. semiconductor
industry for the formation of analogous R&D joint ventures among U.S.
companies, but the first such venture—the Microelectronics and Com¬
puter Technology Corporation (MCC)—was not to come into existence
until January 1983.70 The antitrust concerns that slowed the formation of
MCC were not definitively put to rest until Congress passed the National
Cooperative Research Act of 1984, which greatly reduced these worries
for research joint ventures.

Competition, Collusion, or Predation? The 64K DRAM Wars

Conflict between the U.S. and Japanese industries was to worsen as
the next generation of DRAM, the 64K, was introduced in 1981. Japanese
producers had diverted resources toward early introduction of the 64K
DRAM in mid-1981 and gained a lead over most of their American
competitors.71 Aggressive Japanese pricing in 64K DRAMs and rapidly
falling prices stimulated a new round of industry complaints in the United
States.72 By early 1982, with the Japanese share of the U.S. 64K DRAM

69. See “Integrated Systems Center Begins Work,” Stanford Observer, February 1982,
p. 1. DARPA also funded more traditional academic research programs at universities,
including the relatively large effort at the Massachusetts Institute of Technology (MIT).
See Paul Schindler, “VLSI: Retrieving the Top Spot from Industry,” Technology Review
(January 1983), p. A3, which notes that DARPA funding accounted for $2.7 million of a
$5.1 million budget for the MIT microsystems program.
70. “Control Data’s Push for a Cooperative Chip,” Business Week, April 20, 1981, p.
39; Andrew Pollack, “Singing the Semiconductor Blues,” New York Times, May 24, 1981,
p. F4; Marilyn Chase, “U.S. Electronics Firms Consider Joining in Research Venture to
Counter Japanese,” Wall Street Journal, March 1, 1982, p. 6; Ray Connolly, “Updating
Antitrust Law to Permit Joint R&D,” p. 66, and “U.S. R&D Consortium Takes Shape,”
pp. 97-99, Electronics, March 10, 1982; and William C. Norris, “Cooperative R&D: A
Regional Strategy,” Issues in Science and Technology, vol. 1 (Winter 1985), p. 94.
71. Howard Wolff, “64-K RAM Battle is Murky,” Electronics, September 8, 1981, pp.
89-90; Alan Alper, “Buyers Hedging on Long-Term 64K Pacts Until U.S. Firms Ramp
Up,” Electronic News, February 8, 1982, p. 1; Andrew Pollack, “Japan’s Big Lead in
Memory Chips,” New York Times, February 28, 1982, p. FI; and “A Chance for U.S.
Memories,” Business Week, March 15, 1982, pp. 126-27.
72. See Sabin Russell, “U.S. Suppliers Outnumbered in 64K RAM Competition—for
Now,” Electronic News, August 24, 1981; and “Prices of 64K RAM Drop to One-tenth of
Year Ago,” Japan Economic Journal, September 15, 1981.
THE GENESIS OF AN AMERICAN TRADE POLICY 149

market standing at about 70 percent, some American producers, led by

Motorola, were pressing the Commerce Department to investigate
charges that the Japanese were selling 64K DRAMs below “fair value”
(average cost).73 The particulars of the issues raised are addressed in
chapters 5 and 6.
Officials in the Commerce Department at this point decided to “jaw¬
bone” the Japanese unofficially to do something about tumbling 64K
DRAM prices. MITI officials in Washington were warned that the Com¬
merce Department might begin to “monitor” Japanese import prices.74
In March, however, just as this investigation was beginning, 64K DRAM
prices suddenly doubled, Japanese suppliers began rationing U.S. cus¬
tomers, and it was reported that Japanese companies were cutting back
U.S. exports in order to blunt moves toward trade restrictions on DRAM
imports.75 Within the Japanese semiconductor industry these reductions
in exports are openly acknowledged to have been spurred by MITI guid¬
ance.76 In early April 1982 Japanese DRAM producers actually con-

73. Clyde H. Farnsworth, “Japanese Chip Sales Studied,” New York Times, March 5,
1982, p. Dl; John Eckhouse, “Are Japanese Chip-makers ‘Dumping’?” San Francisco
Examiner, March 5,1982; and Bruce Entin, “Motorola Asks Inquiry into Japanese Pricing,”
San Jose News, March 10, 1982. Motorola’s interest in government action came after a
sharply divided SIA reportedly voted down a Motorola proposal to submit a joint anti¬
dumping complaint. See I. M. Destler and Hideo Sato, “U.S.-Japan High Technology
Trade: Politics and Policy,” University of Maryland, School of Public Affairs, August 1986,
p. 38. Motorola executives also were reported to have wanted the government to take the
lead in order to avoid placing itself in an antagonistic position with Hitachi, the top Japanese
producer of 256K DRAMs, with which it had close ties. Richard Wightman, “SIA Split on
64K RAM ‘Dump’ Action; Expect Members to Petition,” Electronic News, March 22,1982,
p. 1.
74. Clyde V Prestowitz Jr., writes of his and Lionel Olmer’s efforts as officials in the
Commerce Department at that time: “Our only tools were bluff and persuasion. Olmer,
who was responsible for administering the antidumping laws, warned MITI officials in
Washington that the Commerce Department might begin to monitor Japanese chip prices.
Our ability to do so was in fact limited, but we hoped to make the Japanese more cautious
by suggesting the possibility.
It worked for a while. The prices of 64K RAMs began to stabilize, as MITI warned
Japanese companies.” Clyde V. Prestowitz Jr., Trading Places: How We Allowed Japan to
Take the Lead (Basic Books, 1988), p. 49.
75. Alan Alper, “See 64K Levels in Line With Demand,” Electronic News, March 15,
1982; and “Justice Department Investigating Japanese 64K RAM Mktg.; Seek Price, Ship¬
ment Data,” Electronic News, August 2, 1982, p. 1.
76. The Japanese trade publication VLSI Report’s timeline of U.S.-Japan trade fric¬
tions, for example, sets February 1982 as the date “MITI instructions on dumping began.”
The Electronic Industries Association of Japan’s official industry handbook, IC Gaido
Bukku, has a timeline also associating February 1982 with the entry “MITI to Japanese
150 THE GENESIS OF AN AMERICAN TRADE POLICY

firmed to reporters in Tokyo that they were reducing U.S. exports to

alleviate trade friction.77
U.S. Commerce Secretary Malcolm Baldrige immediately informed
the U.S. press of his initial “favorable” reaction to the Japanese voluntary
export restraints, but in an interview for the New York Times he warned:
“They were building much more capacity than they could stand, and we
thought there was possibly some evidence of predatory pricing to take over
the market.”
He added that he thought there was a danger that the nation’s security
could be threatened if the American computer and telecommunications
industries were to become dependent on Japan for supply of the chips,
which have been characterized as the crude oil of the 1980s.
“This is something we would not like to see,” Mr. Baldrige said, noting
that the aim of the Commerce Department investigation was to determine
whether the Japanese were building volume in the 64K RAM market as a
result of subsidies to their industry or because of policies or practices that
insulate it from competition from American companies.78

The very next day, MITI and various Japanese company representa¬
tives officially denied reports of export restraints.79 But U.S. prices were
to continue well above Japanese price levels through early 1983.80
The episode was to take an even more bizarre turn three months later,
in July 1982, when MITI was informed by the U.S. Justice Department
that Japanese producers were being investigated to determine whether a
cartel had been formed to set volume and price levels in the U.S. mar¬
ket.81 (Prices in the United States had continued to hover at almost
double Japanese levels.)82 An NEC spokesman, apparently unfamiliar

industry, no exports that might cause blame for dumping.” See “Japan-U.S. IC Frictions,”
VLSI Report (Tokyo: Press Journal, ca. 1988), p. 62; and Electronic Industries Association
of Japan, IC Gaido Bukku (Tokyo, 1987), p. 62.
77. A. E. Cullison, “Japan Alters Memory Chip Export Policy,” Journal of Commerce,
April 7,1982; and “64K RAM Exports Are Being Held Down by Makers,” Japan Economic
Journal, April 13, 1982.
78. Clyde H. Farnsworth, “Japan to Cut Export of Chips to U.S.,” New York Times,
April 8, 1982, p. Dl.
79. Steve Lohr, “Japanese Deny Any Cut in Chip Exports to U.S.New York Times,
April 9, 1982, p. Dl; see also Associated Press, “Computer Chip Reports ‘Premature,’”
Japan Times, April 10, 1982.
80. Jack Robertson, “Japanese RAM Power,” Electronic News, February 28, 1983.
81. “U.S. Will Probe Japanese Makers of Semiconductors,” Japan Times, July 27, 1982;
Steve Lohr, “6 Japan Concerns Focus of Inquiry,” New York Times, July 27, 1982; “U.S.
Probes Sales of Computer Chips by Six Japan Firms,” Wall Street Journal, July 27, 1982;
“Justice Department Investigating,” p. 1.
82. “U.S. Won’t Indict Japan Semiconductor Makers,” Japan Times, July 30, 1982.
THE GENESIS OF AN AMERICAN TRADE POLICY 151

with U.S. antitrust concepts, responded that “Japanese interests have set
relatively high prices for 64-kilobit RAM chips in the U.S. so as not to
raise suspicions of dumping”!83 Needless to say, the whole sequence of
events left the Japanese somewhat confused. In March, amid loud indus¬
try and Commerce Department complaints about excessively low prices,
Japanese DRAM import prices had suddenly jumped, only to have the
Justice Department announce an investigation of excessively high prices
a short time later.84
A lawyer speaking for the SIA explained the apparent contradiction
as a real-life example of precisely the predatory scenario Robert Noyce
had first warned of back in 1979: “They may have committed violations
of the dumping laws early on, to buy market share, and now they’re
getting the payoff by limiting supplies and raising prices.”85 Noyce himself
even weighed in on the issue in early 1983: “It is probably correct that
the Japanese are selling RAMs in the United States at higher prices. It
is a classic case of competitors using predatory low pricing to take the
lion’s share of a market, and then increasing prices once they dominate
that market.”86
By mid-1983, however, the market for DRAMs was again picking up,
and the industry faced looming shortages. Prices in Japan rose sharply,
pushed up to U.S. levels.87 After depositions were taken from Silicon
Valley representatives of the Japanese chip makers in the spring of 1983,
the Justice Department’s antitrust investigation simply faded away.88 As

83. “U.S. Will Probe Japanese Makers.”

84. “‘In the second half of 1981, the U.S. Commerce Department was told by one of
our competitors that we were dumping 64K RAMs on the U.S. market,’ one semiconductor
company spokesman said in Tokyo last week. ‘Now, the U.S. Justice department is asking
if we are fixing prices and holding back supplies. Well, cartel or dumping, which is it?’”
J. D. Kidd, “Japanese 64K Makers Puzzled by U.S. Probe,” Electronic News, August 22,
1982. See also Andrew Pollack, “Inquiry Puzzles Chip Makers,” New York Times, July 28,
1982.
85. Thomas A. Skornia, quoted in “A New Front in the War Over Japanese Chips,”
Business Week, August 9, 1982, pp. 22-23.
86. Robertson, “Japanese RAM Power.”
87. Sabin Russell and Stuart Zipper, “Motorola Rivals See 64K Woes Pressuring De¬
liveries, Prices,” Electronic News, March 28,1983, p. 52; “Quotations of LSIs Stop Falling,
Start Rebounding,” Japan Economic Journal, May 31, 1983; Sabin Russell, “64K RAM
Revival Ends 3-Mo. Lull,” Electronic News, June 23, 1983, p. 62.
88. Mark Blackburn, “Execs Testify in Computer Chip Probe,” Oakland Tribune,
March 18,1983; “64K RAM Makers Face Possible Antitrust Charges ” Japan Times, March
20, 1983. Apparently, however, no depositions were obtained from Japanese management
based in Japan.
152 THE GENESIS OF AN AMERICAN TRADE POLICY

the industry entered one of its cyclical boom periods, trade frictions
receded as an urgent matter requiring attention in Washington.
What was most notable over this period of renewed friction, as the
64K DRAM entered the market, was a subtle shift in the argument about
predatory behavior in semiconductors. By 1980 a large and highly visible
Japanese R&D subsidy in semiconductors, directly focused on DRAMs
and related manufacturing technology, had been ended.89 Formal quotas
had also been ended, and tariffs were low and dropping rapidly.90 Al¬
though complaints about access to the Japanese market persisted, and
even intensified in early 1983, it could no longer be claimed that higher
prices in the home market enabled Japanese producers to persistently
price below average cost in foreign markets. If Japanese producers were
losing money in the United States on sales of dumped memory chips,
they had to be losing money at home as well.
Thus, the U.S. industry analysis of Japanese predation in 1982 nec¬
essarily had to change from one where the Japanese government’s policies
of R&D subsidy and home market protection created conditions encour¬
aging firms to price exports low, without necessarily requiring explicitly
predatory behavior (since home market profits could offset foreign
losses). Instead, the argument now was that, to sustain massive losses
around the world for a prolonged period of time, Japanese firms must
necessarily have adopted an explicitly predatory strategy, with the expec¬
tation that in the long run, with the exit of foreign competitors, rents
could be collected to offset the initial costs of predation. The appearance
of significantly higher U.S. prices for a period beginning in March 1982,
and the charge that the fruits of predation were finally being harvested,
added an additional element to the mix, namely, collusion: American
producers argued that, having achieved a dominant position in the market
for DRAMs, Japanese companies were cooperating, either on their own,
or with administrative support from their government, to cut back supply
on foreign markets in order to collect monopoly rents. Since 1982 this
mix of strategic government industrial and trade policy with strategic—

89. Substantial (but considerably less visible) support from NTT and a privately funded
cooperative follow-up to MITI’s VLSI project, continued, however. See chapter 2 for
details.
90. In February 1985 Japan and the United States agreed to end all tariffs on semicon¬
ductors, effective the following month. “Japan and US Agree to Abolish Semi-Conductor
Tariff Next Month,” Nihon Keizai Shimbun, February 9, 1985, p. 3.
THE GENESIS OF AN AMERICAN TRADE POLICY 153

and collusive—private firm behavior has consistently been presented as

the U.S. industry’s analysis of Japanese production and pricing practices.

Sectoral Negotiations
As the decibel level mounted in public exchanges between U.S. and
Japanese semiconductor companies, pressure for an increasing level of
government involvement in trade issues also rose. The U.S. government
had already responded to a considerable degree back in the spring of
1981, when several planks of the American chip industry’s public policy
platform—notably, accelerated bilateral tariff cuts and adoption of an
incremental R&D tax credit—had been implemented quite quickly. By
early 1982, however, when Japanese producers had displaced American
suppliers in the 64K DRAM market to a degree visible to even the most
determined optimists, political pressure for further action was building
rapidly. In February 1982 senior Reagan administration officials were
considering a Defense Department proposal to impose restrictions on
imports of 64K DRAMs on national security grounds.91 As previously
discussed, top Commerce Department officials used threats and bluffing
to persuade MITI to intervene with Japanese companies.
Processes set in motion over this period were to result in the first
formal bilateral negotiations over semiconductor trade issues, soon to
become a staple of U.S.-Japan relations. Exchanges between U.S. and
Japanese officials led to the idea of forming an ongoing bilateral “working
group” concerned with trade frictions in high-technology industries, such
as semiconductors, to deal with such issues as market access, export
pricing, national treatment, and elimination of tariffs. The formation of
this forum was also seen by the U.S. government, and presented to U.S.
semiconductor companies, as an alternative to aggressive pursuit of a
dumping case.92 In the spring of 1982 the U.S. and Japanese governments

91. Art Pine, “Computer Chips from Japan May Be Curbed,” Wall Street Journal,
February 5, 1982, p. 6.
92. Author’s discussion with former U.S. government officials, September 1992. High-
technology sectors other than semiconductors in which there had been some history of
trade friction included telecommunications (particularly NTT procurement), the Japanese
satellite market, the Japanese computer market, and cooperation on joint R&D projects.
See John Sullivan Wilson, “The United States Government Trade Policy Response to
Japanese Competition in Semiconductors: 1982-1987,” report to the U.S. Congress (Office
154 THE GENESIS OF AN AMERICAN TRADE POLICY

formally agreed to the formation of a High Technology Working Group.

The HTWG held its first meeting in July 1982, and the participants
reached an agreement to focus on bilateral issues in three sectors: semi¬
conductors, telecommunications equipment, and computers. The prin¬
cipal issue on the U.S. side was access to Japanese markets in these
industries, particularly semiconductors. Subcommittees were to be
formed to work on specific recommendations for each of these sectors.
Semiconductors were the first such sector to be chosen for discussion,
and the only sector in which formal agreements were reached under
HTWG auspices.93
In November 1982 the HTWG completed its first “Agreement on
Principles.”94 This framework, accepted by the Japanese and American
governments in February 1983, was a broad agreement on the importance
of high-tech industries and of free trade and investment in sustaining
them, the need to safeguard rules preventing anticompetitive or pred¬
atory practices, and a call for equal national treatment in access to
markets, and investment promotional programs and subsidies, by both
countries.
The only passages of this agreement that specifically mentioned semi¬
conductors called for the formation of a joint task force on high-tech data
collection, and set out an outline for a long-term work program using
sectoral analyses to identify and resolve sources of trade friction in high-
tech industries. Both measures designated semiconductors as the initial
case to be considered.95 In April 1983 a HTWG subcommittee on semi¬
conductors was formally organized, and in July a joint semiconductor
data collection effort was initiated.96

of Technology Assessment, September 1987), pp. 35-37; Prestowitz, Trading Places,

pp. 49-51.
93. Wilson, “The United States Government Trade Policy Response,” pp. 37-41.
94. See “The U.S.-Japan Work Group on High Technology Industries,” SIA Member
News, vol. 1 (February 1984), pp. 1-2, for what may be the clearest account of the HTWG’s
history in print.
95. See “Recommendations of the U.S.-Japan Work-Group on High Technology In¬
dustries” (Washington, February 1983), p. 3, and “Outline for Long Term Work Program,”
attachments to U.S. Trade Representative, “U.S., Japan Exchange Letters on High Tech¬
nology Trade,” press release, February 10, 1983.
96. Ministry of International Trade and Industry, Subcommittee on Semiconductors,
“Study Process of the U.S.-Japan Work Group on High Technology Industries” (Tokyo,
November 1983); and International Trade Administration, “Recommendations of the U.S.-
Japan Work Group on High Technology Industries, Semiconductors” (U.S. Department of
Commerce, November 1983).
THE GENESIS OF AN AMERICAN TRADE POLICY 155

Most significant, a formal bilateral agreement on semiconductors was

negotiated in November 1983. The agreement covered three main areas.
On trade issues, agreement was reached to take steps toward the elimi¬
nation of tariffs, to continue the ongoing joint data collection effort, and
to pursue a variety of other measures to encourage freer semiconductor
trade. Especially notable was the declaration that the Japanese govern¬
ment should “encourage Japanese semiconductor users to enlarge op¬
portunities for U.S.-based suppliers so that long term relationships would
evolve with Japanese companies.” A confidential “chairman’s note” spe¬
cifically outlined four concrete “measures of encouragement”—includ¬
ing, for example, “special consideration” in access to procurement cer¬
tification tests—to which Japanese chip users would “become subject”
in order “to enlarge opportunities for U.S.-based semiconductor pro¬
ducers.” U.S. negotiators took these portions of the agreement to be a
euphemistic articulation of an oral commitment by MITI to provide “ad¬
ministrative guidance” to Japanese firms to increase their purchases of
U.S. suppliers’ chips, and indeed the language of the confidential note
would seem to seem to support this position.97 Another critical point was

97. Other steps listed in the confidential “chairman’s note” included an agreement to
expand “supply sources ... for some semiconductor products in order to give more business
opportunities to U.S.-based semiconductor producers,” “opportunities to make direct con¬
tacts with semiconductor user divisions of the companies,” and “such measures as dispatch¬
ing to or stationing in the U.S. of staffs responsible for procurement.” These measures
were described as applicable to “Japanese semiconductor users who are also semiconductor
producers,” but a concluding sentence in the note promises “best efforts ... to increase
the number of companies which will become subject to the above-mentioned measures of
encouragement.”
The confidential chairman’s note is reproduced in Semiconductor Industry Association,
Japanese Market Barriers in Microelectronics: Memorandum in Support of a Petition Pur¬
suant to Section 301 of the Trade Act of 1974 as Amended (Washington: Dewey, Ballantine,
Bushby, Palmer and Wood, June 14, 1985), p. 91; and in Wilson, “The United States
Government Trade Policy Response,” pp. 46-47.
Clyde Prestowitz, cochair of the U.S. HTWG representation at the time, argues that
the existence of this “secret side letter,” with its agreement to “encourage” Japan users to
“enlarge opportunities,” was the most significant new aspect of the November 1983 rec¬
ommendations. However, this specific language is also found in the main public text, not
just in the confidential note. See Clyde V. Prestowitz Jr., “The New Chip Agreement: Does
It Go Far Enough?” Electronics, October 1991, p. 25; and Prestowitz, Trading Places,
p. 156.
The reasons for keeping this note confidential were most likely related to another issue.
The note specifically suggests that private business decisions of Japanese firms are to be
“subjected” to “measures of encouragement,” and therefore acknowledges an explicit MITI
role in influencing these decisions. Also, although industry sources say little significance
was read into the language at the time, the note applies to JJ.S.-based semiconductor
156 THE GENESIS OF AN AMERICAN TRADE POLICY

the Japanese government’s agreement to “encourage NTT to hold meet¬

ings in Japan and the United States to explain its certification processes,
this the U.S. negotiators understood to be a commitment by the Japanese
government to enforce previous undertakings by NTT to open up its
procurement to foreign suppliers.98
A second portion of the recommendations dealt with investment. Both
governments pledged to remove barriers and to open participation in
regional investment promotion programs to foreign-owned subsidiaries.
A third section on technology sought to promote technology exchanges,
increase U.S.-Japan interaction on quality and reliability issues, and
improve protection of intellectual property related to semiconductors in
both countries. A final section called for a continuing series of bilateral
implementation reviews, with meetings to occur at least twice annually.
The most tangible immediate gains resulting from the HTWG, from
the U.S. industry’s perspective, were in the areas of tariff liberalization
and intellectual property protection. By March 1985 the United States
and Japan had followed up with an agreement to completely eliminate
tariffs on semiconductors, effective the following month. In October 1984
Congress passed legislation, which went into effect in January 1985, pro¬
tecting semiconductor chip mask layout designs.99 A similar law was
passed in Japan in May 1985.
The main focus of U.S. interest, namely, a larger share of the Japanese
market, proved a more elusive goal. Initial results seemed positive: the
largest Japanese companies reported that, after discussions with MITI,
they had elected to increase their purchases of U.S. semiconductors.100
Amid a boom in chip demand during late 1983 and early 1984, sales by
American companies increased noticeably. Even then, however, Ameri¬
can observers complained that “administrative guidance” from MITI was

producers (as opposed to all foreign semiconductor producers) and might therefore be
interpreted as violating GATT commitments to equal most-favored-nation treatment for all
trading partners.
98. “Recommendations of the U.S.-Japan Work Group on High Technology Industries,
Semiconductors”; and author’s discussion with former U.S. government officials, Septem¬
ber 1992. NTT announced a process for testing and certifying semiconductors based on the
HTWG recommendation in March 1984. Electronics Industries Association of Japan, IC
Gaido Bukku, p. 62.
99. The chip mask protection measures also followed a much publicized dispute be¬
tween Intel and Japan’s NEL over the latter’s copying of an Intel microprocessor.
100. See Wilson, “The United States Government Trade Policy Response,” pp. 47-50;
and SI A, Japanese Market Barriers, p. 68.
THE GENESIS OF AN AMERICAN TRADE POLICY 157

offered only to top and middle management in the six largest companies,
with little or no effort invested in contacting smaller users.101 Moreover,
when overall chip sales veered into sharp decline in late 1984, American
chip sales in Japan declined even more rapidly, and U.S. companies’
market share sagged. Evidence of an active MITI effort to kickstart
foreign company sales, or of success in forging long-term relationships,
was notably lacking in the declining market. Many in both U.S. govern¬
ment and industry argued that, despite an almost three-year investment
in the HTWG effort, little in the way of tangible results had actually
been achieved.
Over the long term, though, the HTWG set several important prece¬
dents. This set of negotiations institutionalized continuing high-level bi¬
lateral negotiations over trade issues in a single high-technology sector.
The initial result of this process, the November 1983 HTWG recommen¬
dations, was the very first “semiconductor agreement.” It marked the
first time MITI explicitly agreed to use its leverage with private Japanese
firms to promote imports of semiconductors—such voluntary import ex¬
pansion (VIE) measures were to occupy an increasingly prominent po¬
sition in the MITI program for dealing with trade frictions as the 1980s
progressed. And it was the first time—but not the last—a “secret side
letter” was used as a device to avoid some of the difficult issues raised
by such commitments. The vagueness of the language in which the
negotiators couched these commitments, and the subsequent debates
over the extent to which they had been honored, foreshadowed even
sharper exchanges over language and meaning in later semiconductor
agreements.
As had been true in the past, and would be again, a decline in the
global semiconductor market and in the fortunes of U.S. producers was
the catalyst for a renewed offensive on trade policy, as the truce forged
by the HTWG fell apart. The last such surge in political activity had
occurred in the spring of 1982, spurred by the Japanese manufacturers’
capture of the lion’s share of the 64K DRAM market, weak chip demand,
and heavy price cutting. A major study documenting the arguments that
the Japanese had subsidized their chip producers, engaged in predatory
pricing, and denied U.S. firms free access to the Japanese market had
been commissioned at this time.102 By the time this study was released,

101. Wilson, “The United States Government Trade Policy Response,” pp. 48-49.
102. Semiconductor Industry Association, The Effect of Government Targeting on World
Semiconductor Competition (Cupertino, Calif., 1983).
158 THE GENESIS OF AN AMERICAN TRADE POLICY

however, in early 1983, demand—and prices—had clearly begun to re¬

bound and to blunt the sharp edges of American industry complaints.
Similarly, in late 1982 the SI A had drafted a Section 301 complaint,
alleging unfair Japanese trade practices, based on the charges found in
its 1983 report. But by the spring of 1983 the combination of recovery in
the semiconductor industry and reasonable progress by the HTWG had
led the SI A to shelve its trade case.103
Bust eventually followed boom, however, and the post-HTWG bust of
1985 was one of the worst in the industry’s history. Trade frictions in
semiconductors escalated to an unprecedented level. The events that
followed led directly to the landmark U.S.-Japan Semiconductor Trade
Arrangement signed in 1986, and a radical realignment of the rules of
the game for global competition.

103. See Wilson, “The United States Government Trade Policy Response,” pp. 41-42;
SIA, “The U.S.-Japan Work Group,” pp. 1-2; and Destler and Sato, “U.S.-Japan High
Technology Trade,” pp. 38-44.
 CHAPTER FOUR

The Semiconductor
Trade Arrangement and
Its Aftermath

The gradually escalating tensions over trade and market access

between the United States and Japan exploded into open confrontation
during the industry slowdown of 1985, triggering a process that culmi¬
nated in the conclusion of the Semiconductor Trade Arrangement (STA)
in 1986. This chapter attempts to construct a coherent account of the
actions taken by the U.S. and Japanese governments over the five-year
life of the original STA. The next chapter evaluates what apparent im¬
pacts of these policies could be observed in the marketplace.

A Thumbnail Historical Sketch

To describe the evolution of the semiconductor trade regime since 1985
requires reviewing a broad range of evidence and accounts in order to
assemble the most accurate possible account of the complex political
economy of U.S.-Japan semiconductor trade. Such a detailed accounting
may exhaust the patience of the casual reader. Readers with a minimal
interest in the hows and whys of the changing rules of the game and the
detailed evidence supporting my reconstruction of the salient events may
wish to skip ahead to the assessment of impacts in the next chapter after
reading the balance of this section.

159
160 THE SEMICONDUCTOR TRADE ARRANGEMENT

Briefly, the historical period from 1985 through 1991 may be broken
into four more or less distinct periods for the purposes of understanding
semiconductor trade policy.

Informal Guidance, 1985 to July 1986

In the first period antidumping petitions for three different types of

memory chips (as well as an unfair trade practices complaint under Sec¬
tion 301 of the 1974 trade act and a private antitrust suit) were filed in
the United States against Japanese chip producers. Over this period
politicians and bureaucrats in Japan applied pressure to their domestic
industry to slow exports to the United States. This policy met with some
success, and prices for the first product acted against—64K DRAMs—
stabilized somewhat and even rose in the U.S. market in late 1985. The
64K DRAM dumping case was formally concluded in April 1986, with
the U.S. Customs Bureau applying antidumping duties to Japanese 64K
DRAM imports. The pressure on 64K DRAMs, an older vintage prod¬
uct, accelerated the shift by both consumers and producers to the next-
generation memory chip, the 256K DRAM, which became embroiled in
even more heated disputes. Dumping cases in 256K and higher DRAMs
and EPROMs (erasable programmable read only memories) and the
Section 301 case were suspended at the conclusion of the STA, which
was negotiated by late July and signed on September 2, 1986.

Jab and Feint, August 1986 to November 1987

Disagreement over the interpretation of the STA was virtually imme¬

diate. Although the Japanese government set up a variety of monitoring
and control mechanisms in the fall of 1986, considerable dispute arose
over the extent to which price floors set by the U.S. Commerce Depart¬
ment were to be applied to sales in third-country markets. This issue also
brought the European Community into the conflict, with the Europeans
submitting a complaint to the General Agreement on Tariffs and Trade
(GATT) which argued that such restraints were illegal.
In response to mounting political pressure from the United States over
third-country dumping, the Japanese government successfully offered
guidance to Japanese producers to reduce DRAM output in the first half
of 1987. Export control mechanisms were also used to pressure companies
to meet minimum export price guidelines. Evasion was initially wide-
THE SEMICONDUCTOR TRADE ARRANGEMENT 161

spread, and the U.S. government responded by imposing sanctions on

imports of selected Japanese products in April 1987.
The Japanese government responded by continuing pressures on firms
to hold production down, in an effort to boost prices by restricting supply.
Guidance also covered investment in new capacity by Japanese firms.
These measures prompted the U.S. government to partially lift the sanc¬
tions in June, and again in November 1987 (the remainder of the sanctions
were maintained to express dissatisfaction with the slow pace at which it
was felt that Japanese companies had increased their purchases of foreign
semiconductors).
By late 1987 demand for chips was tightening with a recovery in the
computer industry, the main market for these products. Production con¬
trols—formally forsworn by the Japanese government in November—had
become irrelevant as the industry approached full capacity utilization.
Guidance of investment reportedly continued into 1988, however, and
Japanese producers reported that a system of regional allocation guide¬
lines for exports was in place by late 1987. Administrative measures were
taken that made it more difficult to export chips without the approval of
both manufacturers and the government, reducing the possibilities for
brokers and arbitrageurs to export chips into the gray market.

The Privatization of Restraints, December 1987 to Mid-1990

By early 1988 a full-fledged shortage of DRAMs was being widely felt

in the United States and Europe. Manufacturers increased their surveil¬
lance of chip sales to reduce supplies filtering into the Japanese gray
market.
By late 1988 and early 1989 semiconductor demand had begun to
weaken. In response to a downward drift in DRAM prices, Japanese
manufacturers began to cut back output to maintain high prices. Japanese
newspapers talked of “coordination structures” among Japanese com¬
panies being used to achieve “high price stability.” Producers elsewhere
continued to expand output aggressively, however, and as Japanese pro¬
ducers cut back, the market share of companies based in third countries
(particularly the Korean producer Samsung) rose sharply. As the industry
slid into a slump, measures to ration scarce output among competing
consumers became superfluous: regional allocation of exports was ended
by mid-1989.
162 THE SEMICONDUCTOR TRADE ARRANGEMENT

Faced with increasing competition in more mature products from for¬

eign producers unrestrained by price floors, Japanese companies seemed
increasingly intent on seeking early entry into next-generation products
requiring advanced technology, where price competition from less so¬
phisticated rivals was less intense. This strategy encountered problems in
the early 1990s, however, as a severe recession slowed the movement
toward next-generation memory chips among chip consumers.

Aftermath: The Second Semiconductor Trade Arrangement

A new semiconductor trade arrangement between the United States

and Japan, signed in June 1991, ended the DRAM dumping cases and
formal Commerce Department calculation of floor prices (the EPROM
dumping case continued in a suspended state). A MITI-supervised, fast-
track antidumping procedure was created, however, which required Jap¬
anese companies to continue to collect the data previously given to the
Commerce Department and keep it available on fourteen-day notice, in
the event of an antidumping case. There is reason to believe that this
effectively created a system of “shadow” floor prices on exports, observed
by companies in their export pricing policies. Furthermore, Korean firms,
which had presented Japanese suppliers their most aggressive and sus¬
tained competition, increasingly became subject to the same price floor
discipline, as the outcomes of new dumping cases in the United States
and Europe were announced.

Evolution of the Semiconductor Trade Regime

The year 1984 marked a cyclical peak in the semiconductor business.

By the late fall of that year, however, semiconductor demand in the
United States was weakening rapidly, and the downturn was mirrored in
the Japanese market. A series of rapid declines in price for the predom¬
inant memory chip of the day, the 64K DRAM, were triggered by the
announcement of a sharp cut in its sales price in October 1984 by Micron
Technology, a small American memory chip manufacturer.1 Other man-

1. See Micron Technology, Inc., “In the Matter of 64K Dynamic Random Access
Memory Components from Japan, Petition for the Imposition of Antidumping Duty,”
presented to the U.S. Department of Commerce, June 21, 1985 (public version), p. 22; and
U S. International Trade Commission, 64K Dynamic Random Access Memory Components
from Japan, USITC Publication 1862 (Washington, June 1986), p. A-67.
THE SEMICONDUCTOR TRADE ARRANGEMENT 163

ufacturers, Japanese and American, quickly followed suit, and DRAM

prices plunged further into a sustained decline by early 1985.
As the U.S. chip market continued to weaken in early 1985 and their
domestic sales faltered, U.S. semiconductor companies began to press
complaints in Washington about limited access to the Japanese chip mar¬
ket, where their sales had fallen off even more sharply. Through the spring
of 1985 the Japanese market remained relatively robust, and American
firms began to worry that their Japanese competitors would continue the
trend of record investments in new capacity logged in fiscal 1984, when
Japanese investments in semiconductor facilities had more than doubled
over fiscal 1983 levels. In May 1985 U.S. trade negotiators reportedly
asked MITI to persuade Japanese companies to restrain their investments
in new capacity. The request was turned down.2

Market Access: The Council of Nine

In the spring of 1985 the debate over the existence of barriers to U.S.
companies’ access to the Japanese semiconductor market heated up con¬
siderably in the United States. Known only to a small number of insiders
in American industry and government, some dramatic reports of collusive
practices within the Japanese semiconductor industry had surfaced and
after investigation been judged credible by the U.S. side.
The failure of U.S. firms to reclaim a larger share of the semiconductor
market in Japan after formal trade barriers had been lifted in the mid-
1970s had long been shrouded in conflicting explanations, allegations,
and suspicions, but no smoking gun had ever materialized. Indeed, as
noted in the chapter 3, there was, in the commodity memory chip market,
at least, some evidence that Japanese firms had managed by the late
1970s to turn superior levels of quality into a competitive advantage and
a reasonable explanation for the Americans’ failure to make headway.
But what of the rest of the market?
From the perspective of American industry, the stubborn fixity of its
Japanese market share coupled with the history of Japanese trade and

2. The U.S. request was reportedly characterized as interventionist and asymmetric by

the Japanese, who argued that companies themselves should judge whether or not it was
appropriate to scale back investment plans, and that it would be one-sided for the Japanese
government to do so when the American government was not imposing similar guidance
on American producers. See “U.S. Requests Japan to Hold Down Increasing of Facilities,”
Nihon Keizai Shimbun, May 10, 1985, p. 1.
164 THE SEMICONDUCTOR TRADE ARRANGEMENT

industrial policy in the 1970s constituted sufficient evidence of continuing

market barriers. U.S. suspicions had been reinforced in the late 1970s
and early 1980s by occasional reports of behind-the-scenes intervention
by the Japanese government to discourage foreign sales.3 But others were
unlikely to accept these as solid evidence of a significant problem.
In 1983 when the SIA had made the case that its members had contin¬
ued to be denied access to the Japanese market, its most dramatic evi¬
dence was the history of sales of certain leading-edge products in the
Japanese market in the postliberalization era. Data on several innovative
new products initially showed explosive growth in U.S. exports to Japan
followed by a downward tumble to virtually zero, which coincided with
the production of a version of this same product by Japanese manufac¬
turers.4 The claim was that the moment a Japanese product came on the
market, U.S. sales dried up or at best represented a residual demand for
what could not be shipped by the new Japanese suppliers. The SIA
commented, “this happens even though U.S. firms have been producing
longer, and thus should enjoy lower costs than the Japanese.”5
But no government guidance, private collusion, or other force with a
deliberate policy of organizing behind the scenes to exclude foreign sup¬
pliers was necessary to explain these facts. The disturbing sales patterns
for U.S. chips in Japan might reflect other, perfectly innocent causes
(better local sales and service, better quality, or even an irrational cus¬
tomer preference for Japanese brands), any and all of which might be at
work.
In 1985, however, a series of reports came forth that had lacked in¬
nocent or benign interpretations by even the most agnostic American
standards. Reports gleaned from a variety of sources in Japan spoke of
the existence of a shadowy Council of Nine consisting of sales represen¬
tatives of Japanese producers of discrete semiconductors who met pe-

3. See note 237 in chapter 2 for examples of such reports. The Consulting Group, BA
Asia Limited, in The Japanese Semiconductor Industry 1980 (Hong Kong, May 1980),
pp. 178-79, cites a case in which Japanese power supply makers refused to buy imported
capacitors, claiming that NTT required that only domestic components be used in equip¬
ment sold to it. Japanese customs practices—capricious misclassification of items and
“uplifts” routinely and arbitrarily added to invoiced values—were also cited as an informal
barrier to imports during this period. See U.S. Department of Commerce, A Report on
The U.S. Semiconductor Industry (1979), p. 96.
4. See Semiconductor Industry Association, The Effect of Government Targeting on
Worldwide Semiconductor Competition (Washington, 1983), pp. 79-84. The examples given
are 8080-type microprocessors and programmable read-only memory chips (PROMs).
5. SIA, Effect of Government Targeting, p. 79.
THE SEMICONDUCTOR TRADE ARRANGEMENT 165

riodically in a social setting to set pricing, agree on market shares, and

allocate sales to individual customers. At least some elements of the
Japanese government apparently knew of these activities, and MITI rep-
resentives were reported to have observed or attended some of the meet¬
ings. Knowledge of the Council of Nine was closely held within both U.S.
government and industry, but senior American officials were well aware
of it by 1985, and satisfied that it existed.6
Because the group was never openly investigated, the extent to which
it actually succeeded in affecting competition in the Japanese market for
discrete semiconductors will probably never be known.7 Never issued
publicly, the reports circulating behind the scenes nonetheless had an
important impact.8 The SI A was reinforced in its determination to file
an official Section 301 trade action. When the government received this
complaint, awareness of the existence of the council played some role in
hardening its resolve to support and pursue the 301 case.9 And one can
only speculate that the Japanese government may have felt some vulner¬
ability on these issues and been more willing to negotiate the agreement
eventually signed.10

Political Pressures, 1985 to July 1986

In June 1985 the Semiconductor Industry Association (SIA), the U.S.

industry trade group, filed a Section 301 complaint with the U.S. Trade

6. My account of this episode draws on confidential letters and drafts from U.S. sources
that I was permitted to review and interviews with former U.S. government officials.
Sources of information about the activities of the Council of Nine, and its very existence,
were considered extremely sensitive by both American government and industry.
7. Some within the government apparently argued that such activities were unlikely to
succeed in the long run in benefiting their organizers and thus could safely be ignored.
Interview with government official, February 1996. As noted in chapter 3, from the early
through mid-1980s, semiconductor prices in the Japanese market generally had been near,
and at times below, U.S. prices.
8. The major reason why the existence of the council never surfaced at the time is
probably the sensitivity of sources of information about its activities. Also, because some
Japanese subsidiaries of U. S. companies as well as major Japanese semiconductor producers
were alleged to have been involved in these activities, the implications for trade negotiations
between the United States and Japan would not necessarily have been clean and sharp.
9. Although by all accounts the existence of the council was only one of several issues,
including concerns about national security, that created a consensus to move forward with
the 301 case.
10. Shortly after the 301 case was filed, a senior Japanese trade official reportedly told
a senior U.S. official that “this case has shaken MITI to the core” because of the history
that might be dug up. Interview with former U.S. government official, January 1996.
166 THE SEMICONDUCTOR TRADE ARRANGEMENT

Representative, alleging that barriers to Japanese market access consti¬

tuted an unfair trade practice and asking for retaliatory sanctions. Later
that month Micron Technology filed a dumping complaint with the Com¬
merce Department against seven Japanese producers of 64K DRAMs.
By this time chip sates in the Japanese market, too, had slowed consid¬
erably, and Japanese producers reported plans to slow their investments
in new capacity, to just under the record amounts logged in 1984.11
Meanwhile conflicts over import competition in EPROMs were also
mounting through most of the spring. The leading edge EPROM at that
time was the 256K, and in early 1985 Japanese manufacturers had begun
selling it in significant volume in the U.S. market. A cause celebre was
stirred up when the so-called 10 percent memo, in which Hitachi America
urged its distributors to undercut all rival price quotes by 10 percent, at
a guaranteed 25 percent profit margin, was unearthed and publicized.
(An often unmentioned but important fact is that this memo specifically
urged Hitachi distributors to target Japanese rival Fujitsu, as well as
American makers Intel and Advanced Micro Devices [AMD].)12 By July
1985, when the Commerce Department began its DRAM dumping in¬
vestigation, trade frictions in memory chips had intensified to the point
of open conflagration.
The top levels of the Japanese political establishment got involved on
July 17, when, at a meeting of politicians from Japan’s ruling Liberal
Democratic Party (LDP) with business organization leaders, a top LDP
politician reportedly urged Japanese semiconductor (and automobile)
companies to restrict their exports “on a voluntary basis, instead of doing
things in a clumsy manner like the Government’s taking the initiative.”13
Within a week Hitachi, the largest producer of 64K DRAMs and the
author of the infamous 10 percent memo, announced voluntary restric¬
tions on semiconductor exports for fiscal 1985. Hitachi’s plan to cut ex¬
ports by 30 percent from 1984 levels was quickly followed by a declaration

11. “Requests for Postponement of Delivery Coming in Succession to Companies Man¬

ufacturing Semiconductor Production Equipment; Reduction of Production Plans, Too, is
under Contemplation,” Nihon Keizai Shimbun, June 23, 1985, p. 4.
12. See Intel Corporation, Advanced Micro Devices, Incorporated, and National Sem¬
iconductor Corporation, “In the Matter of: Erasable Programmable Read Only Memories
(EPROMS) from Japan, Petition for the Imposition of Antidumping Duties Pursuant to
the Tariff Act of 1930, as Amended,” September 30, 1985 (nonconfidential version), pp.
19-20, 25, 29, and app. 4.
13. See “LDP Even Likely to Request Export Self-Restraint; Clarification Toward
Business World; Automobiles and Semiconductors as Pillars: Intertwined with Opening of
Market for U.S.,” Asahi Shimbun, July 18, 1985, p. 9.
THE SEMICONDUCTOR TRADE ARRANGEMENT 167

by NEC that it “plans to reduce exports to the U.S., while increasing its
production in the U.S.,” and an announcement by Toshiba that it too
planned to cut its U.S. chip exports by 20 percent in fiscal 1985.14 Despite
these measures, LDP Policy Board Chairman Fujio publicly stepped up
the pressure at the end of July, declaring it necessary to consider the
“possibility of restricting exports of automobiles and semiconductors and
imposing an export surcharge.”15
U.S. government investigations of the Micron 64K DRAM dumping
case and the SIA’s Section 301 petition continued to move forward rap¬
idly through August 1985, when Intel, AMD, and National Semiconduc¬
tor launched yet another dumping case against Japanese EPROM pro¬
ducers. In September Micron filed a private antitrust case as well, against
Japanese DRAM producers.16
The chip market meanwhile remained mired in recession: by October,
Intel, Mostek, and National Semiconductor had announced their inten¬
tion to close down facilities and phase out their production of DRAMs.17
Three Japanese companies with U.S. manufacturing facilities—Fujitsu,
Hitachi, and Toshiba—announced postponements in plans to expand
their U.S. manufacturing operations, while NEC announced a complete
halt in its fabrication of new 64K DRAMs.18

14. See “Hitachi to Reduce Semiconductor Exports to U.S. for This Fiscal Year by
30%,” Nihon Keizai Shimbun, July 23, 1985, p. 1; “Toshiba to Cut Semiconductor Exports
to U.S.,” Kyodo News Wire story, July 24, 1985; and “Chip Makers to Cut Exports to
U.S.,” Japan Economic Journal, July 30, 1985, pp. 1, 4.
15. See “Also Restriction on Exports of Automobiles and Semiconductors; Policy
Board Chairman Fujio,” Yomiuri Shimbun, July 27, 1985, p. 2.
16. That $300 million suit was dropped in October 1986, a month after the signing of
the STA. See Wall Street Journal, October 2, 1986, p. 4.
17. Most US firms dropped out of DRAMs after the initial cuts in semiconductor
exports by Hitachi, NEC, and Toshiba. This contradicts the assertion by analysts affiliated
with the SIA that “the move toward production regulation by the Japanese producers’
group began . . . [when] Japanese DRAM producers had few competitors left except each
other.” Thomas R. Howell, Brent L. Bartlett, and Warren Davis, Creating Advantage:
Semiconductors and Government Industrial Policy in the 1990s (Washington: Semiconductor
Industry Association and Dewey Ballantine, 1992), p. 117.
Also, Japanese chip exporters had previously cut DRAM exports in response to political
pressure, in 1982, when Japanese world market share was very much smaller than in 1985.
Political pressure, rather than a sudden jump in market power, appears to have stimulated
supply cuts in both cases.
18. “Construction of Very Large-Scale Integrated Circuit Plant in U.S.; Toshiba Post¬
pones Plan; By Half a Year, or One Year, Due to Semiconductor Depression,” Nihon Keizai
Shimbun, October 23, 1985, p. 9; and “NEC to Adjust Production of 64K DRAMs;
Suspends Pre-process Operations; To Digest Semiconductor Products in Stock,” Nihon
Keizai Shimbun, October 31, 1985, p. 9.
168 THE SEMICONDUCTOR TRADE ARRANGEMENT

It was in this atmosphere of crisis that U.S. and Japanese negotiators

held successive, frustrating rounds of talks on semiconductor trade prob¬
lems in August, September, and October of 1985. By November it
was known in Tokyo that the U.S. government was considering initiat¬
ing an antidumping investigation against Japanese producers of latest-
generation 256K DRAMs; for the first time ever Washington itself would
be launching the suit rather than waiting for the industry to come for¬
ward.19 Aware of the rumblings in Washington, Japanese manufacturers
increased their U.S. export prices for 256K DRAMs that November.20
By the end of 1985, though, the worst of the semiconductor recession
finally seemed to have passed, and the Japanese press was reporting that
key manufacturers were even discussing an expansion of their production
of 64K DRAMs. The “production coordination” (as the round of export
and production cuts and price increases were described in the Japanese
press) launched by major Japanese producers earlier that summer was
increasingly effective in pushing up 64K DRAM prices.21 By December
64K DRAM prices were reported to have risen to 85 cents, from a low
of 30 cents in June 1985, and the major Korean producer Samsung, which
had actually suspended production of 64K DRAMs to staunch severe
losses, resumed shipments under the price umbrella created by the Jap¬
anese actions.22 But the successful efforts by Japanese manufacturers to
collectively stabilize prices for 64K DRAMs were too late to halt the
administrative machinery now set into motion by Micron’s dumping pe-

19. Of course, the government was considering this action because of industry com¬
plaints. See “U.S. Heading Toward Antidumping Investigation on 256 Kilobit DRAM of
Japanese Make; Conference of Secretaries Advises President; Hard Fight over ‘Strategic
Goods’ Gives Rise to Crisis Feeling,” Nihon Keizai Shimbun, November 8, 1985, p. 6.
20. “Semiconductor Maker, Full Power for 256K Price Hike, MPU for Price Cut,”
Nihon Keizai Shimbun, January 14, 1986, p. 18.
21. See “Semiconductor Industry Showing Signs of Recovery from Depression,” Sankei
Shimbun, December 5, 1985, p. 6; see also “Following ‘Leather,’ Also ‘Semiconductors’
Have Hard Sailing; Japan-US Consultations; MITI Officials in Charge Impatient Without
Good Idea,” Tokyo Shimbun, December 6, 1985, p. 3; “Semiconductor Companies Remain
Calm Toward Preliminary Ruling of ‘Guilty’ on 64K DRAM; Upper-Grade Item Now
Attached with Major Importance; Consultations between Japanese and US Governments
Are Watched,” Nihon Keizai Shimbun, December 6, 1985, p. 8.
22. See “Int’l Price Recovery Sparks Semiconductor Industries; Samsung Planning to
Expand 64K D-RAM Plant to Meet Rising Demands,” Korea Herald, December 26, 1985;
and “Chip Manufacturers Slate New Facility Investments,” Korea Herald, February 2,1986.
By December 1985 regional price differentials appear to have been created in the 64K
DRAM market, with Japanese domestic prices falling even as U.S. export prices were
rising. See, for example, “Price Reduction Demand for Semiconductors Intensifies,” Nihon
Keizai Shimbun, December 4, 1985, p. 20.
THE SEMICONDUCTOR TRADE ARRANGEMENT 169

tition. In early December the Commerce Department announced a pre¬

liminary finding of dumping in 64K DRAMs.
Higher 64K DRAM prices, coupled with a continuing increase in
supply of and falling prices for 256K DRAMs, accelerated the trend for
chip users to switch to next-generation 256K DRAM in their electronic
systems production. Consumers, Japanese and American producers, and
the American government now shifted their attention toward this prod¬
uct. By mid-December the Commerce Department, as threatened, had
“self-initiated” an antidumping investigation of DRAMs with densities
of 256K or greater.23 As tensions continued to mount and NEC and
Hitachi had “self-reflected upon their excessive competition for mass
production which led to the decline of [256K DRAM] prices,” the two
firms began to cut back on 256K DRAM production as well in late
December 1985.24 As in the case of 64K DRAMs, however, the dumping
machinery, once set in motion, was not to be stopped by anything less
than a formal government agreement.
By early 1986 Japanese producers were announcing 256K DRAM price
increases for their domestic customers (U.S. export prices had been
raised the previous November), in a continuing attempt to reduce trade
friction.25 By late January delivery prices for 256K DRAMs had risen in
Tokyo, in both the spot and the large-scale contract markets.26 Japanese
users complained loudly about the unnatural price increases induced by
gaiatsu (foreign pressure).27 But the administrative gears set in motion

23. This dumping investigation was unusual in two respects. First, it was one of a
handful of antidumping cases ever initiated by the U.S. government itself, rather than an
industry petitioning the government. Second, it covered 256K and higher density DRAMs,
including new products not yet on the market.
24. “Nippon Electric and Hitachi Begin to Curb 256K DRAM Production; Watching
U.S. Dumping Investigations,” Nihon Keizai Shimbun, December 18, 1985, p. 9. See also
“NEC, Hitachi Hold Down 256K DRAM Production; To Avoid U.S. Dumping Charge,”
Japan Economic Journal, December 28, 1985, p. 18.
25. See “Super LSI Domestic Shipment Price Hike a Little Over 10%, Consideration
of Friction with U.S.,” Nihon Keizai Shimbun, January 11, 1986, p. 1; “Semiconductor
Maker Full Power,” p. 18; and “Semiconductor Demand to Recover This Year; From
Demand-Price Survey, Price Crash Period Ends; Bottom Spreads Supports Price,” Nihon
Keizai Shimbun, January 14, 1986, p. 18.
26. “Negotiation for Determining Price, Next Month to Be the Peak; Makers Forceful
for 256K Price Hike; Users Demand Price Reduction Due to Yen Appreciation,” Nihon
Keizai Shimbun, January 29, 1986, p. 20; and “Semiconductor: Due to Japan-U.S. Trade
Friction, Unprecedented Price Rise; U.S.-Made Import Doubtful; Maker Confident in
Profit Maintenance,” Nihon Keizai Shimbun, January 30, 1986, p. 20.
27. “This price increase by Japanese makers, aimed at calming bilateral trade friction,
170 THE SEMICONDUCTOR TRADE ARRANGEMENT

by the dumping cases continued to turn: by March 1986 the Commerce

Department had added preliminary dumping determinations in both the
EPROM and 256K DRAM cases to its December 1985 finding on 64K
DRAMs.28 Following the March EPROM ruling, Intel reportedly raised
its EPROM prices by an average of 25 percent; Japanese producers con¬
cerned about trade frictions followed Intel’s lead by raising domestic sales
prices, exercising “self-restraint” on low-priced sales of EPROMs.29
U.S. and Japanese semiconductor manufacturers also—for the second
time—organized an industry “summit” meeting in March 1986. (The
first such industry-to-industry meeting had been held in Hawaii in No¬
vember 1983.) A speech at this meeting, held in Los Angeles, by MITI
Deputy Director General Tanahashi of the Machinery and Information
Industries Bureau was believed by many Japanese semiconductor execu¬
tives to have been the origin of the 20 percent market share target for
foreign semiconductors later included in a side letter to the STA.30

has its roots in political judgments beyond simple market principles.” “Semiconductors:
Abnormal Price Increase Caused as a Result of the U.S.-Japan Trade Friction,” Nihon
Keizai Shimbun, January 30, 1986, p. 20. The article concludes, “One maker pointed out
that the year 1986 marks the first time all the makers share a sense of cooperation in the
history of the semiconductor industry, unlike in the past when they were competing with
each other for higher production and lower prices. They expect that less price competition
will bring them larger profits in [fiscal] 1986.”
28. The preliminary dumping margins found by the Commerce Department in these
cases were relatively high and undoubtedly created further pressure for a negotiated solu¬
tion to the conflict. For EPROMs, the rates were as follows: Hitachi, 29.9 percent; Toshiba,
21.7 percent; NEC, 188 percent; Fujitsu, 145.9 percent. See “Political Settlement Between
Japan and US Expected by Manufacturers; Tentative Decision of ‘Guilty’ on EPROM,”
Nihon Keizai Shimbun, March 13, 1986, p. 9.
29. “EPROM Prices in Steady Tone—Semiconductors; Japanese Manufacturers Exer¬
cise Self-Restraint on Low-Priced Sales,” Nihon Keizai Shimbun, April 2, 1986, p. 18.
30. In that speech Tanahashi reportedly stated that “it is expected that the total share
of U.S.-made semiconductors used by five large Japanese manufacturers will account for
19.5 to 20 percent in 1990.” This was interpreted in the Japanese press as a proposal from
the Japanese side. See “How Should We Settle Japan-US Semiconductor Friction?” Nihon
Keizai Shimbun, April 2, 1986, p. 2. It was precisely at this point that reports of a 20
percent market share target by 1990 began to circulate within the Japanese semiconductor
industry. See Mikio Fujiwara, “This Is a Side Letter to the U.S.-Japan Semiconductor
Agreement,” Bungei Shunju, May 1988, pp. 124-37 (the author, an executive in a Japanese
semiconductor firm, wrote the article under a pseudonym).
On the U.S. side, this March industry meeting prompted an investigation by the Federal
Trade Commission. See “Suspicions about Violation of Antimonopoly Law—Guarantee
of Share Rate for Semiconductors Imported from US; Fair Trade Commission’s View,”
Asahi Shimbun, April 24, 1986, p. 2. The commission’s probe was not unusual; it has often
examined the outcomes of antidumping agreements.
THE SEMICONDUCTOR TRADE ARRANGEMENT 171

Birth of the STA

Monthly talks on semiconductor trade issues between the American

and the Japanese government began tb pick up speed as they approached
their sixth session, in March 1986. In late February MITI had reportedly
decided to propose a minimum export price system to the American side
at the upcoming meeting.31 With the pressure of a July deadline on the
original Section 301 case hanging over them at the meeting, U.S. and
Japanese negotiators were reported to have been near agreement on a
global price control and monitoring scheme, but a potential deal was
blocked by continuing disagreement over an American chip industry
proposal to specify a minimum share target for the Japanese chip mar¬
ket, and over measurement issues associated with that plan.32 Citing Japa¬
nese “intransigence,” the United States suspended the negotiations on
March 28.
By mid-April the proposed global cost and sales price monitoring
system was being discussed openly by the SI A.33 In Japan MITI also
publicly floated the outline of a proposed price monitoring system shortly

31. The legal basis for the proposal being considered was “flexible application of the
Export Trade Control Ordinance or formation of export cartels on the basis of the Export
and Import Transactions Law.” “MITI Heading Toward Introduction of Minimum Price
System for Semiconductor Exports to US; Proposal Will be Made to US Next Month, at
Earliest,” Sankei Shimbun, February 20, 1986, p. 7.
32. See Louise Kehoe, “US and Japan Poised for Agreement on Semiconductors,”
Financial Times, March 20, 1986, p. 6. Even in its preliminary form the global price
monitoring scheme was an immediate target for its critics: a Financial Times editorial of
March 18 blasted the plan as a “blatantly protectionist” international cartel. Clayton K.
Yeutter, then U.S. Trade Representative, responded that “there is no chance whatsoever
of that occurring, and any such suggestion would be vehemently opposed by the U.S.
government!” See “US, Japan, and Semiconductors,” letter to the editor, Financial Times,
April 25, 1986, p. 21.
It is interesting to contrast Yeutter’s comments with the following assessment of objec¬
tives by Clyde V. Prestowitz Jr., who as an adviser to Secretary Malcolm Baldrige helped
shape the Commerce Department’s position in the semiconductor negotiations: “we would
have to persuade the Japanese government to force up prices in its home market as well as
in third-country markets (that is, markets in countries other than the United States and
Japan). This amounted to getting the Japanese government to force its companies to make
a profit and even to impose controls to avoid excess production—in short, a government-
led cartel.
For the free-traders of the United States to be asking Japan to cartelize its industry was
the supreme irony. Yet it was logical.” Clyde V. Prestowitz Jr., Trading Places: How We
Allowed Japan to Take the Lead (Basic Books, 1988), p. 62.
33. See Yoko Shibata and Louise Kehoe, “NEC Bid to Beat US Dumping Rules,”
Financial Times, April 21, 1986, p. 6.
172 THE SEMICONDUCTOR TRADE ARRANGEMENT

thereafter. MITI’s proposed system had three elements: first, administra¬

tive guidance to producers on output (using as an instrument a supply-
demand forecast committee similar to one used in the Japanese steel
industry); second, a company-specific minimum price floor for U.S.
exports; and third, a uniform export price floor for exports to other
countries.34
Also in April 1986 the Commerce Department issued its final dumping
determination for 64K DRAMs, effectively ending the 64K DRAM
dumping case. Henceforth Japanese imports would pay (company-
specific) duties to U.S. Customs. The dumping margins found were
11.9 percent for Hitachi, 13.4 percent for Mitsubishi, 22.8 percent for
NEC, and 35.3 percent for Oki Electric.
The suspended semiconductor trade talks were resumed in mid-May,
and by the end of that month the essentials of a framework agreement
were reported to have been negotiated by MITI Minister Michio Watan-
abe and U.S. Trade Representative Clayton K. Yeutter. The agreement’s
basic outline was reported in the press: the U.S. antidumping and Section
301 cases were to be suspended in exchange for the implementation of a
system to monitor Japanese pricing of exports (to U.S. and third-country
markets) and a Japanese government commitment to assist in doubling
foreign chip producers’ share of the Japanese market over five years, to
20 percent.35

34. See “Lowest Prices by Enterprises to be Fixed—Semiconductors; Japan’s Plan to

be Proposed for Negotiations with U.S.; Demand-Supply Outlook to Be Announced Four
Times per Year; Goal for Increase in Imports Rejected,” Nihon Keizai Shimbun, April 24,
1986, p. 1.
The fact that MITI proposed the specifics of this system in April, before the STA
negotiations were actually concluded (in late July), constitutes proof that the STA did not
“cause” the creation of a de facto DRAM cartel, according to a recent SIA analysis: “[the
STA causing the organization of a Japanese cartel] was impossible; the joint producers’
activity [the late 1985 ‘production coordination’ described above], as well as the plan for
the MITI guidepost system, existed well before the arrangement.” See Howell and others,
Creating Advantage, p. 121.
But, as described above, American government and industry complaints, legal initia¬
tives, and other manifestations of continuing trade friction were clearly the proximate
“causes” of the political pressure on the industry to jointly cut exports and raise prices in
the fall of 1985. (There had, in fact, been similar actions before, in 1982, under MITI
guidance, in response to U.S. government complaints.) MITI’s April proposal to establish
its three-point price monitoring system was part and parcel of the negotiation process that
culminated in the STA several months later. The STA itself was the legal framework that
regularized and justified these actions as official, bilateral, Japanese and American govern¬
ment policy.
35. See Peter Waldman, “U.S.-Japan Microchip Pact Would Offer Big Boost to Ailing
THE SEMICONDUCTOR TRADE ARRANGEMENT 173

As the negotiations neared a climax, new interests were interjected

into the proceedings. In late April, reportedly concerned that Japanese
memory chip exports would be diverted from U.S. to European markets,
European chip makers set up a working party to bring dumping charges
against Japanese memory chip exports.36 In June, sensing an opportunity
and concerned that their interests were being forsaken, U.S. makers of
application-specific integrated circuits (ASICs), which are customized
versions of standard logic chips, demanded that—as victims of the same
unfair trade practices affecting American memory chip makers—they too
be included in the framework of the STA.37 As the U.S. and Japanese
negotiations approached an end-of-June deadline to suspend the pending
dumping suits, European chip producers rushed to inform the European
Commission that they intended to file an antidumping complaint against
Japanese memory chip imports.38
The final details required considerable further wrangling. Major stick¬
ing points were the scope of the proposed price and cost monitoring
system (more precisely, whether just the United States or Europe and
Asia as well were to be covered), and the details of how costs were to be
determined.39 As the June 30 deadline came and went, the Commerce
Department instituted a “provisional” June suspension of the EPROM
dumping case, and in early July the 256K and higher DRAM dumping
case was suspended. An intense negotiating process continued, largely
centered on the international scope of the proposed price and cost mon¬
itoring system and the issue of whether specific market share targets
would be stated explicitly or implicitly in the agreement.40 By late July,
however, the essential details had been worked out: “secret” side letters

American Producers,” Wall Street Journal, May 30, 1986, p. 40; and Jim Van Nostrand,
“U.S. Claims Trade Breakthrough in Semi Talks,” Electronic Engineering Times, June 2,
1986, pp. 1, 16.
36. Christian Tyler, “Europeans in Chip Dumping Check,” Financial Times, May 1,
1986, p. 6.
37. See Louise Kehoe, “Japanese Targeting Further Chip Market, US Makers Claim,”
Financial Times, June 20, 1986, p. 7.
38. Tim Dickson, “EEC Chip Makers Accuse Japan,” Financial Times, July 1, 1986,
p. 14.
39. Louise Kehoe, “Compromise Likely in US-Japan Semiconductor Talks,” Financial
Times, June 25, 1986, p. 4.
40. Louise Kehoe, “US, Japan Fail to Meet Chip Row Deadline,” Financial Times,
July 2, 1986, p. 6; and Louise Kehoe and Carla Rapoport, Financial Times, “Not Long
Left to Turn a Chip Truce Into a Treaty,” July 9, 1986, p. 5.
174 THE SEMICONDUCTOR TRADE ARRANGEMENT

specified an “expectation” of a 20 percent Japanese market share for

foreign companies within five years, and an agreement in principle to
monitor exports into third-country markets, while the main body of the
arrangement set out the basics of a price and cost monitoring system.41
As a rescheduled July 30 deadline approached, all sides stepped up
their pressure. U.S. chip makers launched a new round of complaints of
price cutting in EPROMs, in the aftermath of the provisional suspension
of the dumping case for that product.42 Both the Japanese and the Amer¬
ican industries reportedly applied eleventh-hour pressure on their gov¬
ernments’ negotiating teams to alter the terms of the provisional agree¬
ments.43 In the end, however, an extensive third-country price monitoring
system was included in the agreement, applying to 90 percent of chips
sold by Japanese producers to sixteen countries worldwide (including the
United States, six European countries, and Asia), as was a “secret” side
agreement in which Japan recognized “the U.S. semiconductor industry’s
expectation that semiconductor sales in Japan of foreign companies will
grow to slightly above 20 percent.”44 Ironically, this controversial and
allegedly secret side letter was quite public right from the start, with the
exact wording of relevant sections reported on at least two occasions, just
before and after the agreement was completed, in the Financial Times of
London.45
Thus, much of the content of the U.S.-Japan Semiconductor Trade
Arrangement of 1986—both public and “secret”—was widely reported
from the moment the negotiations ended on July 31. Much less clear
was what the agreement actually meant for the chip market and its
participants.

41. Louise Kehoe and Carla Rapoport, “Tokyo May Agree to Aid US Chip Sales in
Japan,” Financial Times, July 21, 1986; and “US and Japan Near Accord on Semiconduc¬
tors,” Financial Times, July 21, 1986, pp. 1, 18.
42. See Brenton R. Schlender, “Japanese Dumping of Chips in U.S. Said to Continue,”
Wall Street Journal, July 25, 1986, p. 8; and Louise Kehoe and Carla Rapoport, “Japan
Stepping Up Chip Dumping, Says US,” Financial Times, July 26, 1986, p. 20.
43. See Louise Kehoe, “Stalemate in Chips Trade Talks,” Financial Times, July 28
1986, p. 1.
44. Carla Rapoport, “Japan Yields to US Demands on Microchip Price Monitoring,”
Financial Times, August 4, 1986, p. 1.
45. See Kehoe, “Stalemate in Chips Trade Talks,” p. 1; and Rapoport, “Japan Yields
to US Demands,” p. 1.
THE SEMICONDUCTOR TRADE ARRANGEMENT 175

Initial Implementation of the Arrangement

August 1986 to November 1987

From the start it was clear that some Japanese chip executives opposed
the concessions made by MITI in the STA.46 Indeed, Japanese company
executives and others immediately raised doubts about the legality of
both the third-country price monitoring system and the apparent com¬
mitment to increase foreign chip sales in the Japanese market.47
But even more stringent and direct controls over the Japanese industry
had been built into the STA. An explicit, though little discussed, provi¬
sion of the agreement required MITI to exert direct control over Japanese
256K DRAM exports, to ensure that shipments to the United States
through mid-September took place at the “normal” rate.48 Although this
transitional provision was designed to thwart a last-minute rush of ship¬
ments by Japanese companies in July and August to circumvent the loom¬
ing specter of price floors, the U.S. government had clearly set a prece¬
dent for tolerating—indeed, requiring—direct intervention by MITI in
setting quantities to be shipped by individual companies. Recall also

46. As early as July 7, the chairman of the Keidanren (a business organization repre¬
senting Japan’s largest corporations) expressed “apprehension about the excessively restric¬
tive nature of the agreement.” “Japanese Makers Dissatisfied with Chip Trade Accord,”
Nikkei News Bulletin, July 7, 1986. In August a MITI official was quoted as saying, “We
still don’t think we have [the Japanese industry’s] agreement.” See Carla Rapoport, “Tokyo
Concerned About Setting a Trade Precedent,” Financial Times, August 4, 1986, p. 4. See
also Tim Dickson and Carla Rapoport, “EEC and Japan Criticise Semiconductor Agree¬
ment,” Financial Times, August 2, 1986, p. 1.
47. Brenton R. Schlender and Stephen Kreider Yoder, “U.S.-Japan Semiconductor
Agreement Is Expected to Prove Difficult to Enforce,” Wall Street Journal, August 4, 1986,
p. 19; Rapaport, “Japan Yields to US Demands,” p. 1; and “Tokyo Concerned about
Setting,” p. 4.
48. This agreement was contained in a supplementary “secret” memorandum dated
August 1, 1986, initialed by Yeutter and Watanabe. The exact text is as follows:

Memorandum Regarding July 31, 1986

August 1, 1986
1. MITI will take appropriate actions to control the volume of exports of 256K
DRAMs from Japan to the U.S. for the period between July 2 and September 15.
MITI will ask Japanese semiconductor companies to see that shipments in the U.S.
and indirect shipments to the U.S. of 256K DRAMs will take place at the normal
shipment rate during the said period.
2. For this purpose, MITI has already started the technical work necessary to
find out what is the current normal shipping rate. This rate is to be worked out on
an objective and unbiased basis by experts by August 15. ,
3. MITI is also ready to devise in detail the necessary legal or administrative
procedures to achieve the aim in [point] 1 as soon as possible.
176 THE SEMICONDUCTOR TRADE ARRANGEMENT

that, back in April, MITI had publicly proposed a system of administra¬

tive guidance to producers on output levels, using a supply-demand fore¬
cast committee as an instrument.
At least some MITI officials clearly had even more drastic controls in
mind. After interviews with MITI officials, Carla Rapoport of the Finan¬
cial Times reported in early August 1986 that “Further, MITI realizes
that price monitoring will be ineffectual if it does not force the industry
to cut back on its capital spending plans for enlarging Japan’s already
huge chip-building capacity. Even more radically, it is understood that
MITI may be aiming to force the industry to mothball some of its existing
capacity.”49 These thoughts were shortly to be translated into concrete
actions.
The immediate benefit to Japanese producers was suspension of two
dumping cases (in EPROMs and in 256K and higher DRAMs) and the
Section 301 case, in early August 1986. In exchange, the U.S. Commerce
Department was to “advise Japanese companies exporting to the US of
appropriate fair prices for their products so that they may avoid dump¬
ing.” For other markets and products, the “secret” side letter required
the Japanese government to “monitor company-specific costs and export
prices in order to prevent dumping.”50
It is not widely recognized quite how extensive in scope the original
STA of July 1986 was. In addition to covering DRAMs and EPROMs
exported to the United States, which were to be priced above “fair
values” established by the U.S. Commerce Department, the STA frame¬
work called for prices of DRAMs and EPROMs shipped to other, third-
country markets to be “monitored” by MITI. In addition, MITI was to
monitor “representative” products from seven other classes of semicon¬
ductors—MOS SRAMs, ECL RAMs, microprocessors, microcontrol¬
lers, ASICs, ECL logic chips, and EEPROMs—using company-specific
cost and export price data submitted to the ministry.51 Other products
could be added to the list, after consultation, by either government. The
side letter specified that “based upon monitoring or consultation, the

49. Carla Rapoport, “Tokyo Concerned about Setting,” p. 4.

50. This is a direct quotation from point 2 of section II of the side letter, describing
“Third Country Market Measures.”
51. SRAMs are static random access memory chips; ECL is emitter coupled logic;
ASICs are application-specific ICs; and EEPROMS are electronically erasable program¬
mable read-only memory chips.
THE SEMICONDUCTOR TRADE ARRANGEMENT 177

GOJ [Government of Japan] will take appropriate actions available under

laws and regulations in Japan, including ETC [the Export Trade Control
ordinance], in order to prevent dumping.”52 Thus the STA explicitly
called for the use of Japan’s export control regulations, administered by
MITI, for the purpose of enforcing price floors.53
Virtually immediately, in the first part of August, MITI began to
invoke the export trade control ordinance, and the ministry completely
suspended the issuance of new export licenses for 256K DRAMs, to deal
with the “last minute export contract” problem described above (sales
in export contracts had roughly tripled in July).54 Japanese producers
were advised to either export at the fair value set by the Commerce
Department—not the price set in earlier contracts—or forgo exports.
Implementation of the STA moved quickly in the fall of 1986. In
August the Commerce Department had also issued its initial, company-
specific “fair values” (known as foreign market values, or FMVs) for
DRAMs and EPROMs, and not unexpectedly, Japanese producers faced
large jumps in the prices set for their exports to the U.S.55 On Septem¬
ber 2, 1986, the official signing of the STA took place in Washington. By
mid-September MITI had set up its so-called Forecast Committee, whose
task it was not only to publish forecasts but also to help in “correction
of imbalances” between Japanese supply and demand for products cov-

52. This is point 5 of section II, “Third Country Market Measures,” of the letter.
53. During the negotiations, the American side first suggested that the Japanese use
their export control act for enforcement purposes. Author’s interview with a member of
the U.S. negotiation team, 1989.
54. “Certificate of Approval for Exports to US Withheld, Semiconductors; For Part
Covering Last Minute Contracts; Carrying Out of Promise at Time of Reaching Agreement
at Negotiations, MITI; 256 Kilobit DRAMs as Objects,” Asahi Shimbun, August 6, 1986,
pp. 12-13; “MITI ‘Blocked’ Big Chip Exports Before Price Rise,” Japan Times, August 7,
1986, p. 1; and “MITI to Control July Semiconductor Exports,” Nikkei News Bulletin,
August 7, 1986.
55. The range of FMVs set for individual companies apparently became much more
compressed over time. Company-specific 256K DRAM FMVs initially set in July for the
third quarter of 1986 reportedly ranged from $2.50 to $8.00. The second round of FMVs,
for the fourth quarter of 1986, reportedly ranged from $2.50 to $4.00, while the third
round, set for the first quarter of 1987, spanned $2.50 to $3.00. The FMVs for the second
quarter of 1987 were reportedly identical to the first-quarter floor prices, but those in the
third quarter of 1987 dropped to the $1.89-$2.75 range (averaging $2.34). See Nikkei Top
Articles, December 2,1986; “U.S. Announces No Revision—Fair Prices of Japanese-Make
Semiconductors for April-June Period,” Nihon Keizai Shimbun, March 24, 1987, p. 3; and
“U.S. Dept, of Commerce Announces Fair Market Prices for ICs,” Nikkei News Bulletin,
September 28, 1987.
178 THE SEMICONDUCTOR TRADE ARRANGEMENT

ered by the price monitoring framework.1'6 This twelve-person commit¬

tee, representing manufacturers, users, and outside experts, was to meet
and approve a forecast for the final quarter of 1986, to be released
publicly by MITI on September 30.57 The side letter explicitly called for
this committee to issue forecasts, which were to generate some contro¬
versy (as discussed below).58 On September 22 MITI established a new
office exclusively devoted to running the export monitoring system.59
Bilateral agreement over what the price monitoring system was sup¬
posed to do was absent from the start. The price monitoring mechanism
initially set lower standards for prices in third-country markets than for
sales to the United States. MITI argued that the STA required it to end
“below cost” pricing in third-country markets, but required the use of

56. “Quarterly Price Monitoring of Semiconductors; MITI Announces Prospects for

Supply and Demand as Well,” Nihon Keizai Shimbun, September 23, 1986, p. 1.
57. The Forecast Committee met on a quarterly basis to discuss historical and forecast
data supplied to it by MITI. MITI actually compiled the numbers appearing in the forecast,
based on information requested from-—and furnished by—Japanese semiconductor pro¬
ducers and users. After this quarterly meeting the supply-demand forecast for the next
quarter was then formally issued by MITI.
Three of the committee members came from Japanese semiconductor producers: To¬
shiba (representing the major established producers; Toshiba’s K. Kadono was also chair
of the electronics policy board of the Electronic Industries Association of Japan, or EIAJ),
NMB Semiconductor (a recent, relatively small and independent Japanese entrant into
DRAM production, represented by T. Tamura), and Texas Instruments Japan. Six industry
members represented user industries: Hitachi’s K. Fujiki, chair of the Japan Electronic
Industry Development Association (a computer industry group); NEC’s M. Yamauchi,
chair of the Communication Industry Association of Japan’s management board; Canon’s
K. Yamaji, a member of the board of the Japan Business Machine Makers Association;
Toshiba’s F. Ota, chair of the EIAJ’s consumer electronics board; Yokogawa Electric’s S.
Yamanaka, representing electronic measuring equipment makers; and IBM Japan’s M.
Motobayashi. Two of the three industry “experts” were dispatched from the Nomura Re¬
search Institute (O. Hayama) and Dataquest Japan (S. Morishita), both well-established
consulting firms with close links to the industry. The third outside expert, Professor Tadao
Miyakawa, of Hitotsubashi University, a frequent consultant to government and industry
on electronics industry forecasts, chaired the committee. See MITI, “Semiconductor De¬
mand and Supply Forecast Committee,” unpublished document, n.d., given to the author
in September 1988. Note that TI Japan’s participation on a twelve-member Forecast Com¬
mittee is not mentioned in this document (which lists only eleven names), but was specifi¬
cally cited in an interview with MITI officials, September 1988.
58. Point 5 of section I of the side letter, covering market access, reads as follows: “Both
Governments recognize the importance of discouraging marketing activities which serve to
undercut the intent of the Arrangement. The Government of Japan will compile demand
and supply forecasts on the Japanese semiconductor market in compliance with its domestic
laws and regulations.”
59. “Quarterly Price Monitoring of Semiconductors,” p. 1; and “MITI Creates Chip
Export Price Monitoring Office,” Nikkei News Bulletin, September 22, 1986.
THE SEMICONDUCTOR TRADE ARRANGEMENT 179

the Department of Commerce’s cost accounting procedures only on ex¬

port sales to the U.S. market. For third countries, MITI argued, the
ministry would use its own costing guidelines, which, for example, would
recognize lower marketing and distribution costs in non-U.S. markets.
After further U.S. pressure, however, MITI made some concessions and
agreed to some unification of the two monitoring systems for exports
after November 1986.60 Late that month it notified producers that the
trade control ordinance might be applied to third-country exports. Dis¬
agreement about whether or not the STA required identical pricing of
exports to U.S. and third-country markets was a point of continuing
dispute well into 1987.
Enterprising merchants quickly invented a variety of creative mecha¬
nisms to evade the newly minted MITI controls.61 Separate “sales pro¬
motion rebates” to customers were granted to lower the effective cost of
overseas chip shipments below the invoiced (and monitored) export
prices. Brokers plugged chips into sockets on electronic circuit boards so
that they could be classified as circuit board rather than chip exports,
and thus circumvent MITI scrutiny of chip export prices; the chips could
later be pulled from the sockets and used in other products. Large ship¬
ments were divided into transactions worth less than a million yen each
(the trigger level requiring explicit MITI approval for export prices);
MITI responded by dropping the minimum value of transactions requir¬
ing approval to a mere 50,000 yen after December 1986.
Domestic dissatisfaction with MITI’s monitoring system and produc¬
tion guidelines had intensified in Japan by early 1987, developing into a
major source of continuing semiconductor trade frictions. DRAM output
in the last quarter of 1986 had vastly exceeded the MITI forecast, and
exports to third countries in Southeast Asia had surged. When, at the
end of December 1986, the Forecast Committee put out its forecast for
the first quarter of 1987, it called for a sharp reduction in output by
Japanese producers.

60. See “To Unify Standards for Calculating Original Costs of Semi-conductors; MITI’s
Policy; Aimed at Calming Down Dissatisfaction over Price Differentials in Exports to Third
Nations,” Asahi Shimbun, November 23, 1986, p. 9.
61. ‘“One Product Having Three Prices’ Distorts Markets; Five Months Since Japan-
US Semiconductor Negotiations; Mutilated Restrictions with Loopholes; Third Nations
Single Out Agreement for Criticism,” Asahi Shimbun, December 10, 1986, p. 9; and
“Approval Necessary for Over ¥50,000—Semiconductor Exports; Trade Control Ordi¬
nance to be Revised; to Prevent Evasion of Price Monitoring,” Asahi Shimbun, December
12, 1986, p. 9.
180 THE SEMICONDUCTOR TRADE ARRANGEMENT

Commerce Department officials apparently had hoped for substantial

production cutbacks in Japan, but those hopes were initially dashed.62
Some companies, particularly TI Japan and NEC (then the largest Jap¬
anese producer of 256K DRAMs), were reportedly reluctant to follow
MITI’s new “guidance” on production and export volumes.63 Certainly,
the official MITI forecasts for the first two quarters of the STA’s opera¬
tions tended to be significantly below actual production, and forecast
errors were much larger than in later quarters (see figure 4-1 below).
U.S. merchant semiconductor producers complained bitterly that Japa¬
nese producers continued to export DRAMs to third-country markets at
prices substantially below the newly devised FMVs for U.S. imports of
Japanese chips. In late January 1987 Commerce Department and USTR
officials ended a Tokyo negotiating session with the Japanese government
by announcing at a Japanese press conference that they had presented
evidence of such transactions to Japanese officials.

Controlling Production

By mid-February it was becoming clear that current production levels

would inevitably force further DRAM price reductions on the domestic
market; would add to the increasing volumes of chips sold into the gray
market, made up of difficult-to-control independent brokers and inter¬
mediaries; and would exacerbate the flow of chips around the MITI
monitoring system and into third-country markets. On February 18, 1987,
MITI took the unusual step of revising its December 1986 forecast down¬
ward by 10 percent. (This remains to date the only time that MITI has

62. See John Sullivan Wilson, “The United States Government Trade Policy Response
to Japanese Competition in Semiconductors, 1982-1987,” report to the U.S. Congress
(Office of Technology Assessment, September 1987), p. 119.
63. See “Nippon Electric Switches to Policy of Cutting 256K DRAM Production;
Structure for Monthly Production of Nine Million Will Be Established; MITI’s Guidance
Swallowed for Easing of Friction,” Nihon Keizai Shimbun, March 6, 1987, p. 9; and “NEC
to Cut 256K DRAM Output by 10%,” Nikkei Top Articles, March 6, 1987. In “NEC Cuts
Domestic Output of 256K 40% in March,” the Japan Economic Journal reported on
April 4, 1987, that “while the Government has asked Japanese microchip makers to curtail
production to help alleviate the chip trade dispute with the U.S., NEC did not comply with
the call until February, saying that the domestic market had no oversupply of memory
chips” (p. 19). By March 1987 TI Japan was apparently the lone holdout against production
cutbacks among Japanese DRAM producers. See Robert Ristelhueber, “TI Speeding Up
U.S. DRAM Output,” Electronic News, April 20, 1987, pp. 1, 6.
THE SEMICONDUCTOR TRADE ARRANGEMENT 181

issued a midterm revision to its supply-demand forecast.) The official

reason given for the revision was that “the exportation of those products
remained small and domestic demand remained sluggish.”64 But MITI
officials made it quite clear in open discussions with the Tokyo press
corps that the revised forecast “sends a strong signal of MITI’s desire”
and that MITI was using the forecasts as “administrative guidance” to
Japanese chip producers to cut back production.65 The MITI spokesman
was quoted as saying that “we don’t punish people if our expectations
aren’t met, but I am sure it will play a role as a production guideline.”66
Yukio Honda, director of MITI’s industrial electronics division, even told
Electronic News's Tokyo reporter that MITI’s administrative guidance to
producers had been cleared with the Japanese Fair Trade Commission.67
At least some U.S. officials involved in the chip negotiations welcomed
MITI’s actions. Quoting one such anonymous official, the Wall Street
Journal reported that “the U.S. has resisted telling Japan how many chips
it thinks Japanese producers should be making because of the issue of
sovereignty, but the official conceded that MITI’s production guidelines
‘could potentially be helpful’ in driving up Japanese prices around the
world.”68 Some in the U.S. chip industry were also receptive to MITI-
ordered production cutbacks. Micron Technology Chairman Joseph Par¬
kinson was quick to praise a later round of March cutbacks to a Japanese
reporter as “a correct approach,” and he added, “The problem is how
quickly the Japanese side will achieve results by carrying out the Japan-

64. MITI, “Revision of the Semiconductor Supply-Demand Forecast,” Tokyo, February

18, 1987.
65. See “Actual Production Quotas to be Allocated for Semiconductors,” Asahi Shim-
bun, February 17, 1987, p. 9; “Japanese Chipmakers Asked to Cut Production,” Kyodo
News wire report, February 18, 1987; “MITI to Instruct Microchip Makers to Cut Produc¬
tion,” Japan Times, February 19, 1987; Peter Waldman, “Japanese Chip Firms Told to Cut
Output 10% as U.S. Deadline on Accord Nears,” Wall Street Journal, February 19, 1987,
p. 6; A. E. Cullison and Rose A. Horowitz, “Japan Presses Chip Makers to Cut Back on
Production,” Journal of Commerce, February 19, 1987, P- 1; and “Japan Asks 10% Cut in
Chip Output,” New York Times, February 19, 1987, p. Dl. The Wall Street Journal article
quoted Japanese chip executives as suggesting that MITI was acting less to save the STA
than to reduce the impact of the ongoing recession on Japanese firms.
66. Waldman, “Japanese Chip Firms Told to Cut Output,” p. 6.
67. Minoru Inaba, “MITI Sets DRAM/EpROM Cuts,” Electronic News, February 23,
1987; see also “MITI to Urge Semiconductor Manufacturers to Cut Production, for Pre¬
vention of Sale of Low-Priced Products to Third Countries,” Nihon Keizai Shimbun, Feb¬
ruary 19, 1987, p. 14.
68. Waldman, “Japanese Chip Firms Told to Cut Output,” p. 6.
182 THE SEMICONDUCTOR TRADE ARRANGEMENT

US Semiconductor Agreement, which includes the reduction of produc¬

tion by Japan.”69
In March, complaints about third-country “dumping” of Japanese
DRAMs continued to roll in from American chip makers, and MITI,
worried about American retaliation, stepped up its pressure on Japanese
producers. Approvals of export licenses for memory chips were delayed,
and a new system requiring “export shipment certificates,” issued by the
chip manufacturer directly to the exporter, was introduced in order to
seal off chips sold into the domestic market from resale as exports by
gray market sources.70 (In mid-March new export approvals were re¬
ported to have virtually stopped; by mid-April the wait for an export
license was running one to two months.71) The new measures provoked
complaints from both Japanese semiconductor makers and their cus¬
tomers but did not silence continuing complaints from American chip
makers and their government. Pressure mounted for further actions to
reduce the flow of low-priced chips out of the domestic Japanese market
and into foreign markets. Since tighter export controls were not entirely
successful (or not working rapidly enough), MITI turned its attention
toward further restrictions on supply, in order to raise domestic prices.
In mid-March MITI decided to follow up on its downward February
forecast revision by incorporating even greater production cuts into its
next, regular second-quarter forecast. MITI also reportedly decided to
tighten restrictions on the production of 1M DRAMs, which were just
entering the market in significant quantities, by issuing company-specific
production guidelines for the first time (such targets were already in effect
in other types of DRAMs and EPROMs). MITI officials were quoted in
press reports as believing that “continuation of production curtailment
should give rise to the feeling that there is a shortage of supply in the
market in or about June.”72

69. “Guidance for Reduction of Production Appreciated; U.S. Semi-conductor Man¬

ufacturer Board Chairman Obtains Oki Electric Industry’s Invoice; ‘Quick, Concrete Re¬
sults Urged,’” Nihon Keizai Shimbun, March 22, 1987, p. 3.
70. See “Semiconductor Agreement in Pinch, Complaints Pouring in against Curbing
of Exports; US Side Checking into Concrete Retaliatory Measures,” Asahi Shimbun, March
19, 1987, p. 9.
71. “Semiconductor Agreement in Pinch,” p. 9; and “Production Cut Pushes Up 256K
DRAM Prices,” Nikkei Top Articles, April 21, 1987.
72. See “Curtailment of Semiconductor Production by 20 Percent Will Continue Even
After April; MITI Holds ‘Still Oversupply,’” Asahi Shimbun, March 20, 1987, p. 9.
THE SEMICONDUCTOR TRADE ARRANGEMENT 183

The building atmosphere of crisis deepened on March 19, when Mi¬

cron Technology made public the results of a “sting” operation in Hong
Kong, in which Japanese producer Oki Electric’s sales agents had been
lured into documenting sales to a Hong Kong buyer at less than FMV
prices.73 (The following week, Micron was to announce another such
case, this one involving a Hong Kong distributor associated with Hita¬
chi.74) On March 20 the presidents and vice presidents of the largest
Japanese chip makers were summoned to MITI and notified of the forth¬
coming “guidance” from MITI.75 Shortly thereafter NEC (previously
identified in the Japanese press as resisting MITI production guidance),
announced a 40 percent cut in 256K DRAM output.76 MITI also report¬
edly followed up on the production cuts by ordering large Japanese man¬
ufacturers to raise domestic 256K DRAM prices in early April, and
export prices later that month.77
It was too little, too late, however, to stave off U.S. retaliation on the
third-country exports issue. By early March 1987 a subcabinet group
within the U.S. government had agreed to consider sanctions on Japanese
imports as a punitive measure, and by April these had been announced.
The involvement of MITI in cutting back Japanese DRAM production
was widely reported in the U.S. trade press, since MITI had put pressure

73. See “Evidence of Dumping of Semiconductors; US Company Officially Releases

Copy of Invoice of Oki Electric Industry’s Corporation in Hong Kong,” Asahi Shimbun,
March 20, 1987, p. 2; “Suspicion of ‘Trap’ Becomes Stronger—Semiconductor Dumping
Case Involving Oki Electric Industry,” Asahi Shimbun, March 21, 1987, p. 9; and “Oki
Electronics’ Bargain Sale of Semiconductors; Ignites US Distrust,” Nihon Keizai Shimbun,
March 21, 1987, p. 8.
74. See “Hitachi Is Also Engaging in Dumping; Hong Kong Agent Exports at $1.89,”
Asahi Shimbun, March 26, 1987, p. 1; and “Indicts Hitachi on Semi-conductors, Following
Oki Electric; Interviews with US Micron Corporation and Hitachi Ltd.,” Nihon Keizai
Shimbun, March 27, 1987, p. 9.
75. “MITI to Make Utmost Efforts to Avoid Retaliation; US Senate Resolution over
Semi-Conductor Friction,” Nihon Keizai Shimbun, March 21, 1987, p. 1.
76. See “NEC Cuts 256K DRAM Output by 40% in March,” Nikkei News Bulletin,
March 26, 1987. Toshiba also announced large reductions. “Production Cut Pushes Up
256K DRAM Prices,” Nikkei Top Articles, April 21, 1987.
77. Initially a price of 330 yen was discussed, but a level of 280 yen ultimately appears
to have been chosen. The 280-yen figure compares with an average of 240-250 yen in mid-
March. See “Production Cut Pushes 256K DRAM’s Domestic Price Higher,” Nikkei Top
Articles, April 7, 1987; “MITI Requests Makers to Export 256K DRAM at Over 350 Yen,”
Nikkei News Bulletin, April 25, 1987; and “Semiconductor Market Wobbling Due to For¬
eign Pressure (Part 1); Erroneous MITI Scenario,” Nihon Keizai Shimbun, March 31,1987,
p. 20.
184 THE SEMICONDUCTOR TRADE ARRANGEMENT

on TI Japan to follow its guidance on production.78 The company, which

produced much of its DRAM output at its Miho, Japan, fabrication line,
responded by declaring to the press its willingness to comply with MITI’s
demands. On April 6, 1987, Electronic News reported that MITI had on
two occasions thus far in 1987 requested Japanese firms to cut DRAM
output, and that TI Japan would slash its output of 256K DRAMs by
13 percent to comply with MITI’s wishes. Asked to respond to MITI’s
contention that TI was resisting its requests, Ramesh Gidwani, a TI
group vice president, answered, “We have been asked to reduce produc¬
tion, and we are complying. Does that sound like we are resisting?”79 TI
President Jerry Junkins declared to a stockholders meeting, “Although
we are responding to this Japanese directive, we do not believe that
cutting production, with the attendant risk of creating an artificial short¬
age, is the correct approach.”80
By early April supplies of chips to the Japanese gray market were
declining.81 Prices for 256K DRAMs had begun to rise in Japan despite
complaints from Japanese users; a trading firm representative said, “we
have to accept the new prices because of the direct guidance of MITI.”82
By mid-April DRAM prices had jumped in Singapore and Hong Kong,
and trade officials in those countries added their voices to the chorus of
complaints about the STA to those of increasingly vocal European offi¬
cials.83 As protests from U.S. and Japanese users over higher domestic
and export chip prices began to mount, MITI became considerably more
reluctant to spell out the precise nature of its actions in public.
Although MITI was to later argue that the details of its guidance were
merely “suggestions,” since they were not legally binding, the export

78. It was reported in the Japanese press that TI Japan had been asked by MITI on
March 25 to cooperate by reducing 256K DRAM output. See “MITI to Guide Japan TI
Into Production Reduction—256K DRAM,” Nihon Keizai Shimbun, March 26, 1987, p. 9;
“Semiconductor Manufacturing Coordination Guidance to Be Applied to Foreign-Capital
Manufacturers, Too; MITI,” Asahi Shimbun, March 24, 1987, p. 9.
79. See “TI Japan to Cut Output of 256K DRAMs by 13%,” Electronic News April
6, 1987, p. 4.
80. Ristelhueber, “TI Speeding Up U.S. DRAM Output,” p. 6.
81. “Semi-conductor Market Wobbling Due to Foreign Pressure (Part 2—Conclusion);
Spot Transactions Continue to Decline,” Nihon Keizai Shimbun, April 1, 1987, p. 20.
82. “MITI-ordered Production Cutbacks Raise Local Prices of 256K DRAMs,” Japan
Economic Journal, April 18, 1987.
83. “We Ask EC Representative Denman; Semiconductors,” Nihon Keizai Shimbun,
April 11, 1987, p. 7; and “Japanese Semiconductors, Southeast Asia Crying Out Over
Shortage; Will Not Tolerate Sparks from Japan-US Friction,” Yomiuri Shimbun April 12
1987, p. 7.
THE SEMICONDUCTOR TRADE ARRANGEMENT 185

control apparatus was clearly wielded in a fashion designed to guarantee

compliance.84 A franker interpretation of these activities was offered by
MITI in April 1987, in a memorandum circulated in Washington in re¬
sponse to the sanctions imposed by the U.S. government. Defending
MITTs attempts to raise prices in third-country markets, this memoran¬
dum was later to prove a crucial piece of evidence in the GATT decision
that the third-country pricing provisions of the STA were illegal.85 It
contained the following description by MITI of its own activities over this
period:
In November 1986, MITI invoked the Export Trade Control ordinance in
order to prevent below-cost exports. Thereafter, in January 1987, Japan
lowered the minimum level for export licenses from ¥ 1 million to
¥50,000. In February 1987, Japan increased scrutiny of export license
applications for third country exports in order to prevent gray market sales.
In March 1987, the MITI minister convened an emergency meeting of the
Chairman or President of each of the ten semiconductor companies to
impress upon them the importance of avoiding dumping in third country
markets.
Other actions have been taken aimed at reducing supplies and squeezing
out gray market transactions. In February, MITI exercised administrative
guidance to the companies to reduce production during the first quarter of
1987 by 23 percent below fourth quarter 1986 levels. Last month, MITI
again exercised administrative guidance to the companies to reduce pro¬
duction still further in the second quarter to 32 percent below fourth
quarter 1986 levels.86

Reducing “Excessive Competition”

Thus by April 1987 both the production and the export of DRAMs by
Japanese companies had been placed under fairly tight MITI controls.
The vice chairman of NEC publicly acknowledged MITTs role but called
upon Japanese companies “to extricate themselves from the inclination
toward excessive competition, as can be seen from the rivalry among ten

84. TI Japan, for example, was reported to be experiencing long delays in receiving
MITI export approvals. A spokesperson was quoted as saying that TI “didn’t know if the
delay was intentional or not.” See Jack Robertson, “Japan Export Delays Draw Fire From
U.S. Makers,” Electronic News, April 6, 1987, p. 4. A frank discussion of the use of export
licensing procedures to control export prices may be found in Fujiwara, “This Is a Side
Letter,” pp. 124-37.
85. See General Agreement on Tariffs and Trade (GATT), Japan—Trade in Semicon¬
ductors: Report of the Panel (Geneva, March 24, 1988).
86. MITI, “Japanese Position Paper,” Tokyo, April 10, 1987, p. 8.
186 THE SEMICONDUCTOR TRADE ARRANGEMENT

companies in one market.”87 Although ‘‘semi-compulsory measures, such

as reduction of production under MITl’s guidance and establishment of
virtual export restrictions” were the principal reason why “industry cir¬
cles have begun to show conspicuous moves for orderly marketing, such
as correction of the inclination toward excessive competition,” talk of
manufacturers making their own, private efforts to curb competitive ex¬
cesses increasingly surfaced within the Japanese industry.88 In early May
one “leader of a big semiconductor company” was quoted as saying that
“it is indispensable for manufacturers to make their own efforts hereafter,
in such ways as to establish prices in accordance with the balance between
demand and supply.” Sadao Inoue, a high-ranking Fujitsu executive,
declared, “MITI will probably continue its guidance for some time to
come. However, industrial circles have learned various lessons from re¬
cent events. If a clear interpretation of the Japan-US Semi-conductor
Agreement, including the definition of dumping, is established ... it
may become possible to establish an order in industrial circles in accor¬
dance with this interpretation.”89
The idea of “industry circles” working together in private to avoid
unnecessary trade friction received a blessing of sorts from MITI within
the next several months. The Discussion Council on Future Prospects of
the Machinery and Information Industries, an advisory group reporting
to the Director General of MITI’s Machinery and Information Industries
Bureau, produced a draft report in June 1987 which attracted consider¬
able attention in Japan, by recognizing the need for government inter¬
vention in private sector decisions on production and investment in order
to prevent trade friction.90 The report also called for a considerable

87. The NEC’s Atsuyoshi Ouchi stated in an interview: “Under MITI’s administrative
guidance, the 256K DRAM of Japanese make is subjected to thoroughgoing measures for
the reduction of production and the restriction of exports. Also, the ‘gray market’ for
transactions through brokers, which market is said to be the cause for sales at low prices,
has almost disappeared.” “US Semiconductor Retaliation; Industrial Circles Will Review
International Strategy; Inclination toward Excessive Competition Will Be Improved; Ja¬
pan’s Experiential Rule Shaken,” Nihon Keizai Shimbun, April 19, 1987, p. 4.
88. “Semi-conductor Retaliation; Manufacturers Showing Signs of Self-Reflection!?!),”
Asahi Shimbun, May 3, 1987, p. 8.
89. “Semi-conductor Retaliation,” p. 8.
90. See the description of this report in “Contents of Facilities Investments Also Will
Be ‘Supervised’; MITI Will Change Survey Method for Prevention of Friction; Reduction
of Excessive Investments Will Be Urged,” Asahi Shimbun, January 12, 1988, p. 11.
THE SEMICONDUCTOR TRADE ARRANGEMENT 187

amount of coordination among rival firms in the semiconductor indus¬

try.91 Issued in its final form in late August 1987, this document specifically
called on semiconductor producers to cooperate in planning investments
and in matching production to forecast levels of supply and demand.92
The report was endorsed by MITI officials and later became the rationale
for a new survey and informal MITI guidelines on semiconductor invest¬
ment and production levels, issued in 1988.93
These calls for “privatization” of the implementation of the STA came
as the new MITI enforcement measures were beginning to have a signif¬
icant impact on market conditions. By mid-May 1987 MITI’s more strin¬
gent control regime had helped lift 256K DRAM prices 15 percent on
the domestic market.94 In June 1987, after a meeting with Japanese Prime
Minister Yasuhiro Nakasone, U.S. President Ronald Reagan suspended
a portion of the retaliatory sanctions on Japanese imports.

91. See “MITI’s Council on Machinery and Information Industries Draws up Report
Calling for Promotion of ‘International Cooperation’ by Semiconductor Manufacturers;
Establishment of ‘Prospects for World Demand’ Urged,” Mainichi Shimbun, June 12, 1987,
p. 9.
92. Council on Future Prospects of the Machinery and Information Industries, Pros¬
pects for International Cooperation in the Machinery and Information Industries (Tokyo:
MITI, August 24, 1987) (in Japanese). Concrete proposals for the semiconductor industry
included these two points (p. 41):
1. Coordinating Information Gathering and Providing Information on Demand and
Supply: . . . Also, in light of past failures, we expect to make an accurate forecast of
international demand and supply of semiconductors. It is desirable to promote the idea of
establishing an international practice among related industries so that they can exchange
ideas and information without breaking any laws.
At present, MITI is making a short-term forecast of demand and supply of semicon¬
ductors through discussions at the investigation committee on semiconductor demand and
supply forecasts. It is necessary to listen to the opinions of makers, users, and experts on
relevant items (of semiconductors), make a short and long term forecast of demand and
supply in major markets, and examine the possibility of providing adequate information
that will benefit companies’ planning for manufacturing and investment.
2. Preventing Dumping: On preventing dumping, we should monitor export prices and
costs according to the US-Japan Semiconductor Agreement. Considering the fact that prices
below production cost are due to production in excess of actual demand, we should seek
cooperation of related companies and maintain a production level that matches with actual
demand based on the demand and supply forecast mentioned above. Moreover, we need
to investigate whether we need to improve the current antidumping system.
93. “Contents of Facilities Investment,” p. 11; and author’s interview with Japanese
semiconductor industry analyst, November 1989.
94. “Share of Imported 256K DRAM Chips May be Rising,” Nikkei News Bulletin,
May 13, 1987.
188 THE SEMICONDUCTOR TRADE ARRANGEMENT

European Reaction

European governments, originally worried about excessive supplies

being diverted to their market as a consequence of U.S. import restric¬
tions, instead shifted to requesting that Japan now increase supplies of
chips to the European market. By this time both prices and quantities
for export to third-country markets, as well as the United States, had
been placed under increasingly strict MITI guidelines.95 At midyear com¬
puter industry sales—and chip demand—began to recover, and it was
soon clear that a worldwide chip shortage was developing. Prices contin¬
ued to climb worldwide, and in early June the Reagan administration
announced a partial lifting of the sanctions imposed in March, to reward
the reduction in “dumping.”96
As prices climbed in European markets, Europe’s already negative
reaction to the STA turned even more sour. The perceived injury to
European interests was procedural, economic, and psychological. The
procedural damage involved the secret, bilateral negotiation of a frame¬
work intended to affect international market conditions for semiconduc¬
tors, without consultation with the Community. Fears of economic dam¬
age were provoked by two issues: rising memory chip prices, as MITI
tightened its export price “monitoring” mechanisms for semiconductors
sold in third-country markets to levels prevailing on exports to the United
States; and the implementation of measures designed to increase sales of
American firms in Japanese semiconductor markets. (The latter fear was,
strictly speaking, unfounded: the infamous 20 percent market share was
to be supplied by “foreign-based” firms, not necessarily American com¬
panies.)97 The psychological damage was the implicit message that Eu¬
rope had ceased to be a player that counted in the international semicon¬
ductor industry and could safely be ignored. All three irritations moved
Europe to complain to the GATT in September 1986, and when consul¬
tations with the United States and Japan proved unsatisfactory, to for¬
mally request review of its grievances by a GATT panel in February 1987.

95. “American-Make Semi-conductors Cannot Be Purchased, Even If One Desires to

Buy; Supply Falls Short Due to Recovery of Market,” Yomiuri Shimbun, May 29, 1987,
p. 6.
96. Third-country prices for Japanese DRAMs had reportedly risen to within 85 percent
of the U.S. FMVs. “U.S. to Partially Remove Tariffs against Japan,” Nikkei Top Articles,
June 9, 1987.
97. Of course, since the side letter containing the 20 percent provision was nominally
secret, the Europeans had no way of knowing this for sure.
THE SEMICONDUCTOR TRADE ARRANGEMENT 189

Why was Europe not consulted, in order to head off these complaints?
Europeans certainly realized that the game was afoot in the first part of
1986, and they shot off messages to the Americans and Japanese asking
that strictly bilateral talks affecting third-country markets be broken off.98
The European Community and the United States were at the time em¬
broiled in disputes over agricultural trade issues, and that may have made
cooperation on other issues more difficult. And since European produc¬
ers accounted for only 10 to 12 percent of global chip production, they
may indeed have simply been viewed as a marginal player, an unnecessary
complication in what were already difficult bilateral talks.
An irony in all this was that a movement was already under way in
Europe to lobby for an STA-like outcome for the European market. By
May 1986, European chip producers, in envy of the extraordinary polit¬
ical success of the American SI A, had followed its example by forming
the European Electronic Component Manufacturers Association
(EECA) in order to bring dumping cases against Japanese chip imports.99
In December 1986 the EECA filed a dumping complaint with the Euro¬
pean Community, and by April 1987 the Community had widened the
investigation to include both DRAMs and EPROMs.100 The EECA ar¬
gued that supplies diverted from the U.S. market by the STA were being
shifted to the European market, further depressing prices.
After the STA was signed in September 1986, prices continued to drop
in the European market (as in Japan) through the spring of 1987.101 After
sanctions were imposed and more-restrictive production guidelines im¬
posed by MITI in the second quarter of 1987, prices quickly rose. In the
fall of 1987 Japanese vendors were informing their European customers
that an allocation system had been installed, and by late 1987 prices in
the European market had risen substantially above U.S. levels. Through
1988 and into 1989, Japanese vendors continued to tell European cus-

98. Author’s interviews with EC officials, summer 1989.

99. Christian Tyler, “Europeans in Chip ‘Dumping’ Check,” Financial Times, May 1,
1986, p. 6.
100. “EEC Receives Complaint from SC Manufacturers,” International Herald Trib¬
une, December 6-7, 1986, p. 13; and “EC to Investigate Claims of Japanese Dumping of
EPROMs,” Financial Times, April 10, 1987, p. 5.
101. A more extensive analysis of movements in European DRAM prices may be found
in Kenneth Flamm, “Semiconductors,” in Gary Clyde Hufbauer, ed., Europe 1992: An
American Perspective (Brookings, 1989), pp. 225-92; see also Kenneth Flamm, “Measure¬
ment of DRAM Prices: Technology and Market Structure,” in Murray F. Foss, Marilyn E.
Manser, and Allan H. Young, eds., Price Measurements and Their Uses (University of
Chicago Press, 1993), pp. 174-81, 191-95.
190 THE SEMICONDUCTOR TRADE ARRANGEMENT

tomers that MITI restrictions limited their European shipments.102 Thus

in the fall and winter of 1986 prices fell below American levels, then
caught up by the fall of 1987 (as European purchasers heard about allo¬
cations), and finally rose above U.S. levels by mid-1988.
As a consequence, the STA provoked the extreme ire of both Euro¬
pean chip producers and consumers. Producers protested first, as Euro¬
pean chip prices dropped below American levels. Then, after American
pressure resulted in reduced Japanese shipments to Europe and Asia,
and chip prices in Europe rose well above American levels in 1988,
European consumers began to shout and scream, as they saw themselves
disadvantaged in highly competitive global electronic systems industries.
It is not evident that there was a deeper strategy behind the roller coaster
path of American-European memory chip price differentials: regional
price differentials would inevitably arise under any regional supply ra¬
tioning scheme.
Strategy or not, Europe was angry. The legal vessel in which the
European outrage was deposited was an EC complaint before the GATT
about the third-country export monitoring system. In March 1988 a
GATT panel ruled that the third-country pricing provisions of the STA
were illegal, and the Community stood vindicated. However, a formal
resolution of the GATT complaint was not to be concluded until over a
full year later, in June 1989. At that point, Japan agreed to monitor
retrospectively exports to non-U.S. markets, to abolish the Forecast Com¬
mittee, and to “refrain from interfering with the level of production.”103
The GATT ruling initially seemed to pose a thorny dilemma for the
Japanese. On the one hand, the United States was adamant that it would
regard lower prices in third-country markets than in the American market
as a subversion of the letter and intent of the STA. On the other hand,
the GATT ruling clearly required the Japanese government to cease

102. One cannot, of course, neglect the possibility that this was merely a convenient
excuse. However, according to a number of knowledgeable Japanese involved in that coun¬
try’s electronics industry (who spoke to me in late 1989), MITI continued administrative
guidance on investment through the spring of 1988, continued to guide the regional allo¬
cation of DRAM shipments until mid-1989, and continued to offer opinions on companies’
investment plans through at least late 1989.
103. Commission of the European Communities, “Japan to Implement GATT Rec¬
ommendations on Trade in Semiconductors,” press release, Brussels, June 22, 1989. Japan
had actually issued a proposed response to the GATT panel report on March 7, but it was
apparently not accepted by the Community until June.
THE SEMICONDUCTOR TRADE ARRANGEMENT 191

taking actions—setting export prices or volumes, or fixing production—

that had the effect of setting prices in third-country markets. How could
Japan escape this quandary?
The dilemma was solved by events: chip prices had by mid-1988 risen
far above FMV levels. Even with no “guidance” whatsoever, prices in
Europe (by the measures just offered, inflated above U.S. levels, in turn
well above FMVs) would certainly not fall below FMV levels any time
soon. Thus, ending official monitoring on European pricing and exports
would not lead to clashes with the Americans as long as prices outside
Japan remained above FMV levels.
By mid-1989 when the GATT case was finally settled, however, all but
the final details had been ironed out of a price undertaking ending the
DRAM dumping case brought by the European Community against Jap¬
anese firms. Ironically, the Community was to ask MITI to implement
the Community’s very own set of standards for the monitoring of Euro¬
pean prices, thus replacing the earlier GATT-illegal system with an EC-
blessed floor price structure.
Japan and the European Community agreed in principle on a settle¬
ment of the DRAM dumping case in August 1989. The settlement pro¬
posed that a quarterly reference price on DRAMs be calculated on the
basis of past historical cost data plus a generous 9.5 percent profit mar¬
gin.104 Japanese manufacturers signing the undertaking would be re¬
quired not to sell below this reference price. By early October 1989 the
undertaking had been signed by eleven Japanese manufacturers; it was
then passed on to the Council of Ministers for final approval. At the
United Kingdom’s insistence it was also agreed that the five-year agree¬
ment be reviewed in 1991, when the STA was scheduled to expire.105 Price
undertakings by Japanese exporters were accepted, and the DRAM
dumping investigation was terminated by the European Commission on
January 23, 1990.106

104. The U.S. Commerce Department added an 8 percent profit margin to a con¬
structed cost of production in setting the FMV
105. Lucy Kellaway, “Japan-EC Microchip Agreement Hits Snag,” Financial Times,
October 5, 1989, p. 2.
106. An antidumping duty was also set to apply to gray market imports. See Commis¬
sion Regulation no. 165/90, January 23, 1990, published in Official Journal of the European
Communities, January 25, 1990, pp. L 20/5-30.
192 THE SEMICONDUCTOR TRADE ARRANGEMENT

From Shortage to Crisis

As prices continued to edge up in the spring of 1987, chip users’

complaints became louder. In its June forecast MITI eased up on 256K
DRAM production, allowing producers to increase output by a little over
10 percent.107 For example, after the new forecast was issued, Fujitsu
immediately announced an increase in its 256K DRAM production
for the July-September quarter.108 Some in the Japanese industry at this
time believed that tight controls on 256K DRAM production were ac¬
celerating the trend for producers to switch over to production of the
next-generation product, the 1M DRAM.109
With increased 1M DRAM output and consequent price declines,
MITI had also opted to decree limits on production increases in that
product, in order to stabilize prices.110 In early September both Toshiba
and Hitachi publicly confirmed that they had been scaling back produc¬
tion of 1M parts in response to guidance from MITI. A MITI spokesman
in Washington responded defensively that it set “no limitation or guide¬
line to the electronic industry”; a Toshiba representative in Tokyo “clar¬
ified” the company’s earlier statement by adding that “Toshiba-Tokyo
took serious consideration of the MITI demand forecast and judged it
should change its own manufacturing schedule and plan.”111
Even as prices were edging up, MITI’s control framework was ex¬
tended into new areas. After complaints from U.S. producers of appli¬
cation-specific integrated circuits, MITI reportedly decided to extend the
price monitoring framework to certain types of customized chips in
June.112 By the fall of 1987 the largest-volume type of ASIC chip, gate

107. See “Japan to Raise Output of 256 Kilobit DRAM by 10%,” Kyodo News wire
dispatch, June 24,1987, which reports that MITI officials “instructed” Japanese chip makers
to increase production of 256K DRAMs, and quotes them as saying that the ministry would
“allow” the growth in response to an increase in overseas demand; see also “MITI to Ease
‘Guidance’ for Reduced Chip Production,” Nikkei News Bulletin, June 10,1987; and “MITI
to Partially Lift Restrictions on Chip Production,” Nikkei News Bulletin, June 11, 1987.
108. “MITI Reports Microchip Output Estimate (1),” Nikkei News Bulletin, June 25,
1987.
109. “IC Manufacturers Prepared to Expand Production in Kyushu; No Moderation,”
Asahi Shimbun, June 16, 1987, p. 11.
110. “1M DRAM Price to be Unchanged in June,” Nikkei News Bulletin, June 10,
1987.
111. Rufus Baker, “Toshiba Cuts 1-Mb Chip Ships,” Electronic Buyers’ News, Septem¬
ber 14, 1987, pp. 1, 80.
112. See “MITI to Keep Closer Watch on ASIC Prices,” Nikkei News Bulletin, June
23, 1987.
THE SEMICONDUCTOR TRADE ARRANGEMENT 193

arrays, had joined memory chips on the list of products for which MITI
guidance was holding down production.113
As a recovery in the computer industry continued briskly through the
fall of 1987, prices for chips continued to be pushed upward. In the late-
September forecast for the final quarter of 1987, MITI again raised pro¬
duction quotas for DRAMs: by roughly 5 percent for 256K chips and
80 percent for 1M memories.114 By October a serious shortage appeared
to be on the horizon, and manufacturers were reported to be turning
away large orders. Domestic Japanese prices even began to approach the
FMV export price floors.115

Controls on Investment

The inclination of Japanese chip producers was to expand capacity to

meet the looming shortage, but MITI officials were reported as feeling
that “The removal of the measures which have been taken to curb in¬
vestments, at this time, may lead to an oversupply of products.”116 As
recovery in the industry continued, MITI shifted much of the focus of its
guidance to investment, as companies argued that they should be allowed
to revise their fiscal 1987 investment plans upward. Reported Nihon
Keizai Shimbun, “MITI, which is concerned about the possibility of a
flaring up of the Japan-US semi-conductor friction again, maintains rigid
supervision over the facilities investments of the major companies. As a
result, these companies have drawn up moderate investment plans.”117
The increasing signs of shortage satisfied the American government
that MITI had acted forcefully to increase chip prices in worldwide mar¬
kets. Indeed, by the fall of 1987 the agonized screams of chip users were
being heard loud and clear, and the U.S. government had switched its
public posture to one of encouraging MITI not to restrict chip produc¬
tion. On November 3,1987, in a delicate ballet of communiques, the U.S.

113. See “Anonymous Round Table Discussion on Silicon,” Kinzoku Jihyo, no. 1319,
November 15, 1987.
114. “MITI Lifts Limit on Chip Production,” Nihon Keizai Shimbun, September 25,
1987, p. 3.
115. “Semiconductors; Sudden Increase in Demand Leads to Growth of Strong Con¬
cern Over Shortage of Supply,” Nihon Keizai Shimbun, October 2, 1987, p. 20.
116. “Nine Semiconductor Companies Keep Facilities Investments Moderated in Re¬
vised Plans as Well, In Accordance with MITI’s Guidance,” Nihon Keizai Shimbun, Oc¬
tober 10, 1987, p. 8.
117. “Nine Semiconductor Companies Keep Facilities Investments Moderated,” p. 8.
194 THE SEMICONDUCTOR TRADE ARRANGEMENT

government partially removed more of the spring sanctions and an¬

nounced that it was satisfied that third-country dumping had ended.
MITI announced that it was “imposing no quantitative or other restric¬
tions on the production, shipment, or supply of semiconductors, except
MITI continues to exercise export control from the view point of
COCOM.”118 This last qualification, it turned out, was crucial, since an
extremely rigorous export control system continued in place during 1988,
effectively making it impossible for foreign buyers to purchase Japanese
chips at prices not directly controlled and administered by the chip man¬
ufacturers.119 The wording is also notable in that it does not directly
address the issue of whether extralegal MITI “guidance” or “sugges¬
tions,” in contrast to legally binding “restrictions,” might be influencing
Japanese manufacturers’ production decisions.
Behind the scenes, the U.S. government was less than unequivocal in
its view of the situation. Officials were not eager to see Japanese firms
increasing their chip capacity to meet the looming shortage, preferring
to see American companies “reenter” the DRAM market.120 While pub-

118. MITI, “Statement of Ministry of International Trade and Industry Concerning

Trade in Semiconductors,” Tokyo, November 3, 1987. COCOM, the Coordinating Com¬
mittee on Multilateral Export Controls, was the body through which the Western alliance
coordinated its controls of exports for national security purposes.
119. In 1988 and 1989 foreigners wishing to purchase DRAMs from Japanese vendors
were required to register with MITI; MITI would not consider issuing an export license to
a Japanese vendor until the application was approved (author’s interviews with semicon¬
ductor purchasers, 1989). The policy resulted in tight control over the ability of foreigners
to gain access to the Japanese DRAM market. From personal experience in attempting to
purchase DRAMs on the retail market in Japan, in March and September 1988 in the
Akihabara in Tokyo and Den-Den Town in Osaka, I can report that no retail outlet
appeared willing to sell DRAMs to an obvious foreigner walking in off the street. When
questioned on this point, a MITI official suggested that the vendors may have been con¬
cerned about export licensing requirements.
120. This attitude was aided and abetted by the widespread dissemination of a study
carried out by the consulting firm Quick and Finan, which forecast that “The tight supply
conditions seen in late-1987 [are] only temporary. Several firms have just begun full-scale
production and others will soon follow in the first half of 1988. Thus, throughout most of
1988, potential supply capacity will ramp faster than demand under any plausible market
scenario.” Quoted in Semiconductor Industry Association, One and One-half Years of
Experience Under the U.S.-Japan Semiconductor Agreement (Cupertino, Calif., March 1,
1988), p. 26.
This view conflicted sharply with the consensus view in Japan at that moment, where
semiconductor manufacturers predicted a shortage lasting at least until the spring of 1989.
See Mitsuhiro Takahashi, “Producers Slow to React to Chip Shortage,” Japan Economic
Journal, May 7, 1988, p. 1. Needless to say, the view in Japan—where more than 90 percent
of merchant DRAMs were produced—was the correct one.
THE SEMICONDUCTOR TRADE ARRANGEMENT 195

licly proclaiming that they were acting to remove limits on chip produc¬
tion, American trade negotiators continued to press MITI to limit in¬
vestments by Japanese firms in new capacity well into 1988.121

The Control Regime after November

What was the real content of the changes in the export control regime
made in November 1987? Organizationally, responsibility for approval of
export license applications was moved from MITI’s semiconductor mon¬
itoring office (charged with surveillance of export prices) to a separate
office, so that “there was no feedback from the Monitoring Office to the
office dealing with COCOM screening.” However, “in cases where export
prices were ‘extremely below cost,’ MITI would express its concern to
the companies concerned. . . . Under the old system . . . some misun¬
derstanding seemed to have been created among exporters that delays
had been caused by inappropriate pricing. The new system would elimi¬
nate such misunderstandings.”122
What of the infamous production “forecasts”? After November 1987
the production forecast system was henceforth to be described by
MITI as
a reference for manufacturers in their production schedules. MITI ex¬
plained its objective to manufacturers and impressed upon them the need
to reflect real demand in their production. Individual companies were
expected voluntarily to bring their production almost in line with the fore¬
casts, taking into account the appropriate total production. The forecasts
were not legally binding and the Government did not allocate production
volume to individual companies. For manufacturers to conspire on pro¬
duction volume was against the anti-trust laws in Japan.123

Unfortunately, this description of the operation of the forecast system

contrasted rather markedly with the assessment circulated by MITI itself
in Washington in April 1987 (and diverges as well from the public utter¬
ances of various MITI officials through mid-1987, Japanese press ac¬
counts, and the testimonials of others involved in semiconductor trade).
Indeed the GATT panel surveying the MITI monitoring system rejected
the MITI characterization in its March 1988 report, specifically citing the
April 1987 MITI position paper, and concluded that

121. Author’s interviews with former U.S. government officials, 1989.

122. The quotations are from a GATT panel’s summary of the official Japanese gov¬
ernment position. See GATT, Japan, p. 9.
123. GATT, Japan, p. 10.
196 THE SEMICONDUCTOR TRADE ARRANGEMENT

an administrative structure had been created by the Government of Japan

which operated to exert maximum possible pressure on the private sector
to cease exporting at prices below company-specific costs. This was exer¬
cised through such measures as repeated direct requests by MITI, com¬
bined with the statutory requirement for exporters to submit information
on export prices, the systematic monitoring of company and product-spe¬
cific costs and export prices and the institution of the supply and demand
forecasts mechanism and its utilization in a manner to directly influence
the behavior of private companies. . . . The Panel considered that the
complex of measures exhibited the rationale as well as the essential ele¬
ments of a formal system of export control. The only distinction in this
case was the absence of formal legally binding obligations in respect of
exportation or sale for export of semi-conductors. However, the Panel con¬
cluded that this amounted to a difference in form rather than substance
because the measures were operated in a manner equivalent to mandatory
requirements.124

Indeed, the only immediate, concrete change visible in the operation

of the forecast system after 1987 was that from March 30, 1988, on (after
the GATT semiconductor panel had issued its report) MITI forecasts
were accompanied by text asserting that the forecasts were issued for
“reference” purposes only, and not for the purpose of restricting pro¬
duction.125 Since this assertion had also been made to—and rejected by—
the GATT panel with regard to the 1987 forecasts, it was less than
compelling.

The Zen of Accurate Forecasting

Certainly the statistical record does not indicate a sea change in the
relative accuracy of the “forecasts” of the most politically sensitive prod¬
ucts—DRAMs and EPROMs—before and after November 1987. During
the final quarter of 1986, before MITI clamped down hard on Japanese
producers, the DRAM production forecasts were wildly inaccurate, with
production of both 64K and 256K parts exceeding one-quarter-ahead
forecasts by margins in the range of 30 to 50 percent of actual production.

124. GATT, Japan, p. 42.

125. “Since there seems to be a misunderstanding that this supply-demand forecast is
compiled for the purpose of restricting production, we make it clear that it is not compiled
for that purpose, but for the reference of related parties and that manufacturers are free
to produce more than the production in the forecast. Decision on production fully depends
on producers.” MITI, Machinery and Information Industries Bureau, “Semiconductor
Supply-Demand Forecast for 2nd and 3rd Quarter 1988,” Tokyo, March 30, 1988, p. 1.
THE SEMICONDUCTOR TRADE ARRANGEMENT 197

Since MITI reacted by “revising” the first-quarter 1987 DRAM fore¬

cast downward by 10 percent in midquarter, it is perhaps not surprising
to find that production fell short of the original first-quarter forecast for
DRAM chips by margins clustering rather tightly around 10 percent and
by 15 to 20 percent for EPROMs (figures 4-la and 4-2a). And until MITI
began issuing detailed guidance on company-specific production levels
for the 1M DRAM, in the second quarter of 1987, the one-quarter-ahead
forecast error for the first quarter of 1987 was spectacularly large for that
product, with the forecast falling short by about 35 percent of actual
realized output (figure 4-la). From that point through November 1987, a
period of relatively public MITI controls on production, the forecasts
issued for 265K and 1M DRAM production levels three and six months
out typically fell within 10 percent of actual output, as shown in figure
4-la. (For 1M DRAMs, with production dominated through most of this
period by just one producer, Toshiba, the error was more typically on the
order of 5 percent.)
The truly noteworthy feature of the post-November 1987 DRAM pro¬
duction forecasts is that they continued to be extraordinarily accurate,
after the invisible hand, rather than government’s, ostensibly ruled the
market. Indeed, at least one explanation offered in early 1988 for the
unprecedented jump in memory chip prices clearly flew in the face of
these forecasts’ accuracy. Some analysts argued that unexpected yield
problems were a major factor in the shortage then developing.126 It was
widely believed in the industry that some companies, particularly NEC

126. “At present, manufacturers are hurrying the expansion of [1M DRAM] produc¬
tion. Because of such technical problems as the delay in the improvement of the yield rate,
however, production in the July-September period also is likely to increase by only about
2,700,000.” “Semiconductor Prices Continue Rising; Shortage of Supply Will Reach Peak
in July-September Period,” Nihon Keizai Shimbun, July 16, 1988, p. 18.
‘“It took chip makers longer than they expected to get good yields of one-megabit
chips,’ says Jim Feldhan, vice-president of In-Stat Inc., a Scottsdale Ariz. market research
firm that tracks the semiconductor industry. Mr. Feldhan says as many as 90% of all one-
megabit chips were defective and had to be discarded as recently as early this year, but
that lately yields of good chips have improved to around 50%.” Brenton R. Schlender,
“Chip Prices Fall: Easing of Shortage Seen,” Wall Street Journal, July 18, 1988.
“Currently, major makers are shipping an estimated 12 million units of 1-M DRAM
chips monthly.
The industry had planned to reach that level four to six months earlier. Apart from
Toshiba, however, producers were unexpectedly slow to boost production. Then, compa¬
nies’ average yield rate, a production yardstick that measures the percentage of useable
chips, stood at between 40% and 50%.” Tadashi Tamaki, “Heavy Investment in Chip
Factories Spurs Fear of Glut,” Japan Economic Journal, October 1, 1988, pp. 1, 5.
198 THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 4-1. Errors in MITI Forecasts of Japanese DRAM Production,

Quarterly, 1986-88, and Semiannually, 1987-94a

A. Quarterly forecasts
Percent

B. Semiannual forecasts

Source: MITI, "Semiconductor Supply-Demand Forecast” (quarterly); after 1988, semiannually,

a. Forecast error is defined as forecast less actual as a percent of actual. A positive error represents an excess of
actual over estimated production.
Figure 4-2. Errors in MITl Forecasts of Japanese EPROM Production,
Quarterly, 1986-88, and Semiannually, 1987-94a

Percent A. Quarterly forecasts

- 64K, IQ-ahead forecast

- 256K, IQ-ahead forecast
.64K, 2Q-ahead forecast
^ 256K, 2Q-ahead forecast
\-128K, IQ-ahead forecast
\
\ •••••• 512K, IQ-ahead forecast
\-128K, 2Q-ahead forecast
\ ■ 512K, 2Q-ahead forecast
\

1986:4 1987:1 1987:2 1987:3 1987:4 1988:1 1988:2 1988:3 1988:4

Percent B. Semiannual forecasts

Source: MITl, "Semiconductor Supply-Demand Forecast” (quarterly); after 1988, semiannually,

a. Forecast error is defined as forecast less actual as a percent of actual. A positive error represents an excess of
actual over estimated production.
200 THE SEMICONDUCTOR TRADE ARRANGEMENT

and Hitachi, had had some difficulties raising their manufacturing yields
in their early production of the 1M DRAM.
But if such yield problems, in the aggregate, played a significant role
in creating shortages, of 1M DRAMs in 1988, it is hard to see how they
could have been unexpected, given the accuracy of the MITI forecasts.
Since manufacturing time from starting a silicon wafer on the line to a
finished memory chip is typically 60 to 75 days, to adjust production
scheduling in midquarter to compensate for unexpected yield problems
during that same quarter and still hit the three-month forecast would be
most difficult.127 To accurately hit both three-month and six-month fore¬
cast targets, as happened with both the 256K and the 1M DRAM pro¬
duction forecasts for 1988, would seem to require quite an accurate
estimate of yields over a roughly six-month period. Thus, a continuous
history of coming within 10 percent of three- and six-month production
forecasts, in a product requiring over two months of processing on the
production line, suggests that unanticipated yield problems could not
have been a major issue for the industry as a whole in 1988.
Although yields were undoubtedly low in the initial stages of 1M
DRAM production (as has been true for every generation of chip), it is
hard to reconcile big surprises in yields with the pinpoint accuracy of the
MITI production forecasts. Thus both one- and two-quarter-ahead
DRAM production forecasts exceeded actual output by about 9 percent
in the second quarter of 1988 (figure Ala). At other times during 1988,
forecast output for 1M DRAMS stayed within 2 to 4 percent of actual
production. At their worst, unanticipated yield problems could only have
been a marginal factor.
Japanese chip manufacturers quoted in the Japanese trade press, on
the other hand, gave producers’ restraint in production and investment
the lion’s share of the credit for the 1988 DRAM crisis. Reacting to
reports of a shortage of semiconductors and American user complaints,

127. The “unexpected yield problems” thesis attained its most extreme form in the
assertion that the earthquake that Japan suffered on December 17,1987, by shaking up chip
plants and yields, was a proximate cause of the price surge of early 1988. This story was
given wide circulation by George Gilder in “How the Computer Companies Lost Their
Memories,” Forbes, June 13, 1988, p. 81. MITI’s Forecast Committee actually met on
December 15, two days before the big earthquake, to discuss DRAM production during
the first quarter of 1988. That forecast was extremely accurate: actual production fell short
of the forecast by only -1.6 percent for 64K parts, -7.7 percent for 256K parts, and -1.9
percent for 1M parts.
THE SEMICONDUCTOR TRADE ARRANGEMENT 201

a top Hitachi semiconductor executive offered the following assessment

of the shortage’s causes in March 1988:
It is mainly memory chips which are short. The cause is that, just at the
time when Japanese manufacturers, who are the major suppliers, held
down production out of fear of escalating the semiconductor friction, de¬
mand for office automation equipment, such as personal computers, etc.,
increased sharply. This situation will continue throughout this year. How¬
ever, we cannot tell about the next year and after. From this sense of non-
transparency, various manufacturers are cautious about facilities invest¬
ments, and as a result, the tight situation of demand and supply in the
market is being spurred.128

By historical standards the official MITI DRAM production forecasts

issued from 1987 to 1989 were extraordinarily accurate. Before that pe¬
riod and after, divergence between predicted output levels and actual
ones for either one quarter (figure 4-la) or six months (figure 4-lb) ahead
showed noticeably greater variation.129 The same observation may also
be made for EPROMs (figure 4-2).

The Privatization of Restraints, December 1987 to Mid-1989

Certainly it was a kinder, gentler, and considerably more discreet
control system that continued after MITI’s November 1987 disavowal of
production controls. By this point, however, production controls had
begun to fade as a serious issue, as demand for memories surged. Firms
had been permitted to increase production steadily over the second half
of 1987, and by the end of that year many Japanese semiconductor pro¬
ducers were approaching capacity limits. Increasingly, limitations on new
capacity investments and the allocation of production between domestic
and regional export markets became the focus for scrutiny by Japanese
government regulation of the industry. Administrative guidance of in¬
vestments and a regional allocation system for exports, already visible in
1987, were to become important factors during the emerging electronics
boom of 1988.

128. “One Year Since Retaliation on Semiconductors; Interview with Government and
Industry Circles on Prospects,” Mainichi Shimbun, March 26, 1988, p. 9.
129. The MITI forecasts were available for one quarter ahead only from the first
through third quarters of 1987 and were switched to a six-month forecast at the end of
1988.
202 THE SEMICONDUCTOR TRADE ARRANGEMENT

Regional Allocation of Exports

By the fall of 1987 a regional allocation system for exports had been
put in place. (See the section on “European Reaction” above.) In 1988
this system was to create large price differentials between DRAMs of¬
fered for sale in domestic Japanese markets and those shipped in foreign
markets. The allocation system was not ended until mid-1989.130

Guidance on Investment

As previously remarked, it was widely reported in the Japanese trade

press that MITI had exercised administrative guidance over manufactur¬
ers’ revisions to their fiscal 1987 investment plans in the fall of that year.
That fall had also marked the publication of a report by a MITI advisory
council calling for a greater government role in ensuring that company
production and investment plans did not create trade friction. Top MITI
officials had endorsed the report. In January 1988 MITI replaced its
twice-yearly survey of Japanese industrial investment with a much more
detailed and rigorous questionnaire, in order to better monitor invest¬
ments in trade-sensitive sectors, particularly semiconductors, and advise
businesses when investments were judged to be excessive.131

130. Author’s interviews with Japanese semiconductor executives and analysts, Novem¬
ber 1989.
131. See “Contents of Facilities Investments,” p. 11. The Asahi reporter wrote, “The
surveys until now have been conducted by means of questionnaires with major emphasis
on the monetary amounts of investments. In the coming survey, however, MITI will examine
the contents of investments as well, and ask about such details as the kinds of products and
the amount of production of the respective items. It has decided on revision of its survey
method, from the standpoint that the trade friction, which is represented by the Japan-
U.S. semiconductor problem, has often been caused by excessive investments on the part
of Japanese enterprises. Depending on circumstances, MITI will ask enterprises for reduc¬
tion of the investments, which it has judged to be clearly excessive as a result of its
survey. ... In August of last year, the 'Discussion Council on Future Prospects of the
Machinery and Information Industries,’ which is a personal consultative organ of the Di¬
rector General of the MITI Machinery and Information Industries Bureau, drew up a
report which recognized the necessity for the Government to intervene in the decisions to
be made by enterprises on production and investments, for the purpose of preventing the
occurrence of trade friction.”
MITI’s reaction to the possibility that enterprises might object was the following: “The
survey is to be conducted with the cooperation of enterprises in all cases. The main purpose
of the survey is to grasp the reality. Even if MITI judges investments to be excessive, it will
only take such steps as to urge caution.”
Eight months later, an editorial in the Japan Economic Journal explicitly noted the
THE SEMICONDUCTOR TRADE ARRANGEMENT 203

January 1988 also marked a significant appreciation of the yen. Con¬

scious of continuing scrutiny from the United States, Japanese chip mak¬
ers reportedly raised their U.S. export prices from 5 to 10 percent on
contracts signed in that month.132 Coupled with a recovery in the com¬
puter industry, an increasing demand for memory chips, and the delayed
effects of restrictions on new capacity investments in 1987, these price
increases touched off an extraordinary spiral of rising chip prices. In
early 1988 spot prices for DRAMs in the United States soared to histor¬
ically unprecedented levels—the price of the best-selling 256K DRAM
tripled over a four-month period—and the U.S. computer industry was
plunged into crisis; producers scrambled for supplies of critical memory
chips.
Despite the emerging DRAM crisis, however, Japanese output in¬
creased relatively slowly. In the Japanese business press, sporadic reports
suggested that MITI was continuing to offer occasional informal guidance
to producers on production plans after November 1987.133 Through the
close of the 1987 fiscal year in March 1988, MITI maintained a relatively
tight rein on investment in new capacity. With the approach of summer,
and a worsening shortage, restrictions on capacity increases were appar¬
ently relaxed.134 In the late spring, as plans for fiscal 1988 were disclosed,

existence of the investment control system: “While the Japanese government was busily
setting up an export price monitoring system and instituting a series of production, and
plant and equipment investment controls, the demand from Japanese and American semi¬
conductor users rebounded sharply.” “Editorial: Chip Pact Obsolete,” Japan Economic
Journal, August 13, 1988, p. 22.
132. “Growing Moves to Raise Export Prices, Centered on Exports to US; Semicon¬
ductor Prices Upped by Five to Ten Percent,” Nihon Keizai Shimbun, February 1, 1988,
p. 1.
133. One newspaper (“Prices of Japanese-Make Semiconductors Rise Sharply in East
Asia; Almost Twice as Fligh as Level at Start of This Year,” Nihon Keizai Shimbun,
December 23,1987, p. 18) reported that “manufacturers maintain a cautious attitude toward
expansion of production under MITI’s guidance.” The same paper (“Listen to GATT’s
Decision,” Nihon Keizai Shimbun, March 7, 1988, p. 2) later noted that “MITI is carrying
out strict administrative guidance, such as to allot production quotas for the respective
items to manufacturing companies, etc., in an effort to make the agreement truly effective.”
See also “Semiconductor Shortage; Prices of Personal Computers and Word Processors
Rising Gradually,” Asahi Shimbun, June 23,1988, p. 3. This article, published seven months
after MITI declared that it would not formally restrict production, explains a current
semiconductor shortage in part by saying, “On the other hand, however, it is also strongly
viewed that MITI’s guidance toward various manufacturers for their reduced production—
as a counter-measure for the Japan-US semiconductor friction—took effect to excess.”
134. Tight restrictions were reportedly maintained through March 1988 (author’s inter¬
view with Japanese industry source, 1989). Another Japanese analyst close to the semicon-
204 THE SEMICONDUCTOR TRADE ARRANGEMENT

an average 38 percent increase in semiconductor capital investment was

announced by Japanese chip producers.135 In the face of a mounting
shortage, Japan’s semiconductor business community viewed these in¬
vestment levels as quite restrained.136 The Japan Economic Journal even
explained the continuing shortage of semiconductors in the following
terms: “One reason for this is the makers’ restraint in investment. They
seem to be acting, or not acting, in concert.”137

“Profitability Instead of Market Share”

Around this period, in mid-1988, the Japanese trade press began to

routinely publish analyses suggesting that manufacturers were acting
much more collusively than in the past, and that firms were consciously
acting with restraint in order to increase profitability for the industry as
a whole.138 These two themes—restraint in investment, linked in part to

ductor industry whom I interviewed at this time told me of reports then circulating that
chip makers were submitting investment plans to MITI for informal review.
135. “Big Semiconductor Companies to Increase Facilities Investments for Fiscal 1988
by 38%; Stress to be Laid on 1M DRAM,” Nihon Keizai Shimbun, May 27, 1988, p. 8.
136. Commenting on the proposed increases in investment, the Mainichi Shimbun
wrote, “However, whether or not the shortage of stocks will be dissolved by such increased
production is not clear. As for the reason why, semiconductor industry circles carried out
facilities investments to excess and increased their production on a large scale, centering
on fiscal 1983 and 1984, and they suffered from a big depression in 1985. There was such a
bitter experience, and they generally agree on the view that ‘there was a tragedy in the fact
that various companies were mainly producing memory chips in those days’ (NEC Chair¬
man Atsuyoshi Ouchi). In view of facilities investments, too, the amount of NEC’s facilities
investments is one-third of that in those days (¥140 billion in fiscal 1984), and Toshiba—
about one-half (¥148 billion in fiscal 1984). Moreover, it is because all manufacturers
remain cautious about production leaning toward memory chips, as is represented by such
words as ‘We want to maintain 20% for [production of] memory chips.’ Accordingly, the
shortage of memory chips is likely to be carried over to next year.” “Semiconductor Shortage
Serious; Going beyond Production Increase Setup,” Mainichi Shimbun, June 29, 1988,
p. 8.
137. Naoshi Tamaki and Akira Kikuzumi, “Shortage of DRAMs Causing Headaches,”
Japan Economic Journal, July 2, 1988, p. 4.
138. For example, a May 7 analysis in the Japan Economic Journal gave three reasons
for the sluggish investment in chip production. Said the reporter, “one reason is the fact
that manufacturers are loath to repeat the debacle of 1984-85 [when demand shrank
drastically after a buildup in capacity]. Another reason is the fact that Japanese semicon¬
ductor makers are making comfortable profits from the present arrangements based on the
Japan-U.S. semiconductor agreement signed in the autumn of 1986 at the urging of the
Ministry of International Trade and Industry. Production cutbacks enforced by the agree¬
ment have sharply boosted the market prices of semiconductors and Japanese manufacturers
have come to value profits more than market shares [emphasis added], A third reason, on
THE SEMICONDUCTOR TRADE ARRANGEMENT 205

MITI attitudes, and more cooperative behavior (“profitability instead of

market share”)—were to surface often in Japan over succeeding
months.139
An article in Nihon Keizai Shimbun in June 1988 elaborated on this
theme.

The silicon cycle has completely come undone. The reason for this is the
constraint on investment on the part of semiconductor makers, done in the
name of “cooperation.” Koji Kobayashi, chairman of NEC, asserts that
“The semiconductor manufacturers will not let the tragedy of 1984 happen
again. . . . According to one of the top men at a leading chip manufacturer,
‘I would not want this to be construed as a cartel. Manufacturers have
become more open in sharing information, so that there is more coordi¬
nation. In the old days, we used to send industrial spies to gather infor¬
mation about the activities of our rivals, but recently, that practice has
vanished.’ For example, Toshiba announced that ‘Mass production of the
next-generation 4 megabit chips is not planned for this year, and we can
not predict when it will start.’ Such openness should dispel any lack of trust
within the industry, and is related to efforts to keep excessive competition
under control. ... It does not appear that the chip makers’ move away
from memory chips and their cooperative efforts to avoid excessive com¬
petition are temporary phenomena.” The shortage of memory chips is thus
expected to continue for a long time.140

The same themes were echoed in an August 1988 analysis in the Japan
Economic Journal. After noting that capital investment in fiscal 1988 was
expected to reach 430.8 billion yen, compared with 762.8 billion yen
during the boom year of fiscal 1984, the reporter quoted Yukio Honda,
director of MITI’s industrial electronics division, as saying, “This figure
mirrors the manufacturers’ prudent stance toward capacity expansion for
fear of another recession in the future.” The article continued:

the other hand, is Japanese manufacturers’ fear that any large-scale equipment investments
will rekindle trade friction with U.S. industry.” Takahashi, “Producers Slow to React,”
p. 10.
139. The Japan Economic Journal ran an editorial on May 28, for example, that com¬
mented, “It is unnecessary to note that semiconductor manufacturers are aware that the
government’s supply-demand forecasts mean nothing less than quantitative production
controls, though the ministry takes the position that it is simply one of the many forecasts
it routinely employs. Similarly, the legality of the government’s semiconductor export price
monitoring system is also ambiguous. There is no doubt that the monitoring of exports by
the overseas subsidiaries of Japanese semiconductor manufacturers has no legal justifica¬
tion.” “Editorial: Scrap the Chip Pact,” Japan Economic Journal, May 28, 1988, p. 22.
140. “Supply-Demand Structure for Semiconductor Industry Shifts,” Nihon Keizai
Shimbun, June 8, 1988, p. 9.
206 THE SEMICONDUCTOR TRADE ARRANGEMENT

Chip makers have long been vying for larger market shares, as mass pro¬
duction leads to remarkable cost reductions. Manufacturers embarked on
capacity expansion in peak years which repeatedly caused overproduction
and a resulting recession. Any coordinated action among manufacturers
was unthinkable. But after the 1986 Japan-U.S. chip pact, MITI led the
manufacturers to reduce IC production by about 30% in early 1987. “The
pact helped to virtually create a coordinated production control by chip
makers that we have never seen before,” a broker said. . . . The Japan-
U.S. microchip agreement so far has failed to achieve improved access for
U.S. semiconductor makers to the Japanese market. But it seems certain
that the pact helped Japanese chip makers strengthen their profitability
through production control.141

Japanese semiconductor analysts in Tokyo also stressed both the trend

toward more oligopolistic behavior in the chip industry and the govern¬
ment’s role in encouraging it. Nomura Securities echoed the famil¬
iar “profits instead of production volume” theme in a September 1988
analysis.
In addition, equipment investment for research and development in this
area involves major risks. This leads to an oligopolistic market, in which a
few major firms dominate. The top Japanese firms tend to benefit from this
kind of situation. Moreover, the Japanese-U.S. treaty on semiconductors—
which is expected to be amended slightly—is seen as bringing about a
stabilization of prices and should contribute to ensuring sustained profit
earnings. The switchover from policy of expanding production volume to
a situation where management places an emphasis on profitability will also
contribute to market stability.142

The same rhetoric was repeated in a conversation I had with an official

in MITI’s industrial electronics division on September 26, 1988. After a
long and inconclusive discussion of what sort of exchanges between MITI
and the Japanese industry might be considered “controls” or “guidance”
or even “suggestions,” I was told by the official that it was the position
of the Japanese government in general, and MITI in particular, to en¬
courage chip firms to stress profitability instead of market share.
Probably the closest thing to an official confirmation of a Japanese
government policy of tolerating, if not encouraging, collusive restraint in
production and investment among domestic producers came at a Novem-

141. Shigehisa Shibayama, “Chip Shortage Expected to Last Through ’89,” Japan Eco¬
nomic Journal, August 13, 1988, p. 5.
142. Nomura Securities Co., Ltd., “Economic Insight: Heavy Demand for Memory
Chips,” Japan Times, September 17, 1988, p. 9.
THE SEMICONDUCTOR TRADE ARRANGEMENT 207

ber 1988 Washington meeting between U.S. and Japanese trade negoti¬
ators. The American side’s summary memorandum for that session de¬
scribed the following exchange:
Finally, USG [U.S. government] expressed concern about the role of Fore¬
cast Committee and its impact on market. USG stated many feel Forecast
Committee influences or sets production and investment levels rather than
simply forecasting. USG said language used by GOJ [government of Japan]
to describe Forecast Committee suggests a broader role, and asked how
exactly Forecast Committee can “stabilize” market. USG also asked how
it is that Forecast Committee has been as accurate as it has in projecting
production levels.
GOJ responded that there is no reason for USG concern about Forecast
Committee. GOJ stated forecasts are accurate because they are provided
by companies themselves which are then aggregated by Committee. GOJ
stated Japanese industrial society is very competitive and each member of
Forecast Committee is very interested in how much its competitors pro¬
duce. GOJ said Forecast Committee had “freed the Japanese semiconduc¬
tor companies from unnecessary competition” after period two years ago
during which companies competed in supply and manufacturing.143

As the shortage of DRAMs intensified in the late summer and early

fall of 1988, the major Japanese chip producers ultimately announced
upward revisions of their capital spending plans.144 Even after those re¬
visions, however, capital spending was far (about 40 percent) below out¬
lays during the 1984 boom in absolute terms, despite the substantial
growth in both the sales base and the cost of equipment. October esti¬
mates of fiscal 1988 spending on capital equipment by Japanese chip
companies as a fraction of sales were up somewhat over 1987, but well
short of pre-STA spending levels. Furthermore, a significantly smaller
proportion of this equipment investment went into DRAM capacity than
during the 1984 chip boom.145
Curiously, these relatively modest increases in revised capital invest¬
ment plans for 1988 inspired a campaign in the English-language press
against a forthcoming chip glut. The Japan Economic Journal, in a prom¬
inently displayed, October 1, 1988, front page article, decried the “heavy

143. “Subject: US-Japan Semiconductor Consultations, November 17-18, Washing¬

ton,” memorandum, Office of the U.S. Trade Representative, Washington, n.d.
144. Some analysts in Japan claimed that the investment plans publicly announced by
semiconductor companies exceeded actual investments by a significant amount. See “Chips
in Short Supply,” Tokyo Business Today, September 1988, p. 9.
145. This point was made to me by a MITI official on September 26, 1988.
208 THE SEMICONDUCTOR TRADE ARRANGEMENT

investment in chip factories” as spurring fears of a glut—yet the revised

investment plans given for seven Japanese semiconductor companies were
up only a very modest 6.9 percent over the initial “restrained” plans
described by that same newspaper, back in July, as unlikely to allow
production to keep pace with demand.146 TI Chairman Jerry Junkins, in
an interview with a Japanese reporter, stepped up the pressure by arguing
that “from now on, there will be a slowdown in increases in demand. So
manufacturers should be responsible for their facilities investments so
that they will not disturb the order of prices and market trends.”147
MITI and Japanese producers continued to predict a shortage well
into 1989, however, and manufacturers announced further increases in
revised fiscal 1988 facilities investment.148 By the end of October revised
investment plans for the six largest Japanese chip producers were up 24
percent over initial fiscal 1988 levels.149 The increase prompted prominent
U.S. press coverage of Japanese producers “fighting each other” in a
bruising race to invest in new capacity.150 But through the end of 1988,
and into early 1989, demand in Japan continued at robust levels, relatively
high prices were sustained, and industry insiders predicted continued
shortages.151

146. See Tadashi Tamaki, “Heavy Investment in Chip Factories Spurs Fear of Glut,”
Japan Economic Journal, October 1, 1988, pp. 1, 5; and Tamaki and Kikuzumi, “Shortage
of DRAMs Causing Headaches,” p. 4.
147. See “Concerned About Increase in Japan’s Investments; Interview with US TI
Chairman,” Nihon Keizai Shimbun, October 4, 1988, p. 8; and “IC Maker Fears Output
Rise Might Kindle Dumping,” Japan Economic Journal, October 15, 1988, p. 19.
148. With the release of the fourth-quarter 1988 forecast, MITI officials predicted
further shortages of 1M DRAMs through at least March 1989. “1M DRAM Microchips to
be in Short Supply,” Japan Economic Journal, October 15, 1988, p. 17. In a widely cited
interview, Toshiba’s Tsuyoshi Kawanishi told U.S. industry executives in late September
that memory chip shortages could be expected to last another three years. “Prolonged Chip
Shortage Seen,” Financial Times, September 28, 1988, p. 26; “Chips Shortage Will Last 3
Years,” Business Times (Malaysia), September 28, 1988, p. 13.
149. “Chip Makers to Hike Investment,” Japan Economic Journal, November 5, 1988,
p. 12.
150. David E. Sanger, “A New Japanese Push on Chips,” New York Times, November
9, 1988, p. Dl.
151. See “Semiconductor Memory Chips; High Prices Sustained, Due to Tightness of
Demand and Supply,” Nihon Keizai Shimbun, November 9, 1988, p. 20; “1M DRAM Chip
Prices to Continue to be High in Dec.,” Nikkei Top Articles, November 23, 1988; and
“Semiconductor Prices on Spot Market; High Prices Continuing for Chips of High-Speed
Processing;” Nihon Keizai Shimbun, January 24, 1989, p. 20.
THE SEMICONDUCTOR TRADE ARRANGEMENT 209

Clampdown on the Gray Market

The period from late 1987 through early 1989 saw a continued attempt
by Japanese chip producers, with government backing, to dry up un¬
authorized sales in the gray market. MITI export regulations—which
required foreign chip buyers to register with MITI, and obtain a certifi¬
cate from manufacturers attesting to the origin of exported chips, before
the ministry would grant an export license—were potent bureaucratic
barricades blocking off access to the Japanese gray market by foreign
purchasers.
The major manufacturers also moved on their own in 1988 to reduce
supplies filtering into the Japanese gray market. Producers increased their
surveillance of chip sales, to watch for unauthorized resale of their prod¬
ucts by authorized buyers. Retaliation for such gray market transactions
could serve to discourage them. In 1988 and 1989 reports of these efforts
to reduce gray market sales surfaced with increasing frequency in the
Japanese trade press.152

152. “The spot market prices of semi-conductor memory chips have been sharply rising
from the bottom prices in the first half of last year, as various manufacturers have been
tightening their supply to the spot market because the Japan-US semiconductor friction
came to arise. Especially, that the market prices having been rising since the beginning of
this year reflects the fact that the manufacturers’ side has been strengthening their moni¬
toring of memory chip sales in order to prevent the channeling thereof through their
affiliated routes. At one time, memory chips flowing back from users to the market and
memory chips removed from such products as used personal computers were seen on the
market for sale. In this case, too, it is said that ‘there is almost nothing because manufac¬
turers are strengthening their watch’ (spot market dealer).” “Semiconductor Memories—
Spot Market Prices Continuing to Rise; 256K Prices Double from Beginning of This Year,”
Nihon Keizai Shimbun, October 4, 1988, p. 20.
“On the spot market, centering on Akihabara, Tokyo, the shortage of DRAM chips
has become still more serious than ever . . . there is a voice saying as follows: ‘Manufac¬
turers have been tightening their monitoring of sales, and so the influx [of 256K DRAM
chips] into the spot market through their authorized agents has stopped completely. We
cannot answer inquiries.’ (spot market dealer).” “Semiconductor Memory Priced One Stage
Higher; Demand Remains Vigorous,” Nihon Keizai Shimbun, December 13, 1988, p. 20.
“There is a voice saying that ‘High-speed processing [DRAM] chips will not appear on
the spot market because domestic manufacturers are strengthening their monitoring of
sales’ (spot market dealer in Akihabara, Tokyo).” “Semiconductor Prices on Spot Market,”
p. 20.
“Industry experts said Japanese manufacturers are restricting the outflow of high-speed
256-kilobit DRAMs into the spot market.” “Spot Prices of Microchips Up Slightly,” Nikkei
Top Articles, February 9, 1989.
210 THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 4-3. Year-to-Year Changes in Fisher Ideal Price Indexes for

DRAMs and EPROMs, 1972-89

Percent

Source: Based on data in chap. 5, figures 5-2 and 5-3.

Reaction in the United States

By early 1988 a full-fledged shortage of DRAMs was being widely felt

in the United States and Europe. As prices soared, substantial differen¬
tials between Japanese large-user prices and foreign contract prices ap¬
peared. (On the gray market, inherently much more difficult to control,
available data suggest that U.S. and Japanese spot prices were roughly
equalized.) These differentials persisted throughout 1988 and into 1989,
as Japanese manufacturers increased their surveillance of chip transac¬
tions by their sales agents, in order to reduce supplies filtering into the
Japanese gray market.
Figure 4-3 shows just how extraordinary changes in Fisher Ideal price
indexes for DRAMs and EPROMs over this period look when compared
with earlier years. These indexes refer to average worldwide sales prices;
U.S. contract prices soared by a considerably greater margin, and spot
prices jumped even higher, roughly quadrupling in the first months of
1988 (chapter 5 describes the construction of this chained index of year-
THE SEMICONDUCTOR TRADE ARRANGEMENT 211

to-year Fisher Ideal price comparisons and presents detailed price anal¬
yses). As American users howled in pain over enormous, unprecedented
price increases, criticism of the STA mounted.
American chip producers responded by arguing that reductions in
supply and increases in price were the consequence of a successful cam¬
paign of predation, rather than an outcome created or facilitated by the
STA. One influential defense articulated this view in a particularly con¬
cise way:
Any economist will tell you that we shouldn’t complain about foreigners
dumping, because consumers benefit. The one exception is if foreign firms
can put domestic firms out of business, and then raise prices. If it is costly
to re-enter the business (like it is to restart DRAM production), foreign
firms can gain monopoly profits at the consumer’s expense.
. . . Rather than signaling a bankrupt trade policy, today’s shortages in
DRAMs should remind us that dumped products in an industry like semi¬
conductors usually lead to higher prices and limited availability if domestic
suppliers are allowed to be destroyed.153

As in earlier episodes during the 1980s, the industry’s analysis of rising

U.S. prices was that monopoly power created through earlier expendi¬
tures on a successful campaign of predation was finally being exploited.
In the spring of 1989, SI A vice chairman Wilf Corrigan pointed out to a
House of Representatives subcommittee that, while the sanctions had
compelled Japanese producers to stop dumping, “what [the Japanese]
did then was to begin to double and triple prices, and to act like you’d
expect a cartel to act.” Because earlier dumping had forced other pro¬
ducers out of the market, argued Corrigan, “they could afford then to
raise prices with impunity and say that they were doing it to comply with
the agreement.”154 Such claims of predatory Japanese behavior were fur¬
ther reinforced when it became evident that market prices had stayed
well above the Commerce Department’s floor prices (the FMVs) in 1988
and 1989.

153. David B. Yoffie, “Chip Shortage: Don’t Blame the Pact,” Wall Street Journal, June
21, 1988. Yoffie, a professor at the Harvard Business School, became a member of Intel’s
board of directors in 1989.
154. Wilfred Corrigan, in testimony before the Subcommittee on Oversight and Inves¬
tigations, House Committee on Energy and Commerce, March 1, 1989, as reported in
“Japan Should Be Targeted by Super 301 for Failure to Honor Chips Pact, SIA Says,”
International Trade Reporter, vol. 6 (March 8, 1989), p. 290. The published transcript of
his testimony contains a slightly different, and somewhat garbled, variant of these remarks.
See Unfair Foreign Trade Practices, Hearings before the Subcommittee on Oversight and
Investigations of the House Committee on Energy and Commerce (GPO, 1989), p. 107.
212 THE SEMICONDUCTOR TRADE ARRANGEMENT

“Coordination Structures” and “High Price Stability,”

1989 to 1990

Demand for semiconductor began to weaken in late 1988 and early

1989, and by the late spring of 1989 the semiconductor industry had
entered another recessionary period. This weakening in demand, to¬
gether with the rapid expansion of DRAM production by Korean pro¬
ducer Samsung, was to radically alter the operation of the semiconductor
trade regime.

Price Stabilization

With demand weakening, DRAM prices began to drift downward in

early 1989. Although sharper declines had been expected, contract prices
for 1M DRAMs to be delivered in the second quarter of 1989 remained
relatively stable.155 Demand for memory chips continued to weaken
through the late spring of 1989, yet prices remained relatively high. What
was happening?
Unpublished MITI statistics of monthly semiconductor shipments by
Japanese producers show that 1M DRAM shipments were basically flat
from March through May of 1989.156 The official forecast data show that
1M DRAM production during the first half of 1989 rose to an output
rate roughly 32 percent higher than that of the last quarter of 1988;
inventories of 1M DRAMs jumped by an unprecedented 3.2 million units
over the first half of 1989.157 Thus, Japanese government data suggest
that, after increasing production of 1M DRAMs during the first quarter

155. Use of the expression “high price stability” began appearing in the Japanese press
around early 1989. For example, “it has become definite that the 'high price stability’ will
be sustained for the time being, while being backed by steady-toned demand.” “1M
DRAM; 'High Price Stability’ to be Sustained in April-June Period, Too,” Nihon Keizai
Shimbun, February 1, 1989, p. 9.
156. Monthly statistics on shipments through May 1989 were readily available when I
interviewed a MITI official on this subject in December 1989. These MITI statistics were
even more extensive and detailed than the quarterly numbers assembled for the publicly
released “forecast.”
157. Production of 1M DRAMs during the last quarter of 1988 was roughly 69.6 million
units, compared with an average quarterly output rate of 92 million units during the first
half of 1989. See MITI, Machinery and Information Industries Bureau, “Semiconductor
Supply-Demand Forecast,” September 30, 1988, December 23, 1988, and June 22, 1989.
The largest quarterly increase in producer inventories of 1M DRAMs prior to this, regis¬
tered in the first quarter of 1987 (when MITI had first provided producers with “guidance”
to reduce 1M DRAM shipments), was 0.5 million units.
THE SEMICONDUCTOR TRADE ARRANGEMENT 213

Figure 4-4. Output of 1M DRAMs by Japanese and Non-Japanese

Manufacturers, 1989

Millions of finished chips

Source: Author’s calculations based on unpublished Dataquest data.

a. Toshiba, Hitachi, NEC, Fujitsu, and Mitsubishi.
b. NMB, Sanyo, and Sharp.

of 1989, producers began to cut back production in the late spring, sliced
shipments even more, and increased their inventories of parts.
Unofficial production estimates by the semiconductor industry con¬
sulting firm Dataquest provide an even clearer picture of what was going
on. Figure 4-4 shows Dataquest’s quarterly estimates of DRAM produc¬
tion by several groupings of firms.
It is clear that the major Japanese DRAM producers, as a group, cut
back production after the second quarter of 1989. (Because of steadily
increasing productivity on chip manufacturing lines, constant output ac¬
tually means cutbacks in production runs.) The additional 1M DRAM
chip supply, entering a market increasingly depressed by sagging com¬
puter demand, was mostly coming from non-Japanese suppliers, who
were rapidly increasing production.
At the time this was happening, 1M DRAM prices were far above the
price floors then set for exports (roughly triple the average FMV; see
214 THE SEMICONDUCTOR TRADE ARRANGEMENT

chapter 5). Contemporary press accounts suggested that Japanese chip

makers, as a group, were cutting back production to stabilize prices and
slow their decline. One analysis of this “high price stabilization” ap¬
peared in Nihon Keizai Shimbun in June 1989:

it does not seem that there will be a collapse of prices at one stroke.
Manufacturers have established a “co-ordination structure” among them¬
selves in the last several years, and they unanimously say that they “will
head toward a soft landing, which will gradually follow a downward curve.”
Concerning this artificial price policy, American users, which depend on
Japanese products for the greater part of their semiconductor procurement,
are now beginning to voice criticism. The focus of the Japan-US friction,
in which the low prices were a problem, is likely to be placed on the high-
price stabilization of Japanese products from now on.
... On the other hand, however, users show their dissatisfaction, as
follows: “That good showing [high profits of Japanese semiconductor mak¬
ers] in their settlements of accounts stands on the sacrifice of users.” Users
demand “a price reduction to the fair market value (FMV).”
... In the US too, users came to show obvious moves for taking the
price determination right away from Japanese manufacturers, as they are
irritated at Japanese manufacturers price determination based on their
dango [collusion] constitution. The rising of arguments in the US for cre¬
ating a semi-conductor futures market, is a sign indicating their intention
to leave the price-determination right to the market. Also, seven US man¬
ufacturing companies’ establishment of “US Memories,” which is a DRAM
joint production company, shows their intention of trying to weaken Japa¬
nese manufacturers’ price control.158

As semiconductor demand continued to weaken over the summer and

fall of 1989 and inventories continued to increase, many wondered
whether manufacturers could maintain their discipline and refrain from
lowering prices to sell their output. But chip makers recognized that it
would create a potential threat to the profitability of the industry as a
whole if all succumbed to temptation and a round of vicious price cutting,
like that seen in 1985, broke out. Nihon Keizai Shimbun noted:

“In order to prevent this cutthroat competition from recurring, we decided

not to increase our production,” said Goh Kawanishi, special general man¬
ager of Toshiba, the leading manufacturer in the semiconductor industry.
However, as the general economic environment that surrounds the semi-

158. “Users Dissatisfied with ‘High Price Stabilization’; Semi-Conductor Manufactur¬

ing Companies to Carry Out Price Reduction of 5% from Next Month,” Nihon Keizai
Shimbun, June 28, 1989, p. 2.
THE SEMICONDUCTOR TRADE ARRANGEMENT 215

conductor industry worsens, many experts wonder how long this “coordi¬
nation structure” among manufacturers will last.159

In the early fall of 1989, as demand continued to weaken, production

cuts were widely reported in the Japanese trade press. At the end of July
NEC announced that it had scaled back its plans for increasing produc¬
tion of 1M DRAMs.160 In September Toshiba, Hitachi, and Mitsubishi
each reported that they were cutting back their production levels by about
10 percent.161 In late October NEC announced that it too would actually
cut current production levels in the first quarter of 1990.162
Although clearly unwilling to reveal the precise details, companies
were actually quite open about the trend toward “cooperation.” In early
1990, for example, a U.S. government official asked several executives
from a major Japanese chip producer why the company was cutting back
DRAM production even though the market-clearing price was still several
multiples of production cost. One executive replied, “since the Semicon¬
ductor Agreement, we [Japanese DRAM manufacturers] have moved
from competing for market share to market sharing.”163
Where the government did continue to play an explicit and overt ad¬
visory role to companies was in investments in new capacity. Although
rigorous guidance appears to have ended in mid-1988, MITI continued
to provide “advice” to producers on capacity plans. In November 1989,
for example, MITI was reportedly telling Japanese firms not to increase
capacity, because overcapacity in a downturn was likely to lead to firms
exporting more, exacerbating trade frictions.164 And in December 1989

159. “Falling Spot Price: Over-Production and Less Demand for Semiconductors in
Japan,” Nihon Keizai Shimbun, August 3, 1989, p. 24.
160. NEC withdrew its plan to increase monthly production of 1M DRAMs from
6 million to 8 million chips. Nikkan Kogyo, July 29, 1989, p. 1; Nihon Kogyo, July 29, 1989,
p. 1.
161. “Major Semiconductor Manufacturers to Reduce 1M-DRAM Production by 10%
from September,” Nihon Keizai Shimbun, September 14, 1989, p. 10.
162. The announced cuts were from 6 million to 5 million units per month. Dempa
Shimbun, October 31, 1989, p. 1; Nikkan Kogyo, October 31, 1989, p. 1; Nihon Kogyo,
October 31, 1989, p. 5.
163. Author’s interview with a U.S. government official, May 1990.
164. Author’s interview at a Japanese semiconductor company, November 1989. Re¬
ports of MITI guidance on facilities investments in automobiles and electronics, complete
with a threat to invoke the trade control ordinance, were reported in “Temptations of
MITI’s ‘Managed Trade’ (Part 1)—Request for Expansion of Imports; Balancing of Exports
and Imports Pressed for,” Nihon Keizai Shimbun, December 14, 1989, p. 7. MITI officials
I interviewed during this same period admitted only to talking to producers about the
political dimensions of their investments, such as investment locations in the European
216 THE SEMICONDUCTOR TRADE ARRANGEMENT

several Japanese producers were still getting “opinions from MITI on

how much investment in semiconductor capacity was appropriate.16 As
before, however, U.S. policy appears to have played an important role in
encouraging such action. In early November 1989 the U.S. government
had urged MITI to restrain Japanese companies’ investments in several
industries, including automobiles and electronics.166
MITI also continued to operate its price monitoring system, although
some modifications were made in March 1989, in response to the GATT
panel report.167 Extensive data on exports and unit prices for many dif¬
ferent products, including DRAMs, EPROMs, SRAMs, microproces¬
sors, and microcontrollers shipped to nineteen different countries, were
collected on a monthly basis from Japanese producers. The semiconduc¬
tor monitoring office checked export prices against costs (retrospectively)
and “advised” companies when there appeared to be problems.168 By
some press accounts, MITI occasionally provided guidance to Japanese
companies on pricing of DRAMs and SRAMs in the domestic market,
to prevent rampant price cutting and “share the market with overseas
makers.”169

U.S. Memories to the Rescue

In September 1988, as panicked U.S. consumers scoured the globe for

DRAMs, the SI A and the American Electronics Association (AEA)

Community. Even that level of intervention was officially denied: “MITI . . . has vigorously
denied guiding Japanese companies to spread their investments around European Com¬
munity countries rather than concentrating them in the UK. ‘We have never given any
guidance on this issue. It is a matter that is entirely up to companies themselves to decide,’
MITI said here yesterday.” Ian Rodger, “Tokyo Denies Issuing Anti-UK Guidance,”
Financial Times, December 21, 1988, p. 3.
165. Author’s interviews with Japanese semiconductor executives, November and De¬
cember 1989.
166. See “U.S. Request Contains Contradictions; ‘Concerned over Increase in Japan’s
Facilities Investments,’” Nihon Keizai Shimbun, November 5, 1989, p. 3; “Urges Holding
Down of Expansion of Facilities Investments; On Automobiles, Iron-Steel, Electronic
Equipment, and Ship-Building,” Nihon Keizai Shimbun, November 7, 1989, p. 1; and
“Temptations of MITI’s ‘Managed Trade’ (Part 2)—‘Guidance’ Also for Facilities Invest¬
ments,” Nihon Keizai Shimbun, December 15, 1989, p. 7.
167. See MITI, “Japan’s Coordination Policy in Response to the GATT Panel Report
on Semiconductors,” Tokyo, March 7, 1989.
168. Author’s interviews with a Japanese semiconductor executive and with MITI of¬
ficials, 1991.
169. “MITI Helps Prevent Memory Chip Price Fall,” Nikkei Top Articles, May 25, 1990.
THE SEMICONDUCTOR TRADE ARRANGEMENT 217

created a joint Extraordinary Measures Task Force to formulate a re¬

sponse.170 In March 1989, while DRAM demand remained strong and
prices high, a proposal to form a joint U.S. production consortium made
up of both memory chip users and producers was circulated among the
membership of the SIA and the AEA. The joint project was conceived
as a way to overcome users’ DRAM supply problems and producers’
capital availability problems.171 By late June U.S. Memories, a formal
joint venture, had been launched with the initial participation of three
computer firms (IBM, Digital Equipment Corporation, and Hewlett-
Packard) and four semiconductor producers (LSI Logic, National Semi¬
conductor, Advanced Micro Devices, and Intel). Each partner had
agreed to take 5 to 10 percent equity shares in the new for-profit venture.
In addition, IBM offered its 4M DRAM technology to the consortium,
a senior IBM executive (Sanford Kane) took up the reins as CEO of the
new venture, LSI Logic’s chairman (Wilfred Corrigan) agreed to chair
its board of directors, and 1991 was set as the target date for initial
production of 4M DRAMs.172 A broad consensus—including public
expressions of support from rival U.S. DRAM producers Motorola, Mi¬
cron, and Texas Instruments—for the venture seemed to prevail through
the summer.173 In addition to the original seven committed participants,
who had all signed letters of intent due to expire in December 1989,
additional companies, including AT&T, which actually contributed some
seed money but no firm decision, hovered at the margins of the deal.174
By summer, however, trouble clouds had begun to collect over U.S.
Memories. In late July Cypress Semiconductor Corporation president and
industry maverick T. J. Rodgers, complaining that he had not been in¬
vited to join, blasted the proposed consortium as a threat to his company
before a congressional subcommittee: “As soon as the necessities of the
semiconductor market force U.S. Memories to drift outside the dynamic
RAM market, they will be a company armed with antitrust immunity

170. “DRAM Co-op: Now’s the Time,” Electronic Buyers’ News, July 10, 1989, p. 17.
171. Brian Robinson, “SIA, AEA Urge DRAM Alliance Formations,” Electronic Buy¬
ers’ News, March 20, 1989.
172. Jack Robertson, “Confirm Major IBM Role in Creation of U.S. Memories,” Elec¬
tronic News, June 26, 1989, pp. 1, 58; and John Thompson and Stacey Peterson, “DRAMs:
Made in the USA,” Computer Systems News, June 26, 1989, pp. 1, 8.
173. Patrick Burnson, “DRAM Joint Venture Draws Praise,” Infoworld, July 3, 1989,
p. 32; and Ann Thryft, “Domestic DRAM Makers Applaud U.S. Memories,” Computer
Design, July 17, 1989, pp. 1, 38.
174. David Roman, “DRAM Co-op Catching On,” Electronic Buyers' News, Septem¬
ber 18, 1989, pp. 20-22.
218 THE SEMICONDUCTOR TRADE ARRANGEMENT

and government subsidies—competing with my company and the

hundreds of other semiconductor companies in the United States that do
not enjoy the luxury of immunities and subsidies.”175 Even more impor¬
tant, perhaps, the DRAM market had begun to weaken. In late Septem¬
ber, Apple Computer announced it would not join the consortium.176 By
November, Sun Microsystems and Tektronix had announced their disin¬
terest too, even as Rodgers launched another public campaign against
the consortium, this time volunteering to license IBM’s DRAM technol¬
ogy and deliver products by 1990, well ahead of the consortium.177
By December, amid continuing declines in DRAM prices, U.S. Mem¬
ories managed to conclude its licensing deal with IBM. But as the letters
of intent signed by the seven original backers approached their expiration
date, and with them commitments to invest capital and purchase the
consortium’s output, no new firms could be found to take the remaining
shares of the consortium. Indeed, additional computer companies such
as Unisys, Tandy, Dell Computer, and Everex Systems had gone public
with their disinterest.178
The final blow to U.S. Memories was delivered at a meeting on Jan¬
uary 10.179 Hewlett-Packard scaled back its initial commitment, even as
possible backers AT&T, Compaq, NCR, and Tandem Computers re¬
ported that they were still not ready to commit. Unsuccessful in putting
together a deal after nine months of intense maneuvering, suffering re¬
peated assaults by critics, and faced with a softening memory market,
the consortium was officially terminated and consigned to a historical
footnote. The irony of the episode is that in later years smaller groupings
of companies announced a variety of joint ventures in the chipmaking
business that had many similarities to the U.S. Memories proposal. U.S.

175. David Roman and Brian Robinson, “IC Coops: A Cartel Threat?,” Electronic
Buyers’ News, July 31, 1989, pp. 1, 50.
176. Richard March, ‘‘Clock Running on Stalled U.S. Chip Consortium,” PC Week,
October 2, 1989, p. 139; and Eric Nee, “Apple Shuns DRAM Consortium,” Computer
Systems News, October 2, 1989, pp. 3, 8.
177. David Roman, “U.S. Memories Suffers Rejections”; and Hugh G. Willett, “Cy¬
press Would Speed IBM Design,” Electronic Buyers' News, November 20, 1989, pp. 1, 66.
The IBM technology actually was licensed to Micron, but Rodgers and Cypress ultimately
did not take the plunge into DRAMs.
178. Richard March, “U.S. Memory-chip Group Garners Lukewarm Support,” PC
Week, December 11, 1989, p. 137; and David Roman, “Co-Op Gets IBM License,” Elec¬
tronic Buyers' News, December 25, 1989, pp. 1, 4.
179. David Roman, “U.S. Memories Lades Away,” Electronic Buyers’ News January
22, 1990, pp. 1, 56.
THE SEMICONDUCTOR TRADE ARRANGEMENT 219

Memories was unique, however, in that its principals consisted exclusively

of U.S. companies and that it was responding to a problem conceptual¬
ized as a national problem of vulnerability in supply of a key industrial
commodity.

Korea Forges Ahead

Continued weakness in demand for semiconductors led Japanese pro¬

ducers to again announce a round of production cuts in early January
1990. The public announcements in Tokyo of cutbacks in DRAM pro¬
duction by the three largest memory chip makers were virtually simul¬
taneous, prompting a prominent report in the U.S. press that they were
“acting much like a cartel.”180
The cuts, publicly justified as an effort to prevent drastic price declines,
also roughly coincided with the demise of the U.S. Memories DRAM
production consortium. Some in the U.S. industry even charged—
unjustifiably, in light of the evidence just presented on 1989 cutbacks—
that Japanese companies had flooded the market and forced prices down
in an effort to torpedo U.S. Memories.181

180. “The contrast was striking: On the day last week when American electronics
companies decided to abandon their cooperative venture to make computer chips, their
competitors in Tokyo had already moved in lockstep.
“One by one, within hours, Japan’s biggest chip makers announced plans to cut their
production of one-megabit memory chips. . . . There was far too much supply, each com¬
pany explained in great detail, and unless production was cut, prices would continue to fall
drastically.
'The Japanese companies, despite repeated contentions that they are now each other’s
fiercest competitors, were acting much like a cartel.” David E. Sanger, “Contrasts on
Chips,” New York Times, January 18, 1990, p. Dl.
Sanger’s account is incorrect on one important point—that Japanese producers chose
to reveal their cutback plans just after the proposed U.S. Memories DRAM production
consortium failed. The cutbacks described by Sanger were reported in the Japanese press
on Sunday, January 7, and in the U.S. press on Tuesday, January 9. See “Drastic Drop of
1M DRAM Price: Semiconductor Makers Reinforce Production Cut (10-15%) Starting
This Month,” Nihon Keizai Shimbun, January 1,1990, p. 5; and G. Pascal Zachary, “Japan’s
Biggest Memory-Chip Makers Are Cutting Output in Bid to Ease Glut,” Wall Street Journal,
January 9, 1990, p. B4. The critical meeting at which the U.S. Memories proposal clearly
failed was on Wednesday, January 10, and its failure was publicly announced on Monday,
January 15. See Stephen Kreider Yoder, “U.S. Memories to Abandon Bid for Chip Ven¬
ture,” Wall Street Journal, January 15, 1990, p. B4; “Lessons Linger as U.S. Memories
Fails,” Wall Street Journal, January 16, 1990, p. Bl; and Andrew Pollack, “Memory Chip
Cooperative Is Officially Declared Dead,” New York Times, January 16, 1990, p. Dl.
181. These charges were widely repeated within the U.S. semiconductor industry. They
actually appeared in print in an article by Michael Borrus (“Chips of State,” Issues in
220 THE SEMICONDUCTOR TRADE ARRANGEMENT

In early January major Japanese chipmakers simultaneously an¬

nounced large cutbacks in production of 1M DRAMs.18- Reducing output
did, indeed, stabilize prices, which actually rose in the Japanese market
in early 1990.183 An uptick in demand—widely attributed to increased
demand for memory used in newly introduced notebook computers and
high-performance workstations—coupled with the production cuts,
led to rising 1M DRAM prices and a brief panic over shortages in the
U.S. market in March 1990.184 (Some analysts believe that IBM’s new,
memory-intensive OS/2 operating system for personal computers, intro¬
duced at about this time, suffered in the market as a result of this firming
up of memory prices.) Japanese producers responded by reducing sec¬
ond-quarter production of 1M DRAMs less rapidly, but prices remained
stable in Japan.185 In response, Japanese chip makers actually reversed
course and increased their output of 1M parts over the summer.186

Science and Technology (Fall 1990), pp. 40-48), which claimed that “the Japanese press
reported that Japanese firms were creating a glut in the market to lower prices and thus
discourage the [U.S. Memories] initiative” (p. 44).
Despite considerable effort, I was unable to turn up any such report in the Japanese
press, and in a telephone conversation in November 1990 Borrus, an analyst at the Berkeley
Roundtable on the International Economy, acknowledged to me that he could provide no
specific reference for this claim. The belief that Japanese makers had engineered a shortage
to kill U.S. Memories was, however, fueled by the New York Times, which claimed (incor¬
rectly) that production cuts were announced just after the abandonment of U.S. Memories:
“No one has suggested that Japan’s production cuts and the abandonment of U.S. Memories
are directly related. But both were spurred by a growing glut of chips, after two years of
huge demand.” Sanger, “Contrasts on Chips,” p. D20.
As noted above (and in the Japanese press), Japanese producers were actually cutting
back on shipments of DRAMs in 1989, in order to shore up prices—the precise opposite
of behavior consistent with “creating a glut in the market.”
182. See “Drastic Drop of 1M DRAM Price: Semiconductor Makers Reinforce Pro¬
duction Cut (10-15% Starting This Month),” Nihon Keizai Shimbun, January 7, 1990, p. 5.
“Another Sharp Price Reduction for Semiconductors,” Nihon Keizai Shimbun, January 18,
1990, p. 24, reported an 11 percent cut in 1M DRAM output at Toshiba, 17 percent at
NEC, 22 percent at Mitsubishi, 15 percent at both Fujitsu and Hitachi, and 27.5 percent at
Oki.
183. “Price Reduction in Semiconductors Stopped; Positive Result of Production Cuts,”
Nihon Keizai Shimbun, February 27, 1990, p. 24.
184. G. Pascal Zachary, “Computer Firms Face Shortage in DRAM Chips,” Wall Street
Journal, March 9, 1990, p. A4; Corinne Bernstein-Levy, “Cost Reversal Baffles Buyers,”
Electronic Buyers’ News, March 5, 1990, pp. 1, 50; and Barbara Darrow, “DRAM Dearth
Overblown, Vendors Say,” Infoworld, March 19, 1990, p. 105.
185. See “1M DRAM Wholesale Market Recovered; Major Manufacturers Began
Increasing Production,” Nihon Keizai Shimbun, June 21, 1990, p. 3.
186. See “Makers Once Again Boosting 1M DRAM Production,” Nikkei Top Articles,
June 21, 1990; “Wholesale Prices Cut for 1M DRAM; First Reduction in Six Months,”
THE SEMICONDUCTOR TRADE ARRANGEMENT 221

The resurgence in demand was short-lived in the American market,

where prices soon resumed their descent.187 Because of stronger demand
in the Japanese market, where prices were rising, investment plans for
fiscal 1990, announced in May, remained at fiscal 1989 levels for Japan’s
ten largest chip makers.188 In a relatively tight market, producers’ contin¬
uing efforts to dry up supplies of chips resold into the Japanese gray
market were generally successful.189
Continued weakness in overseas markets finally dragged down buoyant
Japanese DRAM prices in late 1991.190 The real news at this point,
though, was that significant imports of 1M DRAMs into the Japanese
market were forcing Japanese makers to reduce their prices.191 As Japa¬
nese producers had cut back production to maintain prices, other late
entrants into the 1M market overseas—particularly Korea’s Samsung—
had rapidly increased their production and market share. Not subject to
the political and legal pressures faced by Japanese chip makers, they
were able to sell everything they could make by slightly undercutting the
export prices established by their Japanese competitors. By late 1991 this
tactic had even begun to put pressure on Japanese vendors’ prices in their
own home market.
These realities were already evident by the end of 1990. As one re¬
spected source commented,

Nihon Keizai Shimbun, September 15, 1990, p. 24; and “Major Chip Manufacturers Hesi¬
tate in Increasing 1M DRAM Production,” Nihon Keizai Shimbun, October 4, 1990, p. 3.
187. David Roman, “DRAMs Slide Again,” Electronic Buyers’ News, June 25, 1990,
pp. 1, 58.
188. Dempa Shimbun, May 28, 1990, p. 1; Nikkan Kogyo, May 26, 1990, p. 6; “Future
Direction of Sematech,” Nihon Keizai Shimbun, June 5, 1990, p. 11; and Satoshi Isaka,
“Electronics Makers Still Optimistic,” Japan Economic Journal, June 23, 1990, p. 17.
189. “Spot dealing in semiconductor memory chips has decreased rapidly and some of
the buyers are closing their businesses. This is because there is a diminished supply of chips
in the spot market due to the manufacturers’ instructions to their own authorized distrib¬
utors. . . .
“The spot market has been said to reflect the trend of demand and supply of semicon¬
ductors accurately and a leading index for authorized distributors. While this basic belief
has not been destroyed, some point out that ‘free price formation has been threatened as
manufacturers’ influence over distribution has grown, and administrative guidance, backed
by the U.S.-Japan Semiconductor Agreement, has strengthened.” “Rapid Decrease in Spot
Dealing of Semiconductor Memory,” Nihon Keizai Shimbun, June 7, 1990, p. 24.
190. “Wholesale Prices Cut for 1M DRAM; First Reduction in Six Months,” Nihon
Keizai Shimbun, September 15, 1990, p. 24; and “Sharp Wholesale Price Reduction of
1M,” Nihon Keizai Shimbun, October 10, 1990, p. 24.
191. See “Spot Chip Prices Continue to Decrease,” Nihon Keizai Shimbun, October
31, 1990, p. 24.
222 THE SEMICONDUCTOR TRADE ARRANGEMENT

The reason for the price drop of 1M can be said to be that production
capability is larger than actual demand. Although domestic LSI makers
have been trying to stabilize the price of 1M through large production cuts,
since “the Japanese 1M market share in the U.S. has dropped below 50%”
(according to Mitsubishi) production cuts may not be effective. The lower
market share in the U.S. is due to the fact that several other makers such
as Samsung, Siemens, and Micron Technology have improved their supply
capability. Therefore Japanese makers cannot depend on production cuts
to stabilize price and must “wait for demand to recover.”192

By 1991 Samsung was the largest producer of 1M DRAMs in the

world; Siemens and Micron had also climbed rapidly up the ranks of
producers of this product. Japanese producers reacted not by slashing
prices (and profits), but instead cut back on capacity used to produce the
older generation product and focused their resources on rapidly shifting
to the next-generation memory chip, the 4M DRAM.
Four-megabit DRAMs were first shipped in volume in 1989, and by
1990 they dominated the chip making efforts of Japanese makers. In
effect, Japanese producers phased out of the lagging edge product, the
1M DRAM, and shifted into the leading edge product. For 1989 through
1991, Japanese producers’ share of 4M DRAM output could only be
described as overwhelming (see chapter 5).
However, the industry suffered through continuing depressed demand
for computers through most of 1991. Since a large-scale shift toward use
of 4M chips would require new computer designs, persistent slowdown
in the computer industry also tended to put relatively more pressure on
leading edge chip prices, since slower computer innovation meant less
demand for advanced chips. No less than 80 percent of some producers’
4M DRAM production, for example, was reportedly sold to American
computer companies.193
Slowing computer sales in 1991 hit major producers just as they were
ramping up to peak production of the 4M DRAM. Where MITI had
forecast 20 million units would be demanded domestically, actual domes¬
tic sales for the first half of 1991 were only 12.1 million units; MITI’s
forecast of 32.7 million units sold domestically for the second half of 1991
contrasted with actual sales of 19.4 million. Net exports also fell some-

192. “Japanese Promote Strategic Production Formation and Increased Investment of

Plants, Desire an End to US-Japan Semiconductor Agreement,” Nikkei Microdevices,
January 1991, p. 68.
193. “Semiconductor Production Scaled Down, So Is Also Output of Circuit Photo¬
copier,” Japan Economic Review, August 15, 1991, p. 11.
THE SEMICONDUCTOR TRADE ARRANGEMENT 223

what short of the forecasts: 42 million units rather than a forecast

39.6 million were shipped in the first half, and 62.9 million instead of a
forecast 72.5 million in the second.
Through most of 1991, then, Japanese producers were forced to reduce
the rate at which new 4M capacity was to be brought online, and to sell
what they did at lower prices more competitive with the continually
declining prices of older generation chips.194 The so-called crossover
point, where the price per bit of memory for a new chip reaches rough
parity with that of an older generation chip, came in 1991, less than two
years after the crossover between the 1M and 256K DRAMs.195
Thus by mid-1991, as the Semiconductor Trade Arrangement expired,
a new pattern of competition between Japanese memory chip producers
and their foreign competitors had emerged. Constrained by export price
floors in their ability to compete on price in mature products, and redi¬
rected from “competition for market share” into “market sharing” by
political pressures, Japanese semiconductor companies were looking to
the technological leading edge for their profits. Successful competition
increasingly meant being first to achieve significant output of the latest-
generation memory chip, and extracting significant profits from that
product before less advanced and less politically constrained competitors
were able to imitate it, enter the market in volume, and wrest control of
pricing from the technological leaders in Japan. The slump in chip de¬
mand in 1991, just as the leap into a more advanced product was being
made, illustrates that the new strategy has not eliminated the risk that
has always been a part of the memory chip business.

Aftermath: The Second Semiconductor Trade Arrangement

In June 1991 a second U.S.-Japan trade arrangement was signed, re¬
placing the first five-year arrangement due to expire in July. The major
change in antidumping measures was the termination of the FMV floor

194. “How Far Will 4M DRAM Fall; Demand Slower Than Mass Production,” Nihon
Keizai Shimbun, August 14, 1991, p. 18; and “Further Price Reduction of 4M DRAM,”
Nihon Keizai Shimbun, August 8, 1991, p. 20.
195. Since every generation of DRAMs quadruples the number of bits on a chip, and
higher density chips are more desirable because of their lower power consumption, lower
interconnection costs, and reduced space requirements, “crossover” is generally defined to
occur when newer generation memory chips are five to six times more costly than older
generation chips. Because a new generation of chip is introduced roughly every three years,
crossover between generations would also be expected to take place at three-year intervals.
224 THE SEMICONDUCTOR TRADE ARRANGEMENT

price system administered by the Commerce Department and its replace¬

ment with a fast-track antidumping procedure, in which Japanese com¬
panies were required to collect cost and price data and have it available
to the U.S. and Japanese governments on fourteen days’ notice in the
event of an antidumping case. Data on third-country exports were also
expected to be made available to a third-country government should it
(possibly at U.S. request) initiate an antidumping case.
The end of the FMV system, however, was less radical a step than it
at first appeared. The data on export prices and costs that producers now
collected were essentially the same as the data previously requested for
the calculation of FMVs, and the format of the FMV calculations was
now well established; thus Japanese manufacturers were effectively re¬
quired to maintain a system of “shadow” FMVs, to be handed over to
the Commerce Department on fourteen days’ notice should frictions de¬
velop. When interviewed in mid-1991, Japanese semiconductor executives
told me that their companies were continuing to use their FMV-like
calculations as guidelines on export pricing. Through June 1991 MITI
was also continuing to monitor export price and quantity data on ship¬
ments of sensitive products to major markets, and although smaller pro¬
ducers were dropped from this system after this date, the major produc¬
ers continue to report these data to MITI.196 And MITI’s revised system
of twice-yearly forecasts of semiconductor supply and demand continued
unchanged from previous years.197 Furthermore, the European Com¬
munity’s system of reference prices, supervised in Japan by MITI,
continued in effect on DRAMs and EPROMs. Thus a variety of well-
established mechanisms designed to constrain pricing in world semicon¬
ductor markets continued in place in their original or revised form.
One significant development came to a resolution in early 1993. Dump¬
ing cases against Korean exports of DRAMs were filed in the European
Community in 1991 and in the United States in 1992. In the United States
the International Trade Commission made an affirmative preliminary
finding of injury and sent a Micron Technology DRAM dumping petition
on to the Commerce Department for further action. In the European

196. Author’s interview with a semiconductor executive, Tokyo, July 1992.

197. Although no longer released officially by MITI, the semiconductor forecasts con¬
tinued to be compiled through 1995 and with persistence (MITI officials at first insisted the
forecasts had been discontinued after the signing of the second STA) could be obtained
from MITI by the author. EPROMs were dropped from the forecast in mid-1994.
THE SEMICONDUCTOR TRADE ARRANGEMENT 225

Community a price undertaking in DRAMs was quietly negotiated in

consultation with Korean producers and EC member states.
Since Korean producer Samsung’s rapid expansion of DRAM produc¬
tion and willingness to undercut Japanese pricing had played an important
role in undermining the “high price stabilization” of 1988 and 1989,
shackling Korean producers with the same restraints placed on the Jap¬
anese had a potentially important impact on global DRAM pricing. If
stringent restraints had been placed on the pricing of Korean DRAM
imports, it could have had a significant effect on memory chip prices.
Both cases were resolved in the spring of 1993. In the United States,
very modest dumping duties were placed on Korean DRAMs: for Sam¬
sung, which accounts for the vast bulk of Korean chip production, the
duty was only 0.82 percent; Hyundai Electronics was assessed a tariff of
11.45 percent, Goldstar Electron a 4.97 percent levy, and all other South
Korean makers a duty of 3.89 percent. These low tariffs had minimal
effects. The most significant factor constraining Korean pricing as a result
of this finding, in fact, was probably the “monitoring” of Korean prices
that followed.198 In August 1995 the U.S. Court for International Trade
ruled that the dumping margins on Korean DRAMs had been incorrectly
calculated. The ruling had the effect of removing dumping duties entirely
for Samsung and lowering duties for Goldstar and Hyundai to a level of
5 percent or less.199
The price undertaking negotiated between the Koreans and the Eu¬
ropeans in early 1993 followed a provisional 10.1 percent antidumping
duty on Korean DRAMs in September 1992. The agreement, which
required Korean makers to submit detailed quarterly cost data to the
European Community, reportedly differed substantially from the much-
criticized floor price arrangement negotiated with the Japanese in 1990.200
Because continuing high prices for memory chips had rendered these

198. See David Roman, “Korean DRAM Case Fizzles,” Electronic Buyers’ News,
March 22, 1993, pp. 1, 44; and Bob Davis, “Panel Clears Way for U.S. to Impose Import
Duties on Korean Microchips,” Wall Street Journal, April 23, 1993, p. All. Indeed, some
within the industry charged that the mild penalties on Korean chips were the result of
political pressure from computer makers.
199. “Semiconductors Face Lower Antidumping Duties,” Korea Times, August 26,
1995, p. 8; and “USA: Samsung Electronics Cleared from Anti-dumping Charges,” Korea
Economic Weekly, November 9, 1995.
200. See Linda Bernier, “EC Acts on Dumping,” Electronic Engineering Times, March
8, 1993, p. 97. The Korean price floors were company specific like the U.S. FMVs, rather
than the single common floor applied by the EC to Japanese,chips.
226 THE SEMICONDUCTOR TRADE ARRANGEMENT

floor prices irrelevant, the European Commission instituted the first of a

series of suspensions of the requirement for Korean companies to file
cost data on European DRAM and EPROM exports as of June 1995.201
Even where no official finding results, dumping cases can still have a
restraining effect. In early February 1990 Micron Technology took the
lead in publicly raising the possibility of a dumping suit against Korean
vendors, although no suit was actually filed until 1992.202 That same
month, in apparent response, Samsung announced a 20 percent produc¬
tion cut.203 Although Samsung was to later resume expanding production
of 1M DRAMs as prices firmed up somewhat later in the spring of 1990,
a point had clearly been made. That same point had been made earlier,
in the 1980s, in successive episodes of Japanese “production coordina¬
tion” and “self-restraint” triggered by resurgent semiconductor trade
friction with the United States. Although perhaps less effective and more
transitory than formal trade restraints, political “jawboning” and the
threat of official action can have real effects on economic behavior.

201. “EU Suspends Antidumping Rule on Semiconductors,” Korea Times, August 26,
1995, p. 8. The suspension was extended in January 1996. “EU to Suspend Anti-dumping
Regulations on DRAM Chip Imports,” Korea Economic Weekly, January 3, 1996.
202. See Greg Garry and Hugh G. Willett, “Korean Chip Dumping?” Electronic
Buyers' News, February 5, 1990, pp. 1, 58.
203. “Samsung Cuts DRAM Production to 20% of Peak,” Nikkei Top Articles, Feb¬
ruary 27, 1990; and “Japan’s 1-MB DRAMs Yet to Respond to U.S.’s BB Ratio Upswing,”
Nikkei Top Articles, May 9, 1990.
 CHAPTER FIVE

Effects of the Semiconductor

Trade Arrangement on
Semiconductor Markets

It was in September 1986 that the United States and Japan officially
signed the first bilateral Semiconductor Trade Arrangement (STA). Re¬
ports in the Japanese press suggested that the formal arrangement was
preceded by a period of informal pressure on the Japanese semiconductor
industry—what in the United States might be called jawboning—from
political and government circles in Japan, reacting in turn to foreign
complaints. From July 1985 on, for more than year before the formal
signing of the STA, Japanese industry was encouraged to reduce low-
priced DRAM exports to the United States and exercise restraint in
production. This effort was actually successful in raising 64K DRAM
prices, but it came too late to preempt the formal conclusion of the
dumping case involving that product, or to preempt the filing of other
trade cases (in EPROMs and in 256K and higher DRAMs). The STA,
which led to suspension of these later dumping cases, the suspension also
of a Section 301 complaint about market access, and the termination of
a private antitrust suit, intervened in the market at three distinct levels.
First, there was the public document released to the media, which
described a couple of major sets of measures. One section of the agree¬
ment described actions to be undertaken by the Japanese and American
governments to encourage increased sales in the Japanese market of

227
228 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

semiconductors made by foreign-based firms, and both governments

agreed that “expected improvement in access should be gradual and
steady over the period of this Arrangement.”' Also as part of the arrange¬
ment’s market access-related measures, both governments reaffirmed
that there should be fair and equitable access to patents resulting from
government-sponsored research, and that both governments had “every
intention to refrain from policies or programs which stimulate inordinate
increases in semiconductor production capacity.”1 2
The other broad class of measures in the public version of the STA
fell under the rubric of “prevention of dumping.” Dumping cases against
Japanese DRAMs and EPROMs were to be suspended in separate agree¬
ments between the U.S. Commerce Department and Japanese chip com¬
panies. These suspension agreements, referred to within the STA itself,
created a system of company-specific price floors (foreign market values,
or FMVs) on Japanese DRAM and EPROM exports shipped to the
United States; the FMVs were to be set and administered by the Com¬
merce Department using cost data provided by Japanese companies.
The text of the STA itself went on to call for creation of a cost and
price monitoring system for additional products, supplementing the Com¬
merce Department FMVs on DRAMs and EPROMs. This system called
for company- and product-specific cost data to be submitted to MITI,
“in accordance with procedures established by MITI.” If the U.S. gov¬
ernment were to then provide information to Japan supporting a belief
that Japanese firms were selling product in the United States at less than
a company-specific fair value, after a period of consultation the Japanese
government was committed to taking “appropriate actions available un¬
der the laws and regulations in Japan to prevent exports at prices less
than company-specific fair value.” The U.S. government reserved the
right to self-initiate new antidumping actions, and Japan agreed to en¬
courage its companies to provide the data given to MITI to the Com¬
merce Department as well in such an event. Initially this monitoring
system was to cover a certain number of standard chip products,3 but it

1. “Arrangement between the Government of Japan and the Government of the United
States of America Concerning Trade in Semiconductor Products,” September 1986, sec. I,
point 2. (Hereafter STA.)
2. STA, sec. I, point 4.
3. The products to be included were MOS (metal oxide semiconductor, a standard
lower cost chip technology) static random access memory chips (SRAMs), ECL (a high-
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 229

could be extended unilaterally by either government to cover additional

items. A final—and most controversial—provision of this section of the
STA called for the Japanese government to monitor, “as appropriate,”
costs and export prices of these same products exported from Japan to
so-called third-country markets.
At a second level, beyond this “public” STA,*a number of important
“secret” side letters established several particularly sensitive measures.
In a crucial section of the most famous letter, the Japanese government
agreed that the U.S. chip industry’s “expectation that semiconductor
sales in Japan of foreign capital-affiliated companies will grow to at least
slightly above 20% of the Japanese market in five years . . . can be
realized and welcomes its realization.”* * * 4 On the basis of their oral discus¬
sions with Japanese negotiators, U.S. trade negotiators understood this
to be a definite commitment by the Japanese government to a foreign
market share of at least 20 percent by 1991.5 Japanese trade negotiators
later denied the existence of any such side letter.6 Other sections of this
same secret letter decreed that “the Government of Japan will compile
demand and supply forecasts on the Japanese semiconductor market in
compliance with its domestic laws and regulations,” and specifically noted
that Japan’s export trade control ordinance was to be used to enforce the
third-country market provisions of the agreement.7 In another secret side
letter MITI agreed to deal with a last-minute, preagreement surge in

speed, higher cost technology) RAM chips, certain microprocessors, microcontrollers, and
ASICs (application-specific integrated circuits). Quotations in text are from STA, sec. II,
point 2, paras. 4, 5.
4. The precise wording for the 20 percent market share target was reported in Louise
Kehoe, “Stalemate in Chips Trade Talks,” Financial Times, July 28, 1986, p. 1, and Carla
Rapaport, “Japan Yields to US Demands on Microchip Price Monitoring,” Financial
Times, August 4, 1986, p. 1, before the STA was actually signed. Crucial paragraphs from
this “secret” side letter were published verbatim in Mikio Fujiwara, “This Is a Side Letter
to the U.S.-Japan Semiconductor Agreement,” Bungei Shunju (May 1988), pp. 124-37, and
in Inside U.S. Trade, November 18, 1988, pp. 15-16.
5. Author’s interviews with U.S. trade negotiators, 1990, 1992.
6. The top Japanese negotiator described the situation as follows: “A rumor circulated
that the semiconductor agreement included, aside from the published part, a secret side
letter in which Japan promised to raise to 20% the share of American products in the
Japanese semiconductor market within five years. I can state categorically that there was
no such secret letter in the semiconductor agreement.” Makoto Kuroda, “Talking Tough
about Trade,” Journal of Japanese Trade and Industry, no. 6 (1988), p. 31.
7. STA, point 5, sec. 1. The relevant passage (sec. II, point 5) read: “the Government
of Japan will take appropriate actions available under laws and regulations in Japan, in¬
cluding ETC [the Export Trade Control ordinance], in order to prevent dumping.”
230 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

256K DRAM exports by establishing direct quantitative controls over

the volume of these exports.
After the STA went into effect, and for the next five years, there was
a virtually constant stream of meetings and contacts between Japanese
and American government officials, leading to a number of public follow¬
up actions and statements. These included the imposition of sanctions by
the United States in March 1987, their partial removal in June and No¬
vember of and the public disavowal of production controls by Japan, and
in November the disavowal of ex ante review of export pricing for non-
American markets and the dissolution of Japan’s supply-demand forecast
committee in 1989, in response to a negative ruling by the General
Agreement on Tariffs and Trade. These public meetings and actions made
the semiconductor agreement’s institutional structure a living, breathing
creature, whose nature changed substantially over time.
At a third level, beyond any open or secret bilateral agreements be¬
tween governments, was the complex and evolving array of administrative
procedures implemented by both the Japanese government and Japanese
companies in order to further the often vague objectives of the STA. As
described in chapter 4, a variety of sources suggest that government-led
measures included—at various times—controls (nominally, “guidance”
to Japanese companies) on production and investment, allocational
guidelines on exports to major consuming regions, controls on export
pricing enforced through the use of export control procedures to block
or delay objectionable exports, less coercive types of “advice” on ac¬
ceptable pricing levels for monitored products, and export licensing pro¬
cedures designed to shut off uncontrolled chip exports through the so-
called gray market.
Even more controversial is some evidence suggesting that, after 1987,
major Japanese producers may have taken private actions—with some
degree of coordination taking place outside the framework of adminis¬
trative measures or guidance provided directly by the Japanese govern¬
ment—in dealing with the objectives set by the STA, particularly as
foreign pressure forced the Japanese government to take a less directly
interventionist stance. In particular, these actions may have included
some element of coordination and information sharing among Japanese
companies on plans for production and investment in key products (par¬
ticularly in coordinated cutbacks in output levels after the spring of 1989),
and a crackdown on distributors aimed at preventing unauthorized “leak¬
age” of products into the gray market.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 231

Although both these public and private actions were taken within the
political context of the STA, it might reasonably be argued that they were
not part of the STA proper. Implementing actions, after all, were not
formally defined by the two sides within the agreement. All actions either
deliberately implemented (if undertaken or actively encouraged by gov¬
ernment) or passively tolerated (if a private response) by either govern¬
ment, in response to the arrangement's often vague language and sur¬
rounding politics, will be considered here as part of a constantly mutating
STA trade “regime.” An ongoing political dialogue between governments
and companies shaped all these measures. Rather than forming a static
set of policies, the STA was more of a legal and political framework for
a complex, dynamic game involving both public and private players. The
moves in this ill-defined game were a collection of formal and informal
actions—some announced publicly, others undertaken relatively pri¬
vately—evolving constantly over time.

Impacts on Semiconductor Supply and Demand

The natural place to begin examining the impact of this complex of
trade policy measures, and the ensuing maneuvers, on market outcomes
is to sketch out the basic facts of supply and demand, before and after
the agreement. The structure of demand for EPROMs and DRAMs is
considered first, then the basics of supply. In examining demand we must
also briefly digress to consider its structure, particularly the segmentation
of demand by distinct marketing channels. The bulk of the exposition
will deal with DRAMs, whose value (and political visibility) was double
that of EPROMs ($2 billion in DRAMs were sold worldwide by merchant
producers in 1986, compared with $0.9 billion in EPROMs, according to
Dataquest estimates of worldwide merchant revenues). The same basic
description, however, also characterizes institutional arrangements link¬
ing suppliers with consumers in the EPROM market.

The Structure of Demand

DRAMs, along with other memory chips, have a reputation as the

“commodity” product par excellence within the semiconductor industry:
they are high-volume, standardized goods, with almost perfect substitu¬
tion among different manufacturers’ offerings the norm. Certainly indus-
232 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

trywide standardization has been a feature of competition in DRAMs

since the early 1970s.8 Chips from different manufacturers use the same
array of package types and pins and have many common minimal tech¬
nical specifications. They mainly use the same speed classifications (rated
in nanoseconds average access time to a bit). They are generally physi¬
cally compatible, in the sense that products with appropriate specifica¬
tions from different manufacturers may be substituted within a given
piece of equipment. Although DRAMs are in this sense a commodity
product, the actual physical design of the chip’s internal structures and
many subtle aspects of its performance vary by manufacturer.
Because of these subtle but important variations across producers in
DRAM electrical and physical performance parameters, large manufac¬
turers typically put a device through an extensive and expensive qualifi¬
cation process.9 Some retesting is required every time the manufacturing
process for a chip is changed. These costs provide an important economic
incentive for systems manufacturers to limit the number of qualified
suppliers for a particular application. Quality standards maintained by a
manufacturer reduce the need to test components after purchase, and
DRAMs are generally shipped to large customers in boxes with quality
seals to guarantee factory-set standards. (Physical handling of chips is a
major cause of failure or degradation; DRAMs and other high-density
circuit designs are very sensitive to static electricity.) Purchasing chips
from a new supplier, or outside of manufacturer-controlled sales chan¬
nels, will generally require expensive additional testing (unless, as some¬
times happens, the chips can be purchased in boxes with the original
factory quality seals intact).
There are three basic purchasing channels linking the supplier with
U.S. users of integrated circuits. First, large electronic equipment man¬
ufacturers (original equipment manufacturers, or OEMs) who purchase
large volumes of product deal directly with chip manufacturers. Trans-

8. The most prominent exception to this pattern was the 16K DRAM, where two
different and incompatible designs coexisted—one with a single power supply voltage, the
other with dual voltage requirements. Both configurations were produced by multiple
companies, however.
9. Merely qualifying and testing a second source for a part already in use has been
estimated by one industry source to cost $120,000. Qualification costs were large enough
to prompt at least one group of relatively large computer manufacturers to form a coop¬
erative chip qualification joint venture, in order to pool these costs. And within the elec¬
tronics purchasing community, talk of the economic pressure to reduce the number of
suppliers is a staple of everyday conversation.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 233

actions in this market are generally labeled “contract” pricing, though

the nature of these “contracts” is an interesting subject to which I will
shortly return. It is not unknown for OEMs to contract for large purchase
volumes in order to qualify for volume discounts, then resell the excess
over their actual needs to gray market channels.
Second, chip manufacturers maintain a formal distribution network
through “authorized distributors” to service lower volume customers.
Chip manufacturers warrant the product distributed through this chan¬
nel, and they often play an important role in the technical support and
quality assurance programs offered to the customers. Given the historical
fact of continuous yet relatively volatile decline in chip prices, manufac¬
turers have historically offered their authorized U.S. distributors “price
protection,” assuring them that if they lower their sales prices to meet
market competition they will receive a credit reflecting the difference
between the distributor’s purchase price and the lower sales price to the
final consumer. For sales through this channel, the bulk of the risk related
to price uncertainty therefore has been assumed by the chip manufac¬
turer. Pricing generally seems to be on a spot basis, although there can
be a substantial lead time between orders and deliveries in times of
buoyant demand, and distributor prices take on a “contract” aspect.
Finally, there is the so-called gray or spot market. Independent dis¬
tributors, brokers, and speculators buy and sell lots of chips for imme¬
diate delivery. There is also a significant retail market selling directly to
computer resellers and users wishing to upgrade computer systems or
replace defective parts. Supplies of chips on the American gray market
come from chip manufacturers, OEMs, and authorized distributors sell¬
ing their excess inventories, and from Japanese trading companies and
wholesalers purchasing directly from Japanese DRAM manufacturers.10
Gray market products are not warranted by the manufacturer and have
frequently been subjected to unknown handling and quality assurance
procedures. It is not unknown for “pulls” (used chips pulled from sockets
in finished equipment) and even chip manufacturers’ rejects to show up
in misrepresented form in gray market channels. On the other hand,
products are sometimes available in the manufacturers’ original packag¬
ing with quality seals intact, in all respects identical to parts purchased
on contract.

10. See, for example, U.S. International Trade Commission, 64K Dynamic Random
Access Memory Components from Japan, USITC publication 1862 (June 1986), p. A-12.
234 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

In 1985 U.S. industry sources estimated that authorized distributors

accounted for about 30 percent of chip manufacturers’ DRAM sales."
Since gray market sales are often resales of product originally sold
through OEM contracts or authorized distributors and double count
chips flowing into the gray channel from sources other than chip manu¬
facturers, one must be careful in calculating the share of these different
channels in sales to final users. One 1985 estimate held that 20 percent
of “the market” (presumably, end users) is accounted for by the gray
channel in times of shortage. This is roughly in line with more recent
estimates. In early 1989 one industry source estimated that perhaps
15 percent of DRAM sales to final users went through authorized dis¬
tributors, and an additional 15 percent through the gray market.12
An absolutely critical point is that prices typically vary considerably
across these different marketing channels. Price comparisons will there¬
fore be affected in an important way by the precise nature of the sales
transaction.13 Spot prices often rise above contract prices when markets
are tight; they tend to fall below contract price levels when demand is
slack. Figure 5-1, based on data collected by the U.S. International Trade
Commission (ITC), portrays contract, distributor, and spot pricing over
1983 to mid-1985, a period that included both tight (1983 to mid-1984)

11. U.S. International Trade Commission, 64K Dynamic Random Access Memory
Components from Japan, USITC publication 1735 (August 1985), pp. A10-A11.
12. The estimate is that of Don Bell, of Bell Microproducts Inc., whom I thank for
spending the morning of February 16, 1989, attempting to educate me in the intricacies of
the DRAM market.
13. A U.S. International Trade Commission investigation of Korean DRAM sales in
the United States, for example, found that all Korean sales were made on a spot basis, and
found no consistent pattern of Korean products being sold above or below spot sales prices
for U.S.-made products. In contrast, an EC investigation of Korean DRAM exports to the
European Community over roughly the same period found Korean producers’ prices un¬
dercutting those of Community producers by 9.3 to 20.7 percent (weighted-average mar¬
gins). However, the EC report makes no mention of distinctions drawn between contract
and spot sales. In principle, and one suspects in practice, the types of transactions used in
these price comparisons can make a significant difference in the conclusion drawn.
See “Commission Regulation (EEC) no. 2686/92,” Official Journal of the European
Communities, September 17, 1992, p. L 272/19; and U.S. International Trade Commission,
DRAMs of One Megabit and Above from the Republic of Korea, USITC publication 2519,
(June 1992), pp. A-45, A-50. In 100 OEM spot market price comparisons, Korean prices
were below those of the U.S. product in 47 cases, above them in 48, and the two types of
product were priced identically in 5 cases. In 67 comparisons of authorized distributor
purchases, the Korean product was priced lower in 23 cases, above U.S. prices in 42, and
priced identically in 2 cases.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 235

Figure 5-1. Prices of64K DRAMs in Contract, Distributor, and

Spot Markets, 1983-85

Dollars per chip

Source: U.S. Internationa] Trade Commission, 64K Dynamic Random Access Memory Components from Japan,
USITC publication 1735 (August 1985), pp. A40-A41. Prices shown are for DRAMs with an access time of 150
nanoseconds.
a. 10,000 to 100,000 units.
b. Fewer than 10,000 units.

and slackening (late 1984 on) market conditions.14 In four out of six
quarters of relatively robust demand (through the second quarter of
1984), spot prices exceeded contract sales prices to OEMs. In seven out
of nine months of a period of slackening demand (during and after the
last quarter of 1984), spot prices were at or below contract price levels.
As will be seen, spot prices soared much farther above contract price
levels during the DRAM shortage of 1988 than during the tight market
of 1983-84, depicted in this figure.
Further analysis suggests that the quantity commitments that appear
to be the economic heart of OEM “contract” pricing should perhaps best

14. See U.S. International Trade Commission, 64K Dynamic Random Access Memory
Components From Japan (1985), pp. A40-A41. The prices shown are for DRAMs with an
access time of 150 nanoseconds.
236 EFFECTS OF TFIE SEMICONDUCTOR TRADE ARRANGEMENT

be interpreted as a device to minimize additional and significant quali¬

fication and testing costs incurred by chip users, and reduce risks asso¬
ciated with uncertain aggregate demand faced by producers (see ap¬
pendix 5-A). Thus, long-term chip contracts represent the marriage of
quantity commitments to a forward price in force at the beginning of the
contract. Over the remaining life of the contract, however, contract buy¬
ers seem to enjoy something like the “price protection” offered to
distributors.

A Long-Term Perspective on Pricing

For the period up to the mid-1980s there is essentially only one pub¬
lished source of historical price data on DRAMs: Dataquest, an Amer¬
ican market research firm. Dataquest publishes quarterly estimates of
DRAM production, by bit capacity, and an aggregate worldwide “aver¬
age selling price” (ASP) for every capacity chip in large-scale production
at the time of the estimate. The ASP is a “billing” price; that is, it reflects
bills sent out and (one hopes) receipts collected when product is actually
shipped. Because there is often a lag between negotiation of a contract
and actual shipment, current “billing” price is a weighted average of
“booking” prices in both current and earlier periods, when the contracts
were actually negotiated. The Dataquest ASPs are thus probably best
conceptualized as some weighted average (with unknown, variable
weights) of current spot and distributor prices, and current and lagged
contract prices.
There are numerous methodological difficulties in the procedures used
by Dataquest to produce these and other price estimates.15 Nonetheless,
for all practical purposes, these numbers were until the 1990s the only
consistent, long-term source of detailed information on semiconductor
prices and will be used here to analyze pricing trends.
In looking at long-term trends in memory pricing, one frequent prac¬
tice in the industry is to calculate a cost per bit of memory, for example
by dividing total revenues by the total bits of memory produced in all
generations of memory chip at any moment in time. This essentially

15. The producer price index series produced by the U.S. Bureau of Labor Statistics
are not useful in this context. A detailed critique of available statistics may be found in
Kenneth Flamm, “Measurement of DRAM Prices: Technology and Market Structure,” in
Murray F. Foss, Marilyn E. Manser, and Allan H. Young, eds.. Price Measurements and
Their Uses (University of Chicago Press, 1993), pp. 157-97.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 237

weights the price of every type of chip by that chip’s share of total bits
produced. Changes in this price index, unfortunately, confound period-
to-period shifts in demand from one type of chip to another with changes
in chip prices. Even if all prices remained constant, for example, the
aggregate cost per bit might drop significantly if the composition of
demand were to shift toward a chip with a lower price per bit.
A better practice is to use a constant set of weights for prices being
compared in any two periods. One particularly attractive weighting
scheme is the so-called Fisher Ideal price index, which has desirable
properties as an approximation to a theoretically ideal price index.16
Figures 5-2 and 5-3 show annual percentage rates of change for Fisher
Ideal price indices for DRAMs and EPROMs (the products subject to
dumping investigations), constructed using Dataquest worldwide ASP
estimates, along with the widely used (if conceptually less attractive)
calculations of aggregate cost per bit. Although there are some differ¬
ences between the two series, the overall trends are very similar. The
same figures also show annual rates of change in two aggregate measures
of the quantity of chips produced: aggregate bits of memory and an
(again conceptually more attractive) index formed by dividing aggregate
expenditure by a Fisher Ideal price index. Again there are some differ¬
ences, but the overall trends are very similar.
In both DRAMs and EPROMs, annual rates of change in price hit
all-time historical highs in 1987 and 1988, after the STA went into effect.
Indeed, for the first time in the history of memory chip production,
positive annual changes in price were registered in 1988 for both types

16. The Fisher Ideal index I am constructing is the geometric mean of two price indexes
calculated using a single weight on the price of every type of chip in two adjacent time
periods (the weights are constant in the two time periods); this approach does not confuse
the effect of price changes with the shift in usage from one generation of chip to another.
The weights used for the first index are the expenditure shares of chips in the starting
period, while the second index uses expenditure shares for each type of chip in the ending
period: the Fisher Ideal price index giving the price in period 1 relative to period 0 is

/y/ /?"<?!’ \ P'i / y( Pig' \ p"

V i\Dp?*?/ py
i
\pr
i

Diewert has shown that this weighting scheme produces a “superlative index”—a second-
order approximation to a true, exact price comparison between two periods derived
from microeconomic theory. See W. E. Dewert, “Superlative Index Numbers and Con¬
sistency in Aggregation,” Econometrica, vol. 46 (July 1978), pp. 883-900. Indices for
adjacent pairs of time periods have been chained together to produce a single, contin¬
uous price index.
238 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-2. Year-to-Year Changes in DRAM Prices and Quantities

Produced, 1972-89

A. Price changes

Percent

B. Quantity changes
Percent

Source: Author’s calculations based on unpublished Dataquest data.

EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 239

Figure 5-3. Year-to-Year Changes in EPROM Prices and Quantities

Produced, 1972-89

A. Price changes
Percent

B. Quantity changes
Percent

Source: Author’s calculations based on unpublished Dataquest data.

240 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

of chips. The contrast with the historical record was particularly striking
in DRAMs, where previously recorded annual rates of change had stayed
within roughly the —20 to —60 percent range. Prices fell by only 10
percent in 1987, then suddenly jumped up by well over 20 percent in 1988.
The deviation from historical trend was so egregious, in fact, that
some argued that a secular upward shift in the price of DRAMs was
occurring. Within the industry this came to be known as the proposition
that future memory pricing would follow the “bi rule” rather than the
“tt rule.”17 The tt rule refers to the fact that, in the past, DRAM prices
for each generation of chip had tended to decline asymptotically toward
the $3 level (tt = 3.14159 . . .) as mass production of that generation
peaked, then to half that level at the end of its life. Since a new generation
of chip was introduced on average about every three years, and since
each new generation of chip quadrupled the number of bits on a chip,
this amounted to a 75 percent reduction in the cost of a bit of memory
every three years, or an annual rate of decline of about 36 percent per
year. (Remarkably, this is roughly the annual decline in memory bit cost
produced by analyses of actual historical data. See figure 5-4 for data on
DRAM prices through 1985.) The bi rule (“bi” is also, coincidentally,
the pronunciation of the Japanese kanji character meaning “doubling”)
suggests that in the future every new-generation chip will approximately
double in price as mass production peaks. (See figure 5-5 for DRAM
pricing after 1985.) Following the previous logic, this means a 50 percent
decline in bit cost every three years, for an annual rate of decline of
about 20 percent, or about a 50 percent smaller long-run cost decline
than under the tt rule.
The cyclical sensitivity of semiconductor demand is also apparent in
the large swings in price change depicted in figures 5-2 and 5-3.18 Al¬
though 1985—the year when trade friction kicked the political process
leading to the STA into motion—had set a historical record for plunges

17. See M. P. Lepselter and S. M. Sze, “DRAM Pricing Trends—The tt Rule,” IEEE
Circuits and Devices Magazine, vol. 1 (January 1985), pp. 53—54; Yasuo Tarui, “From the
it Rule to the Bi Rule,” Nikkei Microdevices, no. 27 (July 1987), pp. 165-67; and Yasuo
Tarui and Tadaaki Tarui, “New DRAM Pricing Trends: The Bi Rule,” IEEE Circuits and
Devices Magazine, vol. 7 (March 1991), pp. 44-45.
18. Cyclical impacts on semiconductor industry employment are discussed in detail in
Kenneth Flamm, “Internationalization in the Semiconductor Industry,” in Joseph Grunwald
and Kenneth Flamm, The Global Factory: Foreign Assembly in International Trade (Brook¬
ings, 1985), pp. 101-03.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 241

Figure 5-4. Prices for Four Generations of DRAMs, Pi Rule, 1974-89

Dollars per chip

Source: Unpublished Dataquest data.

in DRAM prices, the plunge was only marginally steeper than that seen
in 1981. In EPROMs, on the other hand, the price decline of 1985 actually
fell short of the drop registered in 1981 and was about the same as that
seen in 1978. Rates of increase in chip output also fell close to historical
lows in the years after 1986, in both DRAMs and EPROMs.19

19. At least one study, commissioned by the Semiconductor Industry Association, the
U.S. industry group, argues that the movement of prices in DRAMs after the STA went
into effect had returned price levels to their long-term trend by 1989, and therefore that
the STA had no apparent deleterious effects in raising prices above long-term trends. See
Technecon Analytic Research, Inc., “Impact of the Semiconductor Agreement on DRAM
Prices,” Washington, December 1990. The logic of this argument, however, suffers from
three significant flaws.
First, all forecasts of trend price levels are very sensitive to the choice of base period
used to estimate the trend. By choosing a particular base period, one can raise or lower
trend growth rates arbitrarily. By excluding the period of sharp recession in 1985 from
242 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-5. Prices for Three Generations of DRAMs, Bi Rule, 1987—95

Dollars per chip

Sources: Author’s calculations based on data from Dataquest, “DO First Monday Report”; Electronic Business
Buyer; and Nikkei Telecomm (on-line data service), “Large User Contract Price.”
a. Electronic Business Buyer.
b. Dataquest, “DQ First Monday Report.”

Regional Semiconductor Price Differentials

A significant factor complicating discussion of DRAM pricing was the

appearance of significant regional price differentials in 1987 and 1988,
after the signing of the STA. In response to the STA the Japanese gov-

the period used to estimate the trend, the rate of decline was reduced in this study.
Second, the “correct” trend in this study is estimated by taking price as a function of
cumulative bits of memory produced. Even if this relationship held exactly true to historical
trend after the STA went into effect, and price was exactly as predicted by some historical
function of the cumulative bits of memory produced, producer behavior in response to the
STA would still have raised prices in an “ahistorical” fashion by cutting back on the number
of bits produced. That is, looking at a graph of price versus cumulative bits would lead one
to conclude that nothing representing a break from historical, structural relationships had
occurred—but if producers had indeed cut back substantially on production (and cumula¬
tive bits produced) relative to past trends, prices would indeed have risen extraordinarily.
In other words, even if some structural relation between price and cumulative bits were
unaffected by the events set in motion by the STA, the growth of bits produced over time
certainly was, and this latter channel would have been expected to be the primary effect of
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 243

ernment began to set floor prices for export sales of DRAMs by Japanese
companies. Initially, different standards were set for sales to the U.S.
market and other (third-country) export markets; after U.S. protests,
however, the systems were later unified. (In response to European pro¬
tests, the pricing guidelines were separated again in 1989.)
Regulation of Japanese export sales ultimately involved four elements.
First, an export licensing system was adopted. This system required de
facto government approval of the export price, which was to be set above
minimum norms established by Japan’s Ministry of International Trade
and Industry (MITI). Second, foreign purchasers of Japanese chips were
required to register with MITI. Third, all export transactions required a
certificate from the original chip manufacturer attesting to the fact that
the chips in question were actually manufactured by that producer.
Fourth, MITI established informal regional allocation guidelines, to en¬
sure that supplies were not diverted from one export market to favor
another.
By most accounts, MITI’s guidance was quite effective in setting min¬
imum pricing standards for Japanese DRAM manufacturers’ direct ex¬
port sales. (Because Japanese manufacturers were at this time responsi¬
ble for between 80 and 90 percent of world DRAM sales, this effectively
worked as a floor on price in the global market.) The intent of the second
and third elements was clearly to reduce access by foreign purchasers to
Japanese gray market channels not under the direct supervision and con¬
trol of Japanese chip manufacturers. Predictably, prices in the unregu¬
lated domestic Japanese market soon dropped below foreign export
prices.
The rising differential between U.S. and Japanese DRAM prices was
exacerbated by producer actions that, from the standpoint of U.S. con¬
sumers, made the true price of their DRAMs even greater than the hefty
contract prices reported to market researchers. The practice of tying sales
of DRAMs to consumer purchases of other chips that would not other¬
wise have been bought from Japanese DRAM vendors (particularly ap-

restrictions on supply associated with the STA. The Technecon Analytic Research study
actually notes that the growth in bits produced (and consumed) in the late 1980s was well
below historical trends.
Third, the assumption that some fixed structural relationship exists between price and
cumulative bit production is dubious. As will be seen in chapter 6, although yields, and
therefore marginal costs, may have a relatively simple relationship to cumulative output,
there is no reason to expect prices to follow marginal costs very closely, particularly over
periods when production is constrained by available capacity.
244 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

plication-specific integrated circuits, or ASICs) appears to have surfaced

in 1988 and 1989 and was the source of widespread complaints in the
computer industry.20 Indeed, when the U.S. Memories joint venture was
being organized in 1989 (see chapter 4), Wilfred Corrigan, CEO of ASIC
producer LSI Logic, made it clear that a major reason he agreed to join
the DRAM consortium and chair its board of directors was to acquire
some leverage against the widespread use of these tying arrangements
and free his ASIC customers from the need to shift to Japanese ASIC
suppliers in order to maintain access to needed supplies of DRAMs.21
Functionally, the cost premium faced by U.S. customers was even larger
than the monetary differential alone would suggest.
In apparent response to the appearance of regional price differentials
after the STA was negotiated, Dataquest began reporting regional con¬
tract pricing for a sample of twenty-five semiconductor components, on
a biweekly basis, beginning in early 1987. These data (the Dataquest
“First Monday Report”) are based on a survey of six to ten respondents,
primarily chip manufacturers, in each of six geographic regions.22 Japa¬
nese producers do not report a contract price, and the Japanese price
cited by Dataquest apparently refers to “large volume wholesale” prices.
Figure 5-6 compares the Dataquest “First Monday Report” volume
contract prices for a 120-nanosecond, 256K DRAM in five regions, av¬
eraged over the course of a quarter, with the Dataquest ASPs for all
256K DRAMs through the last quarter of 1988. The increase in regional
pricing differentials in 1987 and 1988 is readily apparent. The extraordi¬
narily sharp rise in European 256K DRAM prices relative to other re¬
gions is particularly notable, and a little puzzling. (Analysis detailed
elsewhere suggests that this sharp rise may be fictitious.)23
20. One well-known U.S. computer vendor’s head of purchasing described in great
detail how they had been forced to order ASICs in order to secure shipments of DRAMs
in 1988. Author’s interview with U.S. computer systems vendor, 1989.
21. Jack Robertson, “Confirm Major IBM Role in Creation of U.S. Memories,” Elec¬
tronic News, June 26, 1989, pp. 1, 58.
22. For DRAMs the survey asks for the current contract price negotiated for three
different volumes: 1,000, 10,000, and “volume” (over 100,000). In some published Data¬
quest reports, however, it is stated that “volume” prices mean greater than 20,000 parts,
not 100,000; the higher definition of “volume” for DRAMs is apparently an exception to
this rule. If producers have not concluded any contracts for a particular volume, they are
asked to estimate the price that would have been negotiated on a contract of that size. I
thank Mark Giudice of Dataquest for this description of the “DQ First Monday Report”
survey in various conversations during 1989 and 1990.
23. See appendix 5-A and Flamm, “Measurement of DRAM Prices,” for evidence on
this point.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 245

Figure 5-6. Quarterly Dataquest Monday Contract Prices and Average

Selling Prices for 256K DRAMs, 1987-92a

Dollars per chip

Figure 5-7 gives the equivalent time series for 1M DRAMs.24 All
available data seem to show unprecedented, sustained increases in
DRAM prices from 1987 on. (Compare this with 64K DRAM pricing
during the shortage of 1983-84, for example, in figure 5-1.)
Weekly data on large-user wholesale prices in the Japanese market are
also regularly published in Nihon Keizai Shimbun (commonly called Nik¬
kei, a Japanese business daily roughly equivalent in terms of reputation
and stature to the American Wall Street Journal or the London Financial
Times).25 Various contract (or large-user) pricing data for both U.S. and

24. The “DQ First Monday Report” contract prices refer to 120-nanosecond, 1M parts,
in plastic cases, which were the cheapest available 1M part. It is surprising, then, that the
Dataquest ASPs—which presumably average these with more expensive parts and with
higher priced parts sold through distributors and in the spot market—for at least three
quarters, during a period of relatively stable prices, are at or below an average of volume
contract prices. This casts some doubt on the reliability of one or the other of these data
sources.
25. These data are reprinted on a weekly basis in the Japan Economic Journal, the
English-language weekly based on Nikkei. When possible the former is used here, supple¬
mented by online data available from Nikkei’s computerized data base, and selected issues
of Nikkei itself. Data sampled on a weekly basis were averaged to calculate a monthly time
series; monthly series were averaged to create a quarterly series. Monthly (quarterly)
246 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-7. Quarterly Dataquest Monday Contract Prices and Average

Selling Prices for 1M DRAMs, 1987-92a

Dollars per chip

Sources: See figure 5-6.

Japanese sales are shown in figures 5-8 and 5-9. Figure 5-8 contrasts the
Nikkei large-user Japanese wholesale price for 256K DRAMs with “DQ
First Monday Report” 256K DRAM contract prices. It is clear that Da-
taquest’s Japanese volume contract price basically follows the Nikkei
wholesale price reasonably closely (albeit with occasional bursts of
noise), and the two are probably different estimates of the same under¬
lying concept. The Dataquest estimates for Japan should probably be
regarded as less reliable than the Nikkei numbers, because they appear
to be subject to frequent retroactive revisions.26

estimated yen per chip was converted to dollars using market average monthly (quarterly)
exchange rates reported by the International Monetary Fund.
26. Indeed, no two sources for the Dataquest price series appear to give exactly the
same numbers, particularly for prices of product sold outside the United States. For this
reason, one may speculate that revisions to the exchange rates used to convert prices to
dollars (the currency in which Dataquest reports all contract prices) may play an important
role in these retrospective changes.
The most dramatic example of retrospective revision of the Dataquest regional contract
price estimates involved the U.S. government’s use of these data in its submission to the
GATT panel investigation of the STA, in order to challenge price data furnished by Euro¬
pean companies. The European data, however, were apparently also supplied by Dataquest!
The data submitted by the U.S. government, in turn, differed from data furnished to me
by Dataquest in February 1989.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 247

Figure 5-8. Monthly Contract Prices for 256K DRAMs in the United
States and Japan, 1985-92
Dollars per chip

Sources: Dataquest “DO First Monday Report”; Nihon Keizei Shimbun (various issues); and Nikkei Telecomm
(on-line service).

Figure 5-8 seems to show quite clearly that a large and significant differ¬
ential between the U.S. and Japanese markets existed in 1988 and most of
1989 in large-volume, direct “contract” sales between Japanese manufac¬
turers and their customers in the two regions. Figure 5-9 compares the
Nikkei time series on wholesale 1M DRAM prices with Dataquest contract
pricing data in the United States and Japan. Figure 5-10 compares a Nikkei
time series on wholesale 64K DRAM prices with Dataquest quarterly world¬
wide average sales prices. (U.S. contract price estimates were unavailable
for 64K parts.) Both seem to tell basically the same story of significant
regional differentials from late 1987 through late 1989. Contemporary ac¬
counts in the Japanese business press also report large price differentials
between Japanese and foreign markets over this period.27

27. For example, on April 9, 1988, the Japan Economic Journal reported that Korean
chip exports to Japan had dropped steeply, because Korean producers had found prices
higher and profitability greater in the U.S. market. Noted the Journal in “Korean Chip
248 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-9. Monthly Contract Prices for 1M DRAMs in the United

States and Japan, 1987-92

Dollars per chip

Sources: See figure 5-8.

The data displayed thus far have largely ignored DRAMs sold by
distributors and in the spot market. This misses an important dimension
of the change in market conditions after the signing of the STA. To

Producers Shun Japan to Harvest Higher Profit in U.S.” (April 9,1988), p. 4, “The reasons
for the shift are obvious. With Japanese semiconductor exports to the U.S. down sharply
as a result of increased trade friction between the two nations, prices on the underfed U.S.
market have begun to skyrocket. ... At the same time, though supplies in Japan were also
dwindling, Japanese companies were reluctant to accept price hikes beyond the typical
¥340 cost for large users. Not surprisingly, Korean makers moved quickly to enter the
more lucrative U.S. markets.”
On July 18, 1988, Nihon Keizai Shimbun's morning edition carried a front-page story
reporting that the price differential between the Japanese and foreign market for a 256K
DRAM had widened from 100 yen, as trade friction heated up, to 200 to 300 yen. (An
English-language summary of the story was carried in “Foreign Semiconductor Prices Even
Higher,” Japan Economic Journal, July 30, 1988, p. 10. Depreciation of the dollar in the
intervening period makes the dollar value of that change in the differential substantially
greater.) On August 9, 1988, Nihon Keizai's morning edition carried a story on page 18
reporting that Japanese semiconductor producers had begun stepping up exports, with
much higher price tags than these products carried in the domestic market. (An English-
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 249

Figure 5-10. Quarterly Japanese and World Contract Prices for 64K
DRAMs, 1985-89
Dollars per chip

Sources: Author's calculations from unpublished Dataquest data; Dataquest, "DQ First Monday Report”; and
Nihon Keizai Shimbun (various issues).

remedy this situation I have constructed time series showing retail spot
prices for memory chips, beginning in the spring of 1985. To do so, I
collected weekly data on sales prices from one of the larger retail vendors
of memory chips in the United States during this period.28 The advertised
prices are dated (an important point, since there is typically a substantial

language summary appears in “Japan Chip Makers Step Up Exports Amid Shortages,”
Japan Economic Journal, August 20, 1988, p. 15. The story also notes that observers feared
this might exacerbate shortages in the domestic market.)
28. The vendor was Microprocessors Unlimited, of Beggs, Oklahoma; the weekly ad¬
vertisements were found in the pages of PC Week and Infoworld magazines. The prices
shown refer to the following parts: the 64K DRAM is a 64K x 1, 150-nanosecond, DIP-
packaged chip; the 256K DRAM is a 256K x 1, 120-nanosecond DIP package; the 1M
DRAM is a 1M x 1, 100-nanosecond DIP package through October 1989, and an 80-
nanosecond chip after that month. For comparison purposes, a 100-nanosecond 1M DRAM
sold for $10.99 in the first week of November 1989, whereas an 80-nanosecond part sold for
$10.99 the following week.
250 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-11. Monthly Spot Prices for 256K DRAMs in the United States
and Japan, 1985-91

Dollars per chip

Sources: Author’s calculations based on published advertisements of Microprocessors Unlimited; Nihon Keizai
Shimbun (various issues); and Nikkei Telecomm (on-line service).

lag between the submission of advertising copy and its publication); con¬
tacts with this vendor have also made it clear to me that these were real
prices, that is, product was actually available in stock at these prices.
Figure 5-11 shows spot retail prices for 120-nanosecond, 256K DRAMs.
The contrast with figure 5-8 is striking: spot retail prices quadrupled
between early 1987 and early 1988!
Some data may also be gathered on spot pricing in Japan. Beginning
in October 1988, spot price data for 256K DRAMs, of unspecified speed,
have been published regularly in Nihon Keizai Shimbun. Weekly prices
have been averaged over the month;29 some occasional reports on spot
prices have also been collected from news articles in Nikkei and are also

29. Prices from the Tuesday or Wednesday morning editions of Nikkei were used,
beginning in October 1988. Data from before that date are taken from an assortment of
occasional Nikkei news reports on DRAM market conditions. Monthly exchange rate data
published by the International Monetary Fund were used to convert yen prices to U.S.
dollars.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 251

Figure 5-12. Monthly Spot Prices for 1M DRAMs in the United States
and Japan, 1986-92

1986 1987 1988 1989 1990 1991 1992

Sources: See figure 5-11.

shown in figure 5-11. The fragmentary data shown here suggest that
Japanese spot prices generally followed the upward trajectory of Amer¬
ican spot prices, but with a lag, so that some differential persisted through
mid-1988. In the fall of 1988 Japanese spot prices seem to have briefly
passed American spot prices, only to fall to approximate parity with U.S.
prices in 1989. These data suggest that MITI’s attempts to wall off the
Japanese gray market from the foreign gray market only slowed conver¬
gence and did not prevent arbitrage from linking prices in the two mar¬
kets. This is not completely surprising, since the gray market is virtually
by definition a bastion of untamed entrepreneurialism.
Figure 5-12 shows an equivalent retail, spot price series I have con¬
structed for 1M DRAMs (with 100-nanosecond access times), along with
data on Japanese 1M DRAM spot prices collected from Nikkei. The
pattern of regional differentials shown by these more fragmentary data
is similar to that for 256K DRAMs. U.S. buyers paid some premium
relative to Japanese buyers through mid-1988; Japanese spot prices may
252 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

have briefly shot past U.S. levels at the end of the year, and 1989 brought
with it rough parity. The increase in price from mid-1987 to late 1988 is
notable, but far less striking than that for 256K DRAMs.
In DRAMs, then, the overall pattern was one of a significant regional
price differential in contract pricing between the domestic Japanese and
overseas markets from 1987 through 1988. The differential disappeared
in 1989, as semiconductor demand slackened and prices began to fall
sharply. In the spot market, in contrast, there was some transitory dif¬
ferential in pricing between the Japanese and American markets, but
after some short lag prices essentially tended to converge in the two
regions. What this suggests is that government price and export controls
were relatively effective in controlling direct transactions between the
major Japanese producers and their large customers, but considerably
less effective in regulating transactions in the secondary spot market,
where arbitrage between the American and Japanese markets approxi¬
mately equalized prices in the two regions.30
Overall regional contract price indexes for DRAMs may be con¬
structed using estimates of contract prices in the United States and Japan
for individual products. A Fisher Ideal price index for the United States
was constructed by taking quarterly Dataquest contract price estimates
in the United States for 256K and 1M DRAMs, and Dataquest ASP
estimates for 64K DRAMs, and using expenditure shares constructed
using relative volumes of worldwide production of these products at U.S.
prices to construct weights for period-to-period comparisons. A chained
index built up from quarter-to-quarter Fisher Ideal comparisons was then
calculated for the United States.31 A similar index was constructed for
Japan but used Nikkei estimates of large-user wholesale prices, and rel¬
ative global volumes, to produce quarterly Japanese expenditure share
weights.32 Since comparisons of contract prices show U.S. and Japanese

30. Export controls were notoriously porous to small-scale evasions. Chips are compact
and light, and a million-yen shipment could fit into a small box or large attache case.
Anecdotes of such irregular traffic abounded in 1987 and 1988.
31. For 256K DRAMs quarterly averages of Dataquest “DQ First Monday Report”
price estimates for volume contracts were used throughout this period. For 1M DRAMs
Dataquest worldwide ASP estimates were used until the first quarter of 1987, when the
Monday contract price series starts; thereafter the latter series was used. For 64K DRAMs
Dataquest worldwide ASP estimates were used throughout, since Monday contract price
data were unavailable for this product.
32. Quarterly averages of Nikkei weekly price quotes for 256K DRAMs were used
throughout. For 64K DRAMs average Nikkei large-user price quotes were used through
the second quarter of 1989, after which Dataquest worldwide ASPs were used. For 1M
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 253
Figure 5-13. Quarterly Fisher Ideal Price Indexes for Large-User
Purchases of DRAMs in the United States and Japan, 1986-89
1986:4 = 100

1987:1 1988:1 1989:1

Sources: Author’s calculations based on unpublished Dataquest data; Nihon Keizai Shimbun (various issues); and
Nikkei Telecomm (on-line service).

prices for all sizes of DRAMs at rough parity in the fourth quarter of
1986, the DRAM price indexes for the United States and Japan were
calculated taking that quarter equal to 100.
Figure 5-13 shows the relative movement in the two regions of an
aggregate price index for large-user purchases of DRAMs from 1986
through 1989. The aggregate differential between the United States and
Japan looks very much like the pattern of differentials observed in each
of the individual varieties of DRAM, with U.S. prices rising well above
Japanese levels in 1987 and 1988, peaking at a level almost 50 percent
higher than in Japan in the second quarter of 1988, then falling to rough
parity by the end of 1989.

DRAMs Dataquest worldwide ASPs were used through the first quarter of 1987, then
replaced with Nikkei average large-user wholesale prices beyond that date.
254 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Fewer data are available on the market for EPROMs (in particular,
data on Japanese spot pricing are infrequently reported in the business
press). Nonetheless, it is clear that a considerably different situation
prevailed. Whereas the overall trend toward rising prices in 1987 and
1988 is similar to that in DRAMs, significant regional differentials in
contract pricing did not appear.
Figure 5-14 compares quarterly averages of Nikkei large-user whole¬
sale prices with Dataquest volume contract prices, where comparative
data are available, for 128K and 256K EPROMs. (As in DRAMs, the
Nikkei numbers probably should be viewed as more reliable than the
Dataquest estimates for the Japanese market.) No persistent price dif¬
ferential appears to have emerged; Japanese EPROM prices generally
tracked American EPROM prices, though often with a lag of roughly a
quarter. The data clearly suggest that Japanese EPROM prices basically
followed American prices downward after late 1988. Contemporary Jap¬
anese press accounts report that prices for imported foreign EPROMs
were pulling domestic price levels down over this period.33

The Supply of Memory Chips

An explanation of why regional contract price differentials appeared

in DRAMs, but not in EPROMs, must consider supply factors. The
industrial organization of the supplier industry for DRAMs and
EPROMs is depicted in tables 5-1 and 5-2, which calculate the Hirsch-
man-Herfindahl index (E1HI) of concentration, and the share of global
supply coming from firms headquartered in different countries, over
time.34 The HHIs calculated here will range in value from one (most

33. “The drop [in EPROM prices] was attributed mainly to price competition among
foreign-affiliated makers hoping to expand their market share in Japan.” “Prices of EPROM
Chips Decline,” Nihon Keizai Shimbun, April 19, 1989.
“The price of EPROMs was pushed down by competition from foreign manufacturers.”
“Semiconductor Prices on the Decline,” Nikkei Top Articles, April 24, 1989.
“In September . . . 1M chips dipped to 1,050-1,100 [yen], . . . Some U.S.-produced
EPROMs are now selling for around 1,000 yen, according to an official at a semiconductor
trading house.” “EPROM Prices Drop for the First Time in Six Months,” Nikkei Top
Articles, October 2, 1990.
“Imported EPROMs are also having an impact on prices . . . excess American output
is entering the Japanese market, said a sales agent.” “1M EPROM Prices Plummet below
900 Yen,” Nikkei Top Articles, November 2, 1990.
34. Because Dataquest identifies chip suppliers by the label on the chip, the reported
U.S. share of the DRAM market and the calculated HHI credit chips manufactured by
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 255

Figure 5-14. Quarterly Wholesale Prices for 128K and 256K EPROMs
in the United States and Japan, Selected Years, 1985-92

A. 128K EPROMs
Dollars per chip

B. 256K EPROMs
Dollars per chip

Sources: Dataquest, “DQ First Monday Report”; Nihon Keizei Shimbun (various issues); and Nikkei Telecomm
(on-line service).
256 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Table 5-1. Market Concentration in DRAMs, 1979-89

Percentage of global production

Hirschman- Top five Other

Herfindahl Japanese Japanese U.S. European Korean
Year index producersa producers producers producers producers

64K DRAMs
1979 0.525 67 0 33 0 0
1980 0.264 59 0 41 0 0
1981 0.178 67 6 28 0 0
1982 0.129 60 6 33 0 0
1983 0.108 53 7 38 2 0
1984 0.092 48 10 38 4 0
1985 0.091 54 7 31 4 4
1986 0.099 56 11 19 5 10
1987 0.106 44 13 20 6 17
1988 0.170 19 22 21 1 37
1989 0.273 4 17 34 0 45

256K DRAMs
1982 1.000 100 0 0 0 0
1983 0.265 92 1 7 0 0
1984 0.213 89 3 9 0 0
1985 0.165 82 3 15 0 0
1986 0.135 77 5 15 1 2
1987 0.102 55 11 24 2 9
1988 0.091 46 18 28 2 7
1989 0.078 43 19 25 3 10
1M DRAMs
1985 0.964 99 0 1 0 0
1986 0.369 87 0 13 0 0
1987 0.347 93 6 1 0 0
1988 0.173 78 10 6 3 2
1989 0.135 54 16 14 5 11
1990 0.110 45 15 21 5 15
Source: Author's calculations based on unpublished Dataquest data.
a. Fujitsu, Hitachi, Mitsubishi Electric, NEC, and Toshiba.

concentrated, with all output produced by a single firm) to zero (ap¬

proached with perfect competition).
In both EPROMs and DRAMs, levels of concentration (generally at
or above the .1 level) were considerably greater than those measured in
the semiconductor market as a whole. In 1987, for example, the HHI for
the fifty largest U.S. semiconductor-producing companies was .0539,

Korea’s Samsung but marketed by Intel under its label to Intel, and Toshiba-made DRAMS
sold by Motorola to Motorola.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 257

Table 5-2. Market Concentration in EPROMs, 1980-89

Percentage of global production

Hirschman- Top five Other

Herfindahl Japanese Japanese u.s. European Korean
Year index producersa producers producers producers producers

64K EPROMs
1980 0.625 0 0 100 0 0
1981 0.389 11 0 89 0 0
1982 0.273 50 0 50 0 0
1983 0.157 60 1 40 0 0
1984 0.143 65 2 31 1 0
1985 0.136 65 2 30 4 0
1986 0.138 64 0 29 6 0
1987 0.089 44 4 38 13 0
1988 0.114 14 3 54 26 3
1989 0.131 7 3 59 29 2

128K EPROMs
1982 0.916 0 0 100 0 0
1983 0.405 22 0 78 0 0
1984 0.165 61 0 39 0 0
1985 0.154 77 0 23 0 0
1986 0.155 65 0 34 0 0
1987 0.112 50 1 43 6 0
1988 0.119 29 2 52 17 0
1989 0.153 17 1 58 24 0

256K EPROMs
1983 0.536 13 0 88 0 0
1984 0.394 18 0 82 0 0
1985 0.202 46 0 54 0 0
1986 0.155 59 0 41 0 0
1987 0.107 47 3 44 5 0
1988 0.097 36 2 50 12 0
1989 0.093 27 1 55 18 0

512K EPROMs
1984 0.500 0 0 100 0 0
1985 0.479 6 0 94 0 0
1986 0.257 32 0 68 0 0
1987 0.138 57 6 36 1 0
1988 0.105 46 3 42 8 0
1989 0.111 26 2 57 14 0

1M EPROMs
1987 0.253 80 0 20 0 0
1988 0.161 82 0 18 0 0
1989 0.133 63 0 34 3 0

2M EPROMs
1988 1.000 100 0 0 0 0
1989 0.882 96 0 4 0 0

4M EPROMs
1989 0.290 80 0 20 0 0

Source: Author’s calculations based on unpublished Dataquest data,

a. Fujitsu, Hitachi, Mitsubishi Electric, NEC, and Toshiba.
258 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-15. Hirschman-Herfindahl Indexes and Cumulative Output for

Five Generations of DRAMs

Index

Cumulative industry output (million of units)

Source: Author’s calculations based on unpublished quarterly Dataquest data.

down from .0597 in 1982.35 What this probably indicates is that semicon¬
ductor firms tend to specialize in particular types of products. It is also
evident that HHIs typically vary enormously over the product life cycle,
with values of one when the very first firm introduces a new-generation
chip, dropping over time as others enter the business, then rising again
at the end of the life cycle as firms phase out their production of the older
product and shift to newer items.
Thus, consideration of whether or not excessive monopoly power ex¬
ists, or whether the industry is more concentrated than the historical
norm, cannot be independent of the stage in the product life cycle. One
convenient measure of how advanced in the product cycle a generation
of product might be is cumulative output over time. Figures 5-15 and
5-16 show the evolution of HHIs over time for various generations of

35. See U.S. Bureau df the Census, 1987 Census of Manufactures: Concentration Ratios
in Manufacturing (U.S. Commerce Department, 1987), pp. 6-38.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 259

Figure 5-16. Hirschman-Herfindahl Indexes and Cumulative Output for

Six Generations of EPROMs

Index

Source: Author’s calculations based on unpublished quarterly Dataquest data.

DRAMs and EPROMs, calculated using annual data.36 Historically,

HHIs in both of these products have tended toward the neighborhood
of .1 as mass production peaked. Normalizing for position in the product
life cycle using cumulative output, across generations of chip, there is
little obvious evidence of a secular trend toward significantly increased
concentration in EPROMs. One might observe that, during the early
years of 256K and 1M DRAM production, the industry was somewhat
more concentrated than the historical norm, but by 1989 concentration
in both these products looked similar to earlier historical levels.
Although variation in producer concentration may not be much of a
potential factor in explaining differences in DRAM and EPROM market
conditions over 1986-89, geographic concentration offers greater prom-

36. These calculations use Dataquest estimates of annual production by all merchant
companies. Production by exclusively captive producers (with no sales on the open mer¬
chant market) such as IBM is excluded.
260 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

ise. For the three most recent generations of DRAM, Japanese market
shares were 67 percent (in 64K parts), 82 percent (256K), and 99 percent
(1M) in 1986, when the STA was signed. In EPROMs, overall, Japanese
company shares were considerably lower: 65 percent (128K), 59 percent
(256K), and 32 percent (512K). Thus, proportionate reductions in pro¬
duction or exports by Japanese producers would have potentially greater
relative impacts on aggregate supply in DRAMs than in EPROMs.
Furthermore, very different adjustments to Japanese production of
DRAMs and EPROMs are revealed in estimates of Japanese chip output
after the STA was implemented. Some of the best and most detailed
semiconductor statistics available cover DRAM and EPROM production
in Japan. As a consequence of the STA and the operation of the MITI
supply-demand forecast committee, Japanese companies reported a large
amount of data to MITI, from which quarterly estimates of Japanese
production, consumption, and net exports were compiled through late
1988. After the dissolution of this committee in 1989, MITI continued to
compile these estimates on a semester basis.37
As in DRAMs, MITI was reportedly issuing guidance to Japanese
producers to reduce EPROM production in 19 87.38 As will be seen
shortly, moreover, foreign producers rapidly expanded EPROM sales in
the Japanese market, and by 1989 Japan’s share of world output for many
types of EPROMs had fallen sharply and Japan had become a significant
net importer of EPROMs. This contrasted with the situation in DRAMs,
where Japan remained a large net exporter and continued to dominate
world supply.
In looking at production statistics, one should note that output in a
fully utilized chip fabrication facility is not fixed over time, but instead
rises on an S-shaped profile (figure 5-17). This occurs for two basic
reasons. First, as fabrication technology gradually improves, smaller fea¬
ture sizes become feasible. As the chip size is shrunk, more chips can be
fit onto the surface of the large silicon wafers on which the processing
actually takes place. Three or more such “die shrinks” may typically

37. In addition to including the worldwide production of Japanese companies, these

statistics also include the Japanese chip production of U.S.-owned affiliates, including those
of IBM Japan, TI Japan, and Tohoku Semiconductor (a Japanese manufacturing operation
jointly owned by Toshiba and Motorola).
38. See, for example, “MITI Authorizes Higher Chip Output for 3rd Quarter,” Japan
Times, June 25, 1987; “Semiconductor Price Negotiations Face Rough Going,” Nikkei Top
Articles, August 15, 1987; and “MITI Is to Start Regulating SRAM Prices,” Electronic
Times, September 17, 1987, p. 6.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 261

Figure 5-17. Yield Plot for IBM’s 64K DRAM

Relative final test yields

Yield
Manufacturing line
improvement
final test yield
due
including half-good
to the
product and redundancy
redundant
Equivalent yield circuits
improvement due to
half-good chips
(without redundancy)

Planned yields

Pilot line yields

Years

Source: C. H. Stapper and others, “Evolution and Accomplishments of VLSI Yield Management at IBM,” IBM
Journal of Research and Development, vol. 26 (September 1982), pp. 532-45.

occur over the life cycle of the product within a single company.39 Second,
for a given die size, better control over the manufacturing process asso¬
ciated with experience and learning-by-doing creates improved yields—
the number of good chips that can be extracted from each batch of
processed silicon. Learning economies—predictable declines in unit cost

39. A desirable effect of incrementally smaller chips is gradually improved speed. Thus
the access time of the standard 256K DRAM produced by most manufacturers went from
150 to 120 nanoseconds over the 1987-88 period, as the result of die shrinks.
262 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

that come with experience—are created as a consequence of the cost

declines associated with improved yields.
One should also note that semiconductor production is a very lengthy
process, with a relatively long lag—typically two to three months—from
the time a batch of chips starts down the production line to shipment to
a customer.40 Changes in production rates are fully reflected in shipments
of finished chips roughly a quarter later.
Figure 5-18 shows published estimates by MITI of quarterly Japa¬
nese DRAM and EPROM production over the period from late 1986
to 1988. Output of both 256K and 64K DRAMs clearly appears to
have been cut back from maximum attainable levels in 1987 and 1988,
with particularly sharp cuts in the first half of 1987. Declines in output
of all types of EPROMs were reversed in 1987, then resumed in 1988.
Figure 5-18 indicates that output of 64K and 256K DRAMs was cut
back quite sharply from 1988 on, supporting anecdotal reports of sig¬
nificant cutbacks in 1M DRAM output in the second half of 1989 and
first half of 1990, described in chapter 4. Output of all varieties of
EPROM declined quite steeply after 1988.
MITI estimates of net exports from Japan are even more revealing
(figure 5-19). Japan remained a large net exporter of the most advanced
varieties of DRAMs throughout the post-STA period, with 50 to
60 percent of 1M DRAM output going overseas, and over 70 percent of
4M DRAM output exported.41 In EPROMs, however, imports steadily
increased, and by 1991 imports of many types of EPROMs—including
the most advanced products—accounted for a substantial portion of the
domestic market, and greatly exceeded domestic output in the case of
the widely used 512K EPROM. Thus Japanese suppliers maintained their
already preeminent position in the global supply of DRAMs; in
EPROMs, in contrast, where the Japanese share of world supply was
initially considerably smaller than in DRAMs, Japan cut back production
sharply and vastly increased imports.
As table 5-3 shows, foreign-based firms’ EPROM market share in
Japan surged upward in 1990 and 1991. (In these Nikkei estimates of

40. In a 1986 submission to the Commerce Department, Texas Instruments gave the
following estimates of time elapsed from the beginning of chip fabrication to sale to the
final consumer, for a 64K DRAM: wafer fabrication, 20-28 days; die inventory, 7-14 days;
assembly and testing, 10-14 days; and finished goods inventory, 20-30 days. See Texas
Instruments (1986), pp. 18-19.
41. The share of older vintage 256K DRAM chips exported dropped sharply in 1990
and 1991, however.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 263

Figure 5-18. Japanese Ouput of DRAMs and EPROMs, 1986-95

A. DRAMs, 1986-883 B. EPROMs, 1986-883

Millions of units Millions of units

C. DRAMs, 1988-95b D. EPROMs, 1988-95b

Millions of units Millions of units

Source: MITI, “Supply-Demand Forecasts.”

a. Quarterly data.
b. Semiannual data.
264 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-19. Japanese Exports of DRAMs and EPROMs as a Share of

Total Output, 1986-95

A. DRAMs, 1986-88“ B. EPROMs, 1986-88“

Percent Percent

C. DRAMs, 1988-95b D. EPROMs, 1988-94b

Percent Percent

Source: MITI, "Supply-Demand Forecasts.”

a. Quarterly data.
b. Semiannual data.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 265

Table 5-3. Shares of U.S. and Japanese EPROM Manufacturers in the

Japanese Market, 1988-92
Percent

Company 1988 1989 1990 1991 1992:1

128K EPROMs
Intel 10 18 29 n.a. n.a.
Hitachi 21 16 3 n.a. n.a.
Mitsubishi Electric 21 16 6 n.a. n.a.
Fujitsu 20 14 7 n.a. n.a.
Toshiba 12 11 7 n.a. n.a.
NEC 10 7 4 n.a. n.a.
Others 6 18 43 n.a. n.a.

256K EPROMs
Advanced Micro Devices11 n.a. n.a. n.a. 10 10
Intel 7 13 20 7 8
Hitachi 20 20 12 3 3
Mitsubishi Electric 21 15 16 4 4
Fujitsu 22 23 18 5 5
Toshiba 19 20 16 2 2
NEC 10 2 2 0 0
Others 2 7 16 69 68

512K EPROMs
Advanced Micro Devices11 n.a. n.a. n.a. 21 22
Intel 7 11 17 18 17
Fujitsu 18 15 27 9 7
Toshiba 16 12 13 4 4
Hitachi 17 11 15 5 5
Mitsubishi Electric 18 13 10 4 4
NEC 24 34 11 2 2
Others 0 3 7 37 39
Source: Unpublished data from Nikkei Telecomm (on-line service). First quarter of 1992 is estimated,
n.a. Not available.
a. Advanced Micro Devices is not broken out separately in 1988-90; in those years it is included in "Others.”

EPROM market share, the foreign-supplied EPROMs show up in the

sales of Intel, Advanced Micro Devices, and smaller suppliers included
in the “other” category.)
All these individual pieces of evidence suggest that, at the margin,
Japanese exports of DRAMs effectively determined prices in overseas
markets. Since Japan was by far the most important source of supply for
this product to world markets, restraint on exports, enforced with signif¬
icant border controls on most export sales, could create a situation in
which foreign prices could rise well above domestic levels.
In EPROMs, in contrast, Japan started with a considerably smaller
share of aggregate world supply. The importance of Japanese supplies
266 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

was to shrink further as Japanese suppliers cut back production of all

types of EPROMs. Since imports of EPROMs were freely permitted—
indeed, encouraged, in order to increase foreign-based semiconductor
sales in the Japanese market, as called for by the STA—and foreign
suppliers accounted for a large and increasing share of aggregate global
supply, unfettered inflows of these products from abroad effectively
pushed Japanese prices to the same levels as in markets outside Japan.

The FMV System

From late 1986 through mid-1991 the U.S. Department of Commerce

administered a system of company-specific price floors (FMVs) for U.S.
imports of Japanese DRAMs and EPROMs. With the announcement of
sanctions against third-country dumping applied to Japan in March 1987,
pressure was placed on the Japanese government to ensure that the prices
of Japanese products sold in other foreign countries met or exceeded the
minimum U.S. price. What effect did the FMVs have on the market price
of the affected products?
Although the FMVs set for each individual Japanese company by the
Commerce Department were never revealed, it is possible to deduce a
range within which these levels varied. Japanese companies were re¬
quired to file public reports every quarter with the Commerce Depart¬
ment which contained “ranged” estimates of various cost concepts, cal¬
culated as the true value plus or minus an error of up to 20 percent (this
was done to preserve a degree of confidentiality). The economic inter¬
pretation of these cost concepts is subject to considerable debate, but
the published estimates nonetheless provide unambiguously useful data.
Because these cost estimates were the basis for the procedures by which
the Commerce Department set FMVs, some insights about FMVs can be
culled from these published reports.
FMVs for a given quarter (call this quarter t) were basically set by the
Commerce Department according to the projected cost for the previous
quarter (t - 1). This in turn was based on actual cost of production in
period t - 2. Projected cost in quarter t - 1 and actual cost in period
t - 2 were contained in a report filed with the Commerce Department
in quarter t - l.42

42. See Semiconductor Industry Association, Antidumping Law Reform and the Sem¬
iconductor Industry: A Discussion of the Issues (Cupertino, Calif., February 1990), for what
appears to be the only published description of this methodology.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 267

Let us consider the “average FMV” for some type of chip, defined as
the unweighted average across companies and minor product variants of
the true FMVs set for Japanese imports. If we treat the “error” added
on to the confidential, true-cost estimates to produce public, ranged cost
estimates as a random error with zero mean, averaging public ranged
cost estimates across minor variants on a specific type of product within
a company, and across companies, will produce an unbiased estimate of
this average FMV. Similarly, adding 25 percent to the maximum public
ranged estimate among the population of ranged average costs for all
companies will produce a number that must always be greater than or
equal to the greatest true FMV among all Japanese companies, and
therefore bounds the true FMV from above. Using these calculations,
then, we can construct an upper bound on the maximum company-
specific FMV and an estimate of the average FMV across all companies.
The picture is slightly complicated by the fact that—particularly in the
initial quarters of the operation of the FMV system—the Commerce
Department reviewed and made changes to the cost estimates submitted
by the companies in their quarterly reports prior to issuing FMVs. After
the system had operated for a while, however, and both the Commerce
Department and the companies had iterated onto a set of procedures
that produced estimates acceptable to the Commerce Department, FMVs
generally were set quite close to the constructed cost projections sub¬
mitted by Japanese companies.43
There is some evidence that actual FMVs had settled down around
companies’ projected cost submissions by mid- to late 1987. Figure 5-20
shows estimates I have constructed of average FMVs across companies
for 256K DRAMs (as well as a bound on maximum company-specific
FMV) based on public cost submissions to the Commerce Department
(of projected cost in period t — l).44 These are shown along with ranges
for actual FMVs for this product reported in the Japanese trade press in
1986-87. As can be seen, after initial large discrepancies between com¬
pany and Commerce Department calculations of cost, the midpoint of
the range for actual FMV—as reported by the press—settles around the

43. Interview with law firm staff responsible for Japanese company submission to the
U.S. Commerce Department, January 1990.
44. Because the type and extent of data reported for every company were different,
and often varied over time, the details of how these estimates were constructed are not
always obvious. These details are discussed fully in appendix 5-B.
268 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-20. Average Foreign Market Value (FMV) across Companies

for 256K DRAMs, 1986-91

Dollars per chip

Sources: Author’s calculations based on data in table 5B-1; Dataquest, DQ First Monday Report; Dataquest
average sales price; and Semiconductor Industry Association, Antidumping Law Reform and the Semiconductor
Industry: A Discussion of the Issues (San Jose, Calif., February 1990), p. 8.

approximate neighborhood of my average FMV around the third quarter

of 1987.
Also shown in this diagram are the midpoints of upper and lower
bounds on constructed actual cost reported in these quarterly submis¬
sions, as compiled by the Semiconductor Industry Association, the U.S.
industry’s trade association. Actual cost in quarter t - 2 may be consid¬
ered an alternative estimate of FMV in period f.45 Although the meth-

45. The SI A is not always consistent in how it associates these estimates of actual cost
with a time period. In Semiconductor Industry Association, Anitdumping Law Reform,
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 269

odological details of the SIA’s work in producing these estimates have

not been published, this figure shows that these estimates are generally
quite consistent with my estimates of FMV in period t based on projected
cost in period t — 1.
This figure shows that by late 1987, and continuing through late 1989,
U.S. contract prices for 256K DRAMs had risen substantially above
FMVs, and therefore that the FMVs were not constraining U.S. DRAM
import prices over this period. From 1990 on, however, at least some
Japanese DRAM imports appear to have been priced out of the U.S.
market, as 256K DRAM prices fell below average FMVs. Indeed, Japa¬
nese 256K DRAM production fell quite sharply from 1990 on. Despite
rapid and deep cuts in Japanese production, U.S. prices dropped below
the FMV levels.
Figure 5-21 shows a similar pattern in 1M DRAMs. From late 1987
through early 1990, U.S. contract prices stayed above average (and, over
most of this period, even the highest-cost Japanese company’s individual)
FMV. Unlike the case in 256K DRAMs, however, U.S. prices roughly
track average FMV over the remainder of the STA’s lifetime, and Japa¬
nese companies cut neither production nor exports of 1M DRAMs to the
degree visible in 256K parts. Cuts in Japanese output were apparently
successful in boosting prices to levels at or above the Commerce De¬
partment FMVs.
An essentially similar story applies to EPROMs. Figure 5-22 compares
Dataquest ASPs with estimated average and upper-bound FMVs for
512K and 1M EPROMs calculated from SIA estimates of upper and
lower bounds on FMVs (the mean of the upper and lower bounds is used
here as an estimate of average FMV). From late 1987 through mid-1989,
market prices exceeded the FMVs, and therefore the FMVs could not
have actively constrained pricing of these EPROMs.
Because of this fact, some have argued that the STA itself was not the
cause of “abnormally” high prices outside Japan.46 What this argument
ignores, of course, is that production cuts and restraint in investment
were clearly associated with the implementation of the STA prior to 1988,

actual cost estimates are given for the quarter in which the report was submitted (although
the actual cost pertained to the previous quarter), whereas in SIA, Four Years of Experience
under the U.S.-Japan Semiconductor Agreement: “A Deal Is a Deal” (San Jose, Calif.,
November 1990), the actual cost estimates are assigned to the quarter in which the cost
occurred (that is, the quarter prior to the submission quarter).
46. See SIA, Four Years of Experience, p. 65.
270 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-21. Average Foreign Market Value (FMV) across Companies

for 1M DRAMs, 1986-91

Dollars per chip

Sources: See figure 5-20; and Semiconductor Industry Association, Four Years of Experience under the U.S.-Japan
Semiconductor Agreement: “A Deal Is a Deal” (San Jose, Calif., November 1990).

and therefore played some role in the increase in price by reducing

aggregate world supply of these chips. It is true, however, that the brakes
applied to Japanese supply apparently greatly exceeded the restraint
required to raise foreign prices above the Commerce Department FMVs
through 1989. The extent to which this reflected mere miscalculation (in
particular, underestimation of the recovery in chip demand in 1988, or
below-average growth in yields in semiconductor production in 1987 and
1988) rather than a deliberate decision to exploit the new framework,
organizing a group of producers accounting for most of global production
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 271

Figure 5-22. Prices and Estimated Foreign Market Value (FMV) for
512K and 1M EPROMs, 1987-90

A. 512K EPROMs

1987 1988 1989 1990

B. 1M EPROMs
Dollars per chip

Sources: Dataquest average sales price; and SI A, Four Years of Experience.

272 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

to seek greater monopoly rents in foreign markets, is probably destined

to remain a controversial subject.

Impacts of Regional Price Differentials on Regional Welfare

No matter what the origin of the restraints on production, it is clear
that measures related to implementation of the STA had a significant
impact in reducing aggregate supply, particularly during and after 1987.
Direct restraints on production were in effect through late 1987. Japanese
press reports suggest that “guidance” supplied to producers in restraining
investment levels continued through at least early 1988, and therefore
probably affected supply through at least 1989. Quantifying the impact
of these measures on supply, however, would be an exceedingly difficult,
if not impossible task, because counterfactual estimates of supply of
different types of chips in the absence of these measures would require
extensive modeling using data that are not available, in which subtle
impacts on the shift of production capacity among types of chips would
also need to be included.
The effects of one subset of these measures—the border controls,
which created significant regional price differentials—can, however, be
assessed in a rough sort of way. For purposes of this analysis I shall
assume that Japanese and foreign production of DRAMs can be taken
as essentially fixed over the 1987-88 period. More specifically, for foreign
producers, output is assumed to be limited by available capacity over
1987-88. For Japanese producers I assume that production controls (in
1987) and capacity (as demand surged in 1988) constrained aggregate
output, and that the regional allocation of that output between the Jap¬
anese and foreign markets was set administratively. The question I will
ask is what the impact on Japanese and foreign welfare would have been
had regional price differentials not been created, and had prices for
DRAMs been equalized in global markets.
This approach assumes away any effect of regional price differentials
in increasing non-Japanese DRAM output over this period. That is, I
shall assume that output of other producers in 1987-88 would not have
been significantly lower had world DRAM prices been equalized across
regions. Given that non-Japanese producers appear to have been close
to capacity over this period, and given the long gestation period associ¬
ated with new investments in semiconductor production capacity (a year
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 273

or more), this is probably not a terrible approximation. However, if such

effects were significant, foreign output would have been lower, and prices
higher, had chip prices then been equalized internationally by the removal
of Japanese border controls.
Table 5-4 shows my calculations of how welfare inside and outside
Japan would have been affected if production and investment restraint
had continued, but border controls had been removed. For purposes of
this calculation, the world is divided into two regions: Japan and the rest
of the world. Unpublished quarterly MITI data on production and export
sales by Japanese producers (that is, excluding foreign affiliates produc¬
ing in Japan) in 1987 and 1988 have been combined with Dataquest
estimates of chip production by non-Japanese producers, to produce
estimates of production by both Japanese and non-Japanese companies,
consumption in Japan and the rest of the world, and the share of Japanese
producers in rest-of-the-world DRAM sales.47
In making these calculations I have assumed a demand curve with a
constant price elasticity of —1.5 (chapter 6 presents evidence on the price
elasticity of DRAM demand supporting this assumption). Aggregate
DRAM revenues in each of the two regions were estimated by taking
quarterly regional consumption, by type of chip (64K, 256K, and 1M),
and multiplying by the contract price using the same data used above to
construct my regional Fisher Ideal price indexes. Regional DRAM reve¬
nues were then divided by the Fisher Ideal price indexes to estimate
quantities (measured in dollars of fourth-quarter 1986 DRAM output)
consumed. The effect of shifts in aggregate demand unrelated to price is
derived on a quarterly basis, as a residual, by combining observed price
changes with the assumed price elasticity of demand.48

47. For 64K DRAMs data excluding foreign affiliates producing in Japan were unavail¬
able. In this case I use published MITI data on shipments and exports, which include the
output of U.S. affiliates in Japan, but for this product Japanese production by IBM and TI
is believed to have been insignificant. The unpublished MITI data were provided to U.S.
government officials and obtained by me. In producing the estimate it is assumed that
Japanese producers supply 100 percent of Japanese chip consumption.
48. The methods used to construct table 5-4 are briefly summarized as follows: start
with constructed time series on DRAM prices in Japan and the rest of the world (ROW),
P/ and Pro„ (the price indexes shown in figure 5-13, with the ROW price the same the as
U.S. price), total DRAM production by Japanese producers, St, and DRAM sales revenues
in Japan (Japanese companies’ DRAM production less exports times Japanese DRAM
price, summed over all chip types) and ROW (ROW companies’ DRAM production plus
Japanese companies’ exports times ROW DRAM price, summed over all chip types).
Dividing regional sales by the regional price indexes produces quantity indexes <2, and Qrow.
 r~». as q <N r- CO q r- o q q q co —;
V sd
X Tj- CO in oo 00 Os
oo On o
co (N in r- sd o in i1 (N
op co co
On | 1
1 1

.
O q q in q q q q co q in q
C^O
in o cd sd sd cd sd GO
oo Os 00
CO
sd
(N m cd
O' co T—• m ii CO 00
Oo CO co
On | 1 T—1
t 1 1

oo t> o CN q q q q o q in q q
00 q r-H
00 csi in cd o
oo Os CO
in CO so so
Os (N (N in | m in o T—i
Oo T"H r-H
CO CO 1 r-H
On I
’"h 1 1
£

’"'■i q rsj o 00 q q q On q 00 q oo q
IT) tj- sd r-H sd in O in o cd oi sd
Co Os (N T-“H (N in <N m CO (N so
00 t-H r-H (N (N 1
Os 1 1
§
1
c
o
oo q q SO in oo q t—H q
5.6

17.4

6.3
o q
K '
IN so CO in o SO in cd cd o (N
-Uri OO t-H o (N r- in CN |1 in CO
°0 r-H i | | 1
c
o ,'H 1
u

"a CO q © in q sO q q q CO q q q q
o t\ CO On q o sd o ud in cd cd o
Oo 00 o H OS i> co i-H i r-H co (N
cq Os t-H
1
i
1
i
|
i
|

so ’-H

>5

C
S3 <N q in 00 00 q (N CO q q q 00 q q
t\ Os 00 oi © (N t—1 t—5 O rt- os cd tj-
t Oo r- OS Os r- SO CN i i (N H
Os | | |
i i
»-H

f
o
r-H q On CO q q (N q CO q CO q q q
^u t\ Os cd cd cd O in cd r-H (N in cd r-H

a; Oo r- 00 00 »n in i
i
i
i
| l
1
Os
1

~o
<j <u
£> S
<u<U
C/5

JU G
13
C/5
o
sg. * X
<U X J£ OJ <U
S I
^ -*—>
c
3
G £ c
03
_
s ^ .2
q$ ° o
cu c > s '5b 2
~
> 1
<U $ o
o
0) CS
Ur C/5 P 2 <U£
-2
13
n
Ur U—1 CAI
S e ^
C X
o

S
Uft
O CU
>N
c
(U
O 0
5 3 CS C/5 T3 03 E kr
O. 0) 0
C/5
<u
0X) O
Ur p
<—
0
C/5
0) 4—*
OX) G
E § c E G 03
tij ^ O
u>
o
u
0
0
2
c
X
0
C
O
O
o i- II S-H
II O
G -O
U) O ts Ur
-o O CD
>-• G G G 03
<L>
0 -r
<U
<u
•2 <u
0 O
«4-( i— q q Ur
x ^
c c
•S
S5
CL Ur T3
l o <2 sd
to N4—1 H-i Ur
<U sd V <D CJ cc 0
<u E <u Ur
in in ■» 00 c O 'HH 00 0Q 00
<D
0JD
s CL
<u
OX) OX) •-
CL O

"5> X) 2 CO On G G C —-a C/5 c c — C/5 X

v> 03 r-H 03 03 03 03 0 03 03 5
O v—^ a. a> <u v— c X X -G
<
X =G
<
3 •2 _CJ 03 03 .O X

C ’C CL u O UUP <9 UUP

£ a. CU Cu C4
 no

97.9
-o-

130.7
sC On

ON mi
nC

(N O n (N
97.0
138.4 oo

ON (N IT) C m nd m
rf ON OO
I

oo
97.7
153.0

cn NO NO o ON NO NO NO 1—H mi
oo 00 ON mi NO NO 00 On
m O ON m i ON
m i T““t
i 1
I I

m r-
95.7
124.2

h Tl CO O fO On r- OO OO
no
^ cn
Ti mi
m rf
1/3 no
n I
On ^ NO
ON ON
I I
I
86.8
113.4

ON oo ON NO NO On
no
o
ON
ON
^
Is
(N on
tT H
rf- nO OO rj
1—( Tf" CO
I
V i i

t"- vq
104.0
83.2

On CO OO
mi jh in On n ON ON O
On r- on ^ h m on
I

n On GO -h cn cn
79.4
98.5

ON ON ©
CO ON © © 0\ O' oi N"
ON r- no on | I ON ^
I I

on oo no
79.5
84.9

NO ON NO ON
ON ON C in d os >n on
GO m> m | |

JD
"c0
/3 13
/3 C
Price before removal of controls

C/3
C
<D <D
u*
— C/3 $ d) *3 <2
3 ,03 2 o c § o
C
o £ C <4_|
o £
d>
c > ^ d>
*5b £ > 1 S *5b
o d> 03
/3 d> * d>£
i-i c
i-H 2 8 t— 8 8
o S d> >s 0)
cCO s 1/5
Price elasticity = — 2

£ <o 5 S O
cm
d> "O
u. g I "O
d> u,
03 Q- d) 3 CL ^
(1986:4=100)

O _ ox) o G 00 OX) O
O © E I a. c 03 X) E 3 03 X
Rest of world

x o o o
1 ii
8 TO d> d> y « s d) d) y § a
.E .£ .S .E 5 Q. y
u o r- -§ c C S. a £
d> d> d> d> £ JJ ■= d) d) cut < «
s £ OX) OX) OX) Of) •= ^'5 OX) OX)
c c ■— c/5 ?
Japan

^
3 ON
T—t
c c c c -p C/5 ^

J= sz 2 <
_ CO CO 03 CO 2 < cd 03
0) w
o
S i £
'G g.u u u u fS u u eS
CO
0-
276 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Removal of border controls was then simulated by calculating the

quantities of aggregate DRAM consumption required in each region to
equalize DRAM prices in Japan and the rest of the world, taking world
DRAM output as fixed. Prices would rise in Japan and fall in the rest of
the world. Japanese consumers clearly lose by paying higher prices, while
Japanese producers gain from increased global profits (despite increased
shipments, revenues fall in export markets; these are more than offset
by increased revenues on smaller quantities sold in the Japanese market).
But when the loss in Japanese consumer welfare is added to the gain in
producer revenues (and therefore profits, since total output is constant),
Japan as a whole loses from price equalization, by an amount generally
ranging from about 5 to 7 percent of Japanese DRAM consumption with
a single world price. Thus, the border controls as a whole worked to
increase Japanese welfare by lowering the price to Japanese chip con¬
sumers at the expense of Japanese producers (who would have received

Assume demand for DRAMs in Japan and ROW is given by a constant price elasticity
demand function with elasticity -1.5, Q: = Z; P~' 5, where Zy is a Japan-specific constant
that varies over time, reflecting shifts in aggregate demand within Japan. A similar equation
describes demand in the rest of the world. Zy is calculated in every quarter by dividing
quantity 2, by Pf15 (and Zrow analogously). Given the two Q’s and the two P’s (corre¬
sponding to border controls and different prices inside and outside Japan), the effect of the
removal of border controls given the same total production is to produce a single new world
price P' and new quantities Q’ and Q’ro„ consumed in Japan and the rest of the world,
which must solve the system of three equations in three unknowns:

Q; = Zy P'-15; Q’row = Zrow P’~' s; Q) + Qrow = Q, + Qnw.

Solving for P’, we have

-1

p, _
\Zj + ZrJ

After border controls are lifted, Japanese companies’ revenues change by

(P1 Q’j - Pj Qj) + [(Sy - Q’)P’ - (Sy - Qj) Prowl

where the first term is the change in domestic revenues, and the second term (in brackets)
is the change in export revenues. For ROW producers, change in sales revenues from their
production is given as

(P’ - ProM, + Q row - Sy).

With the constant elasticity demand function, the effect on consumer welfare of a change
in price from Pa to P’ equals
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 277

greater global revenues by diverting chips into higher-priced foreign mar¬

kets). The benefits to Japanese consumers more than offset profits for¬
gone by producers.
Conversely, the border control system worked to reduce aggregate
welfare in the rest of the world. Non-Japanese producers clearly gained,
but their gains pale in comparison with the loss inflicted on consumers
in the rest of the world market—increased profits for rest-of-the-world
producers amount to about one-third of the welfare loss suffered by
consumers in the rest of the world market as a consequence of the price
differentials so created. The total impact of border controls on the rest
of the world’s welfare is a loss equivalent to roughly 5 to 10 percent of
the value of DRAM consumption with a single world price.
As expected, however, the net gain to Japan is less than the net loss
to the rest of the world. These results are relatively robust to changes in
the assumed price elasticity of demand. Table 5-4 shows that changing
the assumed price elasticity from —1.5 to —2, near the upper bound for
empirical estimates of this parameter, changes the outcome of the simu¬
lation only marginally.
The relatively small magnitudes of the welfare changes induced by
border controls in this exercise provide a dramatic contrast with much
larger estimates by others of monopoly rents associated with STA-
induced price movements. For 1M DRAMs alone, market analysts’ cal¬
culations suggested that the value of sales exceeded costs in 1988 by
anywhere from $1.2 billion (for Japanese companies only) to close
to $2 billion (all suppliers).49 It is probable that 1M parts alone accounted
for one-third of Toshiba’s operating profit in 1988, one-fifth for Oki Elec¬
tric, 17 percent for Mitsubishi Electric, 15 percent for NEC, and
13 percent for Fujitsu (Nomura Research estimates). By 1989 the concept
of “bubble money”—supernormal profits due to abnormal scarcities of
product—was reportedly being used in Japanese industry circles to de¬
scribe the profits being made on DRAMs. Estimates of “bubble money”
being collected in DRAMs by early 1989 hovered around $3 billion to
$4 billion per year.50

49. The figure for Japanese companies is calculated by converting a fiscal 1988 operating
profit of 150 billion yen to dollars at 130 yen to the dollar. See Barclays de Zoete Wedd
Research, Japan-Electronics, Semiconductors (Tokyo, December 1988), p. 17. The figure
for all suppliers is according to the consulting firm In-Stat. See Richard McCausland,
“Semiconductor Makers Concerned Price Cuts Could Hamper Growth,” Electronic News,
January 2, 1989, p. 22.
50. This value is roughly consistent with a report circulating in Japanese industry circles
278 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

These magnitudes are considerably larger—by a factor of roughly

ten—than the increased profits for Japanese producers indicated in table
5-4. If accurate, these estimates would suggest that border controls were
not the most costly aspect of STA-related policies for U.S. and European
DRAM consumers, but rather that the underlying restraints on supply—
production and investment—were the primary underlying cause of the
financial distress experienced by chip buyers in 1988-89.

DRAMs versus EPROMs

Some have argued that the considerably greater market share of Jap¬
anese producers in DRAMs than in EPROMs explains the rather differ¬
ent behavior of these two products over the life of the STA.51 One inter¬
esting question is why Japanese production of EPROMs plunged so
precipitously, and imports into Japan surged, in dramatic contrast to the
situation in DRAMs. The simplest possible answer is that foreign sup¬
pliers were more competitive, lower cost producers of what was clearly a
commodity product, and that Japanese companies, unable to compete,
exited the business. Since by most accounts, however, Japanese compa¬
nies had a cost advantage in the manufacture of commodity semiconduc¬
tors, this reasoning faces some obstacles.52 Another possibility is that
EPROMs were less profitable to manufacture than other products such
as DRAMs. But when DRAM prices and production declined in 1989,
EPROM manufacture continued to drop and imports to increase. Both
of the above explanations are also undermined by the fact that, as the
Japanese EPROM market tightened in 1987 and foreign manufacturers
began to run into capacity constraints, the U.S. chip maker Intel con¬
tracted with the Japanese chip maker Mitsubishi to produce EPROMs

in the spring of 1989, according to which MITI had calculated that “bubble money” was
running around 45 billion yen per month at that time.
51. Tyson argues that “In EPROMs, a product in which American companies still
retained a significant share of the world market at the time the [STA] was signed, the
outcome was very different. . . . American and other buyers did not report shortages in
EPROM supplies, and American EPROM suppliers such as Intel asserted that EPROM
prices never significantly rose above their marginal costs because of competitive mar¬
ket conditions.” Laura D’Andrea Tyson, Who’s Bashing Whom? Trade Conflict in High-
Technology Industries (Washington: Institute for International Economics, 1992), p. 121.
52. Indeed, cost disadvantages faced by U.S. semiconductor manufacturers were one
of the reasons why the Sematech manufacturing consortium was formed to improve the
productivity and competitiveness of American chip producers.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 279

sold in the Japanese market under the Intel brand name.53 Mitsubishi, at
least, could apparently manufacture and sell EPROMs at prevailing
prices at a profit, and found it more attractive to do so than to increase
production of DRAMs, which it also manufactured.
A third possible explanation is that the levels set for FMVs prevented
Japanese producers from selling their output in foreign markets. But, as
we have seen, these price floors were not binding in the U.S. market
through 1988 and were never an obstacle to selling product in the Japa¬
nese market. Nevertheless, Japanese producers steadily cut back their
output of EPROMs, and Japanese consumers increased their use of im¬
ported product in their stead.
This observation and the details of the Intel-Mitsubishi accord suggest
an alternative explanation. Japanese producers as a group may have cut
back production as a deliberate policy aimed at increasing imports and
foreign market share, and assuaging friction with foreign chip makers,
who had continued to market and sell EPROMs even after they dropped
out of DRAMs. Although it is impossible to determine just how impor¬
tant political factors were in explaining all the unusual market develop¬
ments observed after 1985, the review in chapter 4 of the details of the
STA makes it clear that they were present and played some role. While
EPROMs’ role in the memory market declined precipitously in the 1990s
(replaced by so-called flash memory chips, a superior electronically re¬
programmable form of nonvolatile memory), EPROMs provided incum¬
bent producers with a strong technology base for the new products.54

Import Promotion and the STA

Perhaps the most disputed element of the STA was the “secret” side
letter specifying 20 percent as a reasonable expectation for foreign market
share by 1991. After prices rose above FMVs in 1987, the focus for ne¬
gotiations between the United States and Japan increasingly concerned
the interpretation of this letter and the measurement and assessment of
progress toward greater foreign participation in the Japanese semicon¬
ductor market.

53. Andrew Pollack, “Intel to Resell Japanese Chips,” New York Times, July 30, 1987,
p. D4.
54. MITI dropped EPROMs from its semiannual forecast in 1994.
280 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

As a blueprint for a voluntary import promotion measure, the details

of the 1986 side letter left much to be desired. Even if the two sides could
have agreed on whether or not the 20 percent “expectation” constituted
a Japanese commitment, construction of a statistic measuring foreign
market share in Japan soon became embroiled in yet another set of
disputes. Details of a market share measure had not been specified in
the STA or its side letters, and each side quickly adopted a definition
furthering its own purposes. The U.S. government advocated the use of
sales figures compiled by the World Semiconductor Trade Statistics
(WSTS) program, a private data collection organization set up by chip
producers, as the only accurate and internationally accepted figures on
semiconductor sales. MITI, however, chose to construct its own ratio,
measuring Japanese market size through a survey of sixty-three large
Japanese chip consumers (including IBM Japan), and adding internal
captive shipments by IBM to its Japanese subsidiary into its measure of
foreign sales in Japan. The WSTS statistical system did not count IBM’s
output because IBM sold no product on the open market, and because
IBM’s products did not conform to industry standard designs and pack¬
aging. Because the MITI measure of market size in the denominator
excluded a substantial number of Japanese consumers not included in the
surveyed group of sixty-three large users, and because the measure of
sales by foreign companies in the numerator included IBM’s large inter¬
nal shipments, it is not surprising that MITI’s ratio was always consid¬
erably larger than the WSTS number cited by the American side.55
When the STA was renegotiated in 1991, the target of a 20 percent
foreign market share had not been reached, by either the MITI or the
WSTS calculation. The new arrangement raised market access issues to
a new level of prominence and included within its main text the statement
that “the Government of Japan recognizes that the U.S. semiconductor
industry expects that the foreign market share will grow to more than
20 percent of the Japanese market by the end of 1992 and considers that
this can be realized. The Government of Japan welcomes the realization
of this expectation. The two governments agree that the above statements

55. See Electronics Industry Almanac, 1992 (Tokyo: Dempa Publications, 1992),
p. 909 (in Japanese). However, the WSTS statistics also exclude shipments by some small
foreign producers that do not belong to the WSTS, as well as IBM, so that in theory
(although almost certainly not in practice) it would be possible for the MITI number to
exceed the WSTS-based measure of market share.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 281

constitute neither a guarantee, a ceiling nor a floor on the foreign market

share.”56
How this new, formally established market share “expectation” was to
be measured became an official issue. An appendix spelled out the details
of a new statistical system and set out the principles for an official data
collection program (DCP) that would define two distinct measures of
foreign market share. The first formula (FI) basically took sales of semi¬
conductors by DCP merchant market participants, added estimates of
nonparticipant foreign sales, and divided the sum by an estimate of
apparent Japanese consumption, based on official Japanese government
production and trade statistics, less captive consumption (IBM Japan,
for example). A second formula (F2) added in captive consumption to
both numerator and denominator. The two formulas also differed in their
definition of what constituted a “foreign” product: FI assigned country
origin based on the company doing “final assembly,” except where “pure”
assembly was done under contract; F2 assigned it by the nationality of
the producer whose brand was stamped on the chip.57
Although the agreement itself specified neither formula as superior,
predictably the governments of Japan and the United States quickly made
it clear that each had a preferred choice. A press release from the Office
of the U.S. Trade Representative set out its position that FI “accurately
describes changes in foreign market share,” while the second formula
was “not appropriate in measuring market access.”58 MITI took the op¬
posite view, maintaining that the inclusion of both formulas in the pact
made both equally useful measures of market share.
Figure 5-23 shows movements in all of these measures of foreign
market share over time. Note that neither FI nor F2 was officially cal¬
culated prior to the third quarter of 1991; the figures for FI for earlier
periods are SI A estimates. FI appears generally to be slightly larger than

56. “Arrangement between the Government of Japan and the Government of the
United States of America Concerning Trade in Semiconductor Products,” June 11, 1991,
sec. 2, para. 10.
57. "Arrangement,” annex A., pp. 9-10. For production arrangements involving joint
participation by both Japanese and U.S. firms (joint ventures, production under license,
foundry arrangements), these definitions ultimately also involved considerable ambiguity,
and the U.S. Department of Commerce ultimately created a detailed handbook defining
on a case-by-case basis whether the output of a particular factory with such transnational
links was to be considered “foreign.”
58. U.S. Trade Representative, “Questions and Answers on the U.S.-Japan Semicon¬
ductor Arrangement,” June 4, 1991.
282 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Figure 5-23. Alternative Measures of the Foreign Share of the

Japanese Semiconductor Market, 1986-95

Percent

Sources: SIA, Four Years of Experience, app. 4; Electronics Industry Almanac, 1992 (Tokyo: Dempa Publications,
1992), p. 910; "Trend in Foreign Market Share in Japan under the 1991 Semiconductor Arrangement,” graph attached
to Electronics Industry Association of Japan, “UCOM Chairman Comments on Foreign Semiconductor Market Share
in Third Quarter of 1995," press release, December 14, 1995; and "Foreign Share of Japanese Chip Market Soars,"
Computing Japan (on line), March 21, 1996.

its intellectual precursor, the WSTS-based market share estimate. F2 is

considerably smaller than its predecessor, MITFs survey-based market
share estimate, but always substantially larger than FI, as might be
expected.
Both FI and F2 remained well below the 20 percent target through
most of 1992, and bilateral trade talks in semiconductors seemed headed
toward some tense moments as the end-1992 deadline approached. Since
IBM had announced its entry into Japan’s merchant semiconductor mar¬
ket in the summer of 1992,59 MITI took the tack of pressing U.S. nego¬
tiators to include IBM’s shipments to Japan in FI, in order to support its
case that the goals of the 1991 arrangement had been met. Quite to

59. David Roman, “IBM Launches OEM Chip Push in Japan,” Electronic Buyers’
News, August 3, 1992, p. 3; Junko Yoshida, “IBM Japan’s Efforts to Sell Chips Stalled:
Distribs Will Sell Its ‘High-Function’ ICs,” Electronic Engineering Times, August 10, 1992,
p. 94.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 283

everyone’s surprise, however, FI (as well as F2) jumped sharply up at

the very end of 1992, just exceeding the 20 percent level. Even though
FI had dipped slightly below the magic 20 percent mark the following
quarter, in early 1993, tensions were substantially relaxed by this
development.60
Interestingly, the piercing of the 20 percent barrier for foreign market
share in Japan in late 1992 came just as U.S. firms came roaring back in
global chip markets. Extremely strong computer sales had buoyed mar¬
kets for the complex, proprietary microprocessors in which American
firms had a virtual monopoly, while markets for “commodity” memory
chips, where Japanese producers were strongest, seemed mired in low
margins and prices, amid a continuing glut of new capacity. For the first
time since 1985, U.S. firms seemed to pass Japanese companies in sales.61
Was U.S. trade policy finally achieving results, or was it, ironically, the
fact that U.S. firms had been driven out of commodity memory, and
forced by competitive forces into concentrating on the complex logic chips
that were their true forte, that was responsible for both the global come¬
back and the Japanese success story? Put another way, was it the political
pressure put on Japanese firms to buy American, or was it the shift in
the market toward complex logic chips that accounted for the apparent
success? At least some analysts argued that it was market forces, notably
the shift in demand toward niches where technologically innovative
American chip producers were strongest, and not bureaucratic pressure
on Japanese chip consumers, that should be given the bulk of the credit.62
On the other hand, our discussion of EPROM market shares earlier
in this chapter made it clear that there is at least some circumstantial
evidence of a deliberate decision by Japanese producers to yield EPROM
markets in Japan to foreign producers. Other prominent examples of
substantial increases in U.S. market share in the face of formidable
Japanese competitive capabilities may be found. A good example is the

60. The U.S. Trade Representative’s June 1993 projections had suggested that FI would
drop to about 19.6 for the first quarter of 1993. David Lammers, “U.S. Uneasy about Japan
Chip-Share Dip,” Electronic Engineering Times, June 21, 1993, p. 4.
61. Bob Davis, “U.S. Chip Firms Seem to Lead Japanese Ones in Sales This Year, but
Experts Aren’t Sure Why,” Wall Street Journal, December 24, 1992, p. 30; T. R. Reid,
“U.S. Again Leads in Computer Chips,” Washington Post, November 20, 1992, pp. Al,
A42; and Ken Yamada, “Intel Is Ranked No. 1 Chip Maker for 1992 by Study,” Wall Street
Journal, January 5, 1993, p. B6.
62. William R. Cline, “Yen Appreciation Does Reduce Japan’s Surplus,” Wall Street
Journal, May 20, 1993, p. A16.
284 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Japanese market for application-specific integrated circuits (ASICs). The

strongest U.S.-based producer in that market, Nihon LSI Logic, the
Japanese subsidiary of LSI Logic, went from a 0.8 percent share of the
Japanese market in 1988 to a 10.6 percent share in 1991.63 Nihon LSI’s
management, when asked for a qualitative judgement, credited about
half of its growth in the Japanese market to the effects of the STA’s
market access targets.64 Other American semiconductor producers con¬
cur in their assessment of the important role of MITI in encouraging
greater consumption of foreign semiconductors.65 Finally, the frequency
with which Japanese producers announced increases in the share of for¬
eign products in their total chip procurement, and increasing “design-
ins” of foreign product after reports of pressure from MITI, suggest at
least some element of government intervention at work.
Available data shed some light on this question. If the aggregate share
of foreign-based chips in the Japanese market at time tl is Sn, then

where sitl is the foreign share of product i at time 11, and min is the share
of product i in aggregate Japanese semiconductor sales at time tl. Then
the change in foreign share from time tl to time t2, Sl2 - Sn, can be
decomposed into a portion due to increased foreign penetration of par¬
ticular market segments, and a portion due to shifts in the composition
of market demand. That is,

Sl2 — = 2
i
Sit2mit2 ~ Sil\rnit\

= X min(sit2 - s,n) + 2 sil2{mil2 - min).

f /'

Or, in English, change in aggregate foreign market share from time tl to

time t2 (S,2 - S„) is the sum of changes in foreign penetration in market
segment i (sia - s,„) weighted by the share of segment i in total sales at
time tl (ra„,) plus the sum of changes in the share of market segment i
in the total market (m„2 - min) weighted by foreign penetration in that
market segment at time t2 (sit2).

63. Nihon LSI Logic’s manufacturing facilities were a 50 percent joint venture with
Kawasaki Steel. The market shares are calculated from data available on the Nikkei Telecom
online computer database.
64. Author’s interview with K. K. Yawata, Nihon LSI Logic, Tokyo, March 1991.
65. See Jacob M. Schlesinger, “U.S. Chip Makers Find ‘Quotas’ Help Them Crack
Japan’s Market,’’ Wall Street Journal, December 20, 1990, pp. Al, A6.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 285

The first terms summed after the second equality are the changes in
market share due to increased foreign penetration, while the second set
of terms are changes in market share due to changes in the overall com¬
position of demand. So, if there are no changes in market composition
(mIf2 — m,„ = 0 for all i), all change in foreign market share is attributed
to increased foreign penetration; conversely, if foreign penetration is
unchanged in all market segments (s„2 - sin = 0 for all i), all change in
foreign market share is attributed to changes in the aggregate composi¬
tion of demand.
This is not the only possible decomposition of share change into a
penetration component and a market composition component. The de¬
composition just described can be referred to as the “initial market
composition” variant, since changes in foreign penetration within seg¬
ments were weighted by the initial share of that segment in the aggregate
market, while changes in market composition were weighted by final
foreign market share in each segment. An alternative decomposition, to
be referred to as the “initial foreign share” variant, can be given as

where the initial foreign share of each segment has been used to weight
changes in market composition, and final market segment shares in total
sales are the weights on changes in foreign share by segment. Still a third
decomposition may be constructed that takes the mean of the previous
two sets of weights on changes in market composition and foreign pene¬
tration, that is,

This latter variant on the decomposition of changes in foreign market

share has the virtue of using symmetric weights for both foreign market
share and market composition changes.66

66. None of these decompositions is invariant to the level of disaggregation used. More
or less disaggregated product categories will give somewhat different decompositions of the
contribution of foreign penetration by segment, and market composition, to aggregate
foreign market share change.
 <o> oior--ococcoro
Os
r- —■ sc r- — ro sc —•
Os
<3
-s:
Os NCO'fsi'C^nM Os ri in C O to - ro h
r-i — m rq r-i —
<0

O 00 • O <N O O so Os — O CO (N O'
00 OS O O
Foreign Share of the Japanese Semiconductor Market Due to Changes in Foreign Penetration and

rO sO ^ <N CM —< O' so o r-

3 ^ 2 E rn o o ro o (N co o o
<0 I

o
3 »n <N • • m no oo "3- inH\o^THOt
C rr a. ® o in O;oorjfsjion^q
*3 J o c e o o o o 00000000
o
I I I I III
£

c -3 in vo • J h ON O0 Nf
N (N ® ™ -si- t~~ rr ’-H3-Ht^'r<TTOS’-H00
O C e <%i o o ^ O^HT_HOCNOO^t
6 ^ I
eg F
cx.

o ON fN • ■ 00 On in Os so '—1 O' T h O'

-3 3 ON NO . ® O) Ifl O H O' O' CO <N SO CO O
.50 *3
Initial product

O O E E o O O O
o OOOOOOOO
53 I I I I I I
£ s: s
e
.0
c
o c .2 —* NO J, O CN ON O NO 't N ON H so o T OS
<3. .£P 3 (N~) O ra n (v| m n o 1 o so m o 1 so
S 2 ^ E E pi 6 6 Nf o ^r
o o ^
o Cl C
^ su
ex.
Initial foreign penetration

o • 00 no r- cn (NhOfN^fsOOsCC
3 <n w a. -rt O ON OssO(N(N^t(NO^
“3 O E c o d o o
O OOOOOO

%
1 3
3 -2 O in 00 h sOM^hsOOONsO
• SP 3 (N Tf 03 r- so i-H ornoor^iooor-ios
^ 2
Composition, Selected Periods, 1986-94“

O ^
O C <n O^H^-HOCNOO^t-
LT 3 I

T3
3 5s.
3
3 rv1 ^
so 05
- -3 IT) Os
0O «o
O Ol (N
Os ^5
c>inr^ococoom
•2 ^ "3 Os
^
. tT
cs
6 (sj 06 Os (NOOs(Nsd^t-C^(N
(N T-H 1-H y—I r-H t-H
^ CO

«sj £>

SO
o cj
’5b
TO
c _o
03
+
03 c^3 o
e- V-H
0
3 ’5b 0
u, 0
CJ
tS0 O
u O •-
o
Table 5-5.

E
O 3 ’*3 CL) O ’5b E o .2 '5b E
3 2, u c o —
“3 _o SO .2
03 E E 13 o t— 2 >
C3 C — 03
Percent

O O CO CO CO .0 CJ ro Os *0 CO 00 00 n
aOOO
03
C C/3
as u
^ -2 H 1 aOOOi2 c .22
g; £ ^ ^ PQ < Q s s 2 s < Q
lTt
^ >3 S ps Os
•**«<
 .&0
S
0 (N r- <n Os —; Os in so —T os sO 0 cn cn —* r- mososinso —
-s, « rsi 0 0 *—H d ^ Os (N in -h 00 d 06 sd in sd d o r! ri O' ro ri
00 (N m (N in (N in <n m ^h Os m m in —< r- <n
Os
Os

r-
(N
Os Os
J Os
sO
sO
cn
»n
00
rn
CN
m
so
(N
(N
(N
T-H
<N
r-
H
cn (N
T-H T-H
r-
0
soin(nr^ososinso
ooooo(Ninm
1—H 1—H d d d d d d d d d d d d d d 00000 o o

I I I I

O SO SO 00
n GO --h
os o r- 00 so r- in
o in (N m ^ m
HO^in^ONOr^mH^ooin
(Ncnooooomoocomsoso
OOOO 0000000 dddddddddddddd
I I I I I I I I I I I I I I

h m n io cn os os os os m o o hhinhooin»nM(NnnH
Os ro © ^ m os o o so o i—i O <N ■— 1 r t-1 o o m o o gn cn t—* t'" o> O

©odd d o o d d d d oooooooooooooso cx 'y

I I

< -Si

O (N h vD o r- so t-h <n 00 (N HTtinsosoo(NsoNininM(Nso

os (N in h in in n in h m cvjoooooocnoocosor^co §■<

OOOO 0000000 0000(0000000000

CZ u

I I I I I I I I I I I I 1?

os r-
‘5 A

r- r- 00 so o 00 so in 00 t-h o^cocohhoo(Nsonsor>h
00 CN) Os CO in t-h o o 00 o t-h o
OOOO d d d © d d d 00000000000000

u uo
0 —1 in 00 m 00 in 0 SO r- CN (N in (N 0 t-H m (N SO r- <N in 0> CN
(N 00 0 0 (N in (N (N T-H cn (N O O 0 O 0 O (N O O (N SO Os
UJ
d d d d d d d d d d d d d d d d d d d d d d d d d D.
CL
H
1 1 1 1 1 1 1 1 1 1 1 1 1 1
. , O'
C
o :

r- 00 cn 00 so <N <N O oosinsocoooooin(NO(NOs^H

o cn in in o h in h OT-HrHrH^HOOCNJOO^^-lO^t 9 S

t-H ^0 t-H ^0 0000000 dddddddddddddso < S'

I I >> 2

c
bo

§ S; CN in O r- 00 t-H 0 Os cs os (N sO ^H so m in in <N o 00 o o o
£ «
_? -K
00 CN d r 4 SO OS csi
t-H t-H (N
OS r-H t-H
t-H <n
(N
t-H
rsi
Os
(N
T-H
06 d d os d in n
m t-H so t-H
r- (N (N

o
‘5b
E ^
2D
m-Japanese companies

B o X) <D
O
o

o x .am
£3 ’>
C/3 ‘o < A 2 V
3
o
<D
Cl
C/5 D-a
-o
C/5 cd cd C/5 Lh •o
o
C/5
Lh
O X
-•SS
- u cd <u
h a JO o <D

C/5 M in d
cd O
^
4- cn op
o ^ H U ^
cn o
Ui la Cl
3
«s

o O O
^ ^
Q,
Ih
s A 53 o
‘5b
cd 0
c/5
\/
v C
O JJ O
O

0 2
- O Cd
Q.OC ’So
_ 2
^
<L> —
o S 'oi) JO H S
O
o .a E
3 t»
O H-
.S2 rO
H-*
s <->
O v> C/5 O C/5 PC o <L> *rs &O
ft ^ d . o
Os o "O el Cu <1 "3 OX)
o CL
c
cd
2 g C S3 P !Z
I
I

o
.a
2
y cd
1

s
-g ^

is
W
LU CO <
5 CL oc U c/5 o
^ a
P o u £
o 2 X Oh
U UJ O P
O Os
288 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Table 5-5 breaks out the contribution of increases in foreign market

share by segment, and of changes in the share of different product seg¬
ments in aggregate sales, to increased foreign market share. For given
weights, adding across rows gives the total change in aggregate foreign
market share due to changes in both foreign penetration and market
composition in that product segment. Adding down columns gives the
total contribution of changes in foreign penetration, or changes in market
composition, respectively, to the change in aggregate foreign market
share. The sum of the “foreign penetration” column may be interpreted
as the increased share of foreign companies due solely to increased for¬
eign market share in each product segment, taking the share of each
product segment in overall sales as fixed. The sum of the “market com¬
position” column can be interpreted as the increased share of foreign
companies due solely to shifts in the share of each product in aggregate
sales, taking the foreign market share in each product segment as fixed.
An immediate implication of this table is that shifts in the composition
of demand in the Japanese market play only a very small role in explaining
the overall increase in foreign market share over this period.
The calculations reported in the first part of this table, which analyzes
SI A data on U.S. company sales in the Japanese market, are similar
conceptually to the WSTS estimate of foreign market share, used over
the five-year life of the first STA, described earlier. However, detailed
product segment data are only available for U.S. firms participating in
the WSTS program, and the share of these firms in the Japanese market
was lower than the aggregate share of all foreign firms. (Also, the crite¬
rion for judging whether a chip produced through a joint arrangement
between a Japanese and a foreign firm is Japanese or foreign is different
from that used for the “official” market share statistics.)
The vast bulk of the aggregate change in foreign market share is
attributable to increased foreign penetration of markets over either the
1986-90 or 1990-94 subperiod, not shifts in the composition of Japanese
demand, which overall actually reduced market share. These tables
(which, again, include only U.S. producers) show a large change in
market share from 1986 to 1990 attributable to microcomponents and
logic chips, followed by a lesser gain attributable to memory. This may
reflect the fact that the principal non-American suppliers present in the
Japanese market (for the most part, from Korea and Europe) mainly sell
memory chips of various types. Interestingly, over the 1990-94 period,
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 289

when sales by Intel, the U.S. microprocessor leader, soared and for¬
eign market share shot up, increases in U.S. market share in Japan
due to memory chip sales almost exactly equaled those due to micro¬
components.
The second part of table 5-5 breaks out the sources of increased
foreign market share in Japan during 1990-94 in much greater detail.
This much more disaggregated set of data comes from a survey admin¬
istered by the Nomura Research Institute (NRI) in collaboration with
the Users’ Committee of Foreign Semiconductors (UCOM), an organi¬
zation of the sixty-two major Japanese semiconductor users set up in 1988
to encourage increased access by foreign semiconductor suppliers to the
Japanese market.67 The respondents to the NRI survey account for about
two-thirds of Japanese semiconductor consumption.68 The NRI survey
uses a definition of “foreign” that appears to correspond to the F2 variant
favored by the Japanese government (that is, a definition based on the
label appearing on the finished chip).
The specific product areas accounting for the bulk of the foreign mar¬
ket share gain (each contributing half a percent or more and in the
aggregate approaching 7 percent of a 7.4 percent increase in foreign
market share) are ASICs (including gate array, full custom, and standard
cell logic chip types), microprocessors, Mask ROMs and other memory
(nonprogrammable and specialized memory chips), industrial ICs, stan¬
dard logic and application-specific standard products, and microper¬
ipherals. The story told by these data is also one in which increased
foreign penetration in specific product areas, rather than compositional
shifts in the market, is the force driving increased foreign market share—
6.7 percentage points out of a total increase in foreign market share of
7.4 percent. The greater disaggregation adds more complexity to the
picture, however. In both microprocessors and DRAMs, for example,

67. UCOM, affilated with the Electronics Industry Association of Japan, has sent trade
missions overseas to stimulate chip purchases, held seminars on applications and design
opportunities, and established a California liaison office to interface with smaller U.S.
suppliers unable to set up Japan-based organizations.
68. Surveyed companies purchased $13.4 billion in semiconductors in 1990 (based on
the NRI survey), compared with a total Japanese market of about $19.5 billion (based on
WSTS statistics)—or 69 percent of the market. For the second and third quarters of 1994
(the period covered by the NRI survey in 1994), surveyed companies purchased $9.77
billion compared with a total Japanese semiconductor market (WSTS data) of $14.96 bil¬
lion—or 65 percent of the market.
290 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

shifts in overall market demand toward these products were more im¬
portant than increases in foreign market share in accounting for the net
contribution of these products to the overall foreign share.
In fact, the detailed data shown here on the contribution of foreign
memory chip sales in Japan reveal a bit of a puzzle and allow an important
insight into the operation of the system for tracking foreign market share.
Adding up market share changes of the various memory chips on the list
of products in the NRI survey gives a net contribution of roughly
2 percent to foreign market share, about the same as the contribution
that U.S. firms alone made to foreign market share in this segment, as
shown in the analysis of WSTS data in the top part of table 5-5. At first
glance, this is perplexing because it was widely reported that there were
large increases in Korean DRAM shipments to Japan during this period,
so one would have expected the increase in foreign share in memory
chips to have been much larger than that corresponding to U.S. compa¬
nies alone.69
A good part of the answer may lie in the NRI survey definitions,
which, like the Japanese government’s preferred F2 measure, apparently
classified as foreign those products with non-Japanese labels. Newspaper
reports suggest that with increasing frequency Korean DRAMs were
being procured by Japanese companies for sale in Japan.70 If shipped in
assembled form to Japanese companies, then finished, stamped with a
Japanese brand name, and sold in Japan, these products would be class¬
ified as Japanese, using F2 (and in the NRI survey would not be counted
as foreign in table 5-5). They would, however, be counted as foreign by
the U.S.-favored FI formula.
Indeed, there is some good recent evidence that Japanese and Korean
vendors were deliberately structuring deals designed to take advantage
of the definitional complexities hovering over what was defined to be
“foreign” and what “domestic.” In March of 1995, NEC and Samsung
announced a curious deal in which Samsung would ship Korean-made
DRAMs to NEC in Japan, to be sold under NEC’s brand name, while
NEC’s Scottish factory would ship an equal volume of DRAMs to Sam¬
sung’s European assembly facility, where they would then be stamped

69. By 1994 imports from South Korea and Taiwan had risen to 15 percent of the
Japanese memory market. “Memory Chip Imports from ROK, Taiwan Rise,” Nihon Keizai
Shimbun, March 11, 1995, p. 1.
70. See for example Tetsuya Iguchi, “Japanese, Korean Rivals Forging DRAM Alli¬
ances,” Nikkei Weekly, October 18, 1993.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 291

with Samsung’s brand and sold in European markets. When investigated

by a market analyst, NEC sources were forthright in describing the ra¬
tionale for the transaction.71
The Samsung product sold by NEC in Japan would count toward
NEC’s purchases of foreign-made semiconductors under FI, scrutinized
by the Americans, even though NEC’s Japanese customers received the
familiar product under the same familiar NEC logo stamped on product
shipped to Japan from Scotland (which would not have counted as foreign
in the Americans’ eyes). From Samsung’s perspective, the NEC product
shipped by its European facility would be spared the stiff tariffs charged
on DRAMs exported to Europe from Korea, and in addition would curry
political favor with the European Community because of the “locally
made” semiconductor product it had begun shipping. Yet nothing signif¬
icant had fundamentally changed—NEC was producing the same quan¬
tities in its Scottish fab, and Samsung identical volumes on its Korean
production lines.
Similar types of deals had been reported publicly in the past, but none
so clearly targeting the market share calculations carried out under the
STA.72 This particular deal, to have begun in April 1995, is particularly
intriguing because of the sharp jump in both FI and F2 reported in the
third quarter of 1995—more than 3 percent in just one quarter.73 U.S.
industry sources suggested that a significant part of this gain appeared
to have been associated with a jump in Korean imports, as shown in the
trade statistics. One may speculate that branding swaps of locally fabri¬
cated chips from Japanese companies operating in Europe and Japan to

71. Jim Handy, “NEC to Get 4MB DRAMs from Samsung Electronics,” Dataquest,
“DQ Monday Report” (on-line data service), March 13, 1995. See also, “Samsung, NEC
in Euro DRAM Deal,” Electronic Engineering Times, February 13, 1995, p. 16.
72. For example, Korea’s Hyundai and Japan’s Fujitsu announced a 1993 deal under
which the Korean company would license its 4M DRAM chip to Fujitsu in exchange for
the right to acquire DRAM chips made at a U.S. Fujitsu plant and sell them under its own
brand name. Hyundai was thus able to avoid an 11 percent dumping tariff imposed by the
United States on its Korean-made DRAMs. See Don Clark, “Fujitsu and Hyundai Elec¬
tronics Form Unusual Joint Venture in Memory Chips,” Wall Street Journal, October 7,
1993; and Jim DeTar, “Fujitsu, Hyundai In DRAM Talks,” Electronic News, October 11,
1993, pp. 1-2. Korea’s Goldstar also signed a 1993 pact with Hitachi to build chips under
license from Hitachi’s design that it would then supply to Hitachi in Japan. See David
Lammers and Junko Yoshida, “Fujitsu and Hyundai in DRAM Pact,” Electronic Engi¬
neering Times, October 11, 1993, p. 1.
73. As the result of this deal with NEC and of other export arrangements, Samsung
was projecting a 50 percent increase in its exports to Japan in 1995, to reach a total of
130 billion yen. “Memory Chip Imports from ROK, Taiwan Rise,” p. 1.
292 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

be sold with a Korean label for Korean-made product to be sold in the

Japanese market with a Japanese label may have played some role in the
marked 1995 rise in foreign market share in Japan.
To summarize, the available empirical evidence does not particularly
favor the argument that shifts in the composition of Japanese demand,
and a growing U.S. technical lead in microprocessors, were the principal
sources of increased foreign share in the Japanese chip market. In fact,
relatively little of the increase in foreign market share, over any period,
was attributable to shifts in the composition of market demand. In every
period examined, the bulk of the change in aggregate foreign market
share was derived from increased foreign penetration of many individual
product niches. Over 1990-94 the largest contributor to increased foreign
market share seems to have been ASIC sales, an area where foreign firms
had no gross technical advantage over Japanese producers. Thus, if any¬
thing, the available empirical evidence appears to support the anecdotal
suggestions that Japanese consumers consciously substituted foreign
products where they might previously have chosen offerings from domes¬
tic suppliers.

Conclusions

All the evidence reviewed here suggests that a variety of evolving

policies associated with administration of the Semiconductor Trade
Arrangement had important effects on global markets for DRAMs and
EPROMs, the products subject to antidumping actions suspended as
a result of the negotiation of this accord. In 1987 and 1988, government
pressure on Japanese companies led to cutbacks in both production of
and investment in these products. Output dropped and prices rose.
In response to foreign complaints about continued third-country
dumping, a border control regime was established in an effort to reduce
exports and increase export prices. Guidance was reportedly given to
Japanese producers on desirable levels of exports to different foreign
regions. Measures were taken to block flows from the secondary or gray
market inside Japan to foreign customers. The largest producers’ direct
dealings with the largest foreign customers were more easily monitored
and controlled.
Available data show that in the contract market, which accounts for
the bulk of sales, and where dealings between a few large producers and
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 293

a relatively small number of large customers account for most volume, a

persistent differential between Japanese and foreign prices could be ob¬
served from 1987 through early 1989. In the gray market, on the other
hand, in spot sales, which were inherently much more difficult to control,
Japanese and foreign prices after some lag basically converged (but
soared far above contract prices).
It is hard if not impossible to calculate the welfare impact of produc¬
tion and investment restraints, because it is difficult to be precise about
what would have been produced absent these restrictions. It is easier to
produce rough estimates of the welfare impacts of one isolated element
of STA-related administrative measures, namely, border controls, and
the resulting regional price differentials. Our results suggest a negative
impact on welfare in the rest of the world amounting to perhaps 5 to 10
percent of regional consumption, over 1987-88, and a positive impact on
Japanese welfare (with benefits derived from lower prices to Japanese
consumers outweighing export profits forgone by producers).
Differences in the behavior of supply and demand between DRAM
and EPROM markets differ mainly in the emergence of a significant price
differential between Japan and the rest of the world in DRAMs (Japanese
buyers paid less than buyers elsewhere for these products, but not for
EPROMs). I have argued that this phenomenon is probably related to
sharp declines in Japanese EPROM production and soaring imports of
EPROMs into the Japanese market. Japanese producers may have held
a lower market share in EPROMs than in DRAMs, but it is hard to see
how this necessarily leads to reduced production and rising imports in
Japan. The circumstances of this decline are consistent with the sugges¬
tion that the Japanese retreat in EPROMs may have been a deliberate
political decision.
The analysis of available data on the growth of foreign companies’
share of the Japanese semiconductor market supports the notion that
pressures associated with the STA may have played an important role in
increasing foreign sales. Only a small part of the increase in foreign
market share could be attributed to shifts in the composition of Japanese
demand toward products in which foreign firms already had a greater
presence; most of the increase is clearly due to greater foreign penetra¬
tion into individual product niches. Furthermore, a large part of the
growth in foreign market share was associated with ASICs and memory
chips, where foreign firms enjoyed no huge technical advantages relative
to Japanese firms.
294 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Appendix 5-A: The Economics of Contract Pricing

Given that perhaps 70 percent of DRAM sales are initially made as

direct “contract” sales to large OEM users, it is useful to survey the
nature of these contracts in some detail.

The Nature and Function of Long- Term Semiconductor Contracts

Typically “contract” sales are commitments to supply some quantity

of parts, at some specified price, beginning on one specified future date
and ending on another. At first glance, then, they might seem to be a
special form of a continuing forward contract, that is, a legally binding
promise to deliver product on some future schedule of dates at a sin¬
gle specified price. Further investigation suggests this is not the case,
however.
Although these agreements are widely referred to within the industry
as long-term contracts, they rarely seem to be legally binding commit¬
ments. The prices specified generally appear to hold when shipments
under the contract begin, but often they do not persist over the life of
the contract. Many contracts contain explicit provisions for renegotiation
of the price downward, at the purchaser’s option, in response to changing
market conditions. In other cases, purchasers often successfully demand
downward price adjustments even when the contract makes no explicit
provision for such adjustments.74
Furthermore, because the system of price floors for DRAMs put into
place by the U.S. Commerce Department in 1986 specified that the
prevailing floor price at the time a legally binding contract is drawn up
and signed shall remain in force throughout the life of the contract, there
was an additional disincentive to producing a formal, legally binding
document. To do so would have locked a purchaser into current prices,
despite expected future declines in DRAM prices.
On the other hand, suppliers generally seem to respect contract prices
as a de facto ceiling on prices charged their customers. This is generally
irrelevant for products experiencing continuing price declines, such as
DRAMs. But when the historically unprecedented increase in memory

74. An interesting compendium of DRAM contract “horror” stories—users and pro¬

ducers repudiating oral and written price commitments in response to changing market
conditions—may be found in U.S. International Trade Commission, 64K Dynamic Random
Access Memory Components (June 1986) pp. A75-A88.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 295

chip prices of 1987-88 occurred, preexisting contract prices were often

respected and not raised (although some purchasers apparently did face
cancellation or reduction of prescribed contract volumes at the negotiated
price, after being informed of shortfalls in production). Thus, the long¬
term contract price may have served to constrain price increases over the
life of preexisting contracts.
If contract prices are generally not legally or practically binding much
beyond the original beginning date for the contract, what then is the
purpose of entering into these informal, “handshake” commitments?
Interviews with OEM purchasing managers suggest that ensuring the
quantity of DRAMs to be purchased from suppliers is the major objec¬
tive. In fact, purchasers frequently suggest that the critical issue in times
of extreme shortage is not pricing, but getting adequate supplies. How¬
ever, if one is willing to enter the spot market and pay the going price,
one can always bid away scarce supplies from other users. But this may
create other significant costs for the chip consumer that extend beyond
the purchase price. Additional qualification costs or extensive additional
testing may be required for purchases from new sources or gray market
suppliers. Greatly increasing the number of sources of supply probably
also greatly increases purchasing and qualification costs.
Producers of DRAMs face a different logic. Significant learning curve
effects imply a sharp increase in the output of any given initial investment
in DRAM production capacity over the product life cycle of that generation
of DRAM, but there is considerable uncertainty about the exact timing and
magnitude of yield improvements within the industry, and hence about
aggregate supply. The producer must also be concerned about volatility in
demand for the increasing quantities of DRAMs that will be flowing from
existing fabrication lines in future periods. Quantity commitments lock
purchasers into deliveries of a given producer’s output and reduce the odds
that large volumes of chips emerging from ever more productive factories
will have to be liquidated in the gray market at a discount (in part reflecting
new users’ additional qualification and testing costs).

Empirical Evidence on DRAM Contracts

It is possible to examine empirically the nature of DRAM contracts

and possible effects of the STA using a unique data source. Industry
sources provided me with confidential data on OEM DRAM contracts
covering the 1985-89 period. The data are drawn from contracts nego-
296 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Table 5-A1. Distribution of DRAM Contracts to Start (Lead) and

Length, 1984-89
1M Contracts 256K Contracts 64K Contracts

Months Frequency Frequency Frequency

leadllength lead Length lead Length lead Length

July 1986 to Feb. Oct. 1984 to Feb. Jan. 1985 to Feb.

1989 (N = 128) 1989 (N = 174) 1989 (N = 83)
0 0.29 0.41 0.43
1 0.38 0.28 0.14
2 0.04 0.01 0.08 0.01 0.04 0.01
3 0.09 0.30 0.12 0.14 0.19 0.10
4 0.08 0.02 0.09 0.03 0.10
5 0.02 0.02 0.06 0.01
6 0.05 0.42 0.37 0.04 0.22
7 0.05 0.02 0.02 0.08 0.14
8 0.04 0.02 0.01
9 0.01 0.02
10 0.02 0.02 0.01
11 0.03 0.01
12 0.16 0.24 0.46
13 0.01
14 0.01
15 0.03
17 0.01

Oct. 1984 to Aug. Jan. 1985 to Aug.

1986 (N = 66) 1986 (N = 37)
0 0.47 0.54
1 0.29 0.16
2 0.14
3 0.09 0.09 0.08 0.19
4 0.02 0.06 0.11
5 0.05 0.03
6 0.55 0.08 0.24
7 0.06 0.14

tiated by a small number of European and North American OEMs; the

bulk of the reported contracts refer to purchases by European users.
Characteristics of the contracts that were collected include negotiation
date, start date for shipments, period over which shipments are to be
delivered, total quantity commitments over this period, contract price,
nationalities of chip vendors and purchasers, chip organization and pack¬
aging, and chip speed (access time). After discarding contracts for which
speed measures were unavailable (or that covered parts with a mixture
of speeds); parts that used packages other than plastic dual in-line pin
(DIP), plastic leaded chip carrier (PLCC, in the case of 256K DRAMs),
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 297

IM Contracts 256K Contracts 64K Contracts

Months Frequency Frequency » * Frequency
lead/length lead Length lead Length lead Length
Oct. 1984 to Aug. Jan. 1985 to Aug.
1986 (N = 66) 1986 (N = 37)
9 0.02 0.03
12 0.14 0.41
15 0.03
17 0.02

Sept. 1986 to Feb. Sept. 1986 to Feb.

1989 (N = 108) 1989 (N = 46)
0 0.37 0.35
1 0.28 0.13
2 0.05 0.02 0.07 0.02
3 0.14 0.18 0.28 0.02
4 0.13 0.01 0.09
5 0.09 0.02
6 0.27 0.20
7 0.04 0.09 0.15
8 0.03 0.02
9 0.01 0.02
10 0.02
11 0.05 0.02
12 0.31 0.50
14 0.02
15 0.03

Memorandum: Test for same

distribution in both subperiods

Chi-squared statistic 14.6 33 13.7 10.7

Degrees of freedom 5 13 6 9
Source: Author’s calculations based on unpublished data provided by confidential industry sources.

and small outline cases (SO, in the case of 1M DRAMs);75 and chips
with relatively uncommon organizations,76 a sample of 83 agreements for
64K DRAMs, 174 for 256K DRAMs, and 128 for 1M DRAMs remained.
The distribution of these contracts by lead time (months from nego¬
tiation date to start date) and duration (months from start date to end
date) is shown in table 5-A1. For 64K and 256K DRAMs the contract

75. These involved a small number of observations divided among a relatively large
group of other packages. Only 256K DRAMs with access times of 120 nanoseconds or
faster were packaged in PLCC cases in the contracts in this sample.
76. Chip organizations other than 64K x 1, 256K x 1, 1M x 1, or 256K x 4 (the
latter two are 1M DRAM types) appeared only in a relatively small number of contracts
in my sample.
298 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

distributions are shown for periods before and after September 1986,
when the STA went into force. (Only two of the contracts for 1M chips
in this sample were negotiated before September 1986, and no such
disaggregation is shown.) For 64K chips, all contracts in the sample were
for parts with a 64K x 1 organization and plastic DIP packaging.77 For
256K DRAMs, contracts covered both DIP and PLCC packages.78 The
1M contracts include parts with both 256K x 4 and 1M x 1 organiza¬
tions, and both DIP and SO cases, which were all found in abundant
numbers. This proliferation of chip types demonstrates a trend toward
increasing product variety in the DRAM market.
It is readily apparent that delivery on the vast bulk of these contracts
begins within a very short period after their negotiation. The contract
lengths cluster around three, six, and twelve months’ duration. More
than 40 percent of the contracts for 64K and 256K DRAMs could be
considered “spot,” in that shipments were scheduled to begin in the
month they were negotiated; a smaller but still substantial fraction
(28 percent of the 256K shipments; 14 percent of the 64Ks) were to begin
in the month following the contract’s negotiation. The great bulk of the
remaining contracts for 64K and 256K parts began within two to four
months, except for a small number of outliers clustered at five to seven
months.
Contracts for 64K and 256K DRAMs negotiated after September 1986
tended to have longer lead times and to last longer than contracts before
that date. A formal chi-square test comparing pre- and post-STA distri¬
butions generally confirms these casual impressions.79 The period prior
to the signing of the STA was one in which markets saw abundant supplies
and generally declining prices, while 1987 and 1988 were generally
marked by firm or rising prices and tightening supplies. Not surprisingly,

77. Four contracts for parts in ceramic cases were discarded from the sample; the rest
of the discards were also in plastic DIP packaging.
78. A small but nontrivial number of contracts for parts with 64K x 4 organizations
or for ZIP, ceramic, or SIMM packaging were dropped from the sample as well.
79. Both the lead times and lengths of contracts signed for the newer 256K DRAMs
seem to have changed after the STA was signed, whereas lead times shortened but contract
lengths did not seem to change in transactions involving the more mature 64K products.
For 256K DRAMs we reject the hypothesis that these contracts were drawn from identical
populations of lead times at the 5 percent significance level, and we reject a similar hy¬
pothesis on contract lengths at the 1 percent level. For 64K DRAMs we reject the hypothesis
of a single distribution of lead times at the 5 percent level, but we are unable to reject the
hypothesis of a common distribution of contract lengths at even the 10 percent significance
level.
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 299

1M DRAM prices, almost all of which are based on negotiations after

September 1986, more closely resemble the general patterns of distribu¬
tion and clustering seen in contracts for 64K and 256K DRAMs over that
same period rather than the earlier period. Thus DRAM contracts seem
to have been drawn up earlier, and to have covered longer time spans,
after the STA went into effect.
Elsewhere I have studied the prices embedded in these contracts using
a simple econometric model in which these deals are viewed as forward
contracts.KU This econometric analysis of DRAM contract price data for
three successive generations of memory chips supports several general
propositions.
First, the simple model of the term structure of contract prices em¬
ployed seemed quite consistent with these data: formal statistical tests
did not lead to rejection, and estimated coefficients were largely unaf¬
fected by the term structure assumed. The point of departure was a model
in which bias (deviation from the expected price in a future period) in
contract prices serves to compensate purchasers for the transfer of risk
to them by producers. In this model the impact of contract lead time on
contract price reflects the “bias” in forward prices. The empirical results
supported the presence of an expected negative bias (“normal backwar¬
dation,” to use the term coined by John Maynard Keynes) in forward
contract prices for DRAMs.
Second, the econometric results showed a rather dramatic decline of
the bias term from the 64K generation of DRAMs, to the 256K genera¬
tion, to the 1M generation, suggesting that the market viewed prices for
later generation chips as considerably less volatile than previous genera¬
tions of chips over the 1985-89 period. This, of course, was precisely
what the administrative pricing guidelines and mechanisms imposed on
the DRAM market with the advent of the STA would have been expected
to accomplish, so these estimates suggest that the market believed that
the STA actually did make DRAM prices less volatile.
Third, my suggestion that these contracts, beyond the initial purchase
at the contracted price, mainly represent quantity commitments is sup¬
ported by the generally small magnitudes and statistical insignificance of
the effects of contract length as a determinant of contract pricing.
Fourth, analysis of price differentials faced by American and Euro¬
pean purchasers of DRAMs suggested much smaller differentials than

80. See Flamm, “Measurement of DRAM Prices.”

300 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

had been indicated elsewhere, and I could not reject the null hypothesis
of no regional differences between the United States and Europe.
Fifth, estimated quantity discounts were very small and statistically
insignificant. Other available data on contract pricing, disaggregated by
size of transaction, also seem to show little systematic relationship be¬
tween price and contract size.81 Since interviews with purchasing man¬
agers at computer companies in 1989 frequently suggested that the largest
companies often did pay lower contract prices, I interpret this to im¬
ply that the overall size of the account, rather than the size of one
sales transaction for a single type of chip, determined any applicable
discounts.82
Sixth, a general pattern of Korean vendors selling their product at
somewhat higher prices emerged, consistent with anecdotal observations
by market participants.83 In effect, transactions with Korean producers
probably more resembled a short-term, spot market-like exchange than
the less volatile, long-term agreements generally characterizing contracts
between large purchasers and producers. During a period of scarcity, the
Korean producers appear to have charged a higher price, above the long¬
term contract price and approaching the spot gray market price. During
a period of relative glut, on the other hand, we would expect this spot
price to lie below the long-term contract price. This analysis is consistent
with the reports in the trade press on Korean producer Samsung’s deal¬
ings with its American distributors.84 It is confirmed by a U.S. ITC

81. See U.S. International Trade Commission, 64K Dynamic Random Access Memory
Components, 1985, pp. A40-A45.
82. A 1992 U.S. International Trade Commission report suggests that contract sales
may reasonably be divided into large, “Tier 1“ premium OEM accounts, and a second tier
of smaller accounts, including franchise distributors and value-added resellers. The report
actually suggests that premium customers may pay higher prices, because of their more
demanding and lengthy qualification requirements. See U.S. International Trade Commis¬
sion, DRAMs of One Megabit and Above, p. A-43. Since this report covered a period of
slack demand in which prices were generally declining quite sharply, one might interpret
this to mean that second-tier contracts were shorter term, more like spot transactions.
83. In all likelihood a single Korean firm—Samsung—was responsible for all of the
Korean product shipped in this sample of contracts.
84. At the peak of the DRAM shortage, in the summer of 1988, Samsung attempted
to hike its prices to levels that its American distributors protested left them uncompetitive,
and ended its ship-from-stock-and-debit policy, which effectively let distributors hold price-
protected inventory. The distributors had thus been able to capture the benefit from rising
prices by maintaining large inventories of product, without the downside risk that falling
prices would pose with a large inventory. The practice was reinstated when the market
slowed down in early 1989. See Electronic News, August 15,1988, p. 47; February 27,1989,
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 301

investigation, which found that all sales of Korean DRAMs to U.S.

importers were made on a spot basis.85

Appendix 5-B: Construction of Estimated FMVs

This appendix briefly describes the relationship of foreign market val¬

ues (FMVs) to cost concepts reported to the U.S. Commerce Department
by Japanese producers. Producers were required to file quarterly sub¬
missions containing extensive cost data, part of which were reported in
ranged form (reported value within plus or minus 20 percent of actual
value) in public versions available for inspection at the Commerce
Department.86
Cost of Manufacturing (COM). The first major subtotal reported was
cost of manufacturing, equal to the direct materials, subcontract expense,
direct labor cost, and production overhead expense accrued in the suc¬
cessive manufacturing stages of wafer fabrication, assembly, and final
test.87 When divided by number of good chips yielded by the manufac¬
turing process, this generated a per unit manufacturing cost subtotal. To
this was added an imputation for R&D cost calculated by taking the ratio
of all semiconductor-related R&D in that quarter to the total cost of
manufacturing all semiconductors that quarter (excluding R&D) and
multiplying this ratio by the per unit manufacturing cost subtotal just

p. 27; April 3, 1989, p. 35. When prices turned down sharply in early 1990, Samsung
shocked its American distributors by doing away with the customary “price protection”
altogether, so that the effective prices paid by authorized distributors would more closely
follow the ups and downs of the spot market (and more of the price risk would be transferred
to the distributors). American distributors complained bitterly about Samsung’s “broker
mentality” (that it was behaving like a broker, passing on the full volatility of gray market
spot pricing into its contracts with distributors, rather than moderating price swings with
its contract pricing practices). See Bob Ferguson, “Samsung Notifies Distributors: Ship and
Debit Will End Feb. 1,” Electronic News, January 22, 1990, p. 34; Ferguson, “Samsung
Terminates Six Distributors,” Electronic News, February 5, 1990, p. 38; and Ferguson,
“Brajdas Files Suit vs. Samsung over Inventory Returns, Credits,” Electronic News, July
2, 1990, p. 32.
85. See U.S. International Trade Commission, DRAMs of One Megabit and Above,
p. A-45.
86. This appendix is based on SIA, Antidumping Law Reform-, and the author’s inter¬
views with Commerce Department officials and industry sources in 1990. The text describes
the state of practice about 1990; in the first few quarters after the STA went into effect
there was some variation as these procedures were developed.
87. Production overhead expense included depreciation of capital equipment, allocated
across product lines according to generally accepted accounting practices.
302 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Table 5-B1. Foreign Market Value (FMV) Projected Cost Estimates for
Japanese DRAM Manufacturers, by Company, Selected Quarters,
1986-89°
Dollars per chip

Year and _____ Maximum Unweighted

quarter Mitsubishi NEC Fujitsu Hitachi Toshiba + 25%b average

256K DRAMs
1987:1 1.86 2.33 1.86
1987:2 1.87 2.48 2.08 3.10 2.14
1987:3 2.65 1.76 2.57 2.45 3.31 2.36
1987:4 2.68 2.69 2.19 2.10 3.37 2.42
1988:1 2.25 1.49 1.89 1.36 2.81 1.75
1988:2 1.69 2.31 1.98 1.84 2.89 1.96
1988:3 1.72 2.50 1.89 2.19 3.12 2.07
1988:4 2.07 2.40 2.15 2.23 2.99 2.21
1989:1 2.47 2.10 1.95 2.15 3.09 2.17
1989:2 2.39 2.49 2.11 2.20 3.11 2.30
1989:3 2.34 2.31 2.30 2.02 2.92 2.24
1989:4 2.03 2.07 2.19 1.88 2.74 2.04
1.97 1.58 2.06 2.02 2.74 3.43 2.07
1.93 1.78 1.87 2.80 3.50 2.10
2.90 2.14 1.75 2.16 2.03 3.62 2.20
1990:4 2.15 2.16 1.59 1.71 2.71 1.90
2.51 1.78 2.34 1.89 3.14 2.13
2.02 2.98 1.99 3.73 2.33
1991:3 2.13 1.92 4.39 1.80 5.49 2.56

described. The R&D counted by the Commerce Department was gen¬

erally considerably greater than that reported to MITI in compiling Jap¬
anese statistics on IC R&D, including both corporate-level R&D and
research grants to universities. The resulting number was per unit cost
of manufacturing.
Sales, General, and Administrative Expenses (SG&A). The next cat¬
egory of expense was dealt with in a manner analogous to R&D. A ratio
of SG&A to cost of manufacturing was calculated for all semiconductor
sales, then multiplied by COM. This ratio was not permitted to fall below
a minimum floor of 10 percent of COM.
Cost of Production (COP). Adding SG&A to COM yielded an esti¬
mate of cost of production.
Profit. A profit rate of 8 percent was applied to COR This practice
differed from typical antidumping cases (in which an 8 percent profit rate
was generally taken as a minimum).
EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT 303

Company
Year and Maximum Unweighted
quarter Mitsubishi NEC Fujitsu Hitachi Toshiba + 25b% average

1M DRAMs
1987:1 23.75 29.69 23.75
1987:2 22.14 13.31 27.67 17.72
1987:3 15.67 15.96 11.34 9.47 19.95 13.11
1987:4 13.32 9.78 11.85 8.41 16.65 10.84
1988:1 10.21 5.92 13.98 6.81 17.48 9.23
1988:2 8.10 6.90 8.48 6.44 10.59 7.48
1988:3 9.22 7.51 8.20 11.52 8.31
1988:4 8.76 6.20 8.76 10.95 7.90
1989:1 9.45 5.44 6.33 7.10 11.82 7.08
1989:2 9.10 6.09 7.84 6.98 11.37 7.51
1989:3 4.54 6.61 5.06 8.27 5.40
1989:4 3.65 4.68 4.71 5.88 4.35
6.33 3.71 5.62 5.62 7.91 5.32
5.48 5.72 4.87 4.89 7.15 5.24
4.70 5.22 4.89 4.14 2.93 6.52 4.38
1990:4 5.26 4.35 3.90 3.22 6.58 4.18
3.31 3.97 3.79 4.06 5.08 3.78
4.98 3.89 4.44 3.83 3.33 6.23 4.09
1991:3 4.87 4.35 3.95 4.86 6.09 4.51
Source: Based on data taken from unpublished company submissions to the U.S. Department of Commerce.
“FMV” date is the quarter after the quarter to which the projected cost applies and two quarters after the quarter
in which the projection was submitted.
n.a. Not available.
a. Cost concepts used are, by company:
256K DRAMs: Mitsubishi, export sales price (ESP) for M5M4256AP type chip; NEC, average of constructed
value (CV) for 41256P, 41256CF, and 41256L type chips; Fujitsu, CV for MB81256/7P type chip; Hitachi, average of
CV for 50256P NMOS and CV for 51258P CMOS type chips in dual in-line package; Toshiba, average of ESP for
41256AP and 41256P type chips for 1987:2 to 1989:1 and 41256AP only for 1989:2 to 1989:4 and ESP for 256K x 1
NMOS in dual in-line package for 1990:1 to 1991:3.
1M DRAMs: Mitsubishi, ESP for M5M41000P for 1987:1 and ESP for M5M41000AP for 1987:3 to 1991:3; NEC,
average of CV for 421000C/421001C/421002C, 421000LA/421001LA/421002LA and 421000V/421001V/421002V for
1990:2 to 1991:3; Fujitsu, CV for MB81C1000/1P chip type; Hitachi, CV for lMxl CMOS chip type in dual in-line
package; Toshiba, ESP for TC511000P for 1987:2; average for TC511000P and TC511000AP for 1987:3 to 1988:3 and
ESP for 1M CMOS DRAM in dual in-line package for 1989:1 to 1991:2.
b. If reported value is 20 percent less than true value, true value is 25 percent greater than reported value.

Constructed Value (CV). Profits were added to COP to produce a

constructed value (CV). To a close approximation, constructed value was
taken as an estimate of the fair value of the product. Minor adjustments
(which could be positive or negative) for selling expenses were then
made, depending on whether the product was sold to the customer after
entering the United States or before. In the former case the adjusted CV
was called exporters’ sales price (ESP); in the latter, purchase price (PP).
304 EFFECTS OF THE SEMICONDUCTOR TRADE ARRANGEMENT

Because the adjustments were generally small, ESP and PP were essen¬
tially the same value as CV.
Foreign Market Value (FMV). The floor price for a company’s DRAM
exports was related to these cost concepts in the following manner. Thirty
days after the beginning of a quarter, companies were required to submit
data showing actual costs in the previous quarter and projected costs for
the current quarter. The de facto practice in the Commerce Department
was to require the use of the previous quarter’s actual costs as projected
costs for the current quarter, because those were the only numbers that
could be based on objective calculations using actual data and did
not rely on hypothetical assumptions about cost declines and yield
improvements.
Projected ESP and PP for the current quarter were used to set FMVs
for sales delivered to a customer inside or outside the United States for
the next quarter. Thus FMV in quarter t was approximately equal to
projected cost in quarter t — 1. This in turn was approximately equal to
actual reported cost in quarter t — 2.
Estimates of Company FMVs. Public submissions contained at least
some ranged data on the above cost concepts for Mitsubishi, NEC,
Fujitsu, Hitachi, and NEC. FMVs for Matsushita, Oki, and Texas In¬
struments Japan were calculated as a single weighted average of the FMVs
for the five larger producers. Product types and reported data changed
over time for many companies. Although some data were reported in
each quarter for every company, reporting was not consistent. Table
5-B1 reports estimates of FMV based on projected costs for each of these
companies where such a number could be constructed. In all cases the
estimated FMVs are based on prior-quarter projections of CV or ESP.
Where multiple versions of these chips were accounted for on separate
cost tables, the numbers reported are an unweighted average of available
data on unit costs for these different chip types.
 CHAPTER SIX

Dumping in DRAMs

Dumping has traditionally been defined as selling at a lower price in a

foreign market than in one’s home market. Since the mid-1970s, however,
a different concept—that of sales at a cost less than a constructed “fair
value”—has become an alternative standard, embodied in U.S. trade
law, for finding that imports are being dumped in the U.S. market.1 It
has been estimated that, since 1980, about 60 percent of all dumping
cases have been based on charges that foreign firms were selling at a
price below some constructed average cost.2 Perhaps the most widely
publicized application of this standard has been in the case of imports of
dynamic random access memory (DRAM) chips, the most important
type of semiconductor in terms of dollar value of sales. Micron Technol¬
ogy’s 1985 petition for relief from dumping of Japanese 64K DRAM
imports specifically acknowledged that prices for these chips in the U.S.
market may actually have been marginally higher than prevailing prices
in the Japanese market.3 Micron instead based its complaint on the
charge that Japanese chips were being sold at prices that did not cover
their full costs of production.

1. A brief history of the origins of this new standard may be found in Pietro Nivola,
“Trade Policy: Refereeing the Playing Field,” in Thomas E. Mann, ed., A Question of
Balance: The President, the Congress, and Foreign Policy (Brookings, 1990), pp. 229-30;
and Gary N. Horlick, “The United States Antidumping System,” in John H. Jackson and
Edwin A. Vermlost, eds., Antidumping Law and Practice: A Comparative Study (University
of Michigan Press, 1989), pp. 133-34.
2. Horlick, “United States Antidumping System,” p. 136.
3. See Micron Technology’s petition to the U.S. Commerce Department, International
Trade Administration, and the U.S. International Trade Commission, Petition for the Im¬
position of Antidumping Duty (June 1985), pp. 11-14.

305
306 DUMPING IN DRAMS

As described in chapter 4, the U.S. government expanded the inves¬

tigation of Japanese DRAM dumping to include 256K and 1M DRAMs.
Folded into investigations of U.S. industry charges of dumping of eras¬
able programmable read only memory (EPROM) chips, the matter cul¬
minated in the controversial U.S.-Japan Semiconductor Trade Arrange¬
ment (STA) of 1986. One of the outcomes of the STA was a system of
floor prices for Japanese DRAM and EPROM imports administered by
the U.S. Commerce Department, based on the calculation of a so-called
foreign market value (FMV), which was derived from the “fair value”
constructed cost comparisons enshrined in the dumping provisions of
U.S. trade law.4
Although the FMV calculations were dropped from the 1991 successor
agreement to the STA, Japanese producers were required to continue to
collect the relevant data, to facilitate a “fast response” dumping inves¬
tigation should the need arise. Thus, one may surmise that the implicit
threat of a dumping investigation continued to give the FMV calculation
a significant, if shadowy role in determining lower bounds on pricing of
Japanese chip exports to U.S. (and possibly third-country) markets.
The idea of requiring producers to maintain prices at or above some
concept of full long-run average cost is hard to defend, either as a positive
description of what a profit-maximizing producer in a “competitive”
market would choose to do, or as a normative guide for efficient resource
allocation. It is, however, possible to construct an economically coherent
argument that pricing below marginal cost can serve as a warning signal
of “strategic” behavior by producers, which in some circumstances can

4. According to U.S. law, dumping is defined as sales of imports at “less than fair
value” that cause “material injury to a U.S. domestic industry.” To determine the fair value
to which U.S. import prices are to be compared, the Department of Commerce is instructed
by law to determine foreign market value (FMV) using sales prices in the exporter’s home
market. If the data are insufficient, Commerce must then use prices for sales to third
countries. If these data too are inadequate, Commerce must calculate a “constructed value”
to estimate the FMV. The statute says that actual foreign sales data may be disregarded
when, first, they fall below total costs of production over an extended period of time and
in substantial quantities, and second, they do not permit full recovery of all costs within a
reasonable time and in the normal course of trade. The department is reported to take the
“extended period” as its investigation period, “substantial quantities” as 10 percent of sales
by volume, and generally to ignore the second requirement. See N. David Palmeter, “The
Antidumping Law: A Legal and Administrative Nontariff Barrier,” in Richard Boltuck and
Robert E. Litan, eds., Down in the Dumps: Administration of the Unfair Trade Laws
(Brookings, 1991), pp. 71-75. Also see Tracy Murray, “The Administration of the Anti¬
dumping Duty Law by the Department of Commerce,” pp. 23-27 and 38-40, in this same
volume.
DUMPING IN DRAMS 307

justify policy intervention by government. However, in an industry sub¬

ject to learning economies (where unit production cost falls with cumu¬
lative production experience), producers may rationally choose, for com¬
pletely “competitive,” nonstrategic reasons, to engage in “forward
pricing,” that is, to sell at a price that is below current marginal cost.
Is below-marginal-cost pricing for nonstrategic reasons empirically rel¬
evant in the semiconductor industry? Is it reasonable to defend even some
revised version of a constructed cost test for dumping, based on marginal
cost, as a reasonable trip wire for government scrutiny of possible stra¬
tegic behavior by foreign producers? Perhaps most interesting, what can
we deduce about the relationship between price and production costs
using a minimally realistic model of the product life cycle when large up¬
front investments in capacity constrain output, large and relatively fixed
investments in R&D create economies of scale, and learning economies
are likely to be significant? How is an “FMV-like” system likely to con¬
strain producer behavior in these circumstances? Because these charac¬
teristics are typical not just of semiconductor manufacture but of a broad
range of high-technology products, the answers to these questions, and
the methodology used in the inquiry, are of some importance.
This chapter lays out a simple analytical framework that can be used
to compare the time path of output and prices in a nonstrategic, com¬
petitive (open-loop Cournot-Nash equilibrium) semiconductor industry
with variants of constructed fair value that would be associated with the
same path for output. The model is applied with empirically based pa¬
rameters associated with 1M DRAM chip production, to explore how
pricing of semiconductors is likely to be constrained, over the product
life cycle, by constructed values in the form of FMV-type pricing rules.
The basic model should also prove useful in analyzing many other inter¬
esting questions about the potential impact of public policies on high-
technology industries subject to scale and learning economies.

The Economic Rationality of Below-Marginal-Cost Pricing

To a first approximation, stripped of a variety of practically important
administrative and accounting issues, the FMVs constructed by the U.S.
Commerce Department resemble an economist’s concept of average cost
of production, plus a fixed 8 percent markup that ostensibly reflects
308 DUMPING IN DRAMS

“normal” profit.5 (This completely arbitrary markup is ignored in the

rest of this discussion.)6 A result taught in nearly any introductory eco¬
nomics course is that under some circumstances (for example, a downturn
in demand) it can be economically rational for a producer in a competi¬
tive industry to sell at a price less than full average cost, as long as short-
run marginal cost is covered. As long as a firm at least covers the variable
costs of running a production line, and the cost of producing an incre¬
mental unit on that line, it makes economic sense to continue operating,
even if revenues are insufficient to recover the full historical cost of the
initial investment in developing the product and building the factory.
Thus most economists would find it entirely normal that, over at least
some periods, observed prices fall short of full (long-run) average cost of
production.7 Any policy measure that prohibited marginal cost pricing
by foreign exporters, while leaving domestic producers unaffected,
would—if it actually affected market outcomes—increase domestic pro¬
duction at the expense of domestic consumers and (possibly) foreign
producers. It would also arguably deny foreign producers national treat¬
ment, forbidding them the right to engage in economic behavior permit¬
ted of domestic firms.
If price falls short of the full average cost of production over a long
period, of course, one may safely predict that some firms will exit the
industry, and the industry will shrink to the point that the remaining
firms at least recover full average costs over the life of their sunk invest¬
ments. Thus, if sustained dumping according to constructed “fair values”
(pricing below long-run average cost) is observed in a competitive indus¬
try, one may generally infer that excess capacity exists, and that exit will
follow. But the observed pricing behavior may still reflect normal, com¬
petitive behavior on the part of those firms pricing below FMVs.
Is there ever any economic justification for remedial policies triggered
by selling below a constructed FMV? The point at which many econo¬
mists would agree that something other than “competitive,” nonstrategic

5. See chapter 4, appendix B. The administrative and accounting issues so casually

passed over here are widely believed to systematically bias the system toward findings of
dumping and toward higher dumping margins. Richard Boltuck and Robert E. Litan,
“America’s ‘Unfair’ Trade Laws,” in Boltuck and Litan, Down in the Dumps, pp. 13-14.
6. Also ignored is a requirement that overhead costs of no less than 10 percent be
added to materials and fabrication costs in calculating cost of production.
7. See, for example, Alan V. Deardorff, “Economic Perspectives on Antidumping
Law,” in Jackson and Vermlost, Antidumping Law, pp. 30-33, for further elaboration on
this point.
DUMPING IN DRAMS 309

behavior might be going on occurs when a firm’s price falls short of its
short-run marginal cost or, even more obviously, average variable cost
(which bounds short-run marginal cost from below over the relevant
range).8
In considering why a firm might rationally choose to produce and sell
at a price that fails to cover the current marginal cost of production, it is
helpful to distinguish between “strategic” and “nonstrategic” behavior.
I shall label a firm’s behavior “strategic” when it explicitly takes account
of the impacts of its decisions on the behavior of other economic agents,
and “nonstrategic” when decisions consider the actions or choices of
other agents as fixed, unaffected by one’s own actions. Predatory pricing,
as alleged by the U.S. semiconductor industry of its Japanese rivals
(described in chapter 3), can be classified as a form of strategic behavior.
One possible explanation, consistent with nonstrategic behavior, for
pricing below marginal cost is that increased production may lower a
firm’s future production costs, through learning effects (discussed be¬
low). In this case, measured current marginal cost overstates “true”
marginal cost, which should take into account the cost-reducing effects
of current production on future output.9
But another possible explanation for behavior of this sort is a strategic
motive on the part of the dumper: either predation, defined here as
actions intended to encourage other firms to exit the industry; limit pric¬
ing, intended to discourage entry by other firms; or a defensive response
against predatory behavior by others.10 In this case the rents from the

8. See Janusz A. Ordover and Garth Saloner, “Predation, Monopolization, and Anti¬
trust,” in Richard Schmalensee and Robert D. Willig, eds., Handbook of Industrial Or¬
ganization, vol. 1 (North-Holland, 1989), pp. 579-90, for a detailed survey of the literature
on tests for predatory behavior.
9. Such learning economies can also be used as a strategic instrument, with a firm’s
production decisions taking into account the impact of its learning on the actions of its
rivals. See Drew Fudenberg and Jean Tirole, “Learning by Doing and Market Perfor¬
mance,” Bell Journal of Economics, vol. 14 (Autumn 1983), pp. 522-30, for such a model.
Deardorff, “Economic Perspectives on Antidumping Law,” pp. 37-38, points out that
low-priced sales designed to build brand loyalty or otherwise alter consumer preferences
might also rationally lead a producer to sacrifice current profitability for future rents, and
to price below marginal cost. In effect, greater current output shifts future demand sched¬
ules, and current marginal revenue then understates “true” marginal revenue. Such
“demand-side learning effects,” however, may be considered a form of strategic behavior,
since they are designed to alter the behavioral response to price of other economic agents,
in this case consumers rather than rival firms.
10. The modern rehabilitation of the theory of predation focuses on its impact on rival
firms’ expectations about future profitability: as an exit-inducing investment in “disinfor-
310 DUMPING IN DRAMS

exercise of monopoly power must be received later to justify absorption

of a temporary loss on output shipped now.
Many forms of strategic behavior by firms, including predation, are
regulated within national markets by antitrust laws. Thus a constructed
cost test, used in the framework of the dumping laws, might be inter¬
preted as a second-best attempt to remedy behavior by foreign firms that,
if carried out on a purely domestic basis, would be considered the domain
of antitrust policy. Unable to impose domestic policy standards on a
foreign firm’s behavior outside the national market, a national govern¬
ment can instead impose controls on the manifestation of that behavior—
the pricing of sales to importers—in the domestic market.
Since, absent learning effects, pricing below short-run marginal cost
is sufficient (but not necessary* 11) to conclude that a firm is acting strate¬
gically in its pricing policies, it may seem reasonable to review its activities
when such behavior is observed, and to take corrective action if the intent
is deemed to be predation and the potential impact is significant. If the
predatory firm should succeed in its objective, it is at least possible that
monopoly rents paid out later by national consumers, plus deadweight
losses, could more than offset the windfall to national consumers from
the temporary episode of low import prices now; it is these losses that
justify some policy intervention.12
From this point of view, the economic problem with cost-based defi¬
nitions of dumping is not that they are used at all, but that they use the
wrong cost concept (long-run average cost instead of short-run marginal
cost) as the prima facie trigger for considering intervention. This per-

mation” about the predator’s cost structure, for example, or as the consequence of asym¬
metric financial constraints among competing firms created by imperfections in capital
markets. The basic references are Paul Milgrom and John Roberts, “Limit Pricing and
Entry under Incomplete Information: An Equilibrium Analysis,” Econometrica, vol. 2
(March 1982), pp. 443-60; and David M. Kreps and Robert Wilson, “Reputation and
Imperfect Information,” Journal of Economic Theory, vol. 27 (August 1982), pp. 253-79.
Useful interpretations are found in Paul Milgrom, “Predation,” in John Eatwell, Murray
Milgate, and Peter Newman, eds., The New Palgrave: A Dictionary of Economics (London:
MacMillan, 1987), pp. 937-38; and Jean Tirole, The Theory of Industrial Organization
(MIT Press, 1988), pp. 367-80.
11. Criticism of a short-run marginal cost test for predation generally argues that the
rule is not stringent enough; prices above short-run marginal cost may still be associated
with socially costly predatory activity. See Tirole, Theory of Industrial Organization,
pp. 372-73; and Ordover and Saloner, “Predation, Monopolization, and Antitrust,” pp.
579-80.
12. As, for example, Deardorff, “Economic Perspectives on Antidumping Law,” pp.
35-36.
DUMPING IN DRAMS 311

spective also leads one to focus on the close relationship between “fair
trade” laws and competition and antitrust policy. It might be argued that
some binding international standards for competitive business behavior
(and their enforcement) might be offered as a constructive alternative to
national “fair value” dumping tests based on constructed costs, as rem¬
edies for predation.
I will not attempt in this chapter to evaluate whether predation is a
plausible description of what was going on in the DRAM marketplace in
the 1980s; that question is explored—but not resolved—in the next chap¬
ter. I merely note that predatory behavior was one of the allegations
made by the U.S. industry in pressing its case for protection. However,
the modern theory of predation has been interpreted to suggest that high-
technology industries are particularly important places to look for such
behavior.13

Cost Structures and Dumping in the Semiconductor Industry

High-technology industries, which face large sunk costs in research

and development (R&D) relative to sales, are by nature, along with
highly capital-intensive industries, particularly prone to trade friction
involving charges of dumping based on constructed cost tests. When fixed
investments in R&D or factories are very large in relation to a firm’s
sales, a significant gap between average variable cost and long-run aver¬
age cost will exist, and short-run marginal cost may fall significantly below
long-run average cost for a substantial range of economically rational
output levels. In such a case, perfectly competitive behavior may often
trigger pricing below long-run average cost—and thus dumping charges—
in a downturn.
High-tech industries are also particularly prone to dumping allegations
because of the peculiar way in which R&D investments are treated by
trade law (and many companies’) accounting principles. An investment
in a capital facility, for example, is not charged immediately against
company revenues when construction is begun, or even when it is com-

13. Paul Milgrom argues that “policymakers should be especially sensitive to predatory
pricing in growing, technologically advanced industries, where the temptation to discourage
entry is large and the costs of curtailed entry even larger.” (Milgrom, “Predation,” p. 938.)
See Kenneth Flamm, “Semiconductor Dependency and Strategic Trade Policy,” Brookings
Papers on Economic Activity: Microeconomics (1993), pp. 249-344, for further considera¬
tion of the plausibility of strategic behavior in semiconductor competition.
312 DUMPING IN DRAMS

pleted; instead the cost is spread, by means of depreciation charges, over

the period in which the facility is to be used. One may argue that ac¬
counting depreciation is at least an attempt to approximate the profile of
true economic depreciation. The cost of an R&D investment, in contrast,
is generally charged against revenues at the moment it is incurred, not
spread over the investment’s economically useful life.
This “front-loading” of R&D, it is sometimes argued, when processed
through constructed cost calculations, leads to artificially high “con¬
structed values” for high-tech imports—such as DRAMs—when initially
shipped, in effect retarding technological progress. Defenders of this
practice argue that since R&D charges are often allocated on the basis
of sales, rather than identified with some particular product, the practical
effect is to spread R&D charges over generations of products, through
time (although it clearly remains true that a company just entering an
industry after making a fixed R&D investment will necessarily have an
initially high constructed cost).
The semiconductor industry is both technology intensive and capital
intensive: it spends almost 15 percent of sales on R&D; it also typically
spends an even larger fraction of sales—15 to 20 percent annually—on
capital investments. Demand for semiconductors is also notoriously cycli¬
cal, so that it is not surprising to find that constructed cost tests for
dumping were invoked in the 1985 industry downturn.
In addition, semiconductor production is believed to be characterized
by learning economies. Unit production costs are believed to fall sharply
with accumulated production experience. This further complicates our
discussion of the borderline between nonstrategic pricing behavior and
strategic activities. The key result, first elaborated by Michael L. Spence
is that, with learning economies but without strategic interactions with
rivals, a rational firm will generally equate marginal revenue to a value
below its current short-run marginal cost of production, as it takes into
account the cost-reducing effect of current production on future produc¬
tion costs.14
Although the Spence model is not directly applicable to the analysis
of production decisions in the semiconductor industry, the point it makes
greatly complicates the issue of whether or not constructed cost dumping
tests—amended perhaps to use short-run marginal cost rather than long-
run average cost as the indicator for possible intervention—can be jus-

14. “The Learning Curve and Competition,” Bell Journal of Economics, vol. 12 (Spring
1981), pp. 49-70.
DUMPING IN DRAMS 313

tified as a reasonable safeguard against predation by foreign producers.

For in the Spence model, even with nonstrategic behavior, economically
rational firms will engage in forward pricing, producing more even when
marginal revenue falls short of marginal cost in anticipation of the cost-
reducing effects of greater production.

Modeling the Semiconductor Product Life Cycle

The same considerations that make it likely that below-cost dumping
will be observed in semiconductors also complicate any effort to build a
tractable economic model of producer behavior. Any such model needs
to address four distinctive features of semiconductor production:
—learning economies, the manner in which net output from a given
semiconductor fabrication facility rises with accumulated production ex¬
perience. This creates what amount to dynamic economies of scale.
—capacity constraints, due to the long lag between when facilities
investments are started and when they are capable of mass production.
It typically takes a year to a year and a half or more for a new facility to
become operational. Once available for production, debugging manufac¬
turing processes on “pilot production” can take another six months to
a year.15
—short product life cycles, due to accelerated technological change.
In DRAMs, a new, higher density memory chip has been introduced
approximately every three years, rendering older chips obsolete. The
newer chips use technologically more advanced manufacturing processes,
requiring new production facilities (or, at a minimum, extensive retro¬
fitting of older facilities). Coupled with the long gestation periods for
new investments, this effectively means that capacity for a given genera¬
tion of chip is locked in at the very beginning of the product life cycle.
By the time the product is entering large-scale mass production, it is
essentially too late to enter the market.
—large R&D investments. The sunk investment required for product
and process R&D for a new generation of DRAM has typically been
roughly equal to the cost of a high-volume manufacturing facility, and

15. For example, a new state-of-the-art fabrication facility in Taiwan reported in 1992
that it took four months to qualify production processes, followed by five months of further
work to raise production from 1,000 to 10,000 wafers a month, at a facility with a current
production rate of 15,000 wafers per month. See Klaus C. Wiemer and James R. Burnett,
“The Fab of the Future: Concept and Reality,” Semiconductor International (July 1992),
pp. 96-98.
314 DUMPING IN DRAMS

has been increasing rapidly. For the upcoming 256M DRAM, for exam¬
ple, industry sources have estimated both the R&D and plant investment
required to be about $1 billion each.16
In my somewhat stylized depiction of the industry, a DRAM producer
will be assumed to produce a homogeneous commodity, perfectly substi¬
tutable for that of other producers.17 Difficult issues concerning the tim¬
ing of the switchover from one generation of DRAM to another, and
intergenerational externalities, are ignored by assuming that a DRAM
producer faces a fixed period over which the product is sold, and that
costs to develop and produce the product are relevant to that genera¬
tion of DRAM alone. The product life cycle begins at time 0 and ends at
time 1 (hence, the unit of time is the “product life cycle”). Every pro¬
ducer faces revenue function R, giving total revenues at any moment t as
a function of its own production y(t) and the aggregate output of all other
producers x(t). (All revenues and costs are measured in constant-dollar
terms.)
Semiconductor production is believed to be characterized by strong
learning economies. As was explained in chapter 5, output from a fully
utilized chip fabrication facility is not fixed over time, but instead rises
on an S-shaped profile (see figure 5-17 and the accompanying discussion).
Although the Spence model of learning economies is rather unsuitable
for analyzing production decisions in an industry where capacity con¬
straints may be important, as is the case in the semiconductor industry,
it shows, as we have seen, that even with nonstrategic behavior econom¬
ically rational firms will engage in forward pricing. For simplicity in mod¬
eling producer behavior, I will, following Spence, ignore discounting over
time on the grounds that product life cycles are short (typically, a new
generation of DRAM is introduced every three years) and the additional
complexity is substantial.18

16. Comments by an IBM executive at the Workshop on Government Roles in Com¬

mercial Technology, John F. Kennedy School of Government, Harvard University, Septem¬
ber 14, 1992. “According to estimates from IBM, Siemens, and Toshiba, the cost of de¬
signing and qualifying a quarter-micron process is in excess of $1 billion.” Adam Greenberg
and J. Robert Lineback, “IBM, Toshiba, Siemens in 256M DRAM Alliance,” Electronic
News, July 20, 1992, p. 4.
17. This is not an unreasonable approximation. See Kenneth Flamm, “Measurement
of DRAM Prices: Technology and Market Structure,” in Murray F. Foss, Marilyn E.
Manser, and Allan H. Young, eds., Price Measurements and Their Uses (University of
Chicago Press, 1993), pp.157-206, for more detailed discussion of this issue in the context
of semiconductor price indexes.
18. Also following Spence, I will measure all variables in real terms, taking out the
DUMPING IN DRAMS 315

In semiconductor production, plant capacity may be measured in

terms of “wafer starts,” the number of slices of silicon, on which inte¬
grated circuits are etched, that can be processed per unit of time. At any
moment t, w[E(t)] functioning chips are yielded per wafer processed,
where w is an increasing function of E(t), “experience” through time t.
How one defines relevant “experience” is a subject explored below. I will
parametrize the impact of output y on relevant experience E as

p = dE = y_
dt K' ’

where K is capacity and y is a parameter taking on a value between zero

and one. (For notational simplicity, time will sometimes be suppressed
as an argument of time-varying variables.)
Some of the variable cost of producing a chip is incurred with every
wafer processed, but some (the cost of assembly and final testing, for
example) is incurred only with good, yielded chips. Let c be the variable
cost incurred every time a wafer is processed, and d the variable cost
incurred every time a good, yielded chip is taken through subsequent
stages of production (assembly and final testing). If a wafer processing
facility is utilized at rate u(t) (where u is between zero and one), total
variable costs at any moment are dy + cuK. Note that y(t,K) = w[E(t)]
u(t) K.
Let up-front, sunk costs independent of output levels (such as R&D
costs) be equal to F, and let fixed capital investment costs required for a
facility processing K wafer starts be equal to r per wafer start. The
producer’s problem is to maximize (again, undiscounted) life cycle profits
on this product,
1

maxU(r).K I {R[x(0>y(0] - d y(0

0

(6-1) - cu(t)K — rK} dt — F

with y(t) = w[E(t)] u(t) K

s.l. E = ^ = w[E(t)] u(l) K'-\

effects of inflation, so that any applicable discount rate would be smaller than if measured
in nominal terms. This, coupled with the shortness of the product life cycle for DRAMs,
should make my undiscounted life cycle profit calculation a reasonably tolerable approxi¬
mation to the more complex discounted flow.
316 DUMPING IN DRAMS

Assume that firms simultaneously choose initial capacity investments

K and a time path for utilization rates, which give rise to a path for
output over time, which they then proceed to follow. This assumption
that capacity investments in DRAMs are committed at the beginning of
the product cycle is not terribly unrealistic: as already noted, it typically
takes a year or more to get a new fabrication facility up and running, and
a new generation of DRAM is introduced roughly every three years.19
For the moment, take y to equal zero (in other words, absolute cu¬
mulative production is the relevant measure of experience). As in the
Spence model, I will assume a Nash equilibrium in output paths: given
rivals’ actual choices of capacity and a time profile for the utilization rate,
equation 6-1 is maximized by every firm. For the purposes of this chapter,
firms’ behavior in this static game is assumed to be nonstrategic, since
they take their rivals’ capacity and output choices as unaffected by their
own.20

Spence’s Model

If wafer processing capacity K is not fixed over the life cycle but is
continuously variable, as is implicit in Spence’s formulation, then we have
a special case of the above model in which r is zero (capital costs are
included in wafer processing cost c and some arbitrary initial scale for
capacity K is set), capital is a completely variable input, and a producer
is free to choose any nonnegative u (that is, u is unbounded above, not
bounded by one) and to produce any yielded chip output desired. Under
these circumstances, as is easily shown in appendix 6-A, formal maxim-

19. The world record for bringing a new fabrication facility on line is said to be nine
months. See Larry Waller, “DRAM Users and Makers: Shotgun Marriages Kick In,”
Electronics, November 1988, pp. 29-30.
20. An alternative, to be presented in the next chapter, is to set up a two-stage com¬
petition among rival firms, with capacity investment as the initial phase, followed by a
second stage in which firms choose output paths subject to capacity constraints. The
solution of the static game presented here corresponds to the open-loop (nonstrategic)
equilibrium of this two-stage game, in which a firm’s first-period choice of capacity takes
its rivals’ choices in both periods as given. The alternative equilibrium concept will assume
second-period subgame perfectness, that is, that firms take into account the effect of their
first-period capacity choices on their rivals’ second-period output paths. This creates stra¬
tegic interactions among firms. See Avinash Dixit, “Comparative Statics for Oligopoly,”
International Economic Review, vol. 27 (February 1986), p. 114; and Carl Shapiro, “The¬
ories af Oligopoly Behavior,” in Schmalensee and Willig, Handbook of Industrial Organi¬
zation, vol. 1, pp. 383-86.
DUMPING IN DRAMS 317

ization of the objective function in equation 6-1 yields the first-order

condition

(6-2)

that is, u is chosen so that marginal revenue is set equal to current

marginal cost (d + c/w) less a term proportional to nonnegative adjoint
variable 6, which captures the future cost-reducing effects of current
production. Adjoint variable 8, in turn, is determined by the transver-
sality condition

(6-3) 8 (1) = 0

and equation of motion

(6-4) 8 = — — u K wF .
w

By differentiating both sides of equation 6-2 with respect to time, we

immediately see that marginal revenue Ry must be constant over time,
and therefore, by equation 6-3, equal to current marginal cost at the end
of the product cycle, d + c/w[E( 1)].
In short, with continuously variable capacity, a profit-maximizing pro¬
ducer will choose that level of output at which marginal revenue equals
terminal (not current!) marginal cost. In industry parlance this is forward
pricing. With a constant elasticity and autonomous (time-independent) de¬
mand a constant price proportional to terminal marginal cost will result.
Although this model provides an appealing explanation of the phe¬
nomenon of forward pricing, which is a notable empirical feature of
business practice within the semiconductor industry, the actual trajectory
of pricing suggested by this model (with a constant elasticity of demand,
price is fixed at some constant level over the entire product cycle) is quite
inconsistent with observed behavior.21 Chip prices typically drop very
quickly over the first part of the product cycle, drop less quickly as the
product approaches maturity, and fall very slowly, if at all, at the end.
As shall be seen in a moment, a more realistic treatment of capacity
constraints yields a more plausible trajectory for prices.

21. Andrew R. Dick, “Learning by Doing and Dumping in the Semiconductor Indus¬
try,” Journal of Law and Economics, vol. 34 (April 1991), pp. 133-59, for example, invokes
the Spence model to motivate his assumptions about the time path of semiconductor prices
over the product life cycle, but ignores the constant pricing prediction of the Spence model.
318 DUMPING IN DRAMS

The Baldwin-Krugman Conundrum

The pioneering attempt to incorporate learning economies into a styl¬

ized, empirical model of the semiconductor industry is that of Baldwin
and Krugman (B-K).22 The B-K model focuses on regional segmentation
of the U.S. and Japanese semiconductor markets, in order to simulate
the impact of market closure policies, and takes an approach to producer
behavior that differs significantly from that of Spence. B-K constrain
firms to operate at full capacity over the entire product cycle; the choice
variable for the firm is initial capacity, which once set, determines output
levels over the entire product life cycle. The first-order condition for an
optimum is that the life cycle revenue from the addition of a marginal
unit of wafer processing capacity just equals the cost of building and
operating that marginal unit of capacity (since all capacity is always fully
utilized, the distinction my model draws between investment costs and
wafer processing costs is immaterial).
Firms in the Spence model are never capacity constrained; firms in
the B-K model always operate at their capacity constraint. The Spence
model has firms forward pricing— maintaining marginal revenue constant
over the life cycle, equal to their terminal marginal cost. The B-K model
has marginal revenue—and price—falling smoothly over the life cycle.
Thus, while the striking forward pricing behavior of the Spence model
has disappeared (but not all forward pricing)23, a more empirically plau¬
sible path for prices has replaced it.

22. Richard E. Baldwin and Paul R. Krugman, “Market Access and International
Competition: A Simulation Study of 16K Random Access Memories,” in Robert C. Feen-
stra, ed., Empirical Methods for International Trade (MIT Press, 1988), pp. 171-97. A
somewhat different exposition of this model is given in Elhanan Helpman and Paul R.
Krugman, Trade Policy and Market Structure (MIT Press, 1989), chap. 8. The later inter¬
pretation (henceforth referred to as H-K) differs in some significant respects from B-K.
For example, the learning curve in B-K has yields improving with cumulative wafers pro¬
cessed (that is, faulty chips have the same yield-enhancing effects as good ones), whereas
H-K presents a more conventional view of the learning curve, with yield rates rising with
cumulative output of yielded (good) chips. The B-K assumption on yields, although not
the accepted approach to modeling yield improvement within the industry, simplifies the
mathematical structure of the model.
23. There is a great deal of confusion in the industry about what exactly forward pricing
means. Does it mean that learning economies are to be taken into account when forecasting
marginal costs, and prices, for future deliveries in forward contracts with large customers,
or does it mean—as in the Spence model—going even further and producing quantities
such that marginal revenue falls below current marginal cost? Note that even when output
is capacity constrained one can have aggressive “forward pricing” in the sense of price
DUMPING IN DRAMS 319

Unfortunately, when B-K actually calibrated their model to empirical

data, their results indicated that, in the intuitively appealing case of a
(Cournot-) Nash equilibrium in output, only two firms would populate
the industry in zero profit equilibrium, in part because of the steep
declines in unit cost due to learning economies assumed.24 This gross
deviation from actual market structure led them (despite their theoretical
reservations about the approach) to specify firm behavior in terms of
conjectural variations, and even this tactic yielded disturbingly high con¬
jectural variations, fueling doubts about the underlying model.25 As I
point out below (see note 43), B-K actually underestimated learning
economies as reported in the sources they cite, deepening the perplexity
created by their results.
In addition, as Kala Krishna notes, the algebraic tractability of the
B-K specification of firm behavior has been purchased by excluding the
possibility of some interesting forms of strategic competition.26 (Because
B-K empirically calibrate conjectural variations, strategic interactions
among firms exist.) Investments in capacity may be undertaken with
strategic objectives, to convince rivals to reduce output or to exit, or to
dissuade them from entering an industry, creating additional monopoly
power which can then be exploited. Constraining firms to operate at full
capacity over the entire product life cycle may restrict them to suboptimal
output paths, where monopoly power is not fully exploited. It also hinders
analysis of interesting policy questions regarding the potential welfare
impact of strategic policies which may foster the creation and exercise of
monopoly power (explored in chapter 7).
A variant of the B-K model can be fit into the framework outlined
above for the Spence model, after suitable amendments. Utilization rate

(and, necessarily, marginal revenue) falling below current marginal cost: if one examines
the simulations in this chapter, price is below both current marginal and average cost in
the earliest portion of the product cycle (see table 6-4). Thus capacity-constrained output
and aggressive forward pricing are not mutually inconsistent—the firm is producing as
much as it can and reducing the price to whatever level is needed to sell it all.
24. See Baldwin and Krugman, “Market Access and International Competition,”
p. 185.
25. Baldwin and Krugman, “Market Acccess and International Competition,” p. 195;
and Helpman and Krugman, Trade Policy and Market Structure, p. 173. The conjectural
variations approach has each firm assuming that if it increases its output by one unit, each
of its competitors will change its output by v units, where v can take on a range of values.
The case of v = 0 is the Cournot assumption.
26. “Comment on ‘Market Access and International Competition,’ ” in Robert C. Feen-
stra, ed.. Empirical Methods for International Trade (MIT Press, 1988), pp. 198-202.
320 DUMPING IN DRAMS

u is constrained to equal one at all times, and the objective function in

equation 6-1 is maximized with respect to K alone. The right-hand side
of equation 6-4 is replaced by a more complex variant (corresponding to
u = 1), and a new equation determining optimal capacity choice is added:
1

where the B-K specification fixes u equal to one and y equal to zero.

A More Realistic Model of the Semiconductor Product Cycle

It is possible to create a more realistic model of firm behavior, in which

firms can continuously adjust output as in the Spence model, yet also
face capacity constraints on output, as in the B-K model. I briefly sum¬
marize here the more detailed exposition laid out in the appendixes to
this chapter. The firm's problem is to maximize equation 6-1 by choosing
both an initial level of capacity K and a profile for time-varying utilization
rate u(t) for that capacity, which determines output at any moment in
time. The optimal level of capacity chosen satisfies equation 6-5; the left-
hand side of this equation can be interpreted as the net marginal return
on additional investment in capacity. It also must be true that the optimal
path must over some interval be capacity constrained; that is, u(t) = 1.
In general, the optimal path for u(t) will be made up of three types of
segments: interior segments, where 0 < u < 1; lower boundary segments,
where u = 0; and upper boundary segments, where u — 1. Within an
interior segment, equations 6-2 and 6-4 will hold, as in the Spence model,
as will a form of forward pricing: marginal revenue will be held constant,
set equal to current marginal cost less 8IKy—the marginal cost-reducing
value (over the remainder of the product life cycle) of an additional unit
of output—at the endpoint of this interval.
With additional assumptions one can further sharpen the characteri¬
zation of the optimal behavior of a profit-maximizing firm. I shall assume
a symmetric-industry equilibrium with N identical firms, autonomous
demand (that is, demand is not an explicit function of time), and concav¬
ity of total industry revenues in industry output (as would be the case,
for example, with a constant-elasticity demand function and price elas¬
ticity exceeding unity). If we further assume that firm marginal revenue
exceeds the initial value of current marginal cost as industry output
DUMPING IN DRAMS 321

approaches zero (as must be the case with a constant-elasticity demand,

so some production will always be profitable), we can exclude the possi¬
bility of no firms entering the industry in a symmetric equilibrium. Al¬
though a Nash equilibrium in utilization rates is assumed for the moment,
for expositional purposes I will parametrize a firm’s perceptions of other
firms’ reactions to changes in its output in terms of a constant, non¬
negative conjectural variation. (The two interesting cases that motivate
this parametrization are Cournot-Nash equilibrium—conjectural varia¬
tion equal to zero—and a collusive, constant-market-share cartel—con¬
jectural variation equal to N — l.)27
Under these assumptions, optimal u must decline over an interior
segment, and u must be continuous in time. Key features of the behavior
of production and utilization rates over time are that y must be increasing
when u is constant (because of learning economies); u must be decreasing
when y is constant (for the same reason); and y is nondecreasing and u
nonincreasing over time (a consequence of the additional assumptions).
Therefore the optimal path of u must look like an upper boundary seg¬
ment, possibly followed by an interior segment. The Spence forward¬
pricing result of marginal revenue being set equal to terminal marginal
cost will hold whenever the firm is producing but is not capacity con¬
strained (that is, when 0 < u < 1, along an interior segment).
Note that nothing about the specific shape of the learning curve (func¬
tion w) beyond the fact that it is increasing in experience {wE > 0) has
been assumed in arriving at this characterization of optimal policy. Let
the time at which a firm switches from full-blast production to constant-
output production be ts (with full-blast production over the entire product
life cycle an important possibility).
In short, with this simple description of the semiconductor product
life cycle we derive a more realistic specification of firm behavior that
captures both the importance of capacity investments and the ability of
firms to exploit fully what monopoly power they enjoy by varying utili-

27. The major behavioral assumption excluded by a nonnegative conjectural variation

is Bertrand competition in prices. Because DRAMs are essentially a homogeneous com¬
modity sold in well-developed secondary spot markets, specifying that producers sell at a
single market price and choose the quantities they will sell is the natural assumption.
Moreover, David M. Kreps and Jose A. Scheinkman, “Quality Precommitment and Ber¬
trand Competition Yield Cournot Outcomes,” Bell Journal of Economics, vol. 14 (Autumn
1983), pp. 326-27, have shown that in a two-stage game, where first-stage capacity invest¬
ments are followed by a second-stage Bertrand game in prices and a particular (“efficient”)
rationing rule, the outcome is a Cournot equilibrium in output.
322 DUMPING IN DRAMS

zation rates over time. The model is simple enough to be empirically

tractable. Firms will make some capacity investment, operate at that
capacity for some period of time, and then possibly switch to a constant-
output path (with constant marginal revenue, but decreasing utilization
of capacity, as yields continue to rise) over the remainder of the product
life cycle. (A period of full-blast production during which yields rise very
sharply is in fact a pervasive feature of producer behavior in this industry,
and has been given its own special name: “ramp-up”.)

Preliminary Observations on Pricing over the Life Cycle

Even with its relatively general structure, the above analysis provides
several insights into questions concerning pricing and dumping over the
product life cycle. First, pricing below current marginal cost will never
be observed near the end of the product cycle among competitive, non-
strategic, profit-maximizing firms. This follows immediately from the fact
that, if any output is being produced, marginal revenue will never be less
than the right-hand side of equation 6-2 (see appendix 6-A), which at
time 1—the end of the product cycle—equals current marginal cost
(d + ciw). Since price exceeds marginal revenue, price also must exceed
current marginal cost in some neighborhood of time 1, the end of the
product cycle.
Second, it is common practice in the semiconductor industry to note
that chip prices seem to fall with cumulative experience, just as consid¬
eration of learning economies suggests that unit costs should. Indeed,
analysts often identify average unit cost with price, then estimate an
empirical “learning curve” by regressing the logarithm of market price
against the logarithm of cumulative output for the industry. Slopes of this
line typically are found to be around -0.3.28
The analysis just laid out, however, suggests that this implicit interpre¬
tation of prices as proportional to some variant of current unit costs is
generally hard to justify. We have seen that firms may be forward pricing,
with prices set below current marginal costs of production. In this case,
to break even (or make a profit) over the product cycle, prices must, over
at least some other period, be set above current marginal costs. Thus

28. See, for example, Integrated Circuit Economics, Mid-Term 1989 (Scottsdale, Ariz.,
1989); Boston Consulting Group, Perspectives on Experience (Boston, 1972); and Douglas
Irwin and Peter J. Klenow, “Learning-by-Doing Spillovers in the Semiconductor Industry,”
Journal of Political Economy, vol. 102, no. 6 (1994), pp. 1200-27.
DUMPING IN DRAMS 323

prices may in general be below current marginal costs over some period,
and above current marginal costs at other times, and the practice of
treating price as proportional to some definition of unit cost becomes
questionable.29
However, in one important special case, namely, when production
proceeds at full blast over the entire product life cycle, an explicit rela¬
tionship between price and cumulative volume does emerge. In that case,
output is fixed by initial capacity investments and the effects of learning
economies on yields, while price is determined by the parameters of the
demand curve, given whatever output is produced at full capacity. It can
be shown that in this case

dlogP _ dP E _ e
3log£ dE P p’

where e is the elasticity of yield with respect to cumulative output (the

“learning curve” elasticity), and (3 the price elasticity of demand.30 Thus,
in this special case the slope of the relationship between the logarithm
of price and the logarithm of industry cumulative output is the learning
curve elasticity divided by the demand elasticity, and so reflects both
supply and demand parameters.31

29. Needless to say, since average unit cost (in which fixed costs such as R&D and
capacity investments are allocated to different time periods over the product cycle, and the
average of this fixed cost per unit produced is added onto marginal cost) must lie above
current marginal cost, the same criticism applies to identifying price with any definition of
average unit cost.
30. To get this result, assume the industry is made up of N identical firms, each pro¬
ducing output y. Let industry demand be given by inverse demand function g, so that P =
g(Ny). With full-blast, capacity-constrained production, y at any moment in time is given
by w{e)K, with K representing capacity (in wafer starts) per firm and w yielded (good)
chips per wafer, a function of firm cumulative experience e. E (= Ne) is industry cumulative
experience. Substituting for y in the expression for P, and differentiating with respect to
E, we end up with

dgy dw e
dy g de w

where the first expression in brackets is 1/(3, and the second e.

31. In the special case of e = 0.5, (3 = —1.5 (which I shall argue is roughly what real-
world parameters suggest), we then have an elasticity of price with respect to cumulative
volume of about -0.3. With e = 0.5, we also have what is commonly called a “70 percent
learning curve” (referring to the fact that unit cost declines by 30 percent as experience is
doubled; see note 43 in this chapter). The empirical coincidence of a -0.3 elasticity of price
with respect to cumulative output, and a 30 percent decline in cost with a doubling of
cumulative output, has led to enormous confusion, since these two parameters will generally
324 DUMPING IN DRAMS

Closing the Model

Can we say anything about the relationship between price and long-
run average cost? The model sketched out thus far takes the number of
firms in the industry—which will affect profitability and pricing—as
given. One “natural” way to close the model is to specify that firms enter
the industry until rents earned by producers—that is, the integrand in
equation 6-1—just equal zero. The zero profit condition then determines
N, the number of firms entering the industry (we will ignore the difficul¬
ties created if one insists that N be an integer). Zero profits mean that
total life cycle revenues just equal total life cycle costs. Therefore (after
dividing both concepts by total output over the product life cycle) average
life cycle price must equal average life cycle cost per unit.
But is there any clear relationship between current price and current
“fully allocated” average cost at any given moment? Current short-run
marginal cost is a relatively clear concept: it is the additional current cost

not be identical. Indeed e with a 70 percent learning curve is actually equal to 0.5, not 0.3.
Irwin and Klenow's 1994 study, “Learning-by-Doing Spillovers,” which like all other
studies regressing price on cumulative output finds an elasticity of price with respect to
cumulative output of about - .3, interprets this to imply an 80 percent learning curve. For
the reasons just given, this parameter can actually be interpreted as supporting a much
larger, 70 percent learning curve in a capacity-constrained operating environment.
Irwin and Klenow recognize that if firms are capacity constrained, their methodology
is suspect. They argue that an aggregate semiconductor industry capacity utilization mea¬
sure produced by the SIA varied between 43 and 78 percent over 1978-92: “if these capacity
figures apply to DRAM production, then the assumption that capacity constraints do not
bind seems appropriate” (p. 1214). This assumption is unwarranted, however, since the
SIA survey also includes older, economically marginal facilities that produce specialized,
older-vintage, and custom chips on fully depreciated, obsolete lines as demand warrants.
Current-generation DRAMs are generally produced in leading-edge facilities with new
equipment that are basically run at full capacity around the clock for at least the first years
of their lives. Stories in the business press (see chapters 3 and 4) suggest that, at the very
minimum, during periods of cyclical boom in chip demand aggregate DRAM output has
been constrained by available capacity.
The only available data on plant-level capacity utilization for current-generation semi¬
conductor fabrication lines appear to support the notion of a capacity-constrained operating
environment. In their detailed, international study of sixteen semiconductor lines (twelve
of which used current generation 6-inch wafers and equipment, four of which used slightly
older 5-inch wafers and equipment, and one of which used older 4-inch wafers), the Com¬
petitive Semiconductor Manufacturing survey of the University of California, Berkeley,
found that “most but not all of the fabs in our sample were fully loaded [relative to their
capacity] throughout their four-year period of observation.” See Robert C. Leachman,
“Introduction,” in Robert C. Leachman, ed., The Competitive Semiconductor Manufactur¬
ing Survey: Second Report on Results of the Main Phase, report CSM-08 (Engineering
Systems Research Center, University of California, Berkeley, 1994), pp. 17-18.
DUMPING IN DRAMS 325

saved by producing one less unit at any given moment. This is the incre¬
mental cost saved when output is reduced by one unit.32 In my model,
current short-run marginal cost—d + clw—is constant at any moment
and equal to current average variable cost.
To define a fully allocated current average cost, however, it is necessary
first to define an intertemporal cost allocation rule to spread fixed entry
costs F and capital costs rK over the product life cycle. Dividing the
capital and entry cost allocated to some instant in time by output pro¬
duced at that moment yields a current average fixed cost per unit pro¬
duced. If this current average fixed cost is added to current average
variable cost (identical to short-run marginal cost in my model), we have
a long-run average cost (LRAC) concept that satisfies the basic require¬
ments of a long-run average cost: when multiplied by output at that
moment, and when all such products are summed over all moments, total
costs of production over the entire product life cycle are given.
In one special case—when d, the yielded chip output-sensitive part
of variable cost, is “small” (in a sense to be defined in a moment), and
it is assumed that free entry by other competing firms into the industry
forces life cycle profits to zero—a clear pattern in the relationship be¬
tween price and both marginal and average cost over the product life
cycle exists. Differentiating our industry demand function with respect
to time (and assuming determinants of demand other than price remain
constant), we must have

P _ l z
P ~ p z’

linking rates of change in price over time with growth rates in industry
output, where (3 is the price elasticity of demand and z is industry output.
Now, recall that we have defined LRAC as

cuK + rK + F
LRAC = + d,
y

while marginal cost MC is just

32. When a firm operates at less than full capacity, this is identical to the increased cost
incurred in producing one more unit. When operating at full capacity, the incremental cost
of an additional unit is effectively infinite; the marginal cost curve is L-shaped, with a kink
at full-capacity output.
326 DUMPING IN DRAMS

cuK
MC = - + d,
y
and average cost always exceeds marginal cost by the average fixed cost.
Differentiating with respect to time, then, gives33

rK + F
■I- d
LRAC d y
1 - y- + - < 0
LRAC LRAC LRAC u

MC d ly u
1 - < 0
MC MC \y u

and both marginal and average cost must be falling over the entire prod¬
uct life cycle.34 More interestingly, because demand is elastic (p < -1 )
and the industry is populated by identical firms (so z/z = y/y), we must
then have

+ d\
P LR AC (1 d \ y y u
1 - >0,
P ~ LRAC ~ \(3 + LRAC) y " LRAC 1 u
\
if output-sensitive variable cost d is “smaf in the sense of other costs’
share in average cost (AC), 1 — d/LRAC, exceeding 1/p in absolute value.
Because our assumption about entry means that total life cycle costs
are exactly equal to total life cycle revenues, price less the fully allocated
long-run average cost as defined above (that is, profit per unit), multi¬
plied by output and summed over every moment of the product cycle,
must be exactly equal to zero. Thus, if price exceeds the fully allocated
average cost concept at any instant, it must fall below fully allocated
average cost at some other instant over the product cycle, and vice versa.
Under the assumption of “small” output-sensitive variable costs, price
must fall more slowly than average cost, and average cost falls over the
entire product life cycle. Therefore, for there to be zero profit over the

33. Since (see appendix 6A)

rK+F

1 - > 1 - > 0, - > 0, and - < 0.

LRAC LRAC LRAC y u
34. Since the expressions in brackets all are fractions lying between zero and one, y
and u are both nonnegative, y is nondecreasing, u is nonincreasing, and y must be increasing
when u is constant and constant when m is negative.
DUMPING IN DRAMS 327
Figure 6-1. Relationship between Semiconductor Price and Cost,
Assuming “Small” Output-Sensitive Variable Costs

Dollars

Source: See text.

product life cycle, price must originally lie below average cost in the early
part of the product cycle, then rise above it at the end. In this case the
relationship among various cost concepts will appear as in figure 6-1, and
we would expect to observe below-average-cost “dumping” in the early
part of the product life cycle as the consequence of normal competitive
behavior.35
The scenario sketched out in figure 6-1, however, depends critically on
the assumption that costs incurred after wafer processing are a “small”
component of cost throughout the product cycle.36 As we shall see in a

35. This argument is made by example, in the special case of all firms engaging in full-
blast production over the entire product life cycle, in Richard E. Baldwin, “The US-Japan
Semiconductor Arrangement,” discussion paper 387, Centre for Economic Policy Research,
London, March 1990, pp. 17-19; and Baldwin, “The Impact of the 1986 US-Japan Semi¬
conductor Agreement,” Japan and the World Economy, vol. 6 (June 1994), pp. 129-52.
Baldwin implicitly assumes that all variable costs are yield-sensitive (that is, that d = 0).
36. The behavior shown in figure 6-1 is also assumed, without explanation, in Dick,
“Learning by Doing,” pp. 133-59.
328 DUMPING IN DRAMS

moment, our best efforts to describe the semiconductor manufacturing

process empirically suggest that this assumption is incorrect, and that
the precise timing of episodes of below-cost pricing is considerably less
predictable.
More generally, however, if our assumption that entry drives long-run
profits to zero is a realistic one, below-LRAC “dumping” must be oc¬
curring at some point during the product cycle. This follows from the
fact that any “standard” cost allocation rules for fixed costs will define
an average fixed cost, which, when added to current average variable
cost, cannot be exactly equal to actual price at every moment during the
product cycle. For firms to earn zero profits, therefore, prices must some¬
times be above average cost, and sometimes below—with below-cost
dumping observed at the latter moments.
With learning economies present, a cost allocation rule yielding an
average cost coinciding with price must generally be a function of all of
the parameters of the control problem, and the allocated fixed cost will,
in general, take on negative values as well as positive values. Since the
cost allocation rules actually used to spread fixed costs over the product
cycle—by firms or by the U.S. Commerce Department—are generally
functions only of the size of the fixed costs, and of time, and produce
only nonnegative values, it is essentially guaranteed that there will be an
episode of below-LRAC dumping if learning economies are present and
the industry is in a symmetric, zero-profit equilibrium.37

37. Define a cost allocation rule g(z,t), where z is a vector of arguments and t is time,
such that

Define fully allocated average cost (FAAC) by

FAAC = - + d + -.
w y

that is, current average variable cost plus average fixed cost. We know that the optimal
path must contain a capacity-constrained segment, and that along this portion of the optimal
path

P = P'Ny = P'NKweE.

in symmetric industry equilibrium. If price P is to always equal FAAC along this segment,
however, differentiating the expression for FAAC with respect to t, and setting this equal
DUMPING IN DRAMS 329

Some Further Assumptions

If not these, then what paths over time for prices and costs would one
expect to see in the simple model outlined here? Our next step is to take
this simple control model and solve it to explicitly derive an individual
firm’s behavior over time. To sharpen our characterization of a profit-
maximizing firm’s optimal policy, we must address some additional issues.

learning economies. We shall approximate the learning curve by

specifying that:

(6-6) w(E) = (}>£% with £(0) = E0, 0 < e < 1.

This gives yielded chips per wafer as a function of experience E. This

functional form is best regarded as an approximation: mass production
typically starts at initially low yields; after a while yields start to rise
quickly, then flatten out in a pattern closer to a logistic curve. (This
function may be thought of as another view of the data shown in figure
5-17, where the x-axis now shows cumulative volume of good chips pro¬
duced instead of time; good chips increase in a nonlinear way over time
as the result of learning economies.38) Analytical tractability is the
grounds for selecting this approximation. Note that a dummy experience
value E0 is used as an argument in the function to specify some initial

to the last expression, we must at every moment of this interval have

KwEE[(l + p)P - d\.

Since in equilibrium N, P, and other variables will generally be functions of all the param¬
eters of the optimal control problem, a function g that satisfies this last equation must
generally include all parameters of the control problem as arguments, unless wE = 0, in
which case g is constant. (In this case, I note in appendix 6-A that all capacity is utilized
and output is constant over the entire product life cycle.) Thus, as long as there are learning
economies (wE does not equal zero), a cost allocation rule g varying only with F, r, K, and
t cannot satisfy the requirement that P = FA AC, for arbitrary values of the parameters of
the control problem, over this capacity-constrained interval.
Also, we have already noted that it is possible for P less than current marginal cost to
be optimal in the presence of learning economies. (Indeed, the simulations reported below
contain examples of such behavior.) Reexamining the definition of FA AC, it is clear that g
must be negative for P = FA AC to hold true over such an interval.
38. See also Integrated Circuit Engineering, Mid-Term 1988 (Scottsdale, Ariz., 1988),
p. 6-35.
330 DUMPING IN DRAMS

Figure 6-2. Empirical Approximation of the Learning Curve in

Semiconductor Manufacture

Yielded chips per wafer

nonzero yield—without this constant, yields would stay stuck at zero

forever.39 The assumed approximation to the “true” learning curve is
depicted in figure 6-2. The approximation somewhat distorts yields, out¬
put, and pricing in the very earliest portion of the product cycle.
Defining “experience” raises additional issues. It is customary to use
cumulative output as a proxy for experience in empirical studies, and
most published empirical studies of learning economies have taken this
approach. But using absolute, company-wide production experience as
the determinant of any single facility’s productivity implies that running,
say, ten facilities in parallel produces the same yields at the end of a given
period as running a single facility to produce the same total output over
a much longer period. In the semiconductor industry, it is widely believed

39. B-K use the same functional form but do not face the stuck yield problem, because
the argument in their learning curve is experience in processing gross wafers, not yielded
good chips. The latter specification is generally industry practice in estimating learning
curves.
DUMPING IN DRAMS 331

that improved manufacturing yields come from two main sources: refine¬
ments of the operation of the production line (with each new refinement
building on previous experience), and die shrinks (reductions in the fea¬
ture size for chip designs, made possible by improved use of existing
process equipment). These are iterative and sequential in nature. That
is, lessons learned from running a line over some period of time are then
applied to refine the operation of that line over a subsequent period.
However, by this logic, if numerous identical production lines are run
in an identical fashion over the same period of time, then the same lessons
are being learned, in parallel, on each line, and yields at the end of the
period should be no higher than if only a single line were being run. Of
course, if a new line—with no experience and lower yields—is put into
operation after an older line has been running for some time, and it is
possible to completely transfer the fruits of experience across facilities,
then the maximum experience on any one line would be the experience
variable determining production yields. Because all investment occurs at
a single initial moment in my simple model, all lines will have identical
amounts of production experience at any subsequent moment in time,
and cumulative output per facility is the desired measure of experience.
It is possible that the lessons learned on different lines are not the
same, if completely different “experiments” in production refinement are
being conducted at every production facility. If, once again, experience
can be completely transferred across facilities, and there is no duplication
in lessons learned in different facilities, then it might be argued that
company-wide, absolute cumulative output, rather than cumulative out¬
put per unit capacity, is the relevant experience variable.40 Empirical
discussion suggests, however, that the transfer of experience across facil¬
ities is quite costly.41

40. Or perhaps even industry-wide cumulative output, if complete cross-company dif¬

fusion of the lessons of production experience occurs.
41. C. H. Stapper and others, “Evolution and Accomplishments of VLSI Management
at IBM,” IBM Journal of Research and Development, vol. 26 (September 1982), p. 541,
note that IBM, “with hard work on both sides of the ocean” has been able to adapt yield
breakthroughs and transfer them between European and American plants within IBM.
Noboru Ishihara and Hideki Wakabayashi, with Makoto Sumita, note, “Thus, much of the
accumulated knowhow with respect to semiconductor processing technology comes from a
section where it cannot be documented. For this reason, when engineers are unable to
travel, the transfer of technology becomes difficult.” See “Nomura Analyst Report: The
Semiconductor Industry in the 1990s,” Nomura Research Institute, Tokyo, 1991, p. 18. Yui
Kimura, The Japanese Semiconductor Industry: Structure, Competitive Strategies, and Per¬
formance (Greenwich, Conn.: JAI, 1988), p. 50, comments that learning economies “are
332 DUMPING IN DRAMS

A related issue is whether significant intercompany learning effects

are significant, that is, whether yields are influenced by experience not
just on a particular fab line or company, but by the experience of other
companies. Such a specification would significantly alter the details of
the model developed in this chapter.
One way to parametrize these differences in the conceptualization of
how intrafirm learning economies work is to define experience as cumu¬
lative output divided by Ky, where y takes on a value of zero if absolute,
company-wide cumulative output is the correct experience variable, and
one if experience per facility (or unit capacity) is what is relevant.42 This
means

(6-7) E = 77- with some initial E(0) = EQ

Ky

defines E(t).43 My approach will be an agnostic one: I will solve the

model using both zero and one as possible values for y, and then ask
which seems to predict more empirically plausible behavior. While the
“true” value almost certainly lies somewhere between these two ex¬
tremes, it is my prior belief that it should be substantially closer to one.
With plausible empirical assumptions, it turns out that parameter y plays
a critical role in defining the nature of an industry equilibrium and re-

only partially transferable across plants and across firms as the yields often depend on
specific conditions of fabrication processes of a particular plant.”
42. The relationship between this specification and the “per fab” and “per company”
specifications of learning effects may be sketched out as follows. Let Y be total company
output, q output per fab, K company capacity, / capacity per fab (plant size), and m the
number of fabs per company. The basic hypothesis is that E = q m'\ with p an “appropri¬
ability” parameter taking on a value of zero if only the plant’s own experience is relevant
to yields, and one if all company-owned plants’ experience is relevant. Any intermediate
degree of appropriability can be assumed by choosing the appropriate value for p. Since
K , Y ■ Y
m = and q = —, then E = E ' —r, *y = 1 — p. If capacity is measured in units
K Ky
such that / approximately equals one, then my assumed relationship holds as it stands.
If not, rescale experience variable E as £' = E f'~y, and substitute for E in equation 6-6.
43. An alternative specification might make cumulative output, or cumulative output
per unit capacity, the state variable, subject to some initial value; this alternative state
variable times K to some power would then be the argument of w, the function giving yield
per wafer. Such a specification, however, makes initial yield (with no experience) a function
of the scale of capacity investment, which is undesirable. (In that case, increasing or
decreasing capacity simply to raise initial yield on every line will play an entirely artificial
role in determining optimal capacity.)
DUMPING IN DRAMS 333

solves the B-K conundrum.44 An exact solution for E(t), assuming ca¬
pacity-constrained output (see appendix 6-B), is given by

(6-8) E{t,K) = [£’-* + - e)]TT;.

final test and assembly yields. A tested, just-fabricated, “good”

die is not yet a finished integrated circuit. The dice produced on the
wafer fabrication line must then be assembled into a sealed package,
then subjected to a rigorous final testing process. Although yields of
good, tested chips assembled from good dice may also show some
evidence of a learning curve, the impact of learning in this stage of the
integrated circuit production process is thought to be quite small rel¬
ative to learning economies in the wafer fabrication phase of integrated
circuit manufacturing.
I will model assembly and final test yields by assuming a fixed yield of
final good chips from good dice produced on the wafer fab line, v = £y;
v is “net” good, assembled and tested integrated circuits produced from
quantity y of “gross” good dice yielded by wafer fabrication. Thus, after
converting net, finished integrated circuit demand to a gross (defect-
inclusive) demand for fabricated chips, we can pose the optimization
problem in terms of choosing a time path for wafer fab output y (as
opposed to net output y), and otherwise ignore the additional yield
losses in the assembly and final test stages of production.45 In interpreting

44. The existing empirical literature on learning curves gives us little help in deciding
the correct specification. If data on cumulative output from a given facility, or aggregate
data from a group of facilities with fixed capacities, are used to estimate the relationship
y = wK using equation 6-6, we get an equation like

Ln[y(t)] = a + e Ln[Q(t)],

that is, giving the natural logarithm of total output as a linear function of the natural
logarithm of cumulative output Q, even if cumulative output per unit capacity is the relevant
experience variable. The effects of capacity size K have been absorbed into constant a.
Data from different facilities of varying size within a single company, or from different
companies, along with an additional variable controlling for capacity size, are required to
identify and estimate y.
45. The critical assumption is that all good chips coming off the wafer fab line incur all
the costs of assembly and final test before being culled.
If we denote DRAM consumers’ inverse demand function for finished chips by P(&,£y),
then the producer’s maximization problem, taking into account assembly and final test yield
losses, is to

max„(,)X f P&(t,
0
n
£y(0] 4y(0 - 7 £y
^
(0
334 DUMPING IN DRAMS

the results, we must only remember to divide all gross per unit cost and
revenue measures (price, marginal revenue, marginal cost, etc.) emerg¬
ing from the optimization analysis by in order to get the net cost and
revenue measures per good unit observed in the chip marketplace.

dram demand. We must specify a demand function for DRAMs, and

an industry structure, in order to calculate marginal revenue Ry. I shall
assume a constant-elasticity demand function of the form

(6-9) z = a Pp,

where z is aggregate demand for DRAMs, P is the price of DRAMs, and

the industry is made up of N identical firms. With this specification, we
have

(6-10)

where parameter a equals the conjectural variation plus one, divided by

N. With Cournot competition, cr is 1/A; with a constant-market-share
cartel, a = 1.

Model Solution
Next, I briefly summarize the method used to solve numerically for an
optimal policy. Full details may be found in appendixes 6-A and 6-B. It
is useful to categorize optimal policies in terms of two possibilities. One
possibility is that full-blast production is followed by an “interior seg¬
ment” where a firm is producing at less than full capacity. In this case an

with £y(t) = £uwK

E = uwK1^.

Now, if we define an inverse demand function for “gross” chips (including product ruined
in assembly and final test) by

P[x(t), y(t)] = Cy(r)]€

and substitute, we get exactly the maximization problem given earlier in equation 6-1,
where

P[x(t),y(t)} = P[x(t),y(t)]y(t).
DUMPING IN DRAMS 335

optimal policy boils down to picking both an optimal capacity K and

some optimal time ts to switch from full-blast production to constant-
output production. The other possibility is that the firm runs at full
capacity throughout the product cycle. In this regime, necessary condi¬
tions for the firm are only used to solve for an optimal capacity.

Optimal Output Decisions: With Interior Segments

When the firm produces at less than full capacity, an optimal, profit-
maximizing policy must set the difference between marginal cost and
marginal revenue equal to 8IKy, the value of an additional unit of current
production in reducing future production costs over the remainder of the
product cycle. As in the Spence model, marginal revenue along this
interior segment will be constant, equal to terminal marginal cost. We
may solve a differential equation giving experience at time ts, E(ts,K)
after a period of full-blast production through optimal switchpoint ts to
an interior segment, and using this result, derive an equation giving ts as
a function of K, N, and other parameters of the control problem:

(6-11)

•S>m„K) + <t>K'-Xl - <■) E(t„K)‘Y

This is just the condition that marginal revenue at time ts (on the left-
hand side) equals current marginal cost at terminal time 1 (on the right-
hand side).46
A second equation giving optimal capacity may be derived from equa¬
tion 6-5. After solving for 8 over both interior and boundary segments,
and substituting into equation 6-5, we have an expression implicitly giving
as a function of optimal ts, and N. Together with equation 6-11, for
given N and various other parameters, we have two equations in two
unknowns. An optimal ts and K pair must solve these two equations.

46. The expression within the outermost parentheses in the denominator on the right-
hand side of this equation is experience at terminal time 1, after producing from ts to time
1 at a constant output level; the entire denominator is yield at time 1.
336 DUMPING IN DRAMS

Optimal Capacity: With No Interior Segment

In many important cases the optimal path may not contain an interior
segment, so we never switch from full-blast production, and u{t) will
always equal one. The transversality condition (equation 6-3) will still
hold, however, and using this boundary value we can solve an equation
of motion for 8, and derive a different version of equation 6-5, which
implicitly determines A” as a function of N and other parameters.
Since, however, this expression gives us a necessary condition for op¬
timal K conditional on full capacity utilization over the entire product
cycle, we must be careful to ensure that such a path is in fact a Cournot
equilibrium. In searching for Cournot equilibria, then, attempts were
made to solve both the two-equation system characterizing an optimal
policy with interior segments, for a ts and K pair, and the single equation
giving optimal K assuming full capacity utilization throughout the prod¬
uct cycle. Solutions found were then checked as possible Cournot equi¬
libria, by perturbing both firm capacity K and switching time ts by 0.01
in all feasible directions, while maintaining the hypothesized equilibrium
output path for all other firms, and calculating the impact on firm prof¬
itability (which should necessarily be negative if the perturbations are
from a Cournot equilibrium). The procedure assures us that we have
found a local maximum satisfying the first-order necessary conditions.

Plausible Parameter Values

The final step in the simulation of firm behavior is to decide on

empirically plausible parameter values to be used in this model. A signifi¬
cant effort was devoted to constructing realistic and accurate parameter
estimates.

learning economies. I am unaware of any published studies of ex¬

perience curves in the semiconductor industry that control for the effects
of varying facility capacities (that is, that estimate y); however, there are
numerous published estimates of learning curve elasticity e based on the
relationship between the logarithm of output (or of cost) and the loga¬
rithm of cumulative output. For DRAMs there are several published
reports of an empirical 72 percent “learning curve,” meaning that current
DUMPING IN DRAMS 337

unit cost drops by 28 percent with every doubling of output, correspond¬

ing to e = 0.47.47
To estimate the parameters of the learning curve for 1M DRAMs,
estimates of “typical” wafer yields based on historical data and projec¬
tions for the last four years of a five-year product life cycle were used to
derive nonlinear least squares estimates of parameters corresponding to
£0, <tb an<J e in equation 6-6.48 Parameter y was assumed to equal one;
because the unconstrained estimate of E0 was a small number very close
to zero, I imposed a value of 0.01. This was the largest power of 10 that,
when substituted into equation 6-8 to constrain parameter estimation,
left other parameter estimates unchanged from values produced by the
unconstrained estimation procedure. Learning elasticity e was estimated
to be 0.49, while cj> had an estimated value of 31.49

47. With a constant wafer processing cost as the only cost element (the model that
underlies these studies), we have

A learning elasticity e equal to 0.47 is solved from the 72 percent learning curve, since 2~e
= .72. See Robert N. Noyce, “Microelectronics,” Scientific American, September 1977,
pp. 62-69; and Office of Technology Assessment, International Competitiveness in Elec¬
tronics, OTA-ISC-200 (1983), p. 76. Engineers at IBM, on the basis of studies of production
costs for IBM bipolar integrated circuits in the 1960s and 1970s, derived a virtually identical
71 percent learning curve. See William E. Harding, “Semiconductor Manufacturing in
IBM, 1957 to the Present: A Perspective,” IBM Journal of Research and Development, vol.
25 (September 1981), p. 652. Douglas W. Webbink’s 1977 survey of the integrated circuit
industry notes that interviewed companies believed e to generally lie in the 0.32 to 0.52
range, depending on the type of device. Staff Report on the Semiconductor Industry (Federal
Trade Commission, 1977), p. 52.
Note that B-K appear to have erred in interpreting the number reported in the 1983
Office of Technology Assessment study’s report of a 72 percent learning curve—their basis
for assuming that e = 0.28, when it actually corresponds to e = 0.47!
Irwin and Klenow, “Learning-by-Doing Spillovers,” interprets their results as reflecting
an 80 percent learning curve, but this depends critically on the assumption that producers
are not capacity constrained over their product life cycle. In a capacity-constrained envi¬
ronment, their results are consistent with a 70 percent learning curve (see note 34).
48. Since the unit of time is the (assumed five-year) product cycle, yields after two
years correspond to time .4, after three years to .6, and so on. The data are given in VLSI
Research, Depreciation Schedules: Their Impact on the Competitiveness of Semiconductor
Manufacturers, San Jose, Calif., 1990, addendum 2. The data in this addendum correspond
to a “typical” wafer fab running 2,500 wafer starts per week, at full capacity over the
product life of the 1M DRAM. Conversation with Dan Hutcheson, VLSI Research, August
19, 1991.
49. If instead y were set equal to zero, the estimate of <(> would have risen from 31 to
36, but the estimated e would not have changed.
338 DUMPING IN DRAMS

To further check whether this critical parameter seems to accurately

reflect the reality of 1M DRAM production, actual company-specific
quarterly production estimates for the six largest 1M DRAM manufac¬
turers were used to estimate learning elasticity e. “Experience” at time
t is given by

E(t) = E0 +
m
Ky ’

where Q(t) is cumulative production through time t and K is capacity.

Recall that company output at any instant (y) is just the product of yield
(w, described in equation 6-6) times utilized wafer capacity (uK). By
choosing a period of time over which capacity is approximately constant
and fully utilized (u = 1), the last term in this product will just equal a
company’s level of capacity. If E0 (the initial starting value of experience)
is small relative to Q(t)/Ky (current accumulated experience) over the
period analyzed, then

ln[y(t)] = \n($Kl-yt) + dn[Q(0] + (eE0Ky)-^-

must hold.50 The expressions in K in the above equation may be regarded

as part of firm-specific coefficients on two variables—a constant and the
inverse of cumulative output—and e the coefficient of the logarithm of
cumulative output in a regression equation.
We may wish to entertain the possibility of interfirm experience effects
(that is, that w = cj) £€ Gc, where G is the cumulative experience of the
rest of the industry, and £ is a parameter reflecting both the potential
effect of others’ experience on one’s own cumulative output and the
degree of appropriability of the experience of others. In this case we can
develop an augmented relationship between firm i’s output, and own and
rest-of-the-industry experience, through reasoning analogous to that
sketched out above, given by:

ln[y,(0] = K + bMQM] + b2\n[Q,(t)] + b3l~^~ + ,

where b t is an estimate of e, b2 is an estimate of £, overbarred variables

refer to the rest of the industry, and other coefficients (bQ, b3, b4) vary

50. This makes use of the fact that, approximately, ln(l + x) = x, for small values
of x.
DUMPING IN DRAMS 339

across firms. If interfirm learning effects on yields are important, we

would expect b2 to be positive and the coefficients of inverse rest-of-the-
industry cumulative output to be nonzero.
The equation also provides a simple test for the hypothesis that y
equals zero, since in that case the coefficient of inverse cumulative output
should be constant across firms. We would not expect this test to be a
particularly powerful one against the alternative that y equals one, how¬
ever. The coefficients on inverse cumulative output would also be ap¬
proximately constant if firms had roughly similar levels of capacity, or if
the starting value of experience E0 was very small.
The above equation was estimated using data on cumulative output
and current production for the six largest 1M DRAM producers (Toshiba,
Hitachi, Fujitsu, NEC, Mitsubishi, and Samsung) from the second
quarter of 1988 through the first quarter of 1989, a period of booming
demand when trade press accounts suggest that DRAM output was ca¬
pacity constrained.51 Results are shown in table 6-1.
The estimated coefficients provide no evidence that interfirm learning
was significant.52 All of the rest-of-the-industry coefficients were statis¬
tically indistinguishable from zero. The point estimate of £, the elasticity
of yield with respect to others’ experience, was actually negative—a
finding which, if accepted at face value, would indicate that other firms’
experience reduces one’s own yield. Again, however, none of these inter-

51. The data are Dataquest estimates of quarterly output. Reported shipments by
Motorola have been added to Toshiba’s output, and reported shipments by Intel to Sam¬
sung’s output (since it is believed that most Motorola chips were fabricated by Toshiba,
and Intel chips were “private labeled” Samsung output during this period).
52. Irwin and Klenow, “Learning-by-Doing Spillovers,” find some evidence of weak
interfirm spillovers (an additional unit of a firm’s own output contributes three times as
much relevant experience as an additional unit produced by a competitor). Their study is
critically dependent on the assumption of no capacity constraints (see note 34) and the
assumption that a firm’s marginal revenue—proportional to industry price, with a constraint
of proportionality dependent only on firm market share—is set equal to marginal cost. In
their study, however, price is measured using Dataquest’s estimates of average selling price,
which average together long-term contract prices with spot market prices actually observed
in the market during any quarter. Since marginal revenue is defined as the revenue from
an incremental unit produced and sold into the market under prevailing conditions, the
relevant price for marginal revenue calculations is arguably the spot market price. Thus,
their marginal revenue measure almost certainly underestimates true marginal revenue
(and marginal cost, assumed to be set equal by profit-maximizing producer behavior absent
capacity constraints) in tight markets, and overestimates marginal revenue and cost in slack
markets.
 ITi V“i \C «/~, l/“i sC <N
I

00 00
O' ro —
nO
On
o o o o o o

• g^
m -^j- m Tt O
ts I
UJ

•a ©
o o o o o
I

rt n a in 't (N
o oo m oo oo no
ON rn
ON 00
1 s o o o o o © °
2 *> o
</j

I
t.
3
O
On^-ONN^
nO (N| Tt Tf © O'
§ 'S no
•S
i«5
© n n ri oi n n o o o o o o
c
3

m m >n -
n n cn n \o
m
o o o o o o o
I I I I I
£

'o'
*—i m m in in r-oo-^t
rn t)- q rf in in on o
>5 On i-* »T) (N O O ^
-h h
^ h \d ri
MO X O
°)
O
©
3
C
Li_ lO
T 7 ^
tr> On
3
'O ~ a.
6) *■- 00 M r-H Tt r-t c*~)
■*S 3 OnOOOOOOM E
in rf ti- Tt tj- \o o Tf- m cj o o
O
in 6 cn ri Nt h
!j| in o o in o
i I 7 I I <u ~
<38 c £
3
D-
Source: Author’s calculations.

C 3
O
•2£5 3
a. c
3 .o
O '33 (-
c/5 £
?° £ o
oc OX)^
<n
"O <D
c 5 3 <*; 4=
OX) £ OX) O- TO O <4-.
vO 3
0
>) x>as E
5
u, O

_o n- c
•2 •§
r w
S
:- U C •? 03 •~
3 -C
/5 - U V) - Oi -C
u(5 •2- U
n Ifl3 c33 fc2<U *<u«
X5
° * _GJ w •©
C *- •—> frl w
C O '© _ W •*
E °
5
C/i .tJ ‘ZT UJ • - 3
g>0 PC E (2 I t£ z ii J p I u, z S <0
00 < to t2 x £ z 2 $ • OX)
TO qj
£ -J
</) Q
DUMPING IN DRAMS 341

firm effects were statistically significant, and a joint hypothesis that they
were all zero could not be rejected at any reasonable significance level.53
The point estimate of e in the full model and in the model with a
common coefficient on inverse cumulative output was barely over 0.5
(corresponding to a 30 percent learning curve), and 0.67 with no interfirm
effects (corresponding to a 37 percent learning curve). This confirms
other evidence suggesting substantial learning economies. A formal sta¬
tistical test for a common coefficient on inverse cumulative output (cor¬
responding to the hypothesis that y = 0) was not entirely conclusive.54
In summary, all available data seem to point to a large yield elasticity
with respect to production experience, close to 0.5, and at least some
evidence may suggest that absolute cumulative output is not an appro¬
priate choice as the “experience” variable.551 shall use .49 as my estimate
of e, and 31 as my estimate of cj> (the constant that determines chips
initially yielded per wafer).

the demand for 1m drams. There is little reliable information on

the price elasticity of demand for DRAMs. Robert Wilson, Peter Ashton,
and Thomas Egan estimate this price elasticity to range between —1.8
and —2.3 on the basis of a graph of the logarithm of bit price versus the
logarithm of bits sold; William Finan and Chris Amundsen report a price
elasticity of —1.8 on the basis of a simple regression of the logarithm of
bit price on the logarithm of bits sold worldwide.56 Neither of these

53. The F-statistic for this hypothesis was 1.12 with 7, 4 degrees of freedom; the Wald
chi-square statistic was 7.84 with 7 degrees of freedom.
54. The F-statistic leads us not to reject the hypothesis at either the 5 or the 10 percent
significance level, while the Wald chi-square statistic leads us to reject the hypothesis at
the 10 percent, but not at the 5 percent significance level.
55. Estimation of a learning elasticity requires data on either current and cumulative
output, or current average variable cost and cumulative output. The dubious practice
of using price as a proxy for current unit cost will almost certainly lead to incorrect
results, since the simple models of pricing behavior reviewed above suggest that market
prices will diverge from either current average or marginal cost.
56. Robert W Wilson, Peter K. Ashton, and Thomas P. Egan, Innovation, Competition,
and Government Policy in the Semiconductor Industry (Lexington, Mass.: Lexington Books,
1980), pp. 126-27; William F. Finan and Chris B. Amundsen, “An Analysis of the Effects
of Targeting on the Competitiveness of the U.S. Semiconductor Industry,” study prepared
for the Office of the U.S. Trade Representative, the Department of Commerce, and the
Department of Labor, Quick, Finan and Associates (Washington, 1986), p. C-18; and Finan
and Amundsen, “Modeling US-Japan Competition in Semiconductors,” Journal of Policy
Modeling, vol. 8 (Fall 1986), p. 321.
342 DUMPING IN DRAMS

estimates makes any attempt to control for the effect of variation in the
overall level of economic activity on chip demand. In previous work I
estimated an overall price elasticity of demand for semiconductors used
in the computer industry of -1.6, assuming a quality adjustment equiv¬
alent to the improvement in bit density observed in DRAMs, and chip
use in computers fixed in proportion to computer output.57
To get as reliable an estimate as possible for 1M DRAM demand, I
estimated a demand function giving the logarithm of quantity shipped of
1M DRAMs as a linear function of the logarithm of real 1M DRAM
price, logarithms of real prices for 64K and 256K DRAMs (as possible
substitutes), the logarithm of real U.S. GNP, and a linear trend.58 The
last two variables were included to capture the effect of fluctuations in
overall economic activity and intergenerational “transition” effects.59 The
implicit GNP price deflator, rebased so that the fourth quarter of 1989
was equal to one, was used to deflate all monetary values to “real” fourth-
quarter 1989 levels. Deflated GNP and substitute DRAM prices were
converted to indexes with a value of one in the fourth quarter of 1989; as
a result, the constant in a regression equation may be interpreted as the
level of DRAM demand corresponding to fourth-quarter 1989 values for
these variables.
In principle, because of the long time lag in the DRAM production
process (almost a full quarter), we would not expect disturbances in the
demand for DRAMs to have a significant effect on production within

57. See Joseph Grunwald and Kenneth Flamm, The Global Factory: Foreign Assembly
in International Trade (Brookings, 1985), pp. 130-32.
58. This demand equation should be interpreted as a “final” demand for DRAMs. That
is, demand for semiconductor inputs can be derived from a cost function C for semicon¬
ductor-using products of the form C(p,Z), where p is a vector of input prices (including
semiconductors) and Z is some given level of electronic system output making use of
semiconductor inputs. Demand for DRAMs is simply the partial derivative of C with respect
to the price of DRAMs, CPdram(p,Z). The demand for Z is a function of electronic system
prices (Pz), the prices of other finished goods (P0), and the level of aggregate economic
activity (AT, that is, Z = f(Pz,P0,X). Electronics systems prices in turn are related to
their unit cost of production C(p,Z)/Z in competitive markets, so Pz = g[C(p,Z)/Z]. We
then have Z = f{g[C{p,Z)IZ},P0X} implicitly giving Z as a function of p, P0, and X.
Substituting this implicit function for Z in derived demand for DRAMs CPdram(p,Z), we
then have a “final” demand for DRAMs as a function of input prices, prices of other
finished goods, and the overall level of economic activity.
59. The data on quantity are Dataquest estimates and cover quarterly worldwide ship¬
ments from the second quarter of 1985 to the fourth quarter of 1989 by merchant producers.
Data for DRAM prices are also quarterly Dataquest estimates of average sales price over
this same period. Real (deflated) GNP and the implicit GNP price deflator are taken from
Economic Report of the President, various years.
DUMPING IN DRAMS 343

that same quarter, and therefore we would not expect simultaneity be¬
tween price and quantity within any given quarter to be an important
complication in estimating this demand function. Nonetheless, in addition
to ordinary least squares, I used an instrumental variables estimator to
estimate the coefficients of this equation.60 The estimated regression
equations, using both methods, are shown in table 6-2.
All fully specified regressions produced an estimated price elasticity
near —1.5. Dropping either the linear time trend variable or the loga¬
rithm of real GNP as a measure of aggregate economic trends affecting
demand had little effect on the estimated own-price elasticity. Interest¬
ingly, dropping both GNP and the time trend substantially raised the
estimated price elasticity, to - 2.1, near the range where other estimates
of DRAM price elasticities—which, as already noted, ignore variation in
aggregate economic activity—have clustered.
On the basis of these results, —1.5 was used as an estimate of 1M
DRAM own-price elasticity (3, and the value 190,000 as an estimate of
product life cycle demand “level” a.61 To transform this demand function
into a demand for “gross” fabricated dice (prior to test and assembly
losses), it was assumed that net output of tested and finished chips equals
0.9 times the number of good dice produced in wafer fab.62 With the
functional form assumed, a simple transformation of a is merely substi¬
tuted for its original value in order to derive the appropriate inverse
demand function.63

cost parameters. Using estimated 1989 values found in the 1990

VLSI Research study, I estimated r (capital cost per unit of product cycle
wafer capacity) to be $240. Variable cost per wafer processed (including

60. The logarithms of cumulative output through the previous quarter of 1M, 256K,
and 64K DRAMs were used—in addition to the other exogenous variables included in
demand—to instrument DRAM price variables. Cumulative output would be expected to
affect supply of DRAMs, via learning economies, but have no direct effect on demand.
Values lagged one quarter were used as additional instruments.
61. Exp(22.97) multiplied by 20 (= 190,000 million) gives the demand that would be
observed at a 1M DRAM price of $1 over a twenty-quarter (five-year) product cycle, given
real output and substitute price levels prevailing in the fourth quarter of 1989.
62. For estimated test and assembly yields in this general neighborhood, see VLSI
Research, Depreciation Schedules, addendum A; and ICE, Mid-Term 1988, pp. 7-16
to 7-17.
63. That is, P(^z)^ = (£Ny/a)1/p^ = (Nyla')1'13, where a' - aU(1 + fi)-
344 DUMPING IN DRAMS

Table 6-2. Regression Results from Analysis of 1M DRAM Demanda

Ordinary least-squares Two-stage least-squares

Estimated Standard, Estimated Standard

Variable coefficient error t -statistic coefficient error t-statistic

Log of price
64 K 1.23 2.43 0.51 0.78 2.67 0.29
256K 0.63 1.68 0.37 -0.01 1.86 -0.01
1M -1.47 0.49 -2.99 -1.54 0.51 -3.01
Trend 0.29 0.27 1.06 0.20 0.29 0.67
Log real GNP -1.76 36.26 -0.05 10.77 39.67 0.27
Constant 22.97 1.02 22.54 23.24 1.09 21.40
R2 0.95 0.949
Std. error of regression 0.726 0.736
Degrees of freedom 12 12

Log of price
64 K 1.17 2.02 0.58 1.17 2.14 0.55
256K 0.59 1.45 0.41 0.23 1.54 0.15
1M -1.47 0.47 -3.11 -1.53 0.49 -3.14
Trend 0.28 0.10 2.81 0.27 0.10 2.70
Constant 22.97 0.98 23.51 23.22 1.04 22.43
R2 0.9503 0.95
Std. error of regression 0.698 0.702
Degrees of freedom 13 13
Log of price
64K 0.66 2.38 0.28 0.27 2.57 0.10
256K -0.07 1.56 -0.04 -0.58 1.66 -0.35
1M -1.55 0.49 -3.19 -1.61 0.50 -3.21
Log real GNP 33.99 13.77 2.47 35.70 14.32 2.49
Constant 23.01 1.02 22.47 23.33 1.08 21.56
R2 0.946 0.944
Std. error of regression 0.929 0.739
Degrees of freedom 13 13
Log of price
64K 5.40 1.65 3.27 5.46 1.74 3.14
256K 0.87 1.77 0.49 0.36 1.88 0.19
1M -2.17 0.49 -4.44 -2.19 0.52 -4.23
Constant 23.67 1.15 20.51 23.87 1.23 19.42
R2 0.92 0.919
Std. error of regression 0.852 0.857
Degrees of freedom 14 14
Source: Author’s calculations.
a. Prices instrumented with constant, trend, cumulative output of 64K, 256K, and 1M DRAMs, and LRGNP;
LRGNP and cumulative outputs lagged one year.
DUMPING IN DRAMS 345

materials, labor, and wafer probe testing) was estimated to be $390.64

Test and assembly costs were assumed to equal 23 cents for the integrated
circuit package, and about 52 cents for assembly and final testing, for a
total of 75 cents per device produced.65
Overhead is normally a significant part of semiconductor cost. From
aggregate historical data for the 1981-87 period, I have assumed 36 cents
in general, administrative, and selling costs for every dollar of direct
manufacturing cost.66 Thus, estimates for c, d, and r given above were
marked up an additional 36 percent. Table 6-3 shows the assumed em¬
pirical parameter values used.

Baseline Simulations

Table 6-4 gives the optimal values of ts and K derived from numerical
solution of the optimal control problem described above. The roots of a
system of two nonlinear equations in two unknowns (equations B-l and
B-2 in appendix 6-B), or one equation in one unknown (in the case
where full-blast production over the entire product life cycle is the opti¬
mal policy, equation B-2' in appendix 6-B) were sought. Table 6-4 also
shows a “gross rent,” defined as profits net of all costs other than fixed
entry cost F, received by each producer. The columns of table 6-4 cor¬
respond to different assumed numbers of firms in the industry, and the
rows to differing assumptions about parameter y, which defines the ex¬
perience variable relevant to learning economies.
Since identical firms are assumed to make up the industry in equilib¬
rium, one may “close” the model by assuming free entry, that is, that
firms enter the industry up to the point where gross rent per firm just
covers the fixed cost of entry F. Because we are restricted to an integer
number of firms, we define the equilibrium number of firms as that where
one more entrant reduces rent per firm below the entry cost F. As a

64. VLSI Research, Depreciation Schedules, addendum 2, puts material and labor cost
at $381 per wafer processed in 1989; a $10 wafer probe test cost is added based on ICE,
Mid-Term 1988, p. 7-9.
65. The package cost comes from conversations with Dan Hutcheson of VLSI Re¬
search, and the assembly cost is estimated to range from 7 to 20 cents offshore, or from
10 to 50 cents per device onshore (in the United States, Europe, and Japan) according to
ICE, Mid-Term 1988, pp. 7-16 to 7-18. I have used a “typical” value of 32 cents. Final
testing cost is estimated in ICE, p. 7-18 to be 20 cents per unit, for a grand total of 75 cents
for package, assembly, and final testing.
66. The data on which this calculation is based are found in ICE, Mid-Term 1988,
p. 7-20. I have excluded R&D and interest expense as demerits of overhead.
346 DUMPING IN DRAMS

Table 6-3 . Assumed Values for Empirical Parameters

Symbol Parameter Assumed value

“Level” of life-cycle demand for assembled and

a0 tested units at $l/chip 190 billion units

P Price elasticity of demand -1.5

Share of good, yielded chips as a fraction of

6 good dice after assembly and final test 0.9

Level of demand for “gross” fabricated dice

a (including units rejected at final test) a(ir(1 + p)
4> Learning curve wafer fab yield “level” parameter 31

£<> Initial “experience” at time 0 0.01

€ Experience elasticity of wafer fab yield 0.49

7 (Gamma) parameter determining experience variable 0-1

Overhead expense per dollar of direct manufacturing

m cost 0.36

Cost of packaging, assembly, and final testing per

d fabricated unit $0.75 (1 +m)

c Fabrication cost per processed wafer $390 (1 + m)

Capital cost per unit life cycle wafer processing

r capacity $240 (1 + m)
Source: Author’s calculations.

consequence of the integer number of firms, the symmetric equilibrium

so defined will generally be characterized by some small, positive rent
(net of F).
I shall assume that the fixed entry cost (primarily total R&D costs for
the 1M DRAM) that must be invested before mass production of the 1M
DRAM can occur is between roughly $250 million and $500 million.
Thus, for y = 1, if F amounts to $250 million, we would expect to find
fourteen identical firms in the industry, each with facilities capable of
producing 4.66 million wafer starts over a five-year product life cycle.
With F at $500 million we would expect nine producers, each with ca¬
pacity to produce 6.94 million wafer starts over the product cycle. In
either case the optimal policy would involve full-blast production over
the entire life cycle. Thus one observation that emerges immediately from
table 6-4 is that, with y equal to one (which I argued earlier is a heurist-
ically appealing specification), small differences in fixed entry costs can
DUMPING IN DRAMS 347

make a large difference in the industrial structure of the industry (number

of firms observed). The same cannot be said for y much less than one.
Table 6-5 summarizes some characteristics of industry equilibria de¬
rived from table 6-4 under differing assumptions about F. I have taken F
as either $500 million or $250 million; these values are best interpreted
as bracketing a range of feasible values. Alongside the equilibrium num¬
ber of firms, the Hirschman-Herfindahl index of concentration is also
shown.67
To get at the issue of whether or not dumping is observed, I have
calculated observed prices and various cost concepts at 100 equally
spaced points over the product life cycle. One useful cost concept is
current short-run marginal cost, which in my model happens to be con¬
stant at any moment in time, coincides with average variable cost, and is
equal to d + clw. This is the incremental cost saved when output is
reduced by one unit. Another important cost concept is fully allocated
long-run average cost (LRAC). To define this concept I have assumed
straight line depreciation in spreading capital and fixed entry costs over
the product life cycle: an equal amount of these fixed costs is allocated
to every moment in time. Capital and fixed entry cost per unit is then
calculated by dividing fixed costs corresponding to time t by the number
of units y(t) produced at that moment. Adding average variable cost to
average fixed cost gives LRAC = d + clw + Fly + r/uw. Multiplying
LRAC by output at any instant, and summing these costs for every instant
over the product cycle, gives the total cost of producing some time-
varying path of output over the entire product cycle.
Table 6-5 shows that, assuming y = 1, price falls short of short-run
marginal cost over the first 3 percent of the product life cycle, and falls
short of average cost over roughly the first third of the cycle. With y =
0, in contrast, price is less than marginal cost only at the very beginning
of the product cycle; price is less than average cost over two distinct
periods: at the very beginning, and over roughly the second quarter of

67. This index is defined as

HHI = 2 sf,
1= 1

where s, is the market share of company i. The index ranges in value from one, with
monopoly, to zero, with a competitive industry composed of an infinite number of equally
sized firms. In the special case of N identical firms, this index is just equal to 1 IN.
 in
sO —1 On

3*

r- -H
in co 04
04
3-

sO ^h ro
3 sO in
1—4
04
3"

00 T-H CO
co Os OS
04
3-

,
sO 1—< ro On r-H O 3 o
CO 3 O O^ o
04
r-H »n
co
sd 1 sd
in

00 r-
0
so r-H SO
CO
CO r—H SO
Os
r-H in 3 sd
1
0-' 3-
(N r-H r-H 04 r-H m so Os
CO On On r-
O
Table 6-4. Solution of the Optimal Control Problem, Nonstrategic Baseline Simulation

Tf c--
sd 3-
r-*
I
3 ' 3 r-H O 3 r-H 04
Os O On On o~ 3"
On SO
sd 06
3"
Number of firms

On ^H r-H 0- r-H 3" 00 r-

OO
SO SO
r*^
r- O
04
so r-
co
On
04
o
Os
0"* 06 Os

04
SO
00
00
3
00
r-H 04
00
so
00
r-H o
3 04
On
r- 00
OS
0^
CO
©
CM 3"
r™1

On 04 Os T-l SO 04 r-H
0^
Os
CO
co
r-H r-
so
CO o
VO
sO 04
r-H

r-H
CO
3-
Os
in
Os Os
CO
04
3-
Os in
00 r-H r-H
in r-H 04 3“
rH r—l r-H I
3" 04 co O so r-H 00 r-H SO
CO O r- 00 in 3 ■3-
Os r-H
sd 00
04^ r-H

CO
o
lO 3; O 04 04 3 r-H SO
00
Os oo SO
CO
r-H in
0^
00 Os
04
O 0
3 04 co" 04 co"

oo
o
o 04 0 r-H r~H O Os r-H O
00 Os O' Os r-H in
co CO
in 3^
CO so
3"
0 04 O' 04 0"
r-H
r-H

O'v o SO
3- r-
co
3- O
SO
CO
SO
Os
rH O
O'
CO
00
r-H O
3"
in so °o
sd
Os 04" r-H in 0^ 00"
04 CO CO co

3
a.
3a. 3 3
4—• CL
■*—* a.
OX) 3 OX) 3 3
C : O
2. s C : O
OX)
c o
, 1/5
’55
w -2 CTn |gj 00 ^
f*" so wST iH
i
4—> c ° o >>-9 © X)
O 8 >>-9 c
u “3 <D
U. II Q-O-3 U -3 D
U, H feL-Q 3 <D

3 " -n
0 > C/D « s o-T 3 CL „ . _ V5 re a.'~
° C
; o
u. c3E ^
^ 3 T3
0 c ; o u
E o ^ - ; o
3 -n > C/5

v_
P tT3 -g
; oUt
W o rwt 1 E | £ u C
3 a 3 W a re > w a
o a O
 r- CN ON CO r- SO SO in r-~ CO CO
o o <N r- (N o 00 in os 00 CO
r- o CD in
CO co CO © <N d * d
r-H I r-H r-H r-H r-H r-H r-H r-H
1 1 1 1

lO r-H 00 r-H 00 On "3- SO <T) O CN CO SO

r-H lO r—H GO r-H © r- CM 1/1 <N r-H CO OS
<N sO On t>
in d >n d o d. CO d
1 rH 1 r-H 1 r-H r-H 1
1

CO 00 r- r- SO 00 On (N CO r-H t so (N
in OS in 00 CN CO r- On in «n oo C"; CO 00
r- r—H c^.
O d d SO o d
r-H r-H r-H r-H

so SO (N r-H t3- r-H in On 00 Os Tf <N os (N

00 Os CO 00 00 so • (N r-H r-H 00 (N Os
<© o ©^
d O °i © d d Os d SO d
(N co CN co" (N co" r-H CO r-H

SO in o '3’ <N o CN 00o SO r- o <N 00 O

in OS o • 00 in 00 soCO © <N © (N
SO r-H
O (N d CO d <n O 0r OS d
CN o' r-H <N r-H CN CO r-H in
r-H r-H r-H r-H r-H

CO 00 o r- O’ o r- (N o On rH O 00 r-H o
l> On 00 oo m in <N SO in so 00 CO 00
Os 00^
a. Millions of product-cycle wafer starts.

00
© © r-* d sd d "S Tt- d
r-H Os' r-H r-H r-H Tf r-H On r-H t3"
CO in
Source: Author's calculations.

3 3 3 3 3
a CL CL
s* Q-
3 3 3 3 bJO 3
bD b£) bJD bo
O O
b. Millions of dollars.

a © 3 ^ ° C ° c £ ~
'<r> ’35 to *co £ C^T
00
so
C/3
O 3 3 sO
C/3
<U
C/3
in a> w 03
C/3
CO O < ,3 3 C/3
<U W
O o o O
d
o c
d o G
d o
C
d
o 3 o ^
Uh <u u. (U Ui o u. <L) 2 .—
■§
II CL u. II CL u II CL '5 3 u II D- CL o
S *3
3 <4—H i y <4-1
3 3 i 3 <4—1 1 I 03
3 Ui C/3 03 Ui CL , C/3 u, CL „ C/3 03 i-i CL „ . 03
,0) C/3 a> 3 C/3 -3 C/3 ,4) S g-
£ H-i
45 s ■§
O
V-(
£ <4=2
3 S 1
O
i-H
£ <4-1
.3 S -g
O
t-T
£ ^3 s -g • £ <4-1
.03
3
O
e a
w o
c
w o
3
o £03 o
CS PJ o c3 03 03
o o a o a
 (N O' so © r-
q
rsi o
o in
co
VO
cO q
© © o d —< d o d d
I I I I I I
o
I I I O Tj-
o o o o co o r- o O 00
Os co -^t CM

6? .«o o O
£ § u co
o
co
o
<N
q I
•2 5-S5 d © d o o o o o
•£* n C£ I I l l I I
a: 2 ^ o o o o o o

£
•2 =x Si i-H q q ^t q 00 q q q
-c> | 2^ d in oi d CO NO CO 00 On
~ r- CO oo 00 o NO CO rH
I- 3 - rq in q On q o ©^
s £ co" NO r-" oo" r-" oo"
| a- ^ ri
O' § 3 rH rH rH (N (N CO <N
©

>> Si
& ~
Q
c
Q
«
^ q q q q ON q q CO ON
q
NO NO NO NO oi
•Hi ill © 3
•*- oi
C (N CO in in in NO NO NO

c ©

o
o
in
&

"^3
C ~x q q in 00 q oo oo q q
a 1^5 <N CO in CO NO On rH
Co "sf in m NO ON On
O
~ IS ^ in r-
©
©_ ©_ CO q
00
q
(N
q
rH rH rH H cn"

£
ju
h>
o
a
5*.
a

5 <3c 3
1§ rH r-
NO
o
o
o
o
o
o
CO
CO
CO
CO
CO
CO
o
S | S H o
b
■*^i -s e -S T—H
i“<
NO
rH
in
q
in
q
in
q
CO
q
CO
q
CO
CO
in
q
K}
a S*_ ©^ d d d d d d d d d
■o 1 I
£

U
s
■£ X.
£ -C>
5

CO
int
vo
_cu i
3 s oo oo
q os q NO q in CO
e2 o ^ d d d d d d
 in On in in <N © m rN On 00
m (N rg © *-h © © in
f d o d d d *7 ©i ©i ©l ©1 ©
I ii ii ii i ii i i 1 1 1

0.91-

0.71-
© o o o r- o © © © On © <n
°s. "3" fN
d © ©

CO co CN o o o
o © O o o I I I
© © o © o © o o o
I I I I I I
o o o o o o

in © SO SO
SO © © CO co 00 Os
sO i-H VO 00 00 in SO CO
©^ in (N (N ©^ ©_
ri in © oo" ©" ,—T oo"
T-H r-H r-H (N (N CO <N

£ <u
— CJ
so VO so r- so CN r-
in in m sO SO
250 million

i-H 00 (N 00 rn cn
a. — o
i-H
<N o
m
in
CO
in
©
co
sO
CO
r—1
vO
On
On
00 <N o 'is
m © ©_ rn r*^ cn oo c .2
t-H T-H r-T (N rf
I «I
'5b e
D O
£> _
c« 3
u J3 o
M -O >*
£ o SJ
I <u *-
u Cl cj
O <*_ 3
S • O -O

On © o o o co co o
(N O o o o co CO o 5.2 £ c <*- >
o m in in co CO in c o JJ
o <N (N (N (N co CO <N 'ob~o ~
W c • 3
•O « O.
w - u t:
»sEg
(D V
•- ^
« ° ea 3 q O
£ "2
3=0 =
.9 -2 Z%
o tJ £ c
3 3 •“

£ — a «? lift
Cl Cl 2 ,E
^ c/5
w F" O C 3 ~
a 55 w> " Q. 2. « 2
iiEo. L C P U W
l ir v

>* O o o -o c
is) O
c c 2 2
s 5 a.>~
o o c c <u o_
kT 2 « «.S 3
© 2 S C.2
5 > ►5 — uj u.
oo
so so m CO
Os 00 c/3 c/3 s o.y T3 a> ••

© © © O © ©
352 DUMPING IN DRAMS

the cycle. Indeed, for all values of y, given my assumptions about other
parameter values, price falls short of marginal cost only at the very
beginning of the product cycle. Further perusal of this table makes clear,
however, that the timing of periods of sales at less than average cost is
quite sensitive to the specification of the experience variable—depending
on y, such episodes can occur at the beginning of the product cycle, the
middle, or the end, or in some combination.
Table 6-5 also shows that the value of y makes a big difference in the
structure of a symmetric-industry equilibrium. With cumulative output
per facility (y = 1) the relevant experience variable, a relatively large
number of firms (nine to fourteen) populate the industry. With y much
below .9, no more than three or four firms make up the industry.
Figure 6-3 shows the path of price, marginal revenue, marginal cost,
and average cost over time in the case where entry costs are $250 million
and y equals one. Figure 6-4 shows the time path for these variables over
the product cycle when y is instead equal to zero.
Ironically, the specification of firm behavior in the B-K model—full-
blast production over the entire product cycle—turns out to be optimal
if parameter y is close to one (see table 6-4). The irony arises because
the B-K model also specifies absolute cumulative output (y = 0) as the
experience variable, and given realistic choices for other parameters,
optimal behavior would then require cutting back production to levels
below capacity after about the first third of the product cycle.

reality checks. How plausible are these simulations, and do they

suggest anything about the realism of various assumptions about param¬
eters? One straightforward way to evaluate the model is to compare the
predicted industry structure with observed industry structure. Recall that
figure 5-15 showed Hirschman-Herfindahl concentration indexes con¬
structed from Dataquest estimates of annual producer shipments of var¬
ious generations of DRAMs.68 For virtually all generations of DRAM

68. These indexes are calculated from unpublished Dataquest estimates of DRAMs
shipped from 1974 through the end of 1989. Note that there were two distinct varieties of
16K DRAM, one with a single voltage power source, the other requiring dual voltages;
each is treated as a separate product in this figure. In calculating concentration indexes for
1M DRAMs, I have allocated Motorola-labeled product to Toshiba (since virtually all of
Motorola’s product over this period is believed to have been assembled from Toshiba-
fabricated dice, or produced by a Toshiba-Motorola joint venture); 1M DRAMs bearing
the Intel label have been assigned to Samsung, since it is believed that virtually all of Intel’s
sales over this period were “private labeled” Samsung product. Neither of these adjust¬
ments has a particularly significant effect on the pattern of concentration.
DUMPING IN DRAMS 353
Figure 6-3. Time Profile of 1M DRAM Costs and Prices in Simulated
Nonstrategic Equilibrium with Gamma = la
Dollars per chip

Source: Author’s calculations.

a. MC = short-run marginal cost; AC = long-run average cost; P = price; MR = marginal revenue.

the concentration index declines sharply from an very high initial level,
as one producer after another comes on line with volume production.
The index then levels off near .1, rising sharply at the end of the product
cycle as producers drop the product line one after another. Although the
early phases of the 256K and 1M DRAM may have been somewhat more
concentrated than in earlier generations’ life cycles, they too seem des¬
tined to eventually follow this pattern.
If one compares the Hirschman-Herfindahl indexes associated with
my simulations with the pattern depicted in figure 5-15, only the re¬
sults associated with the specification of cumulative output per facility
354 DUMPING IN DRAMS

Figure 6-4. Time Profile of Semiconductor Costs and Prices in

Simulated Nonstrategic Equilibrium with Gamma = 0°

Dollars per chip

Source: Author's calculations.

a. MC = short-run marginal cost; AC = long-run average cost; P = price; MR = marginal revenue.

(7 = 1) as the experience variable fit reasonably closely. Note that my

assumption of symmetric firms means that the associated Hirschman-
Herfindahl index of concentration must be constant over time. While
conceding that my model is at best an approximation to reality, I conclude
that only a value of 7 close to one yields predicted behavior that is
reasonably close to industrial reality.
Another cut at this question may be had by comparing predicted with
actual paths for DRAM prices over time. To do so, I have assumed that
a five-year product cycle for the 1M DRAM effectively began in 1988
(although small quantities were produced as far back as late 1985, quan-
DUMPING IN DRAMS 355

Figure 6-5. Historical Prices for 1M DRAMs Compared with Simulated

Nonstrategic Equilibrium Time Profiles

Dollars per chip

Source: Author’s calculations.

tity production did not really ramp up until 1988). Figure 6-5 charts the
actual behavior of one set of estimates of large-volume contract prices
for 1M DRAMs in the U.S. and Japanese markets through autumn 1993,
along with simulated 1M DRAM price levels associated with assumed 7

equal to one and zero, respectively.69 The period from the first quarter
of 1988 through the first quarter of 1989 was a period of extreme shortage
in DRAM markets, whereas the period after late 1989 was marked by
lackluster demand. Given that the early portion of my empirical approx¬
imation to the learning curve is probably poorer than in later periods
(see the discussion above), and that my assumption of symmetric firms
is probably least appropriate in the early stages of the product cycle, I
am not surprised to find that the very earliest part of the predicted time

69. These data are monthly averages of Nikkei and Dataquest estimates of average
contract prices in these markets. The data are reported in Computer Reseller News and
Nihon Keizai Shimbun, various issues. For more on the strengths and weaknesses of these
data, and a thorough discussion of the segmented spot and contract markets in which
DRAMs are sold, see Flamm, “Measurement of DRAM Prices,” pp. 157-97.
356 DUMPING IN DRAMS

path for prices seems the least accurate. All things considered, the sim¬
ulation with 7=1 seems to do a reasonable job of tracking real 1M
DRAM prices. The simulation with 7 = 0 clearly does not.
Thus, two pieces of evidence—observed and predicted concentration
indexes, and the time path of DRAM prices—seem to suggest that a
value of 7 close to one provides significantly more realistic predictions
than a value close to zero.
A final point to consider is that, according to industry folklore, DRAM
producers have traditionally run their plants at full blast when they were
in operation at all, a behavior that is consistent with the simulations
presented here. However, beginning in mid-1989 Japanese DRAM pro¬
ducers announced production cutbacks. This raises three issues that I do
not explore further here. First, DRAM capacity may be shifted, at some
cost, to production of other types of integrated circuits, a possibility not
explicitly incorporated into my model. Second, DRAM demand is noto¬
riously cyclical, and the consequences of shifts in demand for optimal
producer behavior are, again, not explicitly explored here. Third, pro¬
duction of DRAMs after the conclusion of the 1986 Semiconductor Trade
Arrangement was clearly affected by political constraints, may have led
to a degree of collusive behavior among producers, and otherwise in¬
volved political economic factors not incorporated into my model.

the dumping issue. Given empirical values deemed to be plausible

in the case of 1M DRAMs, the exercises portrayed in tables 6-4 and 6-5
suggest that use of a short-run marginal cost test for dumping as a screen
for potentially predatory behavior is only likely to give false positive
results (pricing below current marginal cost absent strategic behavior) in
the very earliest stages of the product cycle. One might interpret this to
mean that a marginal cost-based dumping test might be defensible if
some sort of exception to a marginal cost-based pricing standard is
granted when a new product is first introduced. But it is not clear how
robust this conclusion is to changes in the empirical parameters used in
my simulations; further sensitivity analysis might shed greater light on
this question.
The same cannot be said for an average cost test for predation. De¬
pending on parameter values, episodes of below-average-cost pricing can
occur in virtually any part of the product life cycle, even when producer
behavior is entirely nonstrategic.
DUMPING IN DRAMS 357

Figure 6-6. Time Profile of Semiconductor Costs and Prices in

Simulation with Very High Initial Yields“

Dollars per chip

Source: Author’s calculations.

a. MC = short-run marginal cost; AC = long-run average cost; P = price; MR = marginal revenue.

Indeed, while the simulations depicted in tables 6-4 and 6-5 all show
an episode of below-average-cost pricing at the beginning of the product
cycle, possibly followed by a later episode, it would be incorrect to assert
that below-average-cost pricing will always necessarily be observed at the
beginning of the product life cycle.70 Figure 6-6 shows that by artfully
changing a single parameter (in this case, by greatly raising initial yields,
making E0 = 500), assuming y = 1 and F = $250 million, one arrives
at a symmetric-industry equilibrium where price never falls below mar-

70. Dick, “Learning by Doing,” pp. 144-46, proposes this behavior.

358 DUMPING IN DRAMS

ginal cost, and price falls below average cost only during the latter half
of the product cycle.71

Conclusions
In recent years, pricing below a constructed long-run average cost has
become the principal grounds for applying the U.S. dumping laws to
U.S. imports of some foreign products. Although this practice has little
obvious economic defense, it is possible to argue that a test based on
marginal cost might serve as a useful screen for potentially predatory
behavior by foreign exporters. However, in the presence of learning econ¬
omies, such as are thought to be present in many high-tech industries,
including semiconductors, below-marginal-cost pricing can be rational
even in the absence of strategic behavior such as predation.
In this chapter I have developed a more realistic model of pricing over
the life cycle of a product in which both fixed costs and learning econ¬
omies are significant. Using empirically plausible parameters for produc¬
tion of 1M DRAMs, and assuming nonstrategic producer behavior, I
have found that below-marginal-cost pricing is likely to be observed only
in the very earliest stages of the product cycle.
The analysis has also shed considerable light on other facets of pricing
and production over the product life cycle. A specification of learning
economies based on cumulative output per facility as the “experience”
variable was found to yield results that were considerably more realistic
than other possible—and widely used—specifications. Contrary to pop¬
ular belief, below-average-cost pricing does not necessarily have to occur
near the beginning of the product cycle.
The model presented here appears to produce fairly realistic predic¬
tions of industry structure and pricing behavior when used with empiri¬
cally plausible parameters. Extensions of this work to consider the pos¬
sibility of strategic, noncooperative behavior on the part of producers,
as well as cooperative or collusive behavior, are explored in the next
chapter. Policy issues to be explored within this framework will focus on
the impact of government policies on industry structure and aggregate
national welfare.

71. The symmetric equilibrium depicted in this figure corresponds to an industry with
eighteen producers, each with a product life cycle capacity of 3.88 million wafer starts,
producing full blast over the entire cycle.
DUMPING IN DRAMS 359

Appendix 6-A: A General Solution to the Problem of Optimal

Capacity and Production Choice
Let:

y® = company’s output at time t

x(t) = output of other companies
F = fixed cost of entry
/?LK0,*(0] = company’s revenues at time t
d = assembly and test cost per chip
wafer processing cost
u{t) = utilization rate at time t
K = fixed capacity
capital cost per unit capacity
w = yielded good chips per wafer
Q = cumulative company output
y = the experience specification parameter, 0 < 7 < 1
E = “yield relevant” experience (E0 + Q!Ky).

The general problem described in the text is:

1

max „(,).* /{^LK0> *(01 - F - dy(t) - cu(t)K - rK}dt,

where y(t) = u{t) w(E) K,

y(0
s.t. E = — = u(t) w(E) K' y w£ > 0,

ue[0, 1], w > 0.

Form the Hamiltonian (suppressing the arguments of functions for

notational simplicity):

H=R-F-dy- cuK - rK + 5 -y-

J Ky

... <5. dH dH dy
d + uweK,
W1,h8= -^= IC*

dR
where Rv denotes —, marginal revenue.
y dy
360 DUMPING IN DRAMS

By the Maximum Principle:

choose u{t) to maximize H, given we [0,1], with 8(1) = 0 (transversality

condition).

There are three possible cases to consider:

0<«<l;^= k-d + l-P cK = 0

du \ y Ky du
8 c\
(A-l.a) (r. d + i-
—-b- -\Kw 0, 8 =-uweK < 0,
Ky w w

Ry-d-- + ^r = 0.
y w Ky

At an interior maximum, we also have a second-order necessary condi¬

tion:

d2H
Ryy (Kwf < 0.
du2

c 8
u ~ 1 R — d-h — >0
(A-l.b) * w Ky

8 = — iyRy ~ d + Y^WeK < °-

c 8
(A-l.c) w = 0; R - d-f —-- < 0 8 = 0.
y w Ky

Together with the transversality condition, this implies that 8 > 0

everywhere.
Along an interior segment, that is, possibility A.l.a,

_8_
Rv — d + — —
y w Ky

(A-l.d) (Ry) I 2_\ _ A.

\ w2 We Ky) Ky

c — uwK l--,
w ~cu~— =0-

which means Ry is constant along an interior segment.

The above analysis holds for any given K. Full optimization requires
that optimal parameter K must be chosen to satisfy:
DUMPING IN DRAMS 361

m 8(q, k, U*(t), £*(01 dt = Q

where u*(t) is the optimal utilization rate and E* the trajectory of ex¬
perience variable E corresponding to u*(t).72 Then
1 r

S {R>-d +
oL
f _ , 8 j dy ^ dx
+ R- jr ~ Kl
cu dt = 0.

For the moment, I will assume nonstrategic behavior—the firm perceives

dx n
6K = °’ S°'

1r
In , 8 8
I /vv — a + —- ww - y — uw - cu - r dt = 0.
i \ y Ky K'1
o L

Thus,
1 r
. C 8 8
IKy - d-+ — ~ 7 — uw - r dt — 0.
i \ y w KV a7
oL

The expression in parentheses above:

= 0 if u is interior;
> 0 if u = 1;
< 0 if u = 0;

while the remaining expressions in brackets sum to a negative number.

This shows that the expression in parentheses must be positive, and
u = 1, over some interval if optimal K > 0 (and any output is produced).
The identical trajectory of output (and variable profit) could otherwise
be produced at lower cost by choosing some smaller K, then choosing
utilization rate u(t) = K/K u(t) < 1.
The result just obtained is perfectly general. To explicitly solve for an
optimal path, we add additional structure and make the following further
assumptions:

72. See George Leitman, An Introduction to Optimal Control (McGraw-Hill, 1966),

pp. 98-100.
362 DUMPING IN DRAMS

—Industry inverse demand (and firm revenue function R) and learning

function w are twice continuously differentiable functions in all their
arguments.
—There are N symmetric firms.
—Industry revenues are an autonomous, strictly concave function of
industry output.
Let R{z) be industry revenues and z(t) industry output, such that
z(t) = x(t) + y(t); then R(z) = P(z)z, where P(z) is industry inverse
demand.
I will assume Rzz = P"z + 2P' < 0, for z > 0, (that is, R is strictly
concave). So

IP'
(A-2.a) P" <-for z > 0.
z

I have also assumed that 7?[y(t),x(t)] is autonomous, not a function of

time other than through x(t) and y(t). Since output is a perfectly homo¬
geneous commodity, with a single market price assumed,

^LKOaCO] = p(x + y)y,

dz
Ry = P' —y + P = P\ 1 + \)y + P,

with X (= dx/dy) the conjectural variation perceived by the firm. I shall

regard X as a constant varying between 0 and N — 1.
These limits parametrize X as lying between two useful limiting cases
of industrial organization:

X = 0 with Cournot competition,

X = N — 1 with a constant-market-share cartel made up of N iden¬
tical firms.

I will be assuming X = 0 for the moment, but I will develop my analysis

of optimal utilization rates including the case of a constant-market-share
cartel.
Note that

p" if) y + 2P' %= (1 + x) [p"(l + x)y + 2p,]

< (1 + X)2P' 1 - ^(1 + X) using (A-2.a).

DUMPING IN DRAMS 363

Now, consider two cases. First, under Cournot competition, X = 0. Since

- < 1, 1 - -(1 + X) > 0, and Rvv < 0.

2 2

Second, with N identical firms in a collusive, fixed-market-share cartel,

X - N — 1,

- = |7, 1 - Z(1 + X) = 0, and Ryy < 0.

zNz yy

In these two cases, firm revenue R is strictly concave in y, with Ryy <
0 everywhere. Now, since functions R and w are assumed twice continu¬
ously differentiable in their arguments, so too will be the Hamiltonian
H, with Huu = Ryy (Kw)2 < 0. H is strictly concave in u. For given K,
then, the necessary conditions for optimal u are also sufficient to guar¬
antee maximization of H. More important, the strict concavity of H in
u, and the constant bounds constraining feasible u, mean that we may
invoke an appropriate theorem to conclude that optimal u is continuous
in other arguments of H,73 Since the other arguments of H are continuous
functions of time, u must be as well.
Note also that u(t) is a continuously differentiable function of time
within an interior segment. This is a consequence of the linearity of E
in u, and the strict concavity of the Hamiltonian.74

73. For example, a result proven by Debreu, found in Kevin Lancaster, Mathematical
Economics (Dover, 1987), pp. 349-50, or a theorem due to Fiacco cited in Garth P.
McCormick, Nonlinear Programming: Theory, Algorithms, and Applications (Wiley, 1983),
pp. 245-46.
More directly, we may note that the strict concavity of H in u means that, at every
moment, the u(t) that maximizes H is unique. Since the set of feasible values from which
u(t) is chosen is compact, we may invoke theorem 6.1 from Wendell H. Fleming and
Raymond W. Rishel, Deterministic and Stochastic Optimal Control (Springer-Verlag, 1975),
p. 75, to conclude that u(t) is a continuous function of time.
74. Consider optimal control u*(t) over some interior segment tt < t < t2. Because of
the strict concavity of the Hamiltonian, a local interior maximum of H must also be a
global maximum of H with no constraints on control u{t). Optimal control u*{t) over this
interval must also be the optimal control for the problem

[/?(jc,y) — F — dy — cuK — rK]dt,

'i

subject to the initial and terminal conditions that E(t) take on the values at times t, and t2
associated with optimal control u*(t) in the original problem, but with no bounds on control
u(t). As Fleming and Rishel, Deterministic and Stochastic Optimal Control (corollary 6.1,
364 DUMPING IN DRAMS

Next I restrict discussion to symmetric-industry equilibria, in which

the industry is made up of identical firms. In this case, I will show that
along an interior segment of such an equilibrium, ii < 0.
We already know (see possibility A.l.d) that along an interior
segment,

(Ry) = 0

= |[/”(1 + % + P]

= P"z( 1 + \)y + P'( 1 + X)y + P'z

= P"Ny(l + X)y + P'(l + X)y + P'Ny = 0

= y[P"( 1+ X)Ny + P'{ 1 + X) + P'N]

Is it possible that the expression in brackets equals zero?

If P"( 1 + X)Ny + P'(l + X) + P'N = 0

P'( 1 + X + N)
P" =
z(1 + X)
N ^
P' 1 +
1 + Xy
P" = >
z
N
since 1 + > 2.
1 + X

But this contradicts (A.2.a) and our assumption that Rzz < 0. So we
must have y = 0.

Now y = uKw
y
y = uKw + uKwe — = 0, which can only be true if
Ky

/. u < 0.

Since u is continuous over time, when u = 0, it cannot jump to 1. Indeed,

u cannot even become positive since ii < 0 as soon as u > 0. So, when
u = 0, the optimal policy must remain u = 0.

p. 77), note, the linearity of the equations of motion in the control variable, and the
concavity of the Hamiltonian, are sufficient to guarantee that the optimal control is
a continuously differentiable function of time for this subproblem, with no constraints
on u{t).
DUMPING IN DRAMS 365

If we add the further assumption that Rz(0) exceeds current marginal

cost at time 0 (as must be true with constant elasticity < — 1), then

A-l.c cannot hold, it can never be optimal for u = 0 and nothing ever
produced.
Note the role of learning economies in this model. If wE = 0 every¬
where (there is no learning), 8 must be constant and equal to zero, and
optimal u must be constant over the product cycle. Optimal capacity
choice implies Ry — d — c/w must be greater than zero, and therefore
u = 1. The first-order conditions then require that K be chosen so
that variable profit generated by a marginal unit of capital
(Ry — d — c/w)w just covers the cost of capital (r) and all available
capacity is fully utilized over the entire product life cycle.

Appendix 6-B: Detailed Solution with Specific Demand and

Learning Curve Assumptions

Specification of the Demand Curve

Let inverse demand be given by P = j with z = x + y, total

industry output.
Consider industry sales, given by

and note that lim^0 Rz =

Now, marginal revenue for any individual firm is given by
366 DUMPING IN DRAMS

*■ 4+§+p P ^ii+D
Pz
+ 1

P - - (1 + X) + 1

where X = ^, the conjectural variation:

dy

X = N — 1 for a constant-market-share cartel;

X = 0 under Cournot competition.

In addition, I assume that the industry is made up of N identical firms.

y 1
In symmetric-industry equilibrium, each firm has market share - = —

1 + X
Let ct =
N

So Ry

1, cartel;
where a =
7-, Cournot.
\N

Thus, in an industry made up of N identical firms,

1

In y\ l(T ^
5 ~ + 1
\ a ) IP /

Specification of Learning Economies and Output

Let w = <j)£e, with E = y/Ky, E{0) = E0, 0 < e < 1, 0 < y < 1.
Initial yield (<j>£o) is independent of capacity choice. We are mainly
interested in two specific cases: y = 0 (learning depends on absolute
production experience), and y = 1 (learning depends on experience per
unit capacity, or facility).
Then y = fyE^uK.
Over the interval from zero to ts, where u — 1,

E = = c}) £* K'~y.
Ky
DUMPING IN DRAMS 367

Solving this differential equation, we have

E(t,K) = 4>t(l - e)]T^7, and y(t,K) = d> E(t,KyK.

We also know

dH
8 = weK
dE

e <j> £e_1 tf,

which can be rewritten as the linear monic differential equation

8 + 8 = f2(t,K), with

= e <}> E(t,KY~1 K1~y,f2(t,K) = ~[Ry(t,K) -

Solving for 8, for some t in the interval (0, ts), given boundary value
8(4) = we have

8„ +
8(() =
,//i(t) di

Define function 8B(t,ts,K,ba) by the right-hand side of this equation.

With some difficulty, and a great deal of tedious algebra, we can then
integrate this expression, and have

p + g /4>a:Mp E(t,K)
8b(44,K,8J + E(ts,K)f 1 -
(3 + 1 \ a / E(ts,K)

- d {l- E(t,K)
vi / E(t,K)'

1 [E(t„K) \ I J / [E(I„K)\

Over the interval from ts to one, optimal u is set such that y{t,k) =
y(ts,k)—in other words, constant output. E(ts,k), y(ts,k) are given by
expressions in the last section. Over the interval from ts to one, then,

z, _ yb’K)
368 DUMPING IN DRAMS

so E(t,k) = (l- o + E(t,,K);

E(t,K) = (I - /,) <\,E(t,,Ky A"-' + E(t„K)

= £(AA)[1 + 0(m,A)],
where = <J)E(ts,K)e~' K'~y(t - ts).
0 is the ratio of incremental experience from time ts to time t, to
experience at ts;

c u wEK
8 = -

w w2
<\>2E(ts,Ky KeE(t,KY
-c = ; >„K);
VE(t,Kf

where

-czE(ts,Ky K
hit ; ts,K)
E(t,Ky + '

Conditional on assumed switchpoint ts, the solution to this differential

equation can be written as

5(1) - 8(1) = J /,(t ; t„K) d t.

or, making use of the fact that 8(1) = 0 (the transversality condition),
define function hE as

I ts’K) f /3(t ; ts,K) d T

cK1
<y[E{ts,K) + <yK'-\t-ts) E(ts,Kyy
cKy
4>[E(ts,K) + ^-^(1-0 £(f, ,*)«]«
cKr> f 1 1
mts,Ky |[i + e(t,ts,K)Y “ [i + 0(1 ,ts,K)Y
In particular, we can solve for 8E(ts ; ts,k), that is, the value of 8 at ts
which, given some assumed ts, solves the equation of motion over an
DUMPING IN DRAMS 369

interior segment and the transversality condition at time 1. In this special

case, we have

Solution of the Model

Our specification has ruled out the possibility that optimal u = 0. The
optimal path for u will consist of a capacity-constrained (u = 1) segment
through time ts, possibly followed by an interior segment. For the mo¬
ment, assume that ts < 1 .

Case 1. ts < 1. At ts, we also know that u(t) will be entering an interior
segment, so

Thus, for given K, optimal ts must satisfy

or, substituting,

—c
<\>[E(ts,K) + ^~^l-ts)E(ts,KYY
(B-l)

This is just the condition that marginal revenue equal current marginal
cost at terminal time 1. If K is chosen nonstrategically to maximize
profits, optimal K must satisfy
370 DUMPING IN DRAMS

Since the expression in brackets is zero after ts, and u is one before ts,
this can be written as

I ts ,

j - y ~ w\>E(t,K)• dt + I W(r,K) - dWE(t,Kr

y)J
ts 0 l

+ Yy <\)E(t,KY(l - dt - cts - r = 0.

We may substitute for 8 with the functions 8B and 8£, described earlier.
Noting that

= y(t„K) = m^KYK = E(t„KY = 1

“ ’ KitFM.Ky KitEU.KY E(t,K)‘ [1 +

over the interval from ts to one, we can solve analytically for the first
integral above, so, substituting,

1
+ e
C7(l~4) e(u,/Q 1
+ j [«„(»,jo - d\
(i-0 [i + e(u,*)]' Q(us,k)

(B-2) ' + 8J t,ts,K, Ey(ts,K) + d + tP\

mts,KY
4> E(t,KY
(1-7) dt — ct, — r = 0.

An optimal choice of ts and K, then, must solve equations B-l and B-2.

Case 2. ts = 1. The other possibility is that ts = 1, and producers fully

utilize available capacity throughout the product life cycle. Since there is
no interior segment, equation B-l does not have to hold at ts. Instead,
the transversality condition means that 8(ts) = 0, so 8s(t, 1, K, 0) gives
the value of 8(f) at any time t. Incorporating this into the first-order
condition for optimal K, I then have

f W(f,K) - d] ■ (j) E(t,KY + hB(t,l,K,0)

0 I

(B-2') •(l-7)6^pj*— c-r = 0.

DUMPING IN DRAMS 371

which can be solved for optimal K.

In searching for Cournot equilibria, then, attempts were made to solve
equations B-l and B-2 for optimal (ts, K), and equation B-2' for optimal
K (assuming ts = 1).

Industry Profits

Total rents per firm, gross of fixed entry cost F, earned in an industry
made up of N identical firms can be calculated as (for given optimal ts
and K):
i

(moo'
1 a
- d y{t,K) — cu{t)K - rK\ dt, or

11 Ny(t,KY
a
- d y(t,K) | dt +
Ny(t„K)Y
a
- d y(ts,K)(l-ts)

1
+ i
cKl -
cK(l-ts) e(U,£) - rK.
(1-0 [1 + 0(1,4, X)]' 6(1 ,t„K)
 CHAPTER SEVEN

Strategic Issues

Today it is frequently asserted in the United States that the semicon¬

ductor industry is a “strategic” or “critical” industry. Indeed the National
Advisory Committee on Semiconductors (NACS), formed by congres¬
sional mandate, titled its first report to the president A Strategic Industry
at Risk.1

Conceptions of “Strategic”
What precisely is meant by this use of the word “strategic”? There
are at least three senses in which the word might apply to the semicon¬
ductor industry.

National Security

The NACS clearly had several concepts of “strategic” in mind. First,

and most simply, the committee asserted that America’s national se¬
curity depended on the capabilities of its domestic semiconductor in¬
dustry. The NACS’s first report repeated the conclusions of a 1987
Defense Science Board task force’s report on defense semiconductor
dependency, arguing that U.S. and NATO forces “rely on a technolog-

1. National Advisory Committee on Semiconductors (NACS), A Strategic Industry at

Risk: A Report to the President and the Congress (Washington, 1989). The NACS issued its
last report in 1992, before disbanding.

372
STRATEGIC ISSUES 373

ical advantage ultimately traceable to semiconductors to offset the

numerical superiority of our adversaries.”2 It is hard to deny the force
of this point, particularly in light of the historical role that the U.S.
military played in driving the initial development and growth of the
American semiconductor industry.
But the national security argument also raises nettlesome questions.
The serious competition for American semiconductor companies comes
from allied nations, not adversaries. Is there a serious threat that imports
from our allies will be denied us?3 Even if they were, the United States
clearly does not lack the technical capability to develop the same products
at home. Certainly, in that event, the United States would be stimulated
to develop domestic sources of supply, and it is hard to envision a sudden
military threat that would last long enough to make the time lag a serious
issue. What is in doubt is not our absolute technical capacity to manu¬
facture this or that semiconductor product, but our ability to manufacture
semiconductors cheaply enough to be competitive in commercial mar¬
kets. Maintaining a captive, technically advanced domestic capability
to produce products for the military, albeit at high cost, rather than solv¬
ing a military challenge, might worsen what is essentially an economic
problem.
Whatever the abstract merits of the national security argument, the
defense semiconductor dependency issue has been closely tied to more
purely commercial concerns. The Defense Science Board’s 1987 argu¬
ment for intervention to strengthen the U.S. semiconductor industry ex¬
plicitly as follows: because “competitive, high-volume production is the
key to leadership in semiconductors,” and “high-volume production is
supported by the commercial market,” and “leadership in commercial
volume production is being lost by the U.S. semiconductor industry,”
“U.S. Defense will soon depend on foreign sources for state-of-the-art
technology in semiconductors. The Task Force views this as an unac¬
ceptable situation.”4

2. NACS, Strategic Industry at Risk, p. 6.

3. Although Shintaro Ishihara, a prominent right-wing politician in Japan, was to fuel
this concern by suggesting in The Japan That Can Say No, trans. Frank Baldwin (Simon &
Schuster, 1991), written with Sony’s chief executive Akio Morita, that Japan had the power
to withhold advanced chips, or make them available to the Soviet Union, and so alter the
global balance of power.
4. Report of the Defense Science Board Task Force on Defense Semiconductor Depen¬
dency (Office of the Under Secretary of Defense for Acquisition, February 1987), pp. 1-2.
374 STRATEGIC ISSUES

Clearly, however, the same logic (because of the shortcomings of do¬

mestic producers in commercial markets, U.S. defense will soon depend
on foreign sources for state-of-the-art technology in product X) could be
used to argue for intervention to maintain commercial preeminence in
any or all high-technology industries (substitute for product X chemicals,
biotechnology, communications, computers, aircraft, or instruments),
virtually all of which have ties to some set of defense systems. Thus,
pushed to its logical extreme, the national security argument calls for
across-the-board market leadership in virtually all high-technology fields,
and an end to the notion that competitive outcomes—on a global scale—
should be left to impersonal market forces in this large and important set
of industries.
The national security argument, if accepted and pushed to its logical
limits, means that competitive market outcomes in high-technology prod¬
ucts are acceptable only if they leave domestic producers (excluding even
those of our allies and trading partners) firmly in command of global
markets. This clearly is a recipe for the breakdown of an open interna¬
tional trading system in high-technology goods. If all nations were to take
the argument seriously, high-tech autarky would be the inevitable result.
If this attitude were to prevail in its most extreme form, there would be
little need for further analysis, and the economic content of discussions
of national policies toward high-technology industries would simply be
about the least costly and most efficient means of attaining national self-
sufficiency in these industries.
Economic realities, however, seem destined to doom this approach.
The range of industries that could be labeled as high technology is so
great, and growing so rapidly, that any policy of broad self-sufficiency
would appear to be insupportably costly, no matter what its perceived
merits from a strictly defense perspective. Furthermore, the revolution
in transportation and communications technologies has pulled the
world together in a way that makes it increasingly difficult to talk about
purely national technological capabilities. Multinational firms based
in different parts of the world link up with one another in a variety of
types of alliances, and pull together specialized technical resources
scattered about the globe, in order to create leading edge, high-tech
products. However militarily appealing in the abstract, high-tech self-
sufficiency is likely to be neither affordable nor feasible. Thus we turn
to other meanings of the word “strategic” for the analytical purposes
of this chapter.
STRATEGIC ISSUES 375

Figure 7-1. “Food Chain’ Concept in Semiconductor Industry

United States World World

1990 1990 2000

Total manufacturing
\ $5.4 trillion $20 trillion $40 trillion /
and service economy

Electronics products
\ $384 billion $751 billion $2 trillion /
and services

Semiconductor
manufacturing \ $21 billion $63 billion $200 billion /

Semiconductor \ .....
. • , . , \ $9 billion $20 billion $60 billion /
materials and equipment \

Fundamental sciences
and processes Software

Gases and Chemicals

Optics, materials, robotics

Source: National Advisory Committee on Semiconductors, Attaining Preeminence in Semiconductors (Arlington,

Va.: February 1992), p. 9.

Strategic Sectors

Another set of arguments advanced by the NACS and others holds

that certain industries are “strategic” because of their linkages to the
rest of the economy. The simplest of these arguments is the “food chain”
theory, often illustrated by something like figure 7-1. Upstream and
downstream industries’ competitive fortunes are interlinked in a complex
ecological system that makes each dependent on the health of the others.
As a report by the Semiconductor Industry Association put it, “The
elements of the electronics industry are analogous to a ‘food chain,’ in
which each component level, from silicon wafers up to finished electronic
products, is dependent on the others. If one link is damaged, the others
are automatically injured.”5

5. Semiconductor Industry Association (SIA), Key Facts and Issues (1989), p. 24. The
376 STRATEGIC ISSUES

The difficulty with this argument in its simplest form may be discerned
by glancing at the input-output table for any major economy. Virtually
all goods (other than final consumer goods) produced by any industrial
sector also serve as inputs into most other sectors. Thus any policy of
labeling as “strategic” industries those that supply significant inputs to
other industries ends up declaring virtually all industries strategic. A
policy arguing for support of “strategic” domestic industries then be¬
comes a recipe for intervening to support virtually anything.
Clearly, if competitive products are available from foreign suppliers at
a competitive price, no harm is necessarily done to other industries up
and down the food chain. Indeed, a favorite response of skeptical econ¬
omists to claims that the U.S. semiconductor industry is in some sense
“strategic” has been to pose the question, “How are computer chips
different from potato chips?”6
One might point out three important differences. First, a potato chip
is a final consumption good, so that poor-quality, high-priced potato
chips, although onerous to the consumer, have no direct impact on other
industries. Semiconductors, on the other hand, are an input to a large
and important array of user industries, for which access to the latest
technology at a competitive price is critical. Second, entry into potato

“food chain” metaphor was widely applied in Washington during the policy debates over
high-definition television of 1988-89: “Loss of consumer electronics has affected today’s
U.S. competitiveness. . . . The U.S. electronics infrastructure is like a biological food chain
with many interdependent segments-when one link is weakened or destroyed, other links
feel injury. Many believe competitive problems experienced today by the U.S. semiconduc¬
tor industry can be traced to this lost segment.” American Electronics Association, ATV
Task Force Economic Impact Team, High Definition Television (HDTV): Economic Anal¬
ysis of Impact (Washington, November 1988), p. i.
“When a personal computer adopts the kind of flat display screen associated with
HDTV technology, it will not only create additional demand for the products associated
with this new industry, but it will also require a large number of semiconductors to support
the new projector and display technology that will be incorporated in the screen itself. The
use of these inputs will give a boost to demand for products from these related industries,
contributing to increases in their output and improving the chances that they will be
profitable and have the funds to support further innovation.
These linkages create a kind of feed-back mechanism that provides its greatest advan¬
tages to our industrial base when the entire ‘food chain’ of the electronics industry is
improved.” Testimony submitted by Robert Cohen in High Definition Television, Hearings
before the Subcommittee on Telecommunications and Finance of the House Committee on
Energy and Commerce, 101 Cong. 1 sess. (Government Printing Office, 1989), p. 226.
6. The comment has been attributed to Michael Boskin, who was chairman of the
Council of Economic Advisers in the Bush administration, but firm evidence that he actually
said it appears to be lacking.
STRATEGIC ISSUES 377

chip making is relatively easy, whereas entry into computer chip manu¬
facture may involve large sunk investments and technology that is difficult
to obtain. Third, computer chips are high-tech goods, requiring large,
continuing investments in R&D; potato chips do not. An empirical lit¬
erature tells us that much R&D is likely to have significant economic
impacts that are not fully captured by the high-tech firm making the
R&D investment.
These circumstances make it credible that one or both of two things
might occur in the computer chip industry that would be unlikely to
happen in the potato chip industry. Barriers to entry may make the
exercise of monopoly power possible, allowing computer chip companies
to earn monopoly rents. Spillovers of knowledge from chip R&D, as well
as production, to other firms and industries may create externalities that
result in economic benefits to firms other than the firm making the in¬
vestment in R&D.
Many economists would agree that these two possibilities create condi¬
tions under which intervention to sustain a “strategic” industry might be
economically justifiable.7 First, if there are significant technological exter¬
nalities in semiconductor production, so that social returns exceed private
returns, policies to correct for presumed underinvestment in these activities
may be appropriate. There is some empirical tradition of economists arguing
that particular capital goods industries have been strategic because of in¬
formational spillovers between producers and users of these goods.8
In semiconductors it is in design, not manufacturing, that it seems
most plausible to argue for technological externalities—that is, that semi¬
conductor users are using the same technologies as semiconductor pro¬
ducers. One is hard pressed to think of examples of other industries that
use the same manufacturing technologies as the semiconductor industry.9
Cross-industry spillovers are more obvious in design. As the density of

7. Paul R. Krugman, for example, identifies these two criteria for “strategic” sectors
as academically respectable economic arguments for selectively promoting specific indus¬
tries. See his “Introduction: New Thinking about Trade Policy,” in Krugman, ed., Strategic
Trade Policy and the New International Economics (MIT Press, 1986), pp. 12-15.
8. See, for example, Nathan Rosenberg, “Technological Change in the Machine Tool
Industry, 1840-1910,” Journal of Economic History, vol. 23, no. 4 (1963), pp. 414-43; and
Bo Carlsson and Staffan Jacobsson, “What Makes the Automation Industry Strategic?”
Economics of Innovation and New Technology, vol. 1, no. 4 (1991), pp. 257-69.
9. Prominent exceptions include flat panel displays (used for the screens of laptop
computers and similar products), which use a similar manufacturing technology, and the
read-write heads used in computer disk drives.
378 STRATEGIC ISSUES

circuit elements that can be inscribed on a silicon chip has increased,

more and more of the proprietary design of an electronic system has been
placed on a single chip, where before systems were built up from non¬
proprietary building block components. As a result, today the border
between chip design and system design is exceedingly blurry. But this is
an argument that systems design and chip design are essentially converg¬
ing into a single industrial activity, undertaken by the same firm, and not
necessarily an argument that there are obstacles that prevent spillovers
from being captured within private firms, which might justify some cor¬
rective policy intervention.
Another kind of technological externality in semiconductor manufac¬
turing might prove a more convincing concern. This involves the transfer
of proprietary information between system designers and semiconductor
manufacturers. With increasing levels of chip density has come the re¬
quirement that system designers transfer proprietary design information
to chip manufacturers. Because Japanese chip manufacturers, in partic¬
ular, are often arms of large conglomerates also involved in the design
and sales of systems, reliance on external chip manufacturers has thus
increasingly meant transferring sensitive design information to one’s own
competitors. In this case, however, the perils of “deindustrialization” in
semiconductor manufacturing boil down to the absence of sufficient com¬
petition in chip manufacturing (forcing one to rely on one’s own com¬
petitors to manufacture needed chips) and the absence of sufficient pro¬
tection for intellectual property.
A similar argument might apply in the more specialized area of sem¬
iconductor production equipment. Eric von Hippel, for example, has
argued that roughly two-thirds of the ideas for innovations in the ma¬
chines used to manufacture semiconductors came from chip makers,
through their close interaction with the machinery makers.10 Conversely,
one might argue that early access to the latest semiconductor manufac¬
turing technology requires close interaction with semiconductor equip¬
ment makers, and that therefore maintenance of a competitive chip man¬
ufacturing industry demands a healthy local semiconductor production
equipment sector. The argument has been made that these sorts of in¬
formational externalities link the semiconductor equipment industry to

10. Eric von Hippel, The Sources of Innovation (Oxford University Press, 1988), pp.
19-26.
STRATEGIC ISSUES 379

the semiconductor industry downstream.11 Indeed, this view was reflected

in the decision of the Sematech, the U.S. semiconductor industry’s R&D
consortium, to focus its resources on improving the technical and eco¬
nomic health of its domestic equipment and materials suppliers.
On the other hand, in an age of drastically cheapened global transport
and communications, and extensive overseas investment, it is not neces¬
sarily the case that a U.S.-based chip company is precluded by distance
from establishing a close relationship with a foreign equipment manufac¬
turer. To make the case for domestic industry support on these grounds,
a further argument is required, namely, the assertion that American
producers are unlikely to be able to obtain timely and competitive access
to foreign chip making technology (this assertion is typically accompanied
by the observation that Japanese chip making equipment companies have
close ties to Japanese chip makers competing with American semicon¬
ductor firms).
This idea, in fact, is the application to semiconductor equipment of a
second argument that is sometimes made as to why the semiconductor
industry itself is strategic. This second argument hinges on the fact that
monopoly profits may be earned in high-tech sectors. The role of semi¬
conductors as an important input to many other sectors makes the poten¬
tial exercise of monopoly power an extremely important concern, since
market power in such an input may be extended downstream into user
industries, by acquisition or vertical integration, allowing even greater
monopoly rents to be collected.12 In particular, if the recipients of mo¬
nopoly rents are foreign nationals, then policies intended to create a
domestic industry capable of seizing a share of these rents, or reducing
their extraction from domestic consumers, may be desirable, even (in the
extreme case) if domestic producers are less efficient.13 As just observed,
this argument is in essence made when contending that a domestic equip-

11. See for example, National Advisory Committee on Semiconductors, Preserving the
Vital Base: America’s Semiconductor Materials and Equipment Industry (GPO, 1990).
12. Otherwise the downstream user would normally substitute other inputs for the
monopolized input as price is raised and dissipate some of the monopolist’s potential rent
in inefficient production. John M. Vernon and Daniel A. Graham, “Profitability of Mon¬
opolization by Vertical Integration,” Journal of Political Economy, vol. 79 (July-August
1971), pp. 924-25.
13. The argument is presented in a simple, diagrammatic form in Kenneth Flamm,
“Semiconductors,” in Gary Clyde Hufbauer, ed., Europe 1992: An American Perspective
(Brookings, 1990), pp. 260-63.
380 STRATEGIC ISSUES

ment industry ought be preserved to limit the extension of foreign mo¬

nopoly power into semiconductors.
Essentially both arguments—the argument from externalities and the
argument from monopoly power—have typically been used to support
the need to sustain the U.S. semiconductor industry as a strategic sector.
On the monopoly power issue, the NACS wrote in 1989, “Our major
competitors have a demonstrated capability—compelled either by policy
or their own strategy and organization—to act in concert and exercise
market power sufficient to control access to technology and its price.
Compelling examples include the DRAM shortage, the inability of U.S.
supercomputer companies to purchase the fastest foreign chips, and the
inability of most U.S. chip producers to gain timely access to the most
advanced foreign semiconductor manufacturing equipment and mate¬
rials.”14 On the externalities question, the NACS wrote in 1992, “The
close linkages between semiconductor and electronics manufacturers are
important to spur innovations in semiconductors, and to incorporate
quickly new technologies into successful products and services. These
strong linkages help to create external economies—economic benefits
that flow between semiconductor firms and their customers and suppliers,
and also between competing semiconductor firms.”15

Strategic Behavior and Strategic Policies

Another use of the word “strategic” comes from game theory, where
strategic actions are defined as those intended to influence the moves of
other players.16 Firms act strategically when they take actions intended
to alter the behavior of their rivals. Governments, too, can make strate¬
gic moves, when they implement policies intended to alter the behav¬
ior of the foreign rivals of their domestic firms, or even other foreign
governments.
For example, suppose a government grants a subsidy, linked to value
or volume of production, to a domestic firm selling against a single foreign
rival in a foreign market. That subsidy will have the direct effect of

14. NACS, Strategic Industry at Risk, p. 14.

15. National Advisory Committee on Semiconductors, Attaining Preeminence in Semi¬
conductors, Third Annual Report to the President and Congress (Washington, February
1992), p. 8.
16. This definition comes from Thomas C. Schelling, The Strategy of Conflict (Harvard
University Press, 1960).
STRATEGIC ISSUES 381

encouraging the domestic firm to increase output and increase its profit.
By itself (taking the output of the foreign firm as given), this does not
mean that domestic society as a whole benefits from the subsidy, since
the increased profits for the domestic producer will generally be more
than offset by the cost of the subsidy to national taxpayers. But it may
also encourage the foreign rival to cut back its output, as news of the
subsidy makes the rival realize that the domestic firm will definitely
increase its output. Given this reality, the foreign rival may find that
producing less output maximizes its own profits, allowing the domestic
producer to expand its output even further and realize some additional
profit. This indirect effect (working through the foreign producer’s out¬
put decision) of the subsidy is the so-called strategic effect of the govern¬
ment policy, and is what makes it possible that such a strategic trade
policy can work to benefit the interests of the subsidizing nation as a
whole.17
Of course, the domestic welfare gains come at the expense of the
foreign country, since the additional profit earned domestically is accom¬
panied by a decline in the foreign company’s profit. The foreign govern¬
ment may then be tempted to intervene to convince the domestic govern¬
ment to abandon its intervention, or to create an outcome more amenable
to its own company’s interest. The result is a strategic interaction between
the two governments’ policies.
The logical link with the concept of a strategic sector, discussed earlier,
is that the literature on strategic trade policy (defined as policy that works
by altering the behavior of foreign companies or governments) has fo¬
cused on cases where benefits to a nation as a whole exist as a conse¬
quence of monopoly profits captured from foreign interests. The exis¬
tence of significant monopoly power was one of the two economically
defensible reasons why a sector might merit labeling as strategic.
But the other rationale, the presence of externalities, might also cred¬
ibly inspire a strategic government policy. Policies used to force domestic
investments by foreign companies, for example, in order to foster tech¬
nological spillovers to domestic companies, might arguably be labeled a
strategic trade policy.

17. This example is the classic model developed in James A. Brander and Barbara J.
Spencer, “Export Subsidies and International Market Share Rivalry,” Journal of Interna¬
tional Economics, vol. 18 (February 1985), pp. 83-100. See also James A. Brander, “Ra¬
tionales for Strategic Trade and Industrial Policy,” in Krugman, ed., Strategic Trade Policy,
pp. 23-46.
382 STRATEGIC ISSUES

Nonstrategic policies (that is, policies that work only through their
effects on domestic economic actors) might also be attractive in strategic
sectors. Even if an economy or sector were entirely cut off from inter¬
national trade, the government might wish to subsidize investments
thought to be characterized by significant externalities. Or, in a strategic
sector in which domestic firms with a unique competitive advantage face
no foreign rivals, official encouragement of the formation of an export
cartel may further national welfare at the expense of foreign consumers.
In a real world linked by pervasive trade, however, virtually any policy
affecting a strategic industry is also likely to have some effect on actual
or potential foreign competitors’ behavior, and hence also qualify as a
strategic policy.
In principle, however, “strategic trade policy” is not conceptually
equivalent to “government policy in strategic sectors.” Strategic trade
policy requires the existence of actual or potential foreign competitors,
operating in a strategic sector, whereas economic justification for policies
to promote strategic sectors, when motivated by externalities, does not
necessarily require foreign competition. As a practical matter, though,
the two ideas are closely identified with one another.
Discussion of government policies toward the semiconductor industry
typically captures basic elements of all three distinct uses of the word
“strategic.” Semiconductors will be a strategic sector, meriting policy
intervention, if the exploitation of monopoly power in a key semiconduc¬
tor input is a credible likelihood, with potentially significant effects on
downstream user industries, or if externalities are important. Strategic
behavior by semiconductor producers (such as preemptive capacity in¬
vestments or predatory pricing) may justify intervention by government.
Strategic policies undertaken by governments (for example, protection,
subsidy, or the toleration or facilitation of cartels) might also inspire a
response by others.

Semiconductor Dependency and Strategic Trade Policy

It is quite remarkable that some key elements of the earliest episode
of U.S.-Japan semiconductor trade friction, described in chapter 3, were
to reappear a quarter of a century later, at the epicenter of controversy
over the U.S.-Japan Semiconductor Trade Arrangement (STA) of 1986,
described in chapter 4. In 1959, as in 1986, the semiconductor industry
STRATEGIC ISSUES 383

based its appeal to the federal government for protection on assertions

about the industry’s strategic nature, and the Japanese response was the
formation of an export cartel.
Over the intervening two and a half decades, as a consequence of
rapid technical innovation within the U.S. semiconductor industry, the
semiconductor threat from Japan at first receded. But in the very late
1970s, as integrated circuits drawing on the large-scale Japanese R&D
effort of the mid-1970s began to be exported in significant quantities,
American producers once again began to raise alarms about their Japa¬
nese competitors. The market for memory chips—specifically, dynamic
random access memory chips (DRAMs), the single most important type
of semiconductor in terms of volumes produced, and erasable program¬
mable read only memory chips (EPROMs)—was the reentry point for Jap¬
anese producers into global semiconductor competition, and the focus
for trade frictions between the U.S. and Japanese semiconductor indus¬
tries. And once more, strategic issues were placed at the center of debate.
The balance of this chapter examines the principal argument for gov¬
ernment intervention in semiconductor trade as it has come to be artic¬
ulated as a strategic economic issue for the United States: is dependency
on foreign suppliers of semiconductors likely to be an empirically per¬
suasive motivation for public policy?18 The analysis will focus on the core
arguments for activist, strategic policy interventions by the U.S. govern¬
ment: that the “dumping” of Japanese chips in the U.S. market in the
early 1980s constituted predatory behavior on the part of Japanese chip
producers, designed to secure monopoly power in key semiconductor
markets, and that collusive behavior designed to exploit that monop¬
oly power, once secured, could create significant costs for the U.S. econ¬
omy, therefore justifying the investment of resources in defensive counter¬
measures.
The essentials of this argument are spelled out in figure 7-2. For
simplicity I have assumed constant returns to scale in chip production
and have assumed that technological investments and past learning have

18. By focusing on monopoly power as the motive for strategic behavior, I ignore the
other possible reason for thinking semiconductors might be a strategic sector, namely,
technological externalities. There is little compelling empirical evidence on this issue, par¬
ticularly on aspects critical to the design of policy: whether spillovers cross industry bound¬
aries as well as firm boundaries, whether they are inherently confined to a local geographic
region, and whether alternative means (such as revised norms for intellectual property
rights, or design and technical standards) can serve to internalize what might otherwise be
externalities for an individual firm.
384 STRATEGIC ISSUES

Figure 7-2. Theoretical Case for Domestic Production of

Semiconductors

created a cost advantage for Japanese chip producers, so that their con¬
stant cost of production is J dollars (and their marginal and average cost
schedule is represented by line JJ'). Domestic companies have higher
unit costs of U dollars (and their cost schedule is given by line UU').
Foreign demand is given by demand curve DD', and marginal revenue
by line MR-MR'. If Japanese producers are then able to act as a profit-
maximizing cartel, without fear of foreign entry, they will produce output
Q0, charge price P0, and collect monopoly profit PQAEJ. If entry and
exit in the industry are costless, however, foreign producers can effec¬
tively contest the market, and this threat of entry places a cap on price
at U. One may argue, however, that sunk costs—in the form of special¬
ized and short-lived capital investments—are substantial in this industry.
No individual domestic firm, then, will be willing to invest in a high-cost
production facility, because in the event of a price war the cartel can
STRATEGIC ISSUES 385

lower its price below U and still fully cover costs, while the high-cost
foreign producer will be forced to produce at a loss or to shut down and
lose an amount equal to its sunk costs.
If the alternative is to face a cartel charging price P0, however, it is
clearly advantageous from the viewpoint of the foreign country to
subsidize high-cost domestic production of output Qx at cost level UU'.
Monopoly profits equal to P0ABP, that would otherwise have been
paid to the cartel are saved, and, in addition, consumer surplus equal
to triangle ABC is gained. Subsidy of high-cost domestic production
is superior to passive acceptance of uncontested, cartelized imports.
One should note that fear of dependency on foreign suppliers is not
an exclusively American preoccupation. Chapter 2 of this book described
an episode in which a similar fear appears to have motivated the Japanese
government to invest in augmenting the domestic capability to supply the
materials used by its semiconductor industry.
The U.S. semiconductor industry’s analysis of its Japanese rivals’ be¬
havior, documented in chapters 3 and 4, has evolved historically. Through
1980 the story told was what might now be described as the “conven¬
tional” account of strategic trade policy: formal and informal barriers
protecting Japan’s domestic semiconductor market against American im¬
ports worked to promote the development of the Japanese industry and
its global market share, to the detriment of the sales (and profits) of U.S.
producers.19 After 1980, however, as it became clear that prices in the
American and Japanese markets were essentially identical, allegations of
dumping—that Japanese producers were pricing exports below cost—
necessarily and explicitly began to include an element of predatory be¬
havior on the part of these companies, and an element of collusion. A
more unconventional story of strategic behavior surfaced: below-cost
pricing, it was asserted, was calculated to induce exit by American pro¬
ducers, after which Japanese producers would jointly raise prices and
extract monopoly rents, which would provide a return on their investment
in predation.
The history of this discussion and the American trade policy response
it provoked could be interpreted skeptically as a self-fulfilling prophecy.

19. This is the “import protection as export promotion” policy described by Paul R.
Krugman, “Import Protection as Export Promotion: International Competition in the
Presence of Oligopoly and Economics of Scale,” in Henryk Kierzkowski, ed., Monopolistic
Competition and International Trade (Clarendon Press, 1984), pp. 180-93.
386 STRATEGIC ISSUES

Clearly, in response to increasing trade frictions with the United States

in the early 1980s, the Japanese government pressured its semiconductor
industry to reduce exports, in essence sanctioning an export cartel. On
the other hand, the U.S. industry has interpreted this same behavior as
the normal second stage of a successful predatory campaign—rent ex¬
traction—thus validating the original assertions about Japanese inten¬
tions. The extent to which creation of export cartel-like market structures
is the handmaiden of a misguided trade policy, or the fruit of successful
predation, is the key issue.
The Semiconductor Trade Arrangement of 1986 drove this debate to
new extremes. The complex evolution of the administrative mechanisms
created under the auspices of this agreement was described in chapter 4.
Perhaps the most interesting element of this most recent episode of U.S.-
Japanese semiconductor trade friction is some evidence suggesting that
after 1988, for perhaps the first time in their history, major Japanese
semiconductor producers were able to maintain a significant degree of
cooperation (or, less charitably, collusion) in a key product market despite
the absence of explicit, overt regulatory pressures from the Japanese
political system. The long-feared predatory threat might finally have sur¬
faced, albeit with considerable support from policies put into place from
1986 to 1988.
In this chapter the model of semiconductor producer behavior devel¬
oped in chapter 6 is modified and applied in order to assess the potential
empirical significance of the threat of collusive behavior that has been
the intellectual underpinning of an increasingly activist U.S. policy in
semiconductors since the early 1980s. Putting aside the issue of whether
collusive behavior was the proximate cause or the unforeseen effect of
American trade policy, just how large an economic threat might it rep¬
resent in the worst case? How much might the United States reasonably
be willing to spend on “anticartel insurance”?
To examine these issues analytically, I have developed what I have
argued is an at least minimally realistic model of pricing and production
over the life cycle of a high-technology industry such as the semiconduc¬
tor industry, in which both learning and scale economies and capacity
constraints are important. I will apply this model using empirical param¬
eters relevant to the production of 1M DRAMs, a product where actual
historical behavior has been interpreted as a real-life episode to which
both claims—first of predatory behavior, and then of collusive organi-
STRATEGIC ISSUES 387

zation of foreign production to increase monopoly rent extraction—are

relevant.

The DRAM Crisis of 1988

To begin, the most interesting data suggesting the proposition that the
behavior of Japanese DRAM producers has not always reflected normal
competitive market forces are drawn from observation of the market for
DRAMs in 1987 through 1990. The relationships between prices and the
foreign market values (FMVs) constructed by the U.S. Commerce De¬
partment raise interesting questions.
Figure 5-8 showed that, by late 1987 and continuing through late 1989,
U.S. contract prices for 256K DRAMs had risen substantially above
FMVs, and therefore that the FMVs were not constraining U.S. DRAM
import prices over this period. From 1990 on, however, at least some
Japanese DRAM imports appear to have been priced out of the U.S.
market, as 256K DRAM prices fell below average FMVs. Indeed, Japa¬
nese 256K DRAM production fell quite sharply from 1990 on. Despite
rapid and deep cuts in Japanese production, U.S. prices dropped below
the FMV levels.
Figure 5-9 showed a similar pattern in 1M DRAMs. From late 1987
through early 1990, U.S. contract prices stayed above average (and over
most of this period, above even the highest-cost Japanese company’s
individual) FMV. Unlike in 256K DRAMs, however, U.S. prices roughly
tracked average FMV over the remainder of the STA’s lifetime, and
Japanese companies cut neither production nor exports of 1M DRAMs
to the degree visible in 256K parts. Moderate cuts in Japanese output
were apparently successful in boosting prices to levels at or above the
Commerce Department FMVs.
By early 1989 semiconductor demand had begun to weaken. In re¬
sponse to a downward drift in DRAM prices, Japanese manufacturers
began to cut back output in order to maintain high prices. Japanese
newspapers talked of “coordination structures” among Japanese com¬
panies being used to achieve “high price stability.” This was a distinctly
different situation from that in 1988, when prices greatly exceeded price
floors as the industry was at full capacity utilization. With all producers
operating close to capacity, it was more difficult to argue that the indus¬
try’s own restraint, rather than an earlier history of politically mandated
388 STRATEGIC ISSUES

restraints on capacity expansion and production, must be causing high

prices. With idle capacity, it became more plausible to argue (as did the
U.S. industry) that collusive firm behavior must be the principal cause
of “abnormally” high prices exceeding FMVs.20

Causes of the Crisis

A more neutral assessment would begin by noting that at least three

explanations for the historically unprecedented run-up in chip prices in
1988 have been seriously discussed. Logically, the 1988 crisis could be
explained by any possible combination of these three distinct arguments:
—Innocent miscalculation by producers (on the demand side, under¬
estimation of the recovery in chip demand in 1988; on the supply side,
unexpectedly slow growth in yield rates in semiconductor production in
1987 and 1988) could explain a sustained shortfall in supply.
—The implementation of the STA in 1987 (when Japanese producers
were being “guided” by their government to reduce both output and
investment) might be expected to have had an effect on chip supply in
1988. That is, we might imagine a purely exogenous, “political” shock to
Japanese suppliers’ production and investment decisions.
—We might imagine a deliberate, collective decision by a group of
suppliers accounting for most of global production to exploit their mo¬
nopoly power in order to seek greater monopoly rents.
Two distinct variants of the third story exist. The first argues that such
organized rent collecting was largely opportunistic, facilitated by the
creation, with the implementation of the STA, of a joint information
gathering and price monitoring framework for Japanese chip supply. The
second argues that the exploitation of monopoly power essentially reflects
the private decisions of a collusive grouping of predatory producers (per¬
haps aided or abetted by the state), but basically independent of the
evolving resolution of semiconductor trade frictions.
Some version of the latter variant has been an element of the story of
Japanese predation told by U.S. chip producers since 1982. The factual
case for this version was reasserted most coherently in a study produced

20. See Semiconductor Industry Association, Four Years of Experience under the U.S.-
Japan Semiconductor Trade Agreement: “A Deal is a Deal” (San Jose, Calif., November
1990), p. 65.
STRATEGIC ISSUES 389

for the SI A and given wide circulation by others citing it.21 The argument
starts from the circumstances of the “production coordination” by Jap¬
anese suppliers in late 1985 and early 1986 (described above), which
attracted little contemporary public comment outside Japan (unlike in
1982, when the U.S. Justice Department became involved). Arguing that
moves to reduce DRAM exports and production predated the STA and
therefore must have been independent of government policy, proponents
of this view instead attribute “production coordination” to Japanese chip
makers passing over some threshold of monopoly power as American
producers withdrew:
the move toward production regulation by the Japanese producers’ group
began in 1985, well before the Semiconductor Arrangement had even been
conceptualized, much less actually put in place. . . . Thus, by the third
and fourth quarters of 1985, Japanese DRAM producers had few compet¬
itors left except each other. It was at this precise moment—in late 1985—
that reports began to appear of joint actions by the Japanese DRAM
producers to stabilize price competition by coordinated curtailments in
output.22

In fact, a careful reading of the historical record refutes the specifics

of these claims. Cuts in exports by the big Japanese producers occurred
after political pressure had been brought to bear on the industry by top
politicians in July 1985. Furthermore, most U.S. firms dropped out of
DRAMs after the initial cuts in semiconductor exports by Hitachi, NEC,
and Toshiba in the summer of 1985. And rather than being “conceptual¬
ized” as a strategy for the first time in 1985, Japanese chip exporters had
previously jointly cut semiconductor exports—as a response to both po¬
litical pressure and government “guidance”—back in 1982, and before
that in 1981, possibly in 1979, and definitely in 1959, when Japanese
market share was very much smaller than in 1985.
On the other hand, business practices that would in the United States
have been characterized as anticompetitive clearly have been observed
with some frequency in Japanese industry. Mark Tilton has recently pro-

21. This analysis is repeated uncritically in Office of Technology Assessment, Compet¬

ing Economies: America, Europe, and the Pacific Region (1991), pp. 11-12; and in Laura
D’Andrea Tyson and David B. Yoffie, “Managing Trade and Competition in the Semi¬
conductor Industry,” in Laura D’Andrea Tyson, ed., Who’s Bashing Whom? Trade Conflict
in High-Technology Industries (Washington: Institute for International Economics, 1992),
pp. 117-18.
22. Thomas R. Howell, Brent L. Bartlett, and Warren Davis, Creating Advantage:
Semiconductors and Government Industrial Policy in the 1990s (Washington: SI A and
Dewey Ballantine, 1992), p. 117.
390 STRATEGIC ISSUES

vided persuasive documentation of the continued toleration (and in some

cases active encouragement) of cartels in Japan’s basic materials indus¬
tries by MITI and the Japan Fair Trade Commission, even after such
activities were explicitly rendered of questionable legality by changes to
laws passed in 1987.23 And from time to time, details have emerged in
U.S. legal proceedings that suggest that Japanese industry associations
continue to serve as the venue for attempts to restrict or limit competi¬
tion. Chapter 2 described how groupings within the Electronic Industries
Association of Japan worked to organize both domestic and export cartels
for televisions in the 1960s and 1970s. In the mid-1980s, a more recent
court case showed, the Japanese Camera Manufacturers Association con¬
tained both a Design Committee and a Price Committee to which man¬
ufacturers were required to submit plans for features and pricing for new
cameras.24
The traces of similar behavior are sometimes visible in semiconductor
trade. In chapter 2, I documented a failed attempt in the 1970s by some
elements in Japanese government and industry to reorganize semicon¬
ductor producers into a smaller number of firms. Designed as a counter¬
measure to the imminent liberalization of foreign imports and invest¬
ment, the proposal had each reorganized firm specializing in particular
chip types. Though some joint tie-ups in R&D ultimately emerged from
the IC Liberalization Countermeasure program, inability to agree on
who got what among Japanese chip companies appeared to prevent any
industrywide cartel-like structure from emerging.
In 1981, an attempt to organize production cutbacks among major
Japanese memory chip makers was reported in the Japanese press. This

23. Tilton studied cartels in the aluminum, cement, petrochemical, and steel industries.
See Restrained Trade: Cartels in Japan's Basic Materials Industries (Cornell University
Press, 1996). As Tilton notes, the Fair Trade Commission is far from independent of other
Japanese government bureaucracies. By informal custom the head commissioner is a former
Ministry of Finance official, while the other four commissioner slots are reserved for a
retiree from the Finance Ministry, MITI, the Justice Ministry, and for a former official
inside the FTC itself (pp. 48-49). On restraints in the Japanese steel industry, see also
Thomas R. Howell and others, Steel and the State: Government Intervention and Steel’s
Structural Crisis (Boulder, Colo.: Westview, 1988), pp. 202-07.
24. See In the United States District Court for the District of New Jersey, Honeywell,
Inc., a Delaware corporation, Plaintiff, v. Minolta Camera Co. Ltd., a Japanese Corpora¬
tion, and Minolta Corporation, a New York Corporation, Defendants, Civil Action 87-4847,
Transcript of Trial Proceedings, vol. 16 (Newark, N.J., October 11, 1991), pp. 2274-2295;
vol. 33 (Newark, N.J., November 6, 1991), pp. 4960-4966; and vol. 63 (Newark, N.J.,
January 2, 1992), pp. 9727-9734. See also Barnaby J. Feder, “Honeywell-Minolta Dispute
Teaches Conflicting Lessons,” New York Times, February 12, 1992, p. Dl.
STRATEGIC ISSUES 391

episode is notable in that several Japanese producers appear to have

attempted to coordinate production of semiconductors among themselves
in the absence of any political initiative to deal with trade friction. But
the attempt apparently failed. In September 1981, as prices for the newly
introduced 64K DRAM plummeted, NEC, Hitachi, and Fujitsu each
announced a freeze on increased production. When producers Oki Elec¬
tric and Mitsubishi broke ranks and announced output increases late that
year, however, “this, together with the rapidly rising demand [for 64K
DRAMs], broke down the tacit agreement among NEC, Hitachi, and
Fujitsu to hold back production.”25
As 64K prices continued to fall in late 1981, trade frictions did intensify.
As recounted in chapter 2, MITI intervened months later, in February
1982, counseling restraint in its “guidance” to Japanese producers. (Jap¬
anese market share in 64K DRAMs actually peaked in the last quarter
of 1981, then went into a sustained decline.) Restraints on exports seemed
to stick only after bureaucratic intervention to resolve a deepening exter¬
nal political crisis.
Chapter 3 described how revelation of the existence of a shadowy
Council of Nine organized to set prices and allocate markets for discrete
semiconductors in Japan played some role in the U.S. government’s de¬
cision to pursue a section 301 complaint against Japan in 1985. What is
unknown, however, is how effective this organization was in accomplish¬
ing its objectives. On the one hand, both this episode and the 1981
attempt at coordinating production cutbacks suggest that in the first half
of the 1980s, at least, collusive and (at least formally) legally questionable
behavior was not unknown in the Japanese semiconductor industry. On
the other hand, the 1981 restraint attempt failed, and the Council of
Nine’s influence does not seem to have pushed Japanese semiconductor
prices by 1985 to notably extreme levels relative to world prices.
On balance, then, frameworks imposed or invented in collaboration
with politicians and bureaucrats in order to deal with trade friction,
rather than sudden changes in market power, seem the more compelling
explanation for historical episodes of successful restrictions on supply. In
all cases prior to 1989, successful producer restraint coincided with active
government initiatives.26

25. Yui Kimura, The Japanese Semiconductor Industry (Greenwich, Conn.: JAI Press,
1988), p. 66.
26. Indeed, Tilton notes that even in highly concentrated basic materials industries,
active MITI involvement was needed in order for cartels to succeed. “In aluminum, cement,
392 STRATEGIC ISSUES

Finally, there is one additional area where continuing American in¬

dustry complaints might be characterized as a story about a more subtle
form of private industrial collusion (fostered historically, to be sure, by
Japanese government industrial policies—formal and informal—of the
sort described in chapter 2). An alleged propensity for Japanese com¬
panies to buy Japanese, even when foreign companies offer equivalent
products at lower prices, has continued to raise hackles through the
present day, and in fact still fuels American arguments that continuing
pressure from government trade negotiations is needed to offer them a
fair shot at the Japanese market.
Such a propensity has sometimes been openly acknowledged within
the Japanese semiconductor industry. A 1985 industry roundtable discus¬
sion of silicon wafer market conditions, for example, noted that Japanese
buyers would not touch foreign wafers at the same price as domestic
products, and would consider them only if prices were 10-15 percent
below the price for Japanese-made product.27 Although one might argue
that delivery problems and quality differentials could account for a con¬
tinuing reluctance to buy foreign inputs, other evidence would seem to
contradict this interpretation. By 1987, for example, Toshiba was saying
that “imported silicon wafers have almost totally satisfied Japanese users
in terms of quality and shipping date, which used to be major problems
of those imported wafers. We will keep the import ratio at 30 percent
this year, considering our subsidiaries’ policy and the effort of domestic

and petrochemicals the policymaking process followed this pattern: MITI pushed industry
to go along with cartels that individual firms privately believed were necessary but that they
could not coordinate on their own.” Tilton, Restrained Trade, p. 207.
27. As anonymous industry executive “B” noted, at a time when the industry was in a
period of peak demand, “Japanese corporations usually don’t buy imported and marketed
wafers even if they are of the same quality and a little cheaper. They may have tried to
obtain ones in the latest shortage, but corporations overseas, then, might well have had no
surplus of supply. Even if Japanese corporations want to buy overseas products on the basis
of quality, ignoring the somewhat different yield involved, the commercial deal may not be
worked out unless the prices are lower than those produced in Japan by 10 to 15 percent;
and even if the prices so favor them, they will never buy very complex super LSIs. Although
the wafers are flown in in only one day, a communications gap still remains a solid barrier
between nations. Overseas corporations, after all, have to build their own plants in Japan
if they are to succeed in selling wafers in Japan, I presume, as the Monsanto Corporation
is going to.” Monsanto, after building such a plant in the late 1980s, was still unable to
significantly increase its share of the Japanese silicon wafer market, and ultimately exited
from the silicon wafer business. “Silicon Semiconductors: Point in Question and Prospects;
A Roundtable Talk With the Names of Speakers Withheld,” Kinzoku Jihyo, no. 1125,
April 5, 1985, p. 171.
STRATEGIC ISSUES 393

silicon producers, but we will request even tougher price reductions of

silicon producers in 1988.”28 Indeed, Japanese silicon wafer producers
refused in 1987 to permit their Japanese customers to import wafers from
the Japanese wafer makers’ U.S. subsidiaries and take advantage of sig¬
nificant price differentials between the United States and Japan.29 As the
Japanese industry coped with the entry of new producers (Japanese steel
companies) and a stronger yen, wafer makers complained that price lead¬
ership was not being properly maintained by the large producers and the
“orderliness” of the market was disappearing.30

28. “Toshiba Trying to Reach 30% Materials Import Target This Year,” Rare Metal
News, no. 1396, April 1, 1987.
29. “According to users, there are even some American [epitaxial] wafers whose prices
are half of Japanese products, including exchange rate differentials and with American
standards. Based on this fact, it is very likely in the future that epi wafers from Japanese
U.S.-based silicon producers, such as Shinetsu Semiconductor, Mitsubishi Metal, and
Osaka Titanium, will be reimported.
Silicon producers point out that ‘the cost of domestic silicon wafers increases in order
to meet the strict Japanese standard, so it is irrational to compare price per surface area
without taking quality into account.’ In addition, Japanese silicon producers operating in
the U.S. also mention that ‘we are currently operating near full capacity in the U.S., and
we cannot sacrifice U.S. users by reexporting the products to Japan.’ ” “Toshiba Trying to
Reach 30% Materials Import Target,” p. 1.
30. At a 1987 anonymous roundtable discussion, when the new fiscal year price nego¬
tiations came up for discussion, executive “E” complained:

The persons in charge of ordering materials for the device manufacturers are trying
to use the high yen value as a weapon. The industries budgeted at 1 dollar equal to
150 yen, but the yen has appreciated to 130 yen per dollar, and they are looking at
the fact that the costs do not match up.
In the past, domestically produced materials and parts have been of high quality.
There had been a long history of transactions, so they were willing to purchase at
somewhat high prices. Recently, however, the struggle to survive is forcing a change
in thinking to move toward the less expensive imported products. There was one
company that was rejoicing at having saved 100 million yen due to such a switchover.
There is less of a consciousness in the industry of taking care of the parts and raw
materials manufacturers.

Executive “D” added:

Up until about the summer of last year [1986], there was quite a bit of difference
in purchase price among different customers for material of the same specifications.
If price cuts were undertaken, they were done while considering the health of the
overall industry; they had a price-leader-type effect.
However, beginning around the fall of last year, the orderliness of the situation
disappeared. The main suppliers were not necessarily the price leaders. Other com¬
panies would cut prices on their own initiative, and a great deal of confusion was
created in the market. It was difficult to make a distinction between the price of
394 STRATEGIC ISSUES

Mark Tilton described a similar willingness to pay a high premium for

domestic inputs in Japanese electronics industries purchases of polypro¬
pylene. According to Tilton’s Japanese sources,
Domestic polypropylene prices in recent years have run from a third to
half again as high as the price of imports. . . . This price disparity is real
and does not merely represent quality differences ... 90 percent of
imported polypropylene is standard-grade and completely identical to
standard-grade domestic polypropylene. . . . Despite this disparity, im¬
ports never accounted for more than 3.1 percent of domestic market share
in 1982-92. . . . The reporter said that the reason is security of supply and
reliability of delivery. Electronics makers generally have a long-term rela¬
tionship with a single petrochemical supplier, although not usually because
of keiretsu ties. Thus, electronics firms buy domestic materials from a single
supplier as a strategy to maximize their own security and efficiency.”31

Even today, there seems to be a similar propensity by Japanese elec¬

tronic equipment makers to favor procurement of domestically made
chips. A 1994 Japanese electronics industry survey found the value of
semiconductors directly procured by major users from overseas sources
to be quite low, but noted that “many users are annoyed by the domestic
overseas price differential for certain products. Thus direct overseas pro¬
curement will likely increase in cases where there is only one source in
Japan, or the price difference is 50-60% for certain products (high per¬
formance linear and operating amps, etc.), or where the difference is
around 20% and the Japanese supplier fails to make efforts to narrow
the gap.”32
Similarly, Mark Tilton notes that an executive in a large Japanese
electronics firm reported to him in 1994 that while his firm continued to
be reluctant to buy foreign steel, it now permitted some of its small
subcontractors to buy imported steel. Although it still did not directly
purchase imports, by threatening to buy foreign steel and giving domestic
steelmakers the opportunity to discount prices to match import prices,
the electronics maker could persuade domestic steel producers to lower

monitor wafers and dummy wafers. Everything was confused.

“Anonymous Roundtable Discussion on Silicon: Japan Consumes Half of the World’s

Wafers; Forecasts of the Outcome of U.S.-Japan Semiconductor Trade Friction,” Kinzoku
Jihyo, no. 1303, June 5, 1987, pp. 188-89.
31. Tilton, Restrained Trade, pp. 162-63.
32. See Osamu Otake, “High-Gear Overseas Production and Semiconductor De¬
mand,” Tokyo Denshi (February 1995), pp. 8-18, in Foreign Broadcast Information Service
Daily Report: Science and Technology, August 30, 1995, electronic version.
STRATEGIC ISSUES 395

prices from the standard “large user” price levels on that portion of their
demand that they could buy from overseas sources.33
In short, evidence of a residual “buy Japanese” preference can still be
discerned within some corners of the Japanese electronics industry. In
many cases—particularly in semiconductors—quality levels for products
produced by major global suppliers, whether Japanese or foreign, no
longer seem to vary significantly aross companies and therefore do not
provide much of an explanation for such a preference.34
There is another set of economic reasons why companies might prefer
to buy from domestic sources at higher prices linked to “relational con¬
tracting.” A long-term relationship with a preferred supplier may be
fostered to obtain superior service and delivery, or partnerships created
with suppliers to develop technologically superior new inputs, at the price
of commitments to maintain procurement even when input prices are
somewhat above market levels. There is also some evidence of a long¬
standing concern in Japan over the vulnerability to external disruption of
supplies procured from overseas, although the overtones of economic
warfare wrapped into this worry make it more a national security issue
than an industrial concern (and it is hard to see how Japan’s vulnerability
in its dependence on foreign industrial inputs—other than natural re¬
sources—is any more severe than that faced by most other industrial
nations). Finally—and often explicitly linked to government concerns
about the security of supplies for key inputs—there is a long history of
government policy formally and informally encouraging “buy Japanese”
attitudes within the industrial sector, although most of the means of
formal encouragement have now been dismantled.35 Separating out what
portion of the explanation for a domestic preference is economic rather
than the result of politics, history, informal government policy, and the
prevailing business culture is likely to remain an impossible task.
Somewhat less hopeless is the analysis of less subtle forms of industrial
collusion. One of the most provocative issues raised by the operation of
the STA is the extent to which cartel-like private behavior was observed

33. Tilton, Restrained Trade, pp. 185-86.

34. Quality differentials are not typically raised as an issue by Japanese companies
today in purchasing foreign semiconductors. Virtually all the major Japanese semiconductor
producers market foreign-manufactured chips in Japan with their own company’s Japanese
brand name stamped on them.
35. Tilton makes a powerful case that informal encouragement remains in the basic
materials industries he studied, primarily because of a continuing government strategic
interest in maintaining the security of supply. See Restrained Trade, chap. 7.
396 STRATEGIC ISSUES

after the arrangement went into operation and what the potential con¬
sequences of such behavior are. I turn next to these issues.

Anecdotal Evidence on Private Collusion

The period from early 1989 through early 1990 is the most interesting
part of the five-year history of the first STA from the standpoint of
allegations of collusive behavior by Japanese companies. Market DRAM
prices in the United States and Japan remained at levels considerably
above U.S. FMVs, at a time of weakening demand, while companies
were reducing their output levels well below capacity. By this time the
Japanese government was under considerable foreign pressure not to
intervene directly and order domestic chip producers to reduce output,
so the manner in which Japanese DRAM producers were reducing their
output prompted intense foreign interest.
It would be helpful to use actual data on pricing and costs to examine
empirically the credibility of allegations of collusive behavior, but unfor¬
tunately no data on price-cost margins for DRAMs are available. Some
noisy information on average cost is available in public submissions to
the Commerce Department filed by companies as part of the process of
setting FMVs, but some possibly contentious assumptions are required
to use these data.36 The only direct evidence on the question, then, is
necessarily anecdotal.
In one widely publicized episode, the three largest Japanese DRAM
producers were reported to have cut output sharply, in rapid succession,
on the day after the failure of an attempt by U.S. chip consumers and
producers to organize a joint DRAM manufacturing venture (U.S. Mem¬
ories) was announced. Reports in the U.S. press fueled assertions that
the Japanese were “acting much like a cartel.” Some analysts close to the
U.S. industry even charged that Japanese companies had flooded the
market and forced prices down in a deliberate effort to torpedo U.S.
Memories (see notes 180 and 181 in chapter 4).
Actually, major Japanese companies had cut shipments and reduced
production long before U.S. Memories failed, in an effort to stabilize
prices going back to the second quarter of 1989. Available data suggest
that, after increasing production of 1M DRAMs during the first quarter
of 1989, domestic producers began to cut back production in the late

36. These data are analyzed later in this chapter.

STRATEGIC ISSUES 397

spring, sliced shipments even more, and increased their stocks of parts
held in inventories.37 In the early fall of 1989, most major Japanese chip
producers announced large production cuts. At the end of July NEC
announced that it had scaled back its plans to increase production of 1M
DRAMs.38 In September, Toshiba, Hitachi, and Mitsubishi each reported
that they were cutting back their production levels by about 10 percent.39
In late October, NEC announced that it too would actually cut current
production levels in the first quarter of 1990.40
Contemporary Japanese press accounts asserted that Japanese chip
makers were collectively cutting back production to slow the decline in
prices, to achieve “high price stabilization.” Talk of a “coordination
structure” also reappeared in the trade press (see chapter 4).
In interviews in November and December of 1989, semiconductor
executives at several Japanese companies were quite frank with me about
their intention to continue to cut back DRAM production in order to
stabilize prices. Asked why, given current prices that were certainly far
above his company’s production costs, he did not cut prices in order to
stimulate sales, one top executive motioned toward a skyscraper visible
through the window, the headquarters of a rival electronics giant, and
remarked that if he cut prices, his neighbor would as well, setting off a
round of continual price cutting. Asked how he could know how much
to cut production in order to stabilize prices, without knowing the plans
of other companies, his colleague responded that the matter was compli¬
cated, with “many aspects,” and declined to elaborate further.41
The subject of coordination among companies is obviously a delicate
one, and while rumors of golf course meetings are a staple of conversation

37. See chapter 4. The basic chronology of events in 1989 was set out as follows in a
presentation by Hitachi’s H. Nakagawa: supply exceeded demand in the second quarter of
1989; booking adjustment by users in the second to fourth quarters of 1989; and production
adjustment by manufacturers after the third quarter of 1989. H. Nakagawa, “Semiconductor
and EDP Market,” slides from a presentation given circa late 1989 (given to the author in
December 1989).
38. NEC withdrew its plan to increase monthly production of 1M DRAMs from
6 million to 8 million chips. Nikkan Kogyo, July 29, 1989, p. 1; Nihon Kogyo, July 29,
1989, p. 1.
39. “Major Semi-Conductor Manufacturers to Reduce 1M-DRAM Production by 10%
from September,” Nihon Keizai Shimbun, September 14, 1989, p. 10.
40. The cut was from 6 million units per month to 5 million units. Dempa Shimbun,
October 31, 1989, p. 1; Nikkan Kogyo, October 31, 1989, p. 1; Nihon Kogyo, October 31,
1989, p. 5.
41. Author’s interview in Tokyo, December 1989.
398 STRATEGIC ISSUES

inside the industry, it is difficult to find someone with first-hand infor¬

mation who is willing to discuss it, even on an unattributed basis. None¬
theless, I did manage to interview one reliable source who had been
present at a number of meetings in Japan where information on individual
companies’ production and capacity was discussed. According to his ac¬
count, these informal and unofficial meetings were attended by managers
from the semiconductor divisions of major Japanese companies. Each
company had available to it computer printouts of other companies’ out¬
put and capacity by product, and all companies had the same data in
their printouts. No government officials were present at these meetings.42
After 1989, observers writing in the Japanese trade press noted the
trend toward increasingly tight oligopoly control of Japanese DRAM
production. As one chip executive remarked to an American government
official in early 1990, “Since the Semiconductor Agreement, we [Japanese
DRAM manufacturers] have moved from competing for market share to
market sharing.”43

Empirical Tests for Competition

Press reports, industry gossip, and the occasional first-hand account
are suggestive and perhaps even believable, but can be made more cred¬
ible by checking the observed empirical behavior of prices for 1M
DRAMs over this period for a certain degree of consistency with the
story. The basic idea is this: profit maximization requires that marginal
revenue, as perceived by the firm, be set equal to marginal cost. With
“competition,” a firm’s marginal revenue curve will lie above that cor¬
responding to a collusive grouping of firms in a constant-market-share
cartel. If we can somehow obtain an upper bound on marginal cost, then
marginal revenue computed under competitive assumptions should al¬
ways lie at or below that upper bound. If marginal revenue calculated
under competitive assumptions lies well above the bound on marginal
cost, it suggests that those competitive assumptions are incorrect, and is
consistent instead with alternative stories, including collusive behavior.

42. The last meeting attended by my source, as of the date of my interview, was in
1990.
43. Personal communication with the author.
STRATEGIC ISSUES 399

Because DRAMs are essentially a homogeneous commodity sold in

well-developed secondary spot markets, the natural assumption is that
producers sell at a single market price and choose the quantities they will
sell. The benchmark adopted here for competitive behavior will be
“Cournot competition,” the assumption that producers take the output
of other firms as unaffected by their own output decisions, at any moment
in time.
For example, in the last quarter of 1989 Toshiba—the largest and
almost certainly the lowest-cost maker of 1M DRAMs—was cutting pro¬
duction of these chips at a time when market prices in the United States
and Japan hovered between $9 and $10 per chip. Its FMV was widely
believed in Japanese industry circles to be about $4.50 at the time.44
Indeed, during the shortage of the summer of 1988, Toshiba’s production
costs had been widely believed in Japan to be in the neighborhood of 500
yen (under $4),45 and its costs in 1989 must surely have declined even
further, as the result of both learning economies and less intensive ca¬
pacity utilization. With Toshiba’s world market share at well under one-
third,46 and a price elasticity of about —1.5, Toshiba’s marginal revenue
would have to have greatly exceeded 80 percent of price with Cournot
competition. With production at levels well below capacity, it is difficult
to believe that marginal cost could have been more than 50 percent
greater than the Commerce Department’s already more than generous
estimate of average cost, as embedded in the FMV. Therefore, unless the
various parameters just used are significantly different from their true
values, Cournot competition must have been a rather poor description
of Toshiba’s competitive environment at that moment.

44. Fourth-quarter 1989 FMVs for 1M DRAMs were reported by one Japanese industry
source to have been set as follows: Toshiba, $4.50; Hitachi, $5.20; Fujitsu, $5.50; Mitsub¬
ishi, $7.00; NEC, $7.50. Author’s interview in Tokyo, December 1989.
45. See Stefan Wagstyl, “Winning When the Chips Were Down,” Financial Times, May
10, 1988, p. 33; Stefan Wagstyl, “Turning Hard Times Into Good Times,” Financial Times,
June 14, 1988, p. 22; and Terry Dodsworth, Louise Kehoe, and Stefan Wagstyl, “The Chase
to Catch the Japanese,” Financial Times, July 25, 1988, p. 18. At the time, 1M DRAMs
were selling for about four times Toshiba’s estimated cost of production, or 2,000 yen.
Toshiba claimed to be achieving 65 percent yields over this period, at a time when the
industry as a whole averaged under 40 percent.
46. Dataquest estimates show Toshiba with under 20 percent of world 1M DRAM
production in the last quarter of 1989, and under 30 percent even if Motorola’s sales are
counted in the Toshiba market share (Motorola sold Toshiba-fabricated chips under its own
name and operated a DRAM manufacturing joint venture with Toshiba at the time).
400 STRATEGIC ISSUES

More precisely, we have perceived marginal revenue of firm i, MRh

given by

(7-1) MR; = P j (X, + 1) + 1

where X, is the so-called conjectural variation, giving the perceived re¬

sponse of other producers’ output to a one-unit increase in the firm fs
output; s, is the market share of firm i; and p is the industry’s price
elasticity of demand. In the case of Cournot competition we have X, equal
to zero; with a constant-market-share cartel we have X, equal to the
inverse of the firm’s share of cartel output less one. Because price elas¬
ticity p is negative, the firm’s perceived marginal revenue in a cartel will
always be less than perceived marginal revenue with Cournot competi¬
tion. Heuristically, we may wish to think of conjectural variation X, as
varying between zero and a positive number equal to the inverse of the
firm’s share of cartel output less one, with the parameter value reflecting
the “degree of competition.”
Before proceeding further, it is necessary to discuss briefly some ad¬
ditional complications peculiar to the economics of semiconductor pro¬
duction. The key issue is learning economies.

Learning Economies

Semiconductor production is believed to be characterized by learning

economies, in which unit production costs fall sharply with accumulated
production experience. A key result due to Spence and generalized by
Fudenberg and Tirole is that, with learning economies, a rational firm
will generally equate current marginal revenue to a value below its cur¬
rent short-run marginal cost of production, as it takes into account the
effect of current production on reducing future production costs. That
is, a profit-maximizing firm sets

MR(t) = MCit) - 8(f) = MC(0,

where 8 is a positive “shadow value” of experience, which declines to

exactly zero at the end of the product life cycle (designated as time 1).
Alternatively, the right-hand side of this equation may be labeled MC,
or “true marginal cost,” equal to current marginal cost MC, adjusted for
the future cost-reducing effects of current output. This analysis assumes
STRATEGIC ISSUES 401

that any capacity constraints are nonbinding.47 (With capacity con¬

straints, MR may exceed MC.)
This relationship allows us to characterize, at least crudely, the rela¬
tionship between prices and costs in an industry with learning economies.
We must have

(7-2) 0 = MR - MC > MR - MC > MR - MC,

where MC is current marginal cost and MC is an upper bound exceeding

MC. Under the null hypothesis of quantity-taking Cournot competition,
we may substitute P(1 + s,/p) for MR in equation 7-2, and after deduct¬
ing off our upper bound on MC we should have a quantity that is une¬
quivocally nonpositive.
Under the alternative hypothesis of cartel behavior, perceived marginal
revenue will be less than under Cournot competition, and

MRcourno, ~ MC > MRcartel - MC = 0,

so that Cournot marginal revenue less “true” marginal cost will be pos¬
itive (although MRCournot - MC may be negative if upper-bound MC
exceeds MC by enough). Our basic procedure for testing for competitive
behavior, then, will be to construct an upper bound on marginal cost,
then deduct it from estimated marginal revenue maintaining competitive
Cournot assumptions. Under competition, the result should necessarily
be less than or equal to zero. Under the alternative hypothesis of collu¬
sion (A., > 0), marginal revenue less an upper bound on current marginal
cost may be positive (although it need not be).

Bounding Marginal Cost

The most difficult issue in an exercise like this is always to find data
that allow one to say something reasonable about marginal cost. For two

47. It does, however, permit discounting of future revenues and costs by the firm when
it makes its output choices. An equation such as this (with current MR set equal to current
MC less a positive term) can easily be shown to hold in asymmetric industries (with firms
of different sizes), as firms continuously make production choices over time in the case of
nonstrategic behavior (in which rivals are assumed to have precommitted output paths). A
similar result has also been shown to hold in a symmetric-industry equilibrium (identical
firms) when firms behave strategically in their choice of current output levels (that is, where
firms manipulate rivals’ output levels in future periods, by signaling lower future costs
through increases in their own current output, via learning effects). See Drew Fudenberg
and Jean Tirole, “Learning-by-Doing and Market Performance,” Bell Journal of Econom¬
ics, vol. 14 (Autumn 1983), p. 527.
402 STRATEGIC ISSUES

Japanese companies—Mitsubishi and NEC—the public cost submissions

discussed in chapter 5 contain a reasonable amount of data on a cost
concept known as cost of manufacture (see appendix 5-B). COM is
essentially equal to average variable cost, plus some allocation of depre¬
ciation charges for the manufacturing facility in which 1M DRAMs are
produced, plus some allocation for R&D costs.
Variable costs of DRAM production essentially have two components.
Briefly, silicon wafers are processed into semifinished chips, then tested.
The good “yielded” chips are then assembled and packaged, then tested
again. The yield of good chips is what is believed to increase significantly
with experience as the result of learning economies. Some portion of
variable cost is sensitive to the processing cost for the silicon wafer start¬
ing down the production line, while another portion varies with the num¬
ber of good, yielded chips that are then assembled and tested. If the
variable processing cost per wafer is c, and the cost per yielded chip is
d, then total variable cost is given by y(c/w + d), where y is finished
chips and w is yield of good chips per wafer. Yield w, in turn, may be
expected to be an increasing function of experience, as measured by
cumulative output Q.
For given experience (cumulative output), then, yield levels are fixed,
and marginal cost is likely to be approximately constant and equal to
average variable cost. This is not an exact description of the cost struc¬
ture, but it is likely to be a decent approximation. Current-generation
1M DRAMs were generally produced in large-scale factories of recent
vintage that were producing a relatively small number of products.48
Marginal decisions to increase output of DRAMs basically require the
producer to cut back on production of some other product (such as
SRAMs or microprocessors), but this is unlikely to have any significant
impact on the variable cost of processing a wafer containing DRAMs or
of testing or assembling a finished DRAM. In effect, the typical short-
run marginal cost curve is likely to be L-shaped: approximately constant

48. See for example, W. C. Holton and others, JTECH Panel Report on Computer
Integrated Manufacturing (CIM) and Computer Assisted Design (CAD) for the Semicon¬
ductor Industry in Japan (McLean, Va.: Science Applications International Corporation,
1988). “The large semiconductor wafer fabrication plants visited by the JTECH panelists
are engaged in high-volume production of VLSI standard products such as dynamic and
static memories and microprocessors. A single wafer fabrication line is typically used to
produce perhaps five to fifteen different products (for example, static and dynamic random
access memory (RAM) devices of different bit capacity, several different microprocessors)
using one or two standard processes” (pp. 14-15).
STRATEGIC ISSUES 403

marginal cost as output of DRAMs is increased, then soars up to infinity

as plant capacity is reached.
Now, the Commerce Department’s measure of the cost of manufac¬
ture, COM, is defined as

COM = {TVC + °WI 11 + p(t)1.

where TVC is total variable cost (assumed proportional to output Y,

given experience Q and a corresponding yield rate), D(t) is the deprecia¬
tion cost allocated to DRAM manufacture at time t, and p(r) is the ratio
of R&D to semiconductor sales for the company as a whole. By the above
reasoning, we then have

COM MC +
m [1 + p(r)],
Y

where MC is current marginal cost. Since depreciation cost per finished

chip, D(t)/Y, is a large positive number (see chapter 6), and p is positive,
we may consider COM an upper bound on current marginal cost.
In fact, COM is likely to be significantly above MC. Depreciation costs
are a significant share of total production cost (likely to exceed 20 per¬
cent),49 and fragmentary data suggest that the estimate of p produced by
the Commerce Department’s procedures was also high (about 0.36 for
Mitsubishi).50 Although the data for Mitsubishi and NEC also suggested
that there was a lot of variance across firms in these ratios,51 it seems

49. A rule of thumb in the semiconductor business is that there is a 1:1 overall ratio
between the capital cost of new capacity and the volume of annual sales generated by that
capacity. With a five-year life for a new production line and straight-line depreciation, this
suggests that depreciation amounts to roughly 20 percent of sales. For a product such as
DRAMs, whose manufacturing costs account for relatively more of the value of sales than
the overall semiconductor industry average, one would expect this ratio to be even greater.
50. For fifteen data points describing Mitsubishi’s 256K and 1M DRAM costs at various
times from 1987 through 1991, the average estimate of p—estimated as COM divided by
the total cost of wafer fabrication, assembly, and testing—was .36. This corresponds to a
cost structure with R&D amounting to 21 percent of constructed value: p/
[(1 -I- p)(1.08)(l +s)] = .21, with s an estimate of sales, general, and administrative expenses
as a share of COM. A single company wide estimate of p was applied by the Department
of Commerce to all Mitsubishi semiconductor products. See appendix 5-B.
51. It was also possible to estimate ratio s—sales, general, and administrative expenses
(SG&A)—as a share of COM using various data from Mitsubishi and NEC. Using ninety-
two observations for various models of DRAMs over the 1987-91 period, and either the
relationship COP = (1 +s) COM or Constructed Value = (1.08) (1 +s) COM, 5 for NEC
averaged 0.39. This corresponds to a cost structure with SG&A amounting to 26 percent
404 STRATEGIC ISSUES

reasonable to conclude that COM is likely to be well above MC, and

serve as quite a high upper bound on marginal cost, when output is not
constrained by capacity limitations.
Both marginal revenue and this upper bound on marginal cost for 1M
DRAMs were then calculated for Mitsubishi and NEC over the period
from the third quarter of 1989 through the fourth quarter of 1990. Since
(see the discussion above) Japanese firms are known to have started to
cut back production of 1M DRAMs in the third quarter of 1989, we can
safely assert that after that quarter production was not capacity con¬
strained. Therefore, profit-maximizing producers at that point would
have been setting perceived marginal revenue equal to marginal cost (and
not exceeding marginal cost as might be the case with output constrained
by capacity). And because DRAM production fell short of capacity, we
would also expect to be on the flat part of an L-shaped marginal cost
curve, with marginal cost approximately constant.
Figures 7-3 and 7-4 plot marginal revenue under the assumption of
Cournot “competition” for Mitsubishi and NEC using several different
sets of price data (a large-user price in Japan and an estimated Japanese
spot price, as reported by Nihon Keizai Shimbun, a U.S. contract price
as estimated by Dataquest, and a Japanese contract price estimated by
Dataquest), and quarterly market share estimates from Dataquest.52 Un¬
der competitive assumptions, these estimates of marginal revenue should
be well below the generous upper bound on marginal cost that COM
should provide.53
They are not in 1990—in either U.S. or Japanese markets, using either
spot or contract prices as the measure of revenues generated by a mar¬
ginal sale—and exceed this upper bound on marginal cost by a substan¬
tial margin. I conclude that the (admittedly imperfect) cost data that are

of constructed value (= s/[(1.08)(l + .?)]). For Mitsubishi, using thirty-one observations on

various types of DRAMs, s averaged just 0.165, corresponding to a cost structure with
SG&A equal to 13 percent of constructed value.
52. The NEC and Mitsubishi fourth-quarter 1990 market shares were used in the first-
quarter 1991 estimates of their respective marginal revenues, since market share estimates
for that quarter were unavailable. A price elasticity of demand of —1.5 was used in con¬
structing these marginal revenue estimates.
53. Actually, the upper bound on COM shown is 1.25 times COM (in the case of NEC,
1.25 times the maximum of the estimates of COM reported for three chip variants). This
is because with these ranged data, reported COM may be up to 20 percent less than the
actual value (and the actual value therefore up to 25 percent greater than the reported
value).
STRATEGIC ISSUES 405

Figure 7-3. Mitsubishi Competitive Marginal Revenue versus Cost of

Manufacture (COM) Reported to Commerce Department, 1989-91

1989:3 1990:1 1990:3 1991:1

Source: Author’s calculations and Dataquest and Nikkei data. Calculations explained in the text. Marginal revenue
based on use of alternative price series is graphed.

available strongly support the contemporary accounts of collusive mar¬

ket-sharing behavior by Japanese producers.

The Costs of Facing a Cartel

Although the STA appears to have played—at the very least—a cat¬
alytic role in the move toward market sharing by Japanese producers,
one can entertain the thought that this particular Pandora, having
emerged from her box, is probably the most persuasive argument to be
mustered for the proposition that declines in U.S. semiconductor man¬
ufacturing capability pose strategic issues for the U.S. economy. Put most
simply (and abstracting from the many nuances that muddy discussion of
what actually has occurred behind the scenes within the Japanese semi¬
conductor industry since the 1986 accord), I wish to ask what the cost to
U.S. chip consumers might be if they faced a hypothetical foreign DRAM
cartel, and therefore what they might reasonably be willing to spend (for
example, by subsidizing entry by U.S. firms) to ensure that such a cartel
could not be formed.
406 STRATEGIC ISSUES

Figure 7-4. NEC Competitive Marginal Revenue versus Cost of

Manufacture (COM) Reported to Commerce Department, 1989-91

Yen per 1M DRAM

I propose to frame this argument in three stages. First I will review a

simple model of the semiconductor product life cycle that assumes non-
strategic, noncooperative behavior on the part of producers. I will then
show how strategic, but noncooperative behavior might be expected to
change producer decisions. Finally, I will sketch out a particular scenario
for cartel formation and simulate its impact on American chip consumers.

Modeling the Impact of Cartelization

My modeling efforts will focus on the 1M DRAM for two reasons:

first, it was the most recent generation of DRAM chip for which relatively
reliable data were available at the time the research was begun; and
second, its product life cycle largely corresponded to the period of the
STA. This second consideration was particularly important, since I
wished to approximate the potential impact of collusive behavior over a
period when it was alleged to have occurred.
The argument that collusive behavior could have significant effects in
DRAMs is not inherently unreasonable. In chapter 5 it was observed that
STRATEGIC ISSUES 407

during the early years of 256K and 1M DRAM production the industry
was somewhat more concentrated than the historical norm, but by 1989
concentration in both these products looked similar to earlier historical
levels. Levels of geographic concentration offered greater promise as a
reason for concern. At the time the STA was signed, reductions in pro¬
duction or exports by Japanese producers could have had potentially great
impacts on aggregate worldwide supply of DRAMs.

Strategic Behavior

Next I modify the model developed in the last chapter in order to

permit noncooperative, strategic behavior on the part of producers. As
in the previous chapter, competition will be thought of as occurring in
two periods. In the first period firms sink resources into capacity invest¬
ments. In the second period firms select a time profile for capacity utili¬
zation rates and output, given the available capacity resulting from their
first-period investment decisions. Given first-period capacities, the second-
stage game in utilization rates results in a static Cournot-Nash equilib¬
rium. Now, however, I assume that this two-stage game is characterized
by subgame perfection. When firms select first-period capacities, they
now correctly anticipate the resulting second-stage equilibrium. (Though
in the first period rivals’ capacity choices continue to be perceived as
unaffected by one’s own capacity choice.) Strategic interactions arise
because first-period capacity choices correctly anticipate effects on one’s
own and competitors’ second-period utilization decisions.54
Since the second-period subgame remains identical to that analyzed
in the nonstrategic case (that is, taking capacities as given, optimal uti¬
lization rates are exactly as analyzed earlier), we need only consider how
the determination of optimal capacity is changed. As before, I assume a
symmetric equilibrium of identical firms.55
To do so, we must add an additional term to equation 6-5 which reflects
the anticipated effects of increasing one’s own capacity on second-stage
outputs. Given the optimal path for the utilization rate u*(t) and the
54. Another possible approach to strategic behavior, not pursued here, would be to
make capacity investments sequential and permit “first-moving” firms to anticipate and
preempt capacity investments by “follower” firms, in a von Stackelberg-type model. A
major component of such an approach would be to describe the economic basis for asym¬
metries among firms.
55. Formal analysis of how strategic competition in capacity investments alters the
model developed in the last chapter is presented in appendix 7-A.
408 STRATEGIC ISSUES

corresponding trajectory E*(t) for the experience variable, the necessary

condition for optimal capacity choice is that
1

(7-3)
1
must hold (see appendix 7-A). Note that this partial derivative is with
respect to K alone, evaluated with u(t) set equal to optimal control u*{t).
This differs from the expression developed from equation 6-5 because,
in the nonstrategic case, initial capacity investments were not viewed by
the firm as affecting rivals’ second-period utilization decisions; that is,
dx/dK = 0 was assumed. With more than two firms in the industry (N >
2), and with symmetric equilibrium, concavity of the industry revenue
function is sufficient to guarantee that rivals will reduce their output
along an interior segment. Since Rx is negative, the additional effect
added on to equation 6-5 must be positive, which means that the marginal
return on additional capacity will be positive at the level of investment
corresponding to the nonstrategic equilibrium. With strategic competi¬
tion, additional capacity will seem attractive.
Along a “boundary” segment (when available capacity actually con¬
strains production), however, the response by competitors to a marginal
increase in one’s own capacity will be zero. Intuitively, when a firm’s
marginal revenue already exceeds its true (corrected for learning econ¬
omies) marginal cost, so that it chooses to operate at maximum capacity,
a small increase in a competitor’s output—which reduces the firm’s mar¬
ginal revenue (with a concave industry revenue function)—will still leave
the firm wishing to produce at maximum output. Thus, if the symmetric
industry equilibrium is one in which firms are everywhere capacity con¬
strained, producing at full blast throughout the product cycle, the stra¬
tegic equilibrium will coincide with a nonstrategic equilibrium. Strategic
behavior will have no effect on the industry equilibrium.
Note that I am allowing only partially strategic behavior with this
modeling strategy. There are two possible instruments of strategic be¬
havior: capacity choice and output (capacity utilization) choice. Output
choice has potential strategic effects via learning economies, insofar as
output at any moment in time affects future cost structure. (Without
learning economies, current output choice has no effect on future cost
structure, and therefore no strategic value.) A firm might choose current
output taking into account the impact of its own current output choice
STRATEGIC ISSUES 409

on rivals’ future output choices, rather than taking the rivals’ future
output path as given. To reduce complexity, I do not allow strategic output
choice, and output paths are chosen as open loop, in which firms precommit
to an output path at the beginning of the second stage of the game.56
However, I do permit strategic competition in capacity. Firms do use
capacity choice strategically, in order to affect rivals’ choice of (open-
loop) output path in the second stage of the game. Thus strategic com¬
petition is permitted in capacity choice, but not in output choice (via
learning effects).
Table 7-1 reworks table 6-4 to reflect strategic behavior, with equation
7-3 used instead of equation 6-5. If no solution for a system based on
equations 6-9 and 7-3 was found, the alternative, corresponding to full-
blast production throughout the product cycle, is identical to (unmodi¬
fied) equation 6-5, since dx/dK = 0 holds along a capacity-constrained
segment. In what I earlier tentatively concluded was the most realistic
set of assumptions about how learning economies work (learning occurs
at the plant level, with parameter y = 1), strategic behavior appears to
have no impact on the symmetric-industry equilibrium, and full-blast
production throughout the product cycle still prevails. In the case where
learning occurs at the company level (y = 0, which had a substantial
interior segment with nonstrategic behavior), capacity per firm increases
by about 50 percent relative to the nonstrategic case, and the equilibrium
number of firms drops from four to three, for the range of fixed R&D
costs considered plausible. The resulting path of DRAM price is shown
in figure 7-5. Strategic behavior thus has no impact with the “per fab
line” specification of learning, but a large impact in the case of compa¬
nywide learning.
For the reasons outlined earlier, the y = 1 case continues to appear
most plausible. Thus I conclude that—within the framework developed
here—my best efforts at discovering realistic values for empirical param¬
eters relevant to 1M DRAM production suggest that strategic competi¬
tion in capacity investments had little potential for shaping market out¬
comes in the industry.

56. One possible (but only partial) empirical justification for this specification is that
there are very long lags between capacity utilization decisions and the resulting output.
The production process for a chip takes almost a full quarter year, so any change in output
rates is observable by one’s rivals only after a substantial lag.
 in
ON

r- r-1 .—1
cn CN
CN

sO rH cn
NO in
CN

00 -h cn
Os Os
CN

NO rH cn Os rH O N’ rH O
cn o r- O
cn in
in NO 1 sd
1

00 rH r- NO rH so cn rH SO
o cn ON
in tj- NO
i Tfr
1

CN rH rH CN rH in NO rH OS
cn Os rH ON
NO
1

Tf rH T^- rH rH O rH CN
ON o ON ON r- N-
Os NO
NO oo
1
oo
ON rH i—H rH Tj" 00 rH rH r-
s NO NO
i>
o NO r- OS
rH rH
o
oo r- (N cn CN ON
00 ON
SS 1 o’
rH
1

•2 j*.
CN rH 00 N- rH CN O
-O SO rH CN rH rH
NO 00
-2 s 00
Os
00
ON
00
cn
00 n*
CN
Os rH
OO
d
rH 1 rH 1
5 I rH
ON rH CN Os rH SO CN rH in cn rH rH
r- cn rH r- cn o
u Os
cn rH
NO
CN NO
rH rH rH 1
<L>
-*h
<3
L. rH rH in rH CN cn rH N- rH rH
■*h cn Os Os ON
Co rH
00
CN
CN
N"
tH ON in
rH
rH rH rH rH rH 1

-2 CN
-o N"
cn
rH
o
cn
r-
rH O
00
SO
r-
rH in
in
00 rH NO
N"
o ON rH 00
cn in NO r-'
rH CN rH CN rH rH rH

O
k. cn rH in rH O CN rH CN NO on r-
o oo rH
C fn sd
ON 00
rH
NO
cn
d
in 00 on r-
d d ™
o H Tt CN cn CN cn
O
00 rH o CN rH o rH rH o in oo o
© oo ON r- Os rH NO ON ON
<N ON
cn
rH cn
cn d ^
rH o' CN o' CN o' CN o
rH rH rH

ON o NO rH O NO rH o cn rH O
<§■ cn
r-
cn
NO
cn
ON r-
in
00
NO
in NO OOr
rH ON rH CN rH in oo"
•s CN cn cn rH cn

3
CL
3 3 3
CL CL CL
0X) 3 0X) 3 3 3
0X)
c o c , o a O OX)
C O
© wo 'Tn ’wo C s -h
tn C/3 C/3 wo O cr>
<u
o "h
Os <L> S s
o
00 <D
CJ
O' **
O •H c
d
C
Hi C
d
O
u.
5^3 d
<D <D
Cl CJ "h U. II CL ■Q 3 Hi II CL
r- i 3 d i 5 H i
uo3l 3
c/3
ju C/3
o
3

£
 r- r-H r- vO 00 Tf 00 rj (N oo r- o r- to NO
© o rf 00 On • r~ o © rf O 00 04
o- CO
CO CO d ^ rH d O d cd d O
i“H 1 rH 04 04
1 1 1 1

lO r- 00 to CO OV ov cn CO rj- Qn rt 04 to
’”H Ov VO 00 VO oo 04 tJ- OV © 04 tJ-
Ov 04 v® S
to © T to o ^ vd d ^ vd d ^
O 1 rH i rH rH 04 04
,""1
1 1 1

rH OO VO Ov 04 CO OV to oo to CO CO ov 00
vO On 0- rH 00 rH VO VO to rH -j ov © rH O
O
d 00 d 7 00 d °° Ov On d ^
1 1 rH rH rH rH
1 1

o- tO VO CO On OV On 04 00 VO 00 to VO On
© O' to ts r- oo 04 VO H rsl
vN
CO O vq H VO
rH d °r rH d 04 d ^ 04 d n 04 d ^
04 CO 04 04 04 H rH 04 rH

VO Tt O Ov O On o r- o 04 O
t-; On 00 On r- co • VO 00 © • Tt rH • to
O V.0
d ^ 'd- d *> 04 d °°r rt- 0 °° 04
04 o 04 o o 04 rH 04 co"
rH rH rH

co 00 o r- vO O o- 04 o Ov rH O 00 t-4 O
ov 00 r- 00 to to r- 04 VO to VO 00 CO 00
a. Millions of product-cycle wafer starts.

d ^ d d °°r vd d Tf d
r-H ov rH rH rH Tj- rH Ov rH Tj-
co TT Tf to
Source: Author’s calculations.

3 3 3 3
a.
4—'
a. CL
4—*
CL
4-J
oc 3 oc 3 oc 3 3 3
C * O
b. Millions of dollars.

C r O c O O g», O

oc C/5
vO IS sjd <u cd u-, £ * J co 4> 4) • iS
o O
o ^ c o c o 8 >.f c 2 5*5 3 o
3

a
o
u> a. 'o 4)
II a'5 3 4)
CL O 3
4)
Q
cd S• 9-
cd
</5 cd
• 3 cd
i cd «4-i
cd
I Cd <4-
cd
cd
CL
E h- cd
> C/5 4) 9" 4) B* **
« S ■§ o TJC w
>■»
4) 9" —“ cd
E .cd ■o ; o - S ^ ; o E
E c t-
E
<J
i| ls 'W
u
E
cd W a cd O cd ^ PJ O cd ^ o cd W o
o o o a
412 STRATEGIC ISSUES

Figure 7-5. Historical Prices for 1M DRAMs Compared with Simulated

Strategic Equilibrium Time Profiles

Dollars per chip

Source: Author’s calculation and Dataquest and Nikkei data. Calculations explained in the text.

Collusion and Cartel

Now it is possible to attempt to answer the question posed at the

beginning of this chapter. Relative to a benchmark of strategic or non-
strategic—but competitive—behavior, can the impact on U.S. consumers
of the systematic exploitation of monopoly power by a foreign DRAM
cartel, that is, of collusive behavior, be estimated?
The simplest variant of the story to be told to motivate this calculation
runs along the following lines. Suppose, for one reason or another, that
firms in country U believe that the fixed sunk cost F required for them
to enter the industry is higher than for firms in country J (that is, that
the firms in country U believe themselves to be less efficient producers).
In a competitive equilibrium where entry ensures zero economic profits,
there will then be no entry by country U firms, and only (more efficient)
country J firms will populate the industry. Country J firms alone then
STRATEGIC ISSUES 413

Table 7-2. Simulated Welfare Effects of Cartelization on the

DRAM Mar kef
Strategic competition in
capacity with:

Competitive
output Second-period.
Simulation choice cartel Change

Gamma = lb
fs, end full-blast output 1 0.237
P(l), final price (dollars) 3.77 9.37
Consumer surplus (billions of dollars) 148.9 117.4 -31.5
Industry net profits (billions of dollars) 0.045 4.564 4.5

Gamma = 0C
ts, end full-blast output 0.16047 0.037
P(l), final price (dollars) 1.61 4.08
Consumer surplus (billions of dollars) 287.9 186.4 -101.5
Industry net profits (billions of dollars) 3.957 34.57 30.61
Source: Author’s calculations,
a. Symmetric industry equilibrium assumed.
b. F equals $250 million; N equals fourteen firms; and K equals 4.655 million product-cycle wafer starts.
c. /-'equals $250 million; N equals three firms; and K equals 22.648 million product-cycle wafer starts.

play the two-stage game described above and select sunk capacity in¬
vestments corresponding to the strategic (or the nonstrategic) case.
Now suppose that, after the first (capacity selection) stage but before
the second (production) stage, an external force intervenes and changes
the rules. The country J firms are to be permitted—perhaps even en¬
couraged—to form a cartel structure, to reduce output and raise prices.
How much will consumers of DRAMs lose? How much will producers of
DRAMs gain?
Carry the argument a step further. Suppose country J firms, if left to
their own devices, are perfectly prepared to form a cartel on their own,
but that to do so requires that some “critical mass” of existing DRAM
producers be country J firms. Then country U can prevent cartel forma¬
tion by subsidizing entry by its own firms (which are assumed to be
behaviorally, or for legal reasons, indisposed toward collusion). What
level of subsidies should country U be willing to pay to ensure against
cartel formation?
Table 7-2 shows the loss to consumers and increased producer profits
associated with a second-stage switch from a noncooperative Cournot-
Nash equilibrium to cartel formation. Initial capacities are assumed to
414 STRATEGIC ISSUES

have been chosen as the first stage of the competitive game described
above, with an anticipated competitive second stage. Now, however, the
conjectural variation parameter a has been reset from a competitive,
Cournot value of 1/N to the collusive, cartel value of one, in calculating
the second-stage output path.
It is striking that the lost consumer welfare is so very much larger than
the profits gained by producers.57 There are two reasons for this. First,
fixed cost per unit produced rises considerably when output is cut back
sharply in the second stage relative to the competitive output levels for
which capacity was initially chosen. Second, the existence of learning
economies means that, when producers cut back output, they also raise
their variable costs per unit over the remainder of the product cycle.
Thus, much of what might have been additional monopoly profit, without
scale and learning economies, is chewed up by higher unit costs incurred
when output is reduced.
Thus, for example, in what I have argued is the more realistic case of
y = 1, a switch from competition to collusion in 1M DRAMs costs over
$30 billion in forgone consumer welfare, over a five-year product cycle,
while producing monopoly profits of only about $4.5 billion over the
same period. (Note that at the height of the real-world 1M DRAM
shortage of 1988-89 some analysts’ estimates put “excess” industry prof¬
its at levels of $1 billion to $2 billion per year.)58 Clearly, if these calcu¬
lations are correct, even if country U accounts for only 40 percent of final
DRAM consumption, its consumers should be willing to pay $2.4 billion
or so annually to prevent the formation of a cartel (40 percent of $30
billion is $12 billion, or $2.4 billion annually). Of course, there may be
other, less costly ways to inhibit cartel formation by foreign producers: if
governments can agree to enforce some common international antitrust
standards, for example, and foreign governments apply such norms to
their producers, the benefits of competition might be available to U.S.
consumers at no cost.

57. In the case of a constant-elasticity demand curve, it can be shown that the total
consumer surplus CS is given by:

58. See Kenneth Flamm, “Internationalization in the Semiconductor Industry,” in Sem¬

iconductor Materials and Equipment Institute (Newport Beach, Calif., January 1989). These
numbers cover both 256K and 1M DRAMs.
STRATEGIC ISSUES 415

More Complex Stories about Cartels

The models developed in this chapter and the previous one, despite
their many simplifications, seem to do a tolerable job in tracking reality
in at least some respects. The extreme sensitivity of my simulations of
symmetric equilibria to the specification of the experience variable rele¬
vant to learning economies was, to me at least, quite unexpected. Grat-
ifyingly, the “per fab” specification of learning economies appears to
solve the somewhat puzzling results of earlier research: Cournot behavior,
coupled with a reasonable empirical approximation to the costs of real-
life DRAM production, then gives much more realistic predictions about
industrial structure.
The story told in the last section raises some interesting questions, not
all of which are easily answerable. First, how realistic is this story? Is the
cartel threat—and the resort to anticartel insurance—or some lesser
degree of collusion a serious possibility? Were the DRAM price and
output manipulations of 1988-89 a purely transitory phenomenon, or has
some threshold been crossed into a less competitive, more cooperative
world of “market sharing”? Continued production cutbacks by Japanese
producers after 1989 may suggest a break with pre-1986 patterns of firm
behavior. Also, efficiency differences between groups of firms could be
tied to a story about intergenerational externalities, where incumbents
have lower costs because they are incumbents.
Second, if country J firms can form a cartel, why not do so in stage
one, and choose more profitable levels of capacity investments? My an¬
swer is that, at stage one, entry by others will be stimulated if reduced
capacity investments imply that positive rents will be earned, while at
stage two new entry is no longer possible.
More important is the issue of whether cartel pricing would attract
additional entrants in later plays of this game. This is a complicated
matter, but one can construct at least the germ of an argument as to why
entry by noncountry J firms might not occur. Let us imagine that nonin¬
cumbent country U firms operate at a cost disadvantage (say in fixed
entry cost F) relative to incumbent country J firms, and that entry by a
single nonincumbent is enough to completely disrupt the cartel. Let us
also imagine that capacity investment is decided as the first stage of a
two-stage competitive game (if it were apparent that the number of firms
and levels of capacity set in the first stage were other than at competitive
416 STRATEGIC ISSUES

levels, such as to imply rents in the second stage, further entry by other
firms would presumably be stimulated). Then we have a terrible dilemma
for the potential country U entrant: if it does not enter, the country J
firms may form a cartel in the second stage, in which case it must sit and
watch as others earn rents. If it does enter, however, it disrupts the cartel,
guarantees competition in the second stage, and thus guarantees itself a
loss in a zero-profit equilibrium (since it produces at a cost disadvantage
relative to the rest of the population of firms).
Third, my consumer welfare calculation assumes that monopoly rents
are extracted by a cartel at arm’s length from a competitive user industry,
which then passes on cost increases directly to final consumers. It is well
known that if an input is not fixed in proportion to output, a monopolist
controlling the pricing of an input maximizes profit by integrating forward
into the user industry that purchases that input, in order to avoid substi¬
tution away from the input.59 Since chip-producing firms in country J
may also be systems producers, there would appear to be few barriers to
this occurring in the long run. The welfare cost to final consumers may
increase if this were to occur.
Fourth, what if a domestic chip-consuming industry is imperfectly
competitive and extracts rents from sales to foreign consumers? Can this
not increase the welfare loss from a domestic standpoint, particularly if
forward integration by foreign chip producers and the exit of domestic
equipment producers accompany monopoly rent extraction?
Fifth, the logic of anticartel insurance suggests that chip consumers as
a group ought to be willing to subsidize entry by cartel-busting entrants.
(The ill-fated U.S. Memories consortium, in which chip users proposed
to fund entry into the DRAM business, comes to mind.) However, there
will be a free rider problem. All consumers gain from the entry of a
cartel buster, whether or not they have actually helped pay for it. A
government role would therefore seem to be needed to handle this “pub¬
lic good” aspect of entry.
Sixth, little attention has been paid to intergenerational externalities.
The fact that several firms—particularly Korean companies and Japan’s
NMB Semiconductor—were able to enter the DRAM business in the
mid-1980s with virtually no prior experience suggests that such external¬
ities, if they exist, are not so significant as to be an insuperable barrier
to entry.

59. See Vernon and Graham, “Profitability of Monopolization,” pp. 924-25.

STRATEGIC ISSUES 417

Finally, the degree of collusion and, indeed, all the details in the
policy-related story I am telling are extreme. The costs calculated here
probably should best be regarded as an upper bound on the costs cor¬
responding to a more realistic scenario. Is there a convenient way to
parameterize less stark forms of behavior? In particular, one might want
to imagine a scenario in which most producers participate in a cartel, but
one or two “outsiders” form a competitive fringe and do not cooperate
with the cartel (but do not break it either). With asymmetric firms,
however, characterization of equilibrium paths becomes much more
difficult.

Conclusion
So-called strategic arguments for an activist U.S. policy in semicon¬
ductors have focused on two issues: the possible existence of technolog¬
ical externalities—spillovers—between semiconductor production and
other types of industrial activity that are not captured by producers, and
the extent to which monopoly profits can be extracted from downstream
chip users by an upstream cartel. This latter issue has been particularly
important in shaping the debate over trade policy in semiconductors.
There is plenty of general anecdotal evidence of Japanese industry
activities that probably would have violated antitrust laws had they
ocurred in the United States (and at least some of which were arguably
also illegal in Japan) in past decades, and continuing up to the present
day. There are even some specific episodes that suggest similar behavior
in the Japanese semiconductor industry. However, in all cases where it
appears that such activities were successful in creating cartel-like indus¬
trial structures, some degree of active or tacit government support for
such organization can also be discerned.
The latter point clearly holds true in semiconductors. Before 1986,
episodes in which Japanese semiconductor producers successfully man¬
aged a reduction in their collective output or exports were in all cases
undertaken with the support of MITI and generally were undertaken as
a response to acute, external trade friction. From 1986 through 1989 the
organization by MITI of formal structures designed to control output
and prices was overt and explicit. Restraint in production and exports
was clearly linked to a deliberate government policy. It is in some sense
418 STRATEGIC ISSUES

circular to argue that the policy was justified by the predatory intent of
restraints that were implemented as part of the policy.
After 1989, however, the visible hand of MITI as organizer of restraints
had disappeared, and it is at least possible to argue that evidence of
collusive behavior reflected wholly private behavior. Available data—
anecdotal and statistical—in fact, seem to show Japanese DRAM pro¬
ducers behaving in a way that does not appear very competitive in 1990.
One is left to wonder, however, whether behind the scenes there was tacit
government support for a policy of private market sharing, to replace the
cutthroat competition sometimes practiced by the Japanese industry, and
minimize the potential for renewed trade friction.
My simulation results, however, made it clear that a foreign chip cartel,
at least in principle, can create a substantial net welfare loss for an
economy making significant use of the cartelized input. Under the right
empirical circumstances, a serious argument can be made for subsidizing
domestic capabilities as a form of anticartel insurance. Whether those
circumstances existed—or continue to exist—in semiconductors is likely
to remain the subject of continued debate.

Appendix 7-A: Solution of the Model with Strategic Behavior

The basic framework is to conceive of competition as occurring in two
distinct stages. In the initial period firms make capacity investment de¬
cisions. In the second period firms decide what utilization rates to apply
to their sunk investments in capacity. I assume that the two-period equi¬
librium is characterized by subgame perfection. Given their first-period
capacity decisions, the second-period game in utilization rates is assumed
to be characterized by a static Nash equilibrium in utilization rates. When
firms select their first-period capacities, they are assumed to be able to
anticipate correctly the effects of their choices on the second-period
equilibrium. In the first period, however, they take the capacity choices
of their rivals as unaffected by their own. The strategic interactions come
from the anticipated effects on their rivals’ second-period output deci¬
sions of their own first-period capacity investments. This contrasts with
the nonstrategic model outlined in appendixes 6-A and 6-B, where the
firm viewed both the first- and second-period actions of competitors as
unaffected by its own actions in either period.
STRATEGIC ISSUES 419

Given first-period capacity investments, the solution for optimal uti¬

lization rates is identical to that given in appendixes 6-A and 6-B. As
before, we restrict our attention to symmetric equilibria of identical firms.
Strategic effects enter the analysis when we consider optimal capacity
choice. Along the optimal trajectory for the utilization rate u*{t) and the
corresponding trajectory E*(t) for the experience variable, a necessary
condition for an optimum choice of capacity K is that

Note that this partial derivative is, with respect to K alone, evaluated
with u{t) set equal to optimal control u*(t). This differs from the expres¬
sion developed in appendix 6-B (equation B-2) because of an additional
term,

(7A-1)

added onto the left-hand side. In the nonstrategic case, initial capacity
investments were not viewed by the firm as affecting rivals’ second-period
output decisions, i.e., dx/dK = 0 was assumed. The remainder of this
appendix develops term C-l explicitly.
First I adopt some additional notational conventions: variables sub¬
scripted with y refer to the firm making a strategic calculation, and
variables subscripted with i to its N — 1 rivals.
A firm sees each rival as maximizing its own Hamiltonian, Hh exactly
analogous to the Hamiltonian shown in appendix 6-A, subject to the
constraint that firm fs utilization rate «, be less than or equal to one. (I
rule out the possibility of a symmetric equilibrium with no output pro¬
duced [«, = 0] for reasons described in appendix 6-A.) The firm takes
its rivals’ capacities as given and is interested in how its rivals’ capacity
utilization rates vary with its own capacity choice.
420 STRATEGIC ISSUES

with x, = u,w(Ei)Kj, |A, a nonnegative Lagrange multiplier, and

5, = - d + Y7julwE,Kr

Applying the maximum principle using this Lagrangean, we have the

following first-order condition:

dL, _ d/f, _ 0
da, da, 1X1

Since

d//. 5..
~ = (P'x, + P - d + ^) *(£,)*, - cK,
K7

for each rival i, a firm knows that the first-order conditions determining
a, and |x, are

(7A-2) (p'x, + P - d + w(E)K, - CAT, - m, = 0

fL/(l ~ U,) = 0,
with |ji, > 0.

The firm does comparative statics by taking derivatives with respect

to Ky, which gives, in the case of the first equation of this two-equation
system (in two unknowns):

w(E,)K,
dlljdK dUjdKv
—

dK I
lx, + r^
dU/dK

p, L dx, dUf dx^d% dy_ djX,

= 0.
\ dUjdKy du,dKy dKy d K.,

Because the firm faces identical symmetric competitors, however,

da. dUj
w(£,) = w(£,), Kt = Kj,
d Ky d k;

and we need solve only for and instead of 2(TV - 1) such terms.
dKy dKy
dy
Let ——- be uyw(Ey), where subscript y denotes the values of these vari-
dKy
ables for the firm varying Ky.
dXj
Since — = w{E)Kt, the comparative statics are thus given by
OU:
STRATEGIC ISSUES 421

[w(E,)K,]2 [N 1 )(P"x, + F) + F] dui

dKy

dll,
~ M-i (1 _dKy_

-[w{E)K]uyW{Ey)(F'x, + F)

Let M be the square matrix on the left-hand side of this equation, and
\M\ its determinant.
Then,

dUj (1 - Uj) F,
dKy
1
dfi,- “ \M\

_dKy_ 1 [w(£,.)A:,.]2[(A^ - l)(P'x, + F) + F]

-w(£,.)^w(^)(T% + F)

where \M\ = [tv(E,.)tf,.]2{(W - \){F'x, + F) + P'}(1 - u) - |x(.

There are two general cases to consider. First, along interior segments,
where q., = 0 and u, < 1, \M\ will be negative in a noncooperative,
symmetric Nash equilibrium when N > 2. In this case the term in braces
must be negative.60

60. By the strict concavity of the revenue function, zP' < -2P'.

Therefore F'x, < —( — 2P')

422 STRATEGIC ISSUES

Second, along boundary segments, where |x, > 0 and u, = 1, \M\ =

- n, < o .61

du, .
The general expression for —— is
dKy

nA ~ dUi_ _ -(1 - u)w{E)KiuMEy){P"xi + P'}

( } dKy \M\

which, in a symmetric Nash equilibrium with N > 2, will be negative

when u, < 1 (because the expression in braces will be negative; see note
60). Noting that, in a symmetric equilibrium, w, = uy = u, Ei = Ey =
X- 1
E, Kj = K = K, and so on, and — = —,
z N

F'x, + P' < f(~2) + 1 P'.

Since P' < 0, the right-hand side of this inequality will be negative as long as

—( — 2) + 1 > 0
z

5
z
X. 1
In a symmetric equilibrium, —' = — .

N
Therefore, F’x, + P' < 0 as long as 1 < — • N > 2 is sufficient for the bracketed term

to be negative.
61. Actually there is a third possibility: that p., = 0 and u, = 1, in which case M is
singular and duJdKy undefined. This can only occur at the transition from a boundary to
an interior segment and can never last more than an instant. (With u = 1, industry output
must be rising and marginal revenue Ry falling. But with p = 0 over some interval with u
= 1,

.• . c y 8 c wK
weK'-<
w2^
(Ry) = :~2We7Z + -7Z = —
W TO /G
= -weK1-* - -Iwp/C1-1'
w
= 0,
which cannot be true in a symmetric-industry equilibrium with all firms at full capacity
over some interval. Therefore, we cannot have u = 1 and p = 0 for more than an instant.)
STRATEGIC ISSUES 423

0, w, = 1
dUj
(7A-4) = u
dK„ -, < 0, ut < 1.
K P" 1 '
1+ ~ »[r*N + \
A key point emerging from this analysis is that—in a symmetric equi¬
librium—over periods when output is capacity-constrained, a marginal
increase in capacity will have no impact on rivals’ output. Intuitively,
when a firm’s marginal revenue exceeds its marginal cost, a small increase
in a competitor’s output—which reduces marginal revenue—will still
leave the firm wishing to produce at maximum output.
Returning to the task of simplifying expression C-l, we then have:

[ R^dt = f P'y^dt = f P'yXw(E,)Ki-~-

Jo *dK Jo 7 dK Jo 7, v ° ‘dKv

(7A-5) P'y(N - 1 )wK

u 44+ dt,
K p" i '
1 + + \

du
since = 0 when w, = 1.
dKy
With a constant-elasticity demand, and p the price elasticity, this
becomes:

(7A-6)
1 -P(A - l)y
pAA
1 + K ■1 * >dt =
\ll
l\
1 + (A - 1) 1 +— 1 -
v [ \p / AJ
424 STRATEGIC ISSUES

1 +
yg. K)P(y{t„ K))
$NK { 1) (1 0-

1 + (N- 1)
N y

This term was added to the right-hand side of equation B-2 in solving
for possible equilibria. Again, it is important to note that strategic in¬
vestment behavior has no effect if the symmetric-industry equilibrium
associated with the empirical parameters of the problem is capacity-
constrained throughout the product life cycle.
 CHAPTER EIGHT

Conclusion:
Mismanaged Trade?

Since 1959 the United States and Japan have been arguing over the rules
for international industrial competition in semiconductors. In some re¬
spects, little has changed.
In 1959 massive Japanese exports of low-priced germanium transistors
ignited concerns of American semiconductor manufacturers. The tran¬
sistor’s importance for computers, communications, and other electronic
systems in an epoch when markets for advanced electronics were domi¬
nated by military applications led to appeals for protection from imports
on grounds that national security was at stake. Japan’s government and
its transistor industry responded by constructing an export cartel de¬
signed to stabilize prices. Transistor consumers around the globe com¬
plained. The U.S. electronics industry was divided: semiconductor pro¬
ducers demanded protection, while systems producers fought policies that
might raise the price of an important component and create a competitive
disadvantage for them in commercial markets. The problem ultimately
was to disappear because of the phenomenal rate of technological ad¬
vance in microelectronics. As individual transistors were replaced by
much more sophisticated integrated circuits, the American semiconduc¬
tor companies pioneering these innovations created entirely new families
of products and rendered the Japanese cost advantage in the older dis¬
crete semiconductors irrelevant.
Two and a half decades later a wave of low-priced Japanese memory
chip exports again threatened a foundering U.S. semiconductor industry.

425
426 CONCLUSION

The crucial function of microelectronics in building high-technology

equipment that provided the U.S. military with its overwhelming tech¬
nological superiority prompted an influential 1987 Defense Science Board
report calling for federal support of a government-industry consortium.
The result was today’s Semiconductor Manufacturing Technology
(Sematech) consortium.
Concerns over national security also precipitated the U.S. govern¬
ment’s pressure on Japan to negotiate the 1986 Semiconductor Trade
Arrangement. The Japanese government worked with its industry to fix
export prices and restrict domestic production and investment. As prices
began to rise, semiconductor consumers around the world again loudly
complained. Ultimately, the American chip industry surged back not
because of an aggressive reentry into memory chip production, but be¬
cause of rapidly improving technology in other areas. U.S. companies
shifted into the design of more technically advanced chips that collapsed
entire electronic systems onto a few chips—whether the microprocessors
and logic chips used in ever more powerful personal computers and
workstations or the specialized logic chips and processors used in tele¬
communications, disk controllers, and multimedia. Trailing behind was
a Japanese semiconductor industry struggling to break out of what
seemed to be an increasingly competitive and decreasingly profitable
memory chip business.
Much has changed over these decades in other respects. In the 1960s
American development of advanced semiconductor technology was
dominated by a cold war defense budget that funded the necessary
research. Over 70 percent of U.S. integrated circuits went into defense
and space applications in 1965. Today the development of semiconduc¬
tors is fueled by an enormous commercial market—the military market
accounts for less than 2 percent of U.S. consumption. Equipment sold
in the marketplace typically uses the latest components, and equip¬
ment shipped to military users (except for very specialized purposes)
often lags behind technologically. What was once a major benefit con¬
ferred by a generous military budget is no longer a significant advan¬
tage for American industry.
In the early 1960s, too, Japanese demand for semiconductors was a
tenth of sales in the U.S. market. Even though American industry under¬
stood that selling in foreign markets was critical to its success, penetration
of the Japanese market was not crucial for an American electronics com¬
pany (although access to the American market certainly was for Japanese
CONCLUSION 427

producers). Indeed, U.S. companies often appeared to settle for unat¬

tractive deals in gaining very limited access to the Japanese market. By
1986, however, the Japanese semiconductor market had become critical:
its size surpassed that of the United States for the first time.
In the 1960s, Japanese companies were in no sense the technological
equals of American companies. Their success in producing inexpensive
transistors was based on low wages, a relatively mature product, and a
labor-intensive manufacturing process. By the late 1980s Japanese com¬
panies generally equaled or surpassed American companies in almost
every area of the rapidly developing technology of semiconductor man¬
ufacture. Only in chip-level systems design and integration did American
semiconductor manufacturers as a group seem to have a significant lead.
And not only was the American monopoly on the technology no longer
a monopoly, it was not even a duopoly. Korean and European competitors
also stood at the frontiers of semiconductor manufacturing. Taiwan and
other Asian countries poured resources into their own developing semi¬
conductor industries and seemed poised to make a similar transition.
In one crucial respect, however, the semiconductor industry today
plays exactly the role it did thirty years ago. In 1959 semiconductors
became the focus for the first major episode of high-technology trade
friction between the United States and Japan. In the early 1970s, semi¬
conductors and computers became the focus for U.S. pressure on Japan
to liberalize access to markets for high-technology products. In 1983
semiconductors constituted one of the first products of concern when the
U.S.-Japan High Technology Working Group began to discuss sectoral
trade and investment issues. In 1986 semiconductors became the first
high-technology industry for which America and Japan agreed on a mar¬
ket share objective to create greater foreign access to the Japanese mar¬
ket. Today, as has been true for almost four decades, disputes over
semiconductor trade continue to be the flash point for efforts to re¬
solve international trade and investment disputes that ultimately will set
precedents for other high-technology industries.

Why Semiconductors?
The economics of the semiconductor industry has caused it to be at
the forefront of high-technology trade frictions. For one thing, the rela¬
tion between product prices and costs of production is enormously com-
428 CONCLUSION

plicated. Economies of scale are significant—there are large fixed in¬

vestments in research and development for increasingly complex chips
and in constructing high-volume production facilities. Learning econ¬
omies are also very important; unit costs fall about 30 percent with every
doubling of cumulative output. Thus the price of an integrated circuit
has historically had only the loosest relationship with some notion of the
average cost of manufacturing a chip at any given time. Companies make
investment decisions based on projected revenues and costs over a five-
year product cycle, knowing that when supplies are limited or demand is
great, prices will soar and profit margins will be fat.
When chip demand falters or supplies are plentiful, prices may be
pushed down toward the incremental cost of producing another chip, and
revenues may not even cover the depreciation charges on the factories
turning out the current generation (not to mention continual R&D ex¬
penses for the next generation). In fact, an economically rational manu¬
facturer might be willing to sell chips for less than the current variable
cost of production if the learning curve is steep enough—and predictable
future decreases in manufacturing costs are large enough to justify con¬
tinuing production through the money-losing times—in order to emerge
a lower-cost producer in a more profitable environment.
These static and dynamic scale economies have also relentlessly driven
manufacturers to sell as many chips as possible wherever possible to
maximize the return on relatively fixed investments in R&D and produc¬
tion capacity and acquire the production experience needed to ride
quickly down their learning curve. From the beginning, then, they have
competed in an international marketplace.
The drive to sell in global markets has inevitably meant that laws
prohibiting dumping can be used to attack foreign imports in times of
slack demand, even though the exporters’ behavior may be economically
rational and no different from the practices of domestic producers. Be¬
havior that might be construed to be a signal of anticompetitive, preda¬
tory intent in an industry lacking these characteristics is a predictable
outcome of normal market forces in semiconductor production. In every
economic downturn since significant U.S. imports of foreign ICs began
in the late 1970s, dumping has been an issue. And it will resurface in
future downturns as long as the current antidumping standards are
in force.
Large R&D investments (historically, 10 to 15 percent of sales) in
semiconductors have made government involvement with the industry
CONCLUSION 429

inevitable in another way. Historically, the U.S. government paid a large

part of America’s semiconductor R&D bill on the grounds of protecting
national security. Even without this rationale, many economists would
agree that government subsidies are legitimate and desirable. As in other
high-technology industries dependent on large R&D expenditures (and
with society’s return on these investments typically much greater than
what a private investor can fence off and take claim to), governments can
and should stimulate greater investment in developing new technology.
But research subsidies can also be explicitly designed to serve strategic
economic ends and help domestic manufacturers defeat foreign compet¬
itors in world markets. There is no way to distinguish between govern¬
ment support for R&D to make a nation’s producers more productive
and innovative contributors to the domestic economy, and support pri¬
marily intended to capture profits from foreign producers.
Although other high-technology industries may also have significant
economies of scale associated with steep R&D requirements, large sunk
investments required to enter production, or steep learning curves in
manufacturing, few have them all simultaneously. Furthermore, semicon¬
ductors have experienced one of the steepest sustained rates of techno¬
logical advance ever recorded for an important economic commodity
(with prices for a bit of memory falling continuously for decades at an
average rate of more than than 35 percent a year). This situation has
been coupled with increasing importance in the economy (and interest
on the part of governments) and a boom and bust cyclicality of demand
that is extreme even by the standards of the rest of the electronics
business. As a result, the semiconductor industry has seemed destined
to be the pioneer in working out trade rules for all high-technology
industries. Other high-technology industries emerging in the global
market in the second half of this century often have had some of these
characteristics, but rarely all, and never to the extent seen in the chip
business.

Strategic Policy in Semiconductors

None of the factors shaping today’s competitive environment was very
clear before the late 1970s. The United States effectively dominated world
semiconductor markets, primarily becaused of its unchallenged techno¬
logical strength. Public policy during the cold war helped build the foun-
430 CONCLUSION

dations for this supremacy, but it had never explicitly grappled with the
economic implications of this advanced infrastructure, nor was economic
benefit ever the purpose of its construction. U.S. investment in devel¬
oping semiconductor technology was strategic only in the very narrow
sense of the pursuit of qualitative military advantage.
The development of the Japanese semiconductor industry, however,
was shaped by very different policies. Initially, semiconductors were
merely an incidental appendage on a different beast. Through the 1960s
the primary focus for Japanese industrial policy in electronics was the
domestic computer industry. To sell in the Japanese market, American
computer companies were forced into joint ventures with Japanese coun¬
terparts, endured sales ceilings, or accepted other restrictions. By the
early 1970s, however, several U.S. partners of Japanese electronics com¬
panies had dropped out of the computer business. It had become clear
that a frontal assault on the established mainframe computer market
dominated by IBM was not going to work unless Japanese companies
could offer lower prices to offset the American advantages in virtually
all other aspects of the computer business.
By then the general design principles for a commercial business com¬
puter were well understood. The extraordinary decrease in the cost of
computing power continued, primarily due to ever cheaper and more
powerful components rather than design innovation.1 So Japanese pro¬
ducers realized that by manufacturing state-of-the-art chips at lower cost
and incorporating them into their machines, they might be able to com¬
pete in established markets for computer systems.
Meanwhile, the semiconductor manufacturing divisions of these same
Japanese companies faced equally rocky prospects. In the mid-1960s,
when American companies first began to ship computers based on ad¬
vanced chips, the Ministry of International Trade and Industry undertook
a small program to improve Japanese semiconductor technology. Japan’s
first semiconductor memory chips were built as part of a related MITI
project, the development of a high-speed mainframe computer.
But these efforts had not closed the gap with American companies.
Cheap American chips flooded world markets, including Japan, and Jap¬
anese producers complained loudly about American dumping. Despite

1. See Kenneth Flamm, “Technological Advance and Costs: Computers versus Com¬
munications,” in Robert W. Crandall and Kenneth Flamm, eds., Changing the Rules: Tech¬
nological Change, International Competition, and Regulation in Communications (Brook¬
ings, 1989), pp. 16-19.
CONCLUSION 431

stiff trade and investment barriers, including the refusal to allow foreign
producers to establish Japanese subsidiaries, the share of U.S. imports
in Japanese integrated circuit consumption steadily rose, peaking at
35 percent of the market in 1971.
As a further attempt to combat this trend, MITI introduced more
informal means to discourage the use of foreign chips. Imports of ad¬
vanced American calculator and memory chips—to be incorporated into,
and often reexported as, Japanese calculators—were increasing quickly.
Calculator manufacturer Sharp had started the rush to American chips,
and MITI responded by making an example of the company, denying
necessary licenses to allow it to use imported chips. American companies
attempting to recruit Japanese semiconductor executives to help build
their Japanese sales even found MITI lobbying the executives to dissuade
them from breaking ranks.
MITI’s tactics proved successful. Although Japan abolished trade and
investment restrictions in the mid-1970s, the import share of Japanese
chip consumption continued to shrink. Meanwhile, government agencies
funded much larger programs than they had previously to improve Jap¬
anese semiconductor technology. From 1975 to 1980 the government-
owned telephone company sponsored two consecutive development pro¬
grams for advanced semiconductor equipment and devices, the so-called
Very Large Scale Integration (VLSI) projects. In 1976 MITI began its
own four-year VLSI project. After organizing Japanese companies into
research consortia, these projects pumped large amounts of money into
semiconductor R&D. Japanese and American semiconductor companies
spent comparable shares of their sales revenues on R&D between 1976
and 1985, but from 1976 to 1979 the VLSI subsidies added resources half
again the size of the Japanese industry’s own funding of integrated circuit
R&D.
The VLSI projects helped Japanese companies develop expertise in a
new generation of semiconductor production technologies. The key to
American success in semiconductors had been Silicon Valley and its
American counterparts, dense clusters of technical expertise where chip
manufacturers could buy the latest capital goods developed for use in
production systems. Japanese manufacturers hoping to produce better
and cheaper chips than their American competition would need produc¬
tion machinery superior to that used by American producers. When
products developed by the VLSI projects came to market in the 1980s,
Japanese companies began shipping the most advanced and reliable sys-
432 CONCLUSION

terns in key areas of the chip manufacturing process. With first access to
this equipment, they would henceforth have a head start toward produc¬
ing the best chips at the lowest cost.
The next logical step was to invest aggressively in production facilities
that could use the new systems. Through the 1980s, Japanese semicon¬
ductor companies’ ratio of capital spending to revenues far outpaced that
of American companies. Together, the investment in new manufacturing
technology and the capital spending blitz that followed made Japan the
world leader in semiconductor manufacturing. U.S. companies’ share of
world sales peaked in 1975 then plunged while the Japanese share soared.
And Japanese imports, which constituted one-third of Japan’s apparent
integrated circuit consumption in 1974, fell to 12 percent in 1986.
Alarmed at the earliest of these developments, U.S. chip producers
formed an industry association in 1977 and began to complain to their
government about their inability to secure access to Japanese markets
and about Japanese government industrial policies. A few years later they
began making frequent allegations of Japanese dumping in the U.S.
market.
In response to American complaints, MITI reacted in two ways. It
adopted a decidedly lower public profile. For example, the VLSI program
was abruptly shut down in 1980 (although a private sector version contin¬
ued through 1986). This lower profile was aided by the success of the
Japanese effort: producers could stand on their own and government
funds for R&D became relatively less important as Japanese chip pro¬
duction soared on a self-sustaining trajectory. At the same time, however,
the U.S. complaints gave the ministry ample justification for continuing
to provide “guidance” to Japanese chip makers as it attempted to manage
semiconductor trade frictions with America in the 1980s.
In late 1985, as the semiconductor industry sank into a frighteningly
deep downturn, American chip producers urged their government to
respond to the perceived threat from Japan. The government reacted in
two ways. With financial support from the Department of Defense,
Sematech was formed to improve U.S. chip manufacturing. Unlike pre¬
vious Pentagon investments in semiconductor production, Sematech fo¬
cused on the technological and economic health of the infrastructure of
the entire industry, not just on specialized producers and products cater¬
ing to unique needs of the military. Meanwhile, the U.S. government
pressed suits to stop dumping and other trade violations. The result was
ultimately the Semiconductor Trade Arrangement of 1986.
CONCLUSION 433

Effects of the Semiconductor Trade Arrangement

The trade arrangement had three major features. First, Japan agreed
to a set of pricing floors administered by the U.S. Department of Com¬
merce on exports of DRAMs and EPROMs, and set up strict border
controls to ensure that exports were not being priced under the floors.
Second, Japan agreed to “monitor” an even broader variety of chips,
based on company-specific cost and price data, and “take appropriate
actions” to prevent dumping in export markets. Because of the inherent
difficulty of controlling flows of such an easily transportable commodity
as semiconductors, the appropriate actions evolved into administrative
guidance to companies on production and investment levels. Third, MITI
(in a “secret” side letter) agreed to a numerical benchmark for foreign
market share, and took actions to encourage Japanese electronics man¬
ufacturers to buy more foreign semiconductors.

FMVs and Border Controls

The price floors, called foreign market values (FMVs), set by the
Commerce Department and the border controls constructed by MITI to
make them stick hit hard. Significant memory chip price differentials
existed between the Japanese and American markets from 1987 through
1989. These differentials put U.S. computer and electronic equipment
producers at a disadvantage because they had to compete against Japa¬
nese companies purchasing identical memory chips at considerably lower
prices. From the standpoint of the aggregate welfare of the world outside
Japan’s borders, the impact of the regional price differentials was notice¬
able but not staggeringly large. In DRAMs, for example, I estimate that
the net welfare impact of the FMVs and border controls was 5 to 8 percent
of what would have been the value of rest-of-world DRAM consumption
in the absence of border controls (see chapter 5).
Moreover, there were clear limits to how tightly markets could be
controlled. Large regional price differentials existed during these years
in the large user “contract” market, which amounted to 70 to 80 percent
of consumption. In the much more freewheeling and entrepeneurial grey
or spot market, however, regional price differentials were transitory (see
chapter 5). The conclusion, perhaps, is that the more diffused, decen¬
tralized, and competitive the market, the tougher it is to swim upstream
434 CONCLUSION

against fundamental economic forces, even for the most powerful of

government bureaucracies.
In large part because of computer industry objections, the FMV sys¬
tem was dropped when a second semiconductor trade arrangement was
signed in 1991. Instead, Japanese companies were required to collect and
maintain the same data used for FMV calculations so as to create a faster
antidumping procedure. In effect, a self-policing shadow FMV system
was established (see chapter 4). MITI continued to collect data on export
prices from companies and to issue semiannual forecasts of production
and exports of monitored products. Whether this new framework would
have maintained relatively high prices without additional government
intervention during an economic downturn was never tested: from 1992
to 1995 the semiconductor industry enjoyed a sustained expansion of
historic proportions.2

Production and Investment Controls

The effect of the system of restraints on production and capital in¬

vestment that MITI created in 1987-88 is the most difficult outcome of
the STA to assess. Any analysis would require hypotheses about levels
of production and investment in the absence of restraints, and there is
little data that can be used to construct a credible alternative to observed
history. Under restraints, Japanese semiconductor investments surpassed
20 percent of sales in 1989 and 1990 (see figure 1-3). It is not unreasonable
to suspect that cutbacks in production and investments in 1987-89 were
the most economically significant element associated with implementa¬
tion of the STA and clearly an important factor in the unprecedented
increases in memory chip prices in 1988.
More indirectly, the Japanese government seems to have facilitated
some “privatized” restraint by Japanese producers in 1989 and 1990
(chapter 7), after the official constraints were removed. This facilitation
resulted from the government’s sometimes open and sometimes tacit
approval of “coordination” of Japanese production and investment, and

2. Driven by annual growth rates in PC shipments that averaged more than 23 percent
from 1993 to 1995 and rapidly rising semiconductor content in electronic equipment, PC
circuit boards alone accounted for one-third of world consumption in 1994. Lee Gomes,
“Growth in Computer Shipments Slowing,” San Jose Mercury, January 30, 1996; and Ron¬
ald A. Bohn and Mary A. Olsson, “Semiconductors: An Industry in Transition,” Red
Herring, no. 23 (September 1995).
CONCLUSION 435

its creation of an official apparatus to monitor production and investment

through supply-demand forecasts, surveys of company prices, and ques¬
tionnaires on investment plans. As in the past, an attitude of tolerance—
and possibly indirect support—from the bureaucracy seems to have been
necessary for the successful organization of coordinated action within
Japanese industry.
In the end, however, this attempt to organize the Japanese industry
may have backfired. High prices for memory chips and restraints on
production and investment allowed South Korea’s young semiconductor
industry to charge into the global chip market. From 1987 to 1994 South
Korea’s three major chip producers quintupled their output of DRAMs,
producing more than a quarter of world output by 1994. Built largely on
memory chip sales, their market share for all semiconductors went from
1 percent in 1987 to 8 percent in 1994.3 When the semiconductor boom
started in 1992 and U.S. and South Korean companies responded with
massive investments in a new generation of fabrication lines, Japanese
companies held back and invested a considerably smaller share of their
revenues in new facilities (see figure 1-3).4 This inertia may also be
explained by the collapse of Japan’s speculative economic bubble and the
damping effects on the Japanese economy that also occurred over roughly
this same period.

Increased Foreign Market Share

In many respects the least ambiguous outcome of ten years of semi¬

conductor trade arrangements was the effects on foreign sales of semi¬
conductors to Japanese customers. The record, laid out in chapters 2 and
3, makes it clear that through the mid-1970s there was a determined and
systematic, officially sanctioned effort to block, disrupt, and slow down
foreign chip sales in Japan. Yielding to strong foreign pressure, formal
trade barriers were removed by the mid-1970s and investment controls
brought into conformity with OECD standards by the early 1980s. But
from the mid-1970s through the mid-1980s, there remained some informal

3. Dataquest, “South Korea’s Semiconductor Giants Increase DRAM Share Fivefold

in 1994,” online research highlight, November 27, 1995.
4. In 1995 VLSI Research estimated that Korean firms were reinvesting 30 to 55 percent
of their semiconductor revenues in new plant and equipment, compared with 22 percent
for U.S. firms and 15 percent for Japanese firms. Julie Chao and David P. Hamilton, “Bad
Times are Just a Memory for DRAM Chip Makers,” Wall Street Journal, August 28, 1995,
p. B4.
436 CONCLUSION

actions and attitudes that hindered foreign sales. The sanction or toleration
of dubious (from an antitrust perspective) private activities also made it
more difficult for most foreign firms to break into Japanese markets.
With the implementation of the STA, the Japanese official attitude
toward semiconductor imports publicly reversed course. Instead of re¬
moving obvious obstacles only after foreign pressure had been brought
to bear, MITI became a promoter of foreign imports. And the change
extended far beyond the “import now” placards (in English) attached to
luggage carts at Tokyo’s international airport. MITI officials visited users
of chips to spread the new gospel. It supported industry overseas trade
missions, and set up organizations and infrastructure to make it easier
for foreign producers to reach Japanese chip consumers. Many large
companies set, under government pressure, “voluntary” targets for in¬
creasing their use of imported chips. As industrial giant Matsushita’s
executives noted in interviews with the New York Times, “the 1986 trade
agreement between the United States and Japan was instrumental in
prompting the company to consider American chips.”5
Other factors, of course, also explain the increase in the foreign share
of the Japanese chip market from 9 percent in 1986 to almost 30 percent
in 1991. An appreciating yen made foreign chips considerably cheaper.
The quality of American chips improved: by the late 1980s defect and
failure rates for both American and Japanese chips met the standards set
by demanding multinational customers. And (as noted in chapter 5)
Japanese companies seemed to learn how to cut deals with foreign com¬
panies that helped meet their “import share” targets without significantly
changing their production and sourcing patterns.
But some attributions of the causes of the increasing foreign share of
the Japanese market are wrong. It is not true that Japanese companies’
increasing tendancy to use chips that U.S. manufacturers tended to spe¬
cialize in was responsible for the bulk of the increase. An analysis (see
chapter 5) of two different sets of data concluded that increases in the
rate of foreign penetration in different market segments, as opposed to

5. Andrew Pollack, “The Surplus That Wouldn’t Die,” New York Times, February 27,
1996, p. D8. ‘“In 1988 we imported virtually nothing except some semiconductors,’ said
Toshihiko Murota, managing director of Matsushita Communication Industrial Company.
. . . ‘Even the semiconductors,’ he conceded, ‘were imported only because of the 1986
semiconductor trade agreement. Purchasing managers were comfortable buying from Jap¬
anese companies. So Matsushita went over their heads, requiring its division chiefs to come
up with annual import targets.’ ”
CONCLUSION 437

shifts in Japanese demand between segments, accounted for most of the

change.
And contrary to some assertions (particularly by European critics of
the STA), the increases in foreign market share were not limited to
American producers.6 In 1986, when the first STA took effect, U.S.
producers accounted for 93 percent of foreign sales of semiconductors in
Japan (of a total foreign market share of 8.5 percent). By 1994 they
accounted for 78 percent of a much larger (20.2 percent) foreign share.
U.S. producers thus captured two-thirds of the expansion, European
producers 5 percent, and Korean and Taiwanese companies 28 percent.7
It is not surprising that some Asian producers are considerably less neg¬
ative than European critics are about U.S. pressure to open the Japanese
semiconductor market.8

Where Do We Stand?

The concerns that drove U.S. policy when the STA was negotiated in
1986 confront very different realities a decade later, in 1996.

Technological Competition

It is a much more competitive American industry that faces its inter¬

national competitors today. One crude measure of this renaissance is the

6. Douglas Dunn, the president and CEO of Philips Electronics N.V. of the Nether¬
lands, was reported to have told a conference in Tokyo that although the agreement was
“aimed at boosting the share of foreign products in general, Japanese customers are pres¬
sured to buy U.S.-made products. Dunn said the share of European semiconductors in
Japan remains at a mere 1 percent [which] shows that the pact has functioned to shut out
third-country suppliers.” AP wire report, February 2, 1996. See also “SIA to Japan: Let’s
Renew Pact,” Electronic Buyers’ News (February 12, 1996), p. 70.
7. These statements are based on unpublished Dataquest estimates of foreign compa¬
nies’ sales in the Japanese market. From 1986 to 1994 U.S. companies’ share rose from
7.9 percent to 15.7 percent, European companies’ from 0.5 percent to 1.1 percent, and
Korean and Taiwanese companies’ from 0.1 percent to 3.4 percent. Dataquest assigns sales
to companies based on the brand name stamped on the chip.
8. For example, when Miin Wu, President of Macronix International, Taiwan’s largest
supplier of nonvolatile memory chips, was asked in late 1995 if the STA should be allowed
to lapse, he replied: “If every country opens up its market for people to freely compete,
there’s no need for any agreements. Right now, the Japanese believe they are buying enough
from overseas, so there’s no need for an agreement. We should look at the facts, and
whether or not Japan is really doing that. If things go the wrong way and Japan closes its
door, then I’m sure the U.S. government will come back again.” “Plain Talk On What’s
Ahead for Taiwan’s Chip Business,” Electronic Buyers’ News (January 2, 1996), p. T2.
438 CONCLUSION

share of the world market held by the U.S. semiconductor equipment

and materials industry. After years of steady decline to less than
45 percent in 1990 (slightly smaller than the share held by Japanese
producers that year), the U.S. industry rebounded and exceeded 50
percent in 1992.9 There is widespread agreement among Sematech
member companies that the organization has been instrumental in im¬
proving the U.S. industry’s manufacturing infrastructure and stimulating
productive cooperation between equipment and materials suppliers and
chip producers. In 1997 federal funding of Sematech will end, but mem¬
ber companies will continue to fund its activities at or above their current
contribution.
A less persuasive but equally interesting measure of Sematech’s
performance is the perceptions of others. In the fall of 1995, MITI re¬
quested funds for the largest government semiconductor R&D effort
since its 1976 VLSI project. Modeling the cooperative program on
Sematech, the ministry organized Japanese semiconductor makers and
equipment and material manufacturers into the Association of Super
Advanced Electronic Technologies (ASET). MITI was to begin funding
with a fiscal year 1996 subsidy of $100 million (10 billion yen) a year.
Expected to run for 5 years, the program’s goal is to develop basic
processing technologies needed for 1 gigabit DRAMs. Japanese subsi¬
diaries of three U.S. firms (IBM, Merck, and Texas Instruments) are
among the twenty-one firms taking part.10 Japanese wafer producers and
MITI have banded together in an associated effort, the Super Silicon
Crystal Research Institute (with a $70 million tab split equally between
MITI and the companies), to develop ultralarge 16-inch (400 mm) silicon
wafers for the next century.11
9. VLSI Research data, as cited in William J. Spencer and Peter Grindley, “SEMA¬
TECH after Five Years: High Technology Consortia and U.S. Competitiveness,” California
Management Review, vol. 35 (Summer 1993), pp. 13, 20. In the narrower category of wafer
fabrication equipment, Dataquest data show U.S. producers’ share steadily declining to a
low approaching 35 percent in 1991, then increasing after 1992 (although continuing to fall
short of Japanese producers’ market share). See Clark Fuhs, “Worldwide Wafer Fab Equip¬
ment: Status, Tends and Forecast,” presentation to the SEMI-Industry Strategy Sympo¬
sium, Pebble Beach, California, January 1994.
10. “Ministry to Promote Advanced Electronics R&D,” Asahi Shimbun, August 29,
1995, p. 11; “Unity of 10 Semiconductor Companies,” Asahi Shimbun, October 13, 1995,
p. 1; David Lammers, “Japanese Team Up for Semiconductor R&D,” Electronic Engi¬
neering Times, October 30, 1995, p. 1; and Yoshiko Hara, “Japanese Set 12-inch, 16-Mbit
Research Research Projects,” Electronic Engineering Times, February 19, 1996, p. 10.
11. Unpublished photographic slides furnished to the author by the Semiconductor
Equipment and Materials Institute, April 1996.
CONCLUSION 439

The research initiative was one of a number of new cooperative efforts

to “strengthen Japan’s inferior design power while further raising process
technology strength in which Japan is already strong.”12 Japanese com¬
panies had become anxious after 1993, when their share of world output
dropped below the U.S. share for the first time since 1985.13 According
to one Japanese analysis of the new initiatives, Sematech’s establishment
helped U.S. manufacturers to regain power. “Also, South Korean and
Taiwanese manufacturers, who have become active in the market and are
rapidly closing in on Japanese manufacturers, owe their rapid growth to
their governments’ assistance. ... It is obvious that the Japanese semi¬
conductor industry is preparing to launch a counterattack against foreign
manufacturers in full cooperation with the government.”14
In addition to the Super Advanced Electronics project, in 1995 MITI
began a ten-year national R&D project on femtosecond technology—
electronic devices that generate and manipulate ultrafast pulses and
the basic science, technology, and materials needed to develop them.
The Ministry of Education has also begun to try to remedy some long¬
standing problems in coupling academic science and education to indus¬
trial technology. In 1996 the ministry will set up at Tokyo University a
center to design and evaluate complex chips; about one hundred other
universities have requested similar centers.15 More than twenty univer¬
sities and ten large semiconductor companies have set up a four and one-
half year, $10 million joint program to develop an advanced micropro¬
cessor design to be used in distributed and parallel computer systems,
the architecture for such systems, and an operating system to be used
with it.16 This project is intended to be a prototype for future business-
university cooperation.
The other key elements of the counterattack included establishment
of the Semiconductor Industry Research Institute of Japan (SIRIJ) in
1994 and a cooperative Japanese industry effort to develop a new gener-

12. Taro Okabe, “Semiconductor Industry Research Institute Resolves Policy Frame¬
work,” Nikkei Microdevices (December 1994), pp. 120-21.
13. “Ten Japanese Companies Together Try to Develop Semiconductor Production
Equipment ” Nihon Keizai Shimbun, July 8, 1995, pp. 1, 9.
14. Emi Yokota, “Chip Makers Preparing to Form United Front,’” Ekonomisuto, Oc¬
tober 31, 1995, p. 42.
15. “Ministry of Education Seeks Practical Education in Semiconductors,” Nihon Kei¬
zai Shimbun, September 4, 1995, p. 17.
16. “Research On State-of-the-Art Technology with the Cooperation of Businesses and
Universities,” Nihon Keizai Shimbun, October 2, 1995, p. 17.
440 CONCLUSION

ation of semiconductor production equipment using 12-inch (300 mm)

diameter wafers. The SIRIJ was set up by the ten large Japanese chip
producers as a think tank for “reactivating Japan’s semiconductor indus¬
try.”17 One of institute’s functions is to foster closer research ties with
Japan’s universities—it mimics the Semiconductor Research Consortium
(SRC) administered by the U.S. Semiconductor Industry Association in
this respect. The Semiconductor Technology Academia Research Center
is pairing universities with companies on modestly funded research proj¬
ects.18
What will probably become one of the SIRIJ’s most controversial
efforts is a joint Japanese industry project to develop next-generation
production equipment and processes for 12-inch wafers, which was widely
estimated to require $500 million to $1 billion in funding.19 The MITI
Super Advanced Electronic Technology project seems designed to pro¬
vide funding equal to about half this amount for basic technology devel¬
opment. In early 1996 the ten large chip makers in SIRIJ announced the
formation of Semiconductor Leading Edge Technologies Inc. (SELETE)
to help fund and execute the development of 12-inch (300 mm) produc¬
tion equipment.20 Funded by its membership at $350 million over five
years, SELETE is oriented toward nearer-term equipment evaluation and
development.21
An interesting feature of the Japanese 12-inch wafer effort is that it
explicitly rejected a cooperative effort with American and other foreign
companies. Sematech had begun organizing an international effort to
accelerate 12-inch equipment development in 1994 and had invited Jap¬
anese companies to participate. After some discussion the Japanese de-

17. Okabe, “Semiconductor Industry Research Institute.” The ten companies are
NEC, Toshiba, Hitachi, Fujitsu, Mitsubishi, Matsushita, Oki, Sanyo, Sharp, and Sony.
See also “Semiconductor Makers to Jointly Develop Advanced Equipment,” August 23,
1995, in Foreign Broadcast Information Service, Pacific Rim Economic Review: Japan,
August 24, 1995.
18. Okabe, “Semiconductor Industry Research Institute”; Asahi Shimbun, October
22, 1995, p. 3; and unpublished Semi photographic slides, April 1996.
19. David Lammers, “Japanese May Form Equipment Consortium,” Electronic Engi¬
neering Times, July 17, 1995, p. 132; and “Ten Japanese Companies.”
20. “Electronic Firms Form Alliance,” AP wire story, February 13, 1996; Peter N.
Dunn, “Japanese R&D Co-ops Proliferate; Some Confusion over Who Does What,” Solid
State Technology, April 1996, pp. 58, 61; and Hara, “Japanese Set 12-inch.”
21. One Japanese executive has said that SELETE is “for tomorrow’s real business
while ASET is for the primary technology development needed in the next century.” Hara,
“Japanese Set 12-inch.”
CONCLUSION 441

cided to go it alone, although some information will be exchanged be¬

tween the two projects.22 A clue to the logic behind this choice was
contained in a report released by the Semiconductor Equipment Asso¬
ciation of Japan in October 1995 that made recommendations on R&D,
standards, and business practices, and ratified the emerging consensus
for a major cooperative effort by government, industry, and universities
to improve Japanese semiconductor competitiveness.23 After noting the
importance of further standardization and calling for creation of a per¬
manent standards organization, as well as keeping “foreign semi-
conductor-related laws and regulations under observation,” the report’s
recommendations also included “establishment of a Japanese-created
world standard” as an industry objective.24
What all this shows is that despite the growth in strategic alliances
among semiconductor manufacturers headquartered in different coun¬
tries, the often fierce competition among Japanese chip producers in their
home market, and a globalized market for semiconductor equipment and
materials, at least some within government and industry continue to
impute a national economic advantage to control over development of
the most advanced semiconductor manufacturing equipment. They see
enough of an advantage to turn down direct cooperation on R&D and
thus saving real resources through cooperative efforts with rivals if those
rivals are foreign. This view is by no means unique to Japan (Sematech’s
charter, for example, restricts membership to U.S.-based companies),
but it contrasts markedly with much that is written about an increasingly
borderless world of global alliances in high technology.25

22. See Naoyuki Mikami, “Next Generation Semiconductor Development Noted,”

Shulcan Toyo Keizai, August 26, 1995, pp. 80-82; and author’s interviews with Sematech
officials, March 1996. Korean, Taiwanese, and European firms have joined the Sematech-
sponsored international initiative.
23. The report called for the “establishment of a research and development organiza¬
tion composed of semiconductor, equipment, and materials manufacturers” and “the
establishment of a government-academia-industry R&D organization for critical tech¬
nologies.” It also recommended changes in taxes on capital investment and revisions
of some semiconductor and equipment business practices unique to Japan. “Proposal about
Semiconductor-Manufacturing Equipment,” Nikkei Sangyo Shimbun, October 3, 1995,
p. 9.
24. See “MITI to Subsidize Semiconductor Research Association,” Pacific Rim Eco¬
nomic Review, November 1, 1995, in FBIS, Science and Technology, November 2, 1995,
cd-rom version, citing a Nikkan Kogyo press report.
25. Sematech sponsorship of an international 12-inch program is clearly a move toward
greater international cooperation in technology development. And, with legislative con¬
straints associated with federal funding disappearing (along with the funding), the organi-
442 CONCLUSION

Pricing and Dumping

The availability and pricing of semiconductors was an issue for two

groups of American companies in the 1980s. Facing stiff competition
from low-priced Japanese memory chips, semiconductor producers ar¬
gued that Japan had organized a strategy to drive them from the market.
Ultimately, they argued, dependence by U.S. users on a Japanese indus¬
try riddled with anticompetitive practices was dangerous. It would ulti¬
mately free Japanese suppliers to form a cartel and extract huge monopoly
profits from U.S. electronic companies. Worse yet, dependence could
create a competitive advantage for Japanese companies that would allow
them to penetrate U.S.-dominated electronic equipment markets by de¬
laying U.S. companies’ access to advanced components or charging high
prices for them.
These sorts of warnings were self-serving when they came from Amer¬
ican chip producers seeking protection from Japanese imports. But they
were not rejected out of hand by U.S. user industries, particularly the
computer manufacturers, who might have been expected to be deeply
interested in maintaining access to low-priced components. In the 1980s,
U.S. supercomputer manufacturers had sometimes privately expressed
concern about their dependence on a small group of Japanese suppliers
for ultra-high-performance semiconductors, at a time when the super¬
computer divisions of these suppliers were aggressively marketing new
computer hardware based on superior performance against American
systems.26 (In addition, the U.S. computer industry had its own history
of market access grievances in Japan.)
Concerns over access to components were reinforced by the “DRAM
crisis” of 1988 (see chapter 4).27 With only a minimal competing U.S.
production capability for these memory chips remaining in place, U.S.
computer makers saw prices for them soar and serious shortages develop.

zation may be reconsidering affiliations with foreign-based companies. The U.S. semicon¬
ductor industry, including Sematech, and the U.S. government supported a proposal for a
joint venture between Japan’s Canon and the U.S.-owned Silicon Valley Group to develop
and produce semiconductor lithography equipment, but negotiations were not successful.
26. References to the occasional public discussion of this issue may be found in General
Accounting Office, International Trade: U.S. Business Access to Certain Foreign State-of-
the-Art Technology, GAO/NSIAD 91-278 (September 1991), p. 21.
27. A concise summary of U.S. computer industry thinking at the time may be found
in National Research Council, Computer Science and Technology Board, Keeping the U.S.
Computer Industry Competitive: Defining the Agenda (Washington: National Academy
Press, 1990), chap. 2.
CONCLUSION 443

Compelled to cut back production and hold up introductions of new

computers, American companies watched as Japanese competitors un¬
dercut their prices with access to cheaper chips, introduced new systems
using the parts they were having trouble obtaining, and cut into their
share of global markets. U.S. computer makers even went to the extraor¬
dinary length of sitting down with U.S. chip makers and organizing a
crash joint venture to manufacture DRAMs.
This project never got off the ground, perhaps because of its sheer
size and complexity. But it dramatized the tangibility of concerns over
the potential impact of a semiconductor supplier cartel on user industries.
The economic damage from a supplier cartel could be large (see the
simulation models constructed in chapter 7), and if realized, would be
an important concern for public policy.
Concerns about organized restrictions on supplies of semiconductors
are not currently an issue for computer producers. New competition in
DRAMs, stimulated by the large-scale market entry of Korean and Tai¬
wanese producers and increased investments by the remaining U.S. and
European producers, has created enough diversity in supplies to make a
successful cartel impossible to construct within Japan and difficult even
to visualize.28 For other more specialized semiconductors the concerns
have also receded, primarily because most U.S. producers of high-
performance computers have adopted hardware strategies built around
the high-volume commodity microprocessors in which U.S. companies
are undisputed world leaders.29
Interestingly, many of the proposals to lessen U.S. companies’ vulner¬
ability from dependence on uncertain supplies of critical components,
initiatives that were controversial when included in the proposal for a
U.S. Memories venture, are today a routine feature of the semiconductor
business. Long-term, five-year contracts between DRAM suppliers and

28. Although concentration in DRAM production remained high enough to prompt

occasional thoughts that a small number of companies continued to have the ability to set
prices in the short term, were they able to successfully coordinate their actions? In March
1996 Samsung, the world’s largest DRAM producer, announced amidst plunging memory
chip prices that it would be halving its production of 4 M DRAMs to fix its “price problem.”
Added to cuts announced weeks earlier by NEC and Hitachi, this action would cut monthly
supplies from 50 million chips to 31 million by the end of 1996. See Newsbytes News
Network, “Asia Technology Newsbriefs,” on-line data, March 29, 1996; and Vision Multi-
media Technologies, “Chip Talk,” on-line data, April 15, 1996.
29. See Kenneth Flamm, “Controlling the Uncontrollable: Reforming U.S. Export
Controls on Computers,” Brookings Review, vol. 14 (Winter 1996).
444 CONCLUSION

computer firms are reported in the trade press.30 Customers routinely

buy equity stakes in chip suppliers’ operations to ensure access to crucial
supplies.31 Joint ventures to share the enormous costs of developing and
fabricating new chips are also common.32
What remains a concern for both U.S. chip producers and con¬
sumers—and a potent potential source of schisms between them—is the
application of antidumping laws and price floors to semiconductor im¬
ports. From 1986 to 1989, Japanese government measures to implement
the pricing provisions of the STA had the ironic effect of helping create
the very cartel-like structure that had long been considered the main
threat driving U.S. trade policy. The U.S. semiconductor industry argued
that these measures had not been sought and strictly speaking were not
part of the agreement. But without either the strictest of border controls
or reductions in production, higher prices in world markets could not be
sustained. Given the porousness of borders, the ingenuity of entrepe-
neurs, and the ease with which large volumes of chips can be transported,
the Japanese government was probably correct in concluding that both
sets of measures were needed to comply with the STA. The U.S. govern¬
ment, formally and informally, at first encouraged Japan to attempt to

30. See Mark LaPedus, “5-Year DRAM Deals,” Electronic Buyers’ News, December
18, 1995, p. 1.
31. See for example, Don Clark, “In the Hot World of Chips, Tradition Melts,” Wall
Street Journal, October 30, 1995, p. B4; and Mark LaPedus, “HP May Join Taiwan Ven¬
ture,” Electronic Buyers’ News, January 8, 1996, p. 12. One pronounced trend in the
semiconductor world has been for so-called fabless semiconductor companies—design
houses that contract with outside foundries to manufacture their designs—to invest in the
foundries to ensure the availability of capacity to manufacture their products. Examples in
1994-95 include disk controller producers Adaptec and Opti; graphics controller companies
S3, Cirrus Logic, Trident, and Oak; specialized memory producers Alliance and Lanstar;
and programmable logic makers Actel, Altera, Lattice, and Xilinx. See Mark LaPedus,
“Alliance, S3 Sign Fab Deal—To Build Foundry in Taiwan with UMC,” Electronic Buyers’
News, July 17, 1995, p. 1; Ismini Scouras, “Lanstar Lines Up DRAM Fab,” Electronic
Buyers’ News, February 12, 1996, p. 1; “Fabless Deals,” Fabless Semiconductor Associa¬
tion, downloaded April 1996; and Dataquest, “The Fab Four,” on-line abstract, Data-
qauest, October 1995.
32. Recent announcements include joint ventures to build DRAM production facilities
in the United States by partnerships between Siemens and Motorola, Toshiba and IBM,
and Texas Instruments and Hitachi. Motorola is also considering joining an existing three-way
IBM-Siemens-Toshiba R&D effort to work on technology for 1 gigabit DRAMs. See Clark,
“In the Hot World of Chips”; Darrell Dunn and Mark Hachman, “Motorola Jumps into
DRAM Fray,” Electronic Buyers’ News, October 30, 1995, p. 3; and Loring Wirbel, “TI,
Hitachi Tip Twinstar Setup,” Electronic Engineering Times, January 23, 1995, p. 92.
CONCLUSION 445

cut production and seal its borders. When protests erupted and the gov¬
ernments were forced to abandon production controls, sticking to the
objectives of the arrangement meant driving the controls underground,
in essence by tolerating private cartel-like behavior.
The contradictions built into these policies were ultimately resolved
when buoyant demand and high chip prices between 1992 and 1995 made
price floors irrelevant. Demand was so strong, in fact, that for the second
time in the history of the industry the average cost of a bit of memory
increased slightly in 1993. Memory costs on average seemed to be drop¬
ping at half the historical rate.
This is unlikely to continue. According to the U.S. semiconductor
industry’s projections for the next fifteen years, the cost of producing a
bit of memory will decrease about 26 percent a year, about halfway
between the earlier trend and the record of the early 1990s.33 Further¬
more, long-standing historical patterns are unlikely to have abruptly
ended for the semiconductor industry. The sustained expansion of the
early 1990s could not continue forever, and a cyclical plunge into weak
demand, excess supply, and sharply dropping prices was virtually inevi¬
table. Indeed, in early 1996 the industry seemed poised for just such a
decline.
When downturns come, the rules for semiconductor pricing will again
become an issue. American producers will likely urge limits on how low
import prices can plunge and will launch antidumping cases to accomplish
this end. Chip users will protest, with considerable justification, that the
antidumping laws establish price floors that lack a sound economic basis
and are biased against foreign producers (see chapter 6). Unless (as in
1986) some arrangement is negotiated that effectively sets a global floor
on chip prices (rather than a U.S. floor), U.S. chip users will be tempted
to move their manufacturing operations to take advantage of lower prices
outside the United States. And if a worldwide undertaking to raise global
chip prices uniformly is sanctioned by government actions, the paradox
of U.S. trade policy facilitating the creation of the very cartel-like struc¬
ture that has been the principal threat driving other trade, competition,
and technology policies will unfold.

33. This is the implicit annual rate of decline calculated from Semiconductor Industry
Association, The National Technology Roadmap for Semiconductors (San Jose, Calif.:
1994), p. 11.
446 CONCLUSION

Market Access

From the vantage point of early 1996, the pressure exerted to gain
greater access to the Japanese market has paid off. The target of
20 percent market share written into the trade arrangements of 1986 and
1991 today seems of historical interest only as the foreign share of the
Japanese market approaches 30 percent.
But there was a real history of Japanese discrimination against foreign
products. When, despite considerable obstacles, foreigners established
initial relationships with Japanese customers in the late 1960s and early
1970s, the Japanese government actively worked to disrupt them. The all-
important long-term relationships were broken up. Thus, calls in recent
years for “affirmative action” to assist foreign companies in gaining easier
access to the Japanese market resonated in American ears: the barriers
in Japan appear higher than elsewhere, and the ladders thrown up to
scale them were deliberately smashed by the state.
Given the current situation, however, is it now time to declare what
seems mainly historical to have been resolved? One is tempted to say
yes. Today, all seems well. Under continuing political urging, a strong
web of relationships between foreign and Japanese companies has been
spun. It is hard to see how this network can be dissolved when, by most
accounts, it seems to benefit both sides.
Still, some reservations must be expressed. First, much of the discus¬
sion of the continuing growth in strategic alliances between Japanese and
foreign semiconductor firms seems to overstate the case. Precise and
consistent definitions of what a strategic alliance is are critical in deter¬
mining trends, and little such consistency exists. Collaborations between
a foreign supplier and a Japanese customer resulting in the incoporation
of a foreign chip into the design of a Japanese product certainly seem to
have increased greatly in the past decade.34 But because these collabo¬
rations may not extend beyond production of a very specific product, it
is not clear whether they should be counted as strategic alliances.
Some data portray a volatile process of alliance formation. One source
shows that strategic partnerships between U.S. firms and Japanese chip
makers doubled in the years just before 1987 then decreased sharply in

34. See Electronics Industry Association of Japan, Semiconductor Facts 1995 (Wash¬
ington, 1995), p. 4-1.
CONCLUSION 447

1989 and 1990.35 Other data show that after rising sharply (from 3 in 1980
to a peak of 120 in 1987) new Japanese semiconductor alliances have
fluctuated considerably from year to year. From 1985 through 1994, ac¬
cording to industry consulting firm Dataquest, Japanese companies
formed new semiconductor alliances with other companies an average of
86 times a year. In 1993 the number sank to 71, in 1994 it was 99, and in
1995 it dropped to 83.36
A second cause for concern is the potential impact of a sharp cyclical
decline in the semiconductor market. The major Japanese producers
would be faced with the choice of letting machines and men stand idle
(or with even greater difficulty, reducing their workforce) or cutting back
on purchases of semiconductors from outside vendors and using excess
internal capacity to substitute for purchased parts. When demand is high
and capacity fully utilized, as it has been for the past four years, this has
not been an issue. In slack times there will be a strong economic incentive
to switch to producing products internally. Although proprietary micro¬
processors and custom logic designs may not be easily switched to internal
production lines, there are other chips procured from foreign compa¬
nies—flash memories, static memories, DRAMs, and ASICs, for exam¬
ple—that are not going to be difficult to switch.
Thus it seems safe to predict that in a serious industry downturn, many
alliances with outside vendors—strategic and nonstrategic—that are as¬
sociated with purchases of foreign chips will be put under stress, perhaps
to the breaking point. Although the economic logic driving such a con¬
traction in foreign sales is readily apparent, foreign complaints and re¬
newed trade friction are a very predictable consequence.
Third, the current absence of severe complaints over Japanese prac¬
tices that impede foreign access to the Japanese market does not reflect
a more general consensus that access as a whole is no longer an issue.
Significant disputes exist in other parts of the Japanese economy. The
key point for the semiconductor industry is that rather than pointing to
formal trade barriers, most of the complaints by foreign businesses relate

35. See National Research Council, U. S.-Japan Strategic Alliances in the Semiconductor
Industry (Washington: National Academy Press, 1992) p. 14.
36. Sheridan Tatsuno, “Beyond the Chip Wars: The Boom in Japanese Semiconductor
Alliances,” Venture Japan, vol. 1, no. 2 (1988), pp. 24-25; Dataquest, “1995 Japanese
Semiconductor Alliances: A Slowdown,” on-line abstract, Dataquest, January 1996; and
“Japanese Semiconductor Alliances Taper Off in 1995,” on-line research highlight, March
1996. Dataquest has revised the 1987 peak downward to 117.
448 CONCLUSION

to private anticompetitive practices, government competition policy, the

continued use of informal administrative guidance by government offi¬
cials, and the selective enforcement of antitrust laws.
Strictly speaking, competitive practices and antitrust matters are not
trade issues. A playing field tilted against outsiders affects the would-be
Japanese entrant as well as the foreigner. But insiders are likely to be
Japanese, just as foreign entrants will almost certainly be outsiders. The
effect of hidden barriers to competitive entry into the Japanese market
will be to hinder foreigners and protect Japanese businesses.
Because there is no baseline for judging the adequacy of competitive
conduct or antitrust policy that has been agreed to internationally and
no responsibilities defined or enforced in the World Trade Organization,
anticompetitive private behavior, arbitrary or selective enforcement of
laws, and informal government guidance that impedes market access have
no agreed international venue for settling disputes. Foreign companies
can go to national authorities with complaints, but if anticompetitive
behavior is tolerated by custom or law, or if national laws are selectively
enforced by national authorities, or if bureaucrats issue undocumented
guidance to manufacturers, there is no framework for resolving griev¬
ances except government-to-government negotiation.
In the Japanese semiconductor industry successful collusive behavior
among firms seems to have occurred only with the explicit or implicit
support of MITI.37 Although government toleration of anticompetitive
behavior in the semiconductor industry is not currently an issue, there is
persuasive evidence that such behavior is sometimes tolerated and even
encouraged in other areas of the Japanese economy, in some cases in
apparent conflict with the letter of Japanese law (see chapter 7). Thus
the fear sometimes expressed by American semiconductor producers that
without some formal government-supported framework Japanese manu¬
facturers will be tempted to slide back into the practices of the past is
not without foundation. If the semiconductor industry were to face a
serious downturn, the Japanese government might be tempted to permit
(or even encourage) electronics producers to collude in order to keep

37. See chapter 7. In recent years, MITI has responded to foreign grievances about
anticompetitive practices by referring the complainant to the Japan Fair Trade Commission,
historically an ineffectual arm of the government. MITI recently displayed its power even
within the JFTC’s nominal sphere of antitrust enforcement by issuing an unprecedented
notice to auto dealerships requesting “information about possible violations of the Anti-
monopoly Law.” “Please Tell Us about Violation of an Antimonopoly Law,” Nikkei Sangyo
Shimbun, November 7, 1995, p. 14.
CONCLUSION 449

factories from closing in what has long been regarded as an economically

strategic industry. Is it unreasonable to worry when one can find evidence
that this happens elsewhere in the economy?

Where Do We Go from Here?

In the spring of 1996, after ten years of interaction through semicon¬

ductor trade arrangements, the American and Japanese chip industries
and their respective governments seemed again to be on a collision
course. The Americans seemed reasonably happy—although ready to
argue that market access in specific product sectors continued to be
restricted. The Japanese, however, were distinctly unhappy.
Their concerns were not hard to understand. With the STA apparently
succeeding in improving American chip firms’ access to the Japanese
market, U.S. trade negotiators had since the closing days of the Bush
administration increasingly looked at its use of numerical benchmarks as
the proven formula for making headway on difficult matters of market
access. Japanese officials, aware that they had created a Frankenstein’s
monster by agreeing to benchmarks in semiconductors and later to some
extent in autos, were adamant in their determination to erase numerical
targets or benchmarks from the trade policy map. Much of the conflict
in U.S.-Japan trade negotiations of the early and mid-1990s centered on
whether quantitative measures of market access were even to be col¬
lected, and if so, what purpose they would serve. Bilateral agreements
with the United States that contained specific targets were particularly
troublesome from Japan’s perspective. Despite the World Trade Orga¬
nization’s procedures, put into place in 1994, for settling multilateral
disputes, the United States would have a legal basis for using its unilateral
powers to retaliate against Japan—section 301—if it could be argued that
a bilateral agreement had been broken.
In the fall of 1995, with Ryutaro Hashimoto serving as minister of
trade and industry (as well as deputy prime minister), MITI publicly
declared that “the agreement between the governments has ended its
historic mission.” At a quadrilateral trade ministers’ meeting in late Oc¬
tober, then U.S. Trade Representative Mickey Kantor had called for its
extension. Hashimoto refused, declaring, “We do not need to go to the
trouble of signing a bilateral accord because the semiconductor market
450 CONCLUSION

has been rapidly turning international in recent years.”38 In 1996, with

Hashimoto now prime minister and Trade Representative Kantor contin¬
uing to stress the need to negotiate a new agreement, both sides seemed
locked into public positions that could only end in painful confrontation.
Both sides, it might be argued, were missing a unique opportunity to
strengthen the institutional structures supporting international commerce
in high-technology products. By recognizing the problems raised during
ten years of semiconductor trade arrangements, and remedying them
while reinforcing their contributions toward creating more open and com¬
petitive international markets, the two governments could strengthen the
foundations of an open international trading system in high-technology
products.
The most important criticism of the STA is of its implicit definition of
a 20 percent market share as a floor for foreign sales in the Japanese
market. Shares in an open, competitive market reflect not only the ability
of firms to gain access to the market, but also the underlying competi¬
tiveness in price, quality, service, and support of the products being sold.
Before the 1980s the Japanese government often took action—directly
and indirectly, formally and informally—on behalf of its domestic pro¬
ducers to disrupt the ability of foreign firms to sell semiconductors in
Japan. Unquestionably there were ample grounds for U.S. firms to seek
compensatory measures. But after close to a decade of successfully pres¬
suring the Japanese government to pursue aggressive affirmative action,
the United States needs to ask whether it needs to return to what should
be its long-term objectives for the international trading system. If open,
transparent, competitive markets are a basic tenet of the structures we
build to support global high-technology industry, a negotiated market
share floor contradicts that goal. In fact, the U.S. government indicated
in early 1996 that it was willing to drop this feature of the STA.39
A second criticism of the STA was that as a bilateral agreement it did
not recognize the increasing globalization of semiconductor production.
Asian producers, especially Korea and Taiwan, account for more than 10
percent of world chip output and European makers a little less than 10

38. “Collision Foreseen over Semiconductor Accord,” Nihon Keizai Shimbun, Novem¬
ber 3, 1995, p. 3.
39. In early 1996 chief U.S. trade negotiator Ira Shapiro declared to a San Francisco
business conference that the United States would no longer seek numerical targets
in semiconductors from Japan. Newsbyte News Network, “Japan Technology Newsbriefs
3/1/96” (on-line news feed), March 1, 1996.
CONCLUSION 451

percent. As China, Singapore, Thailand, and Malaysia enter the market,

their contribution must be recognized and their importance embraced in
defining the future rules of international competition. Multilateral dis¬
cussions of semiconductor trade and investment issues are inevitable.
While the United States and Japan may no longer account for 90
percent of world output as in 1985, with 80 percent (and 60 percent of
global consumption) they remain the giants of the industry.40 If they can
agree on a direction for semiconductor trade, they will have a powerful
influence in shaping new multilateral institutions.
In a similar vein, it is sometimes argued that the web of strategic
alliances and complex sourcing arrangements between international com¬
panies makes it increasingly difficult to define what is national and what
is foreign. But this was also true in 1986 and in 1991. (Qualitatively, some
of today’s relationships may be more complex, but it is not clear that
U.S.-Japan semiconductor alliances are quantitatively more significant
than they were a decade ago.) As long as the words foreign and national
continue to have operational significance for governments—as they
do, for example, in technology, procurement, investment, and antitrust
policies—in resolving disputes we will be compelled to struggle to define
in a pragmatic way what is foreign and what is national.
A third critique of the STA is that with the creation of the World
Trade Organization in 1994, a multilateral mechanism now exists to re¬
solve the sorts of trade disputes that led to creation of the STA. The
WTO may now be the proper mechanism to settle the semiconductor
trade issues addressed by the STA. But although the WTO can deal with
trade issues for which guiding principles have been agreed to in the
General Agreement on Tariffs and Trade, many of the concerns ad¬
dressed by the STA are not addressed in the GATT. Increasingly, anti¬
competitive practices (tolerated or even encouraged by government),
informal and undocumented administrative guidance, and formal regu¬
lation that limits the ability of new companies to compete against those
already in a market form the core of U.S. complaints about access to
Japanese markets. There simply are no agreed upon international stan¬
dards of conduct in these areas, and the suggestion that they be adjudi-

40. In some respects the industry has become less globalized. In 1970 U.S. and Japanese
firms produced less than 75 percent of world output and European companies the balance.
The sharp decline in the European industry is as notable as the rise of East Asia. For an
analysis of European decline, see Kenneth Flamm, “Semiconductors,” in Gary Clyde
Hufbauer, ed., Europe 1992: An American Perspective (Brookings, 1990), pp. 225-92.
452 CONCLUSION

cated by the WTO begs the question of what standard is to be used in

resolving them.
In the long run, international norms in competition and antitrust policy
will be discussed and perhaps adopted.41 But this will be a painstaking
process that is likely to take decades given the varying approaches and
attitudes among different nations. For the moment, to resolve basic prob¬
lems, there would seem to be no alternative to direct discussions among
governments.
The fact that the GATT-WTO structure does not address some im¬
portant matters, particularly competitive practices and antitrust enforce¬
ment, that can create functional barriers to the entry of U.S. firms into
foreign markets creates some serious tensions for U.S. policy. The GATT-
WTO apparatus severely limits the extent to which trade policy can be
used to apply pressure to resolve other issues. It is improbable to suppose
that, stymied by the absence of effective trade policy leverage, the United
States will accept that it can do nothing about privately administered (or
informally constructed) barriers around foreign markets. What is more
likely is the construction of new unilateral tools (for example, extrater¬
ritorial application of U.S. antitrust standards), which in the long run
will be a step away from multilateral governance of an international
trading system.
A final critique of the STA has been that the frictions it addressed are
ancient history. With foreign chips (by whatever definition) taking
30 percent of the Japanese market, market access is no longer a major
sore point. The worldwide industry is relatively healthy, and markets
continue to grow rapidly. As a press release from MITI stated, “com¬
petitive foreign-based semiconductors have been firmly incorporated as
indispensable products in the Japanese market. . . . Thus, the Japan-
U.S. Semiconductor Arrangement has fully achieved its objectives and
shall expire.”42
If the semiconductor market slips into a major decline, however, this
situation will change. Given the enormous increase in world chip pro¬
duction capacity expected in the next few years and the economic pres-

41. For a useful analysis, see F. M. Scherer, Competition Policies for an Integrated
World Economy (Brookings, 1994). As Keio University professor Jiro Tamura comments
in this book (after noting that Japanese antitrust law is modeled on U.S. law), Japan’s
“discrepancy in antimonopoly activity [with the United States and Europe] is based less on
the law itself than on the weakness of the enforcement of the law” (pp. 112-13).
42. MITI, “Comment on the Announcement of the Market Share of Foreign-based
Semiconductors,” press release, Tokyo, March 19, 1996.
CONCLUSION 453

sures to keep the new factories running, prices are likely to come down
steeply in a recession. There will almost certainly be accusations of unfair
competition, acute trade frictions, and dumping cases. What are still
largely nationally defined standards (subject to some loose guidance from
the GATT) governing fair values for imports will be used to punish
offending exporters. Within Japan at least some of the foreign semicon¬
ductors so “firmly incorporated as indispensable products” are likely to
prove dispensable as electronics producers consider using idle factories
to make products previously purchased from outside. If, along with the
STA, reliable data on foreign chip purchases disappear, bitter disputes
over what precisely is going on in the Japanese market are likely to erupt.
In the absence of some stipulated facts, disputes over what is happening
will add additional fuel to a furor over why it might be happening.
This brings up some of the potentially positive elements the STA has
contributed to the dialogue on semiconductor trade issues. Gathering
some agreed upon data on what has been a contentious issue for thirty
years serves one very useful purpose. Facts prevent some unhappy com¬
panies from complaining to their government that their experiences are
typical of a broader trend when they are not. Certainly, well understood
data have permitted Japan to argue without dispute that what was once
a problem is one no longer. Without such facts, additional disagreement
is likely whenever trade issues periodically erupt—as they surely will in
this boom and bust industry.
A second positive contribution of the STA is the establishment of a
forum to facilitate speedy resolution of disputes. In the next downturn,
both market access and dumping will certainly be issues. An agreed upon
format for quickly talking about them can avoid time-consuming, acri¬
monious, and possibly inconclusive debates over how to structure talks
and who is to sit at the table.
Third, from the U.S. perspective the STA has illuminated shadowy
corners of the Japanese electronics industry. In an environment of mis¬
trust built by decades of behind-the-scenes exclusionary maneuvering,
the regular dialogue created by the STA has built confidence in the
United States that attitudes have indeed changed. Questions can easily
be asked and answers given. The private industrial relationships that have
now been constructed may be able to sustain the atmosphere of openness
and cooperation. But given the signs of continued official tolerance of
legally dubious insider relationships in at least a few Japanese industries,
there is some justification for U.S. chip companies’ worries that if the
454 CONCLUSION

lights go out, they will again be in the dark, on the outside. A continuing,
regular channel for discussion and consultation could dispel much of the
suspicion that might otherwise poison relations.
How, then, can we take the best features of a decade’s experience with
semiconductor trade policy, eliminate the negative elements, and build a
world trading system in microelectronics that contributes to a prosperous,
competitive, and technologically innovative global industry? One ap¬
proach would be to extract bits and pieces of ideas already under discus¬
sion by government and industry and combine them in what might be
called the Global Semiconductor Conference (GSC).

The Global Semiconductor Conference

The Uruguay Round and earlier rounds of trade talks pulled nations
already committed to an open trading system and membership in the
GATT into a regular forum that gradually enabled the world trading
system to transform abstract principles into rules to deal with the real
problems of a maintaining a living, open system. The GSC would have a
similar purpose. It would create a forum that could gradually attempt to
resolve some of the difficult policy issues that are an integral feature of
international competition in semiconductors. In the process, if the solu¬
tions worked out in the rough-and-tumble laboratory of the semiconduc¬
tor industry are useful, they might be incorporated into the more general
GATT-WTO apparatus and address similar issues in other industries
(particularly high-technology sectors with many of the same industrial
characteristics).
In its simplest incarnation the GSC would consist of regularly sched¬
uled multilateral meetings whose sole objective would be to work out the
mechanisms needed to ensure a transparent and competitive interna¬
tional market for semiconductors. Just as the GATT makes adherence to
some minimum set of commitments and responsibilities a condition for
participation in decisions about further evolution of the system, mem¬
bership in the GSC would also require adherence to a minimum set of
ground rules. Those might reasonably be expected to include free trade
in semiconductors, unrestricted foreign investment in semiconductor pro¬
duction and sales, a minimum set of guarantees for intellectual property,
a commitment to collect common data needed to assess the transparency
of markets, and a willingness to remove both formal and informal barriers
CONCLUSION 455

to entry into national markets when they are identified in an agreed upon
fashion.
The work of the GSC would be to add the detail needed to make these
general rules into practical ones for competitive international semicon¬
ductor markets. In the future the agenda for the GSC might include other
difficult issues. Given that national tests for dumping and other pros¬
cribed trade practices in the semiconductor industry vary considerably,
as do remedies, and can still be manipulated to create disadvantages for
foreign competitors selling in national markets, international harmoni¬
zation of procedures is needed to define the rules of the road for com¬
petition. Current antidumping laws may be adequate for cement or pasta,
but they do not address important features of the semiconductor industry
and other high-technology industries. Procedures that take into account
sunk R&D costs, economies of scale, and learning curves in an econom¬
ically defensible way are needed.
Similarly, an R&D and technology group within the GSC could en¬
courage greater cooperation on government-funded R&D programs to
increase the productivity of global investments in basic microelectronics
research and technology. A standards group of the GSC could define a
framework to facilitate harmonization of international standards for de¬
sign, manufacturing, equipment, and materials in the semiconductor in¬
dustry, when there was broad agreement that uniform standards might
be useful.
Finally, the difficult question of harmonization of antitrust rules and
their enforcement is probably best addressed in small pieces. By working
on this question within the restricted venue of the semiconductor indus¬
try, some experience useful to broader rules would likely be accumulated.
In a sense, the semiconductor industry is already serving as such a labora¬
tory, since the ongoing discussion of dumping and of antidumping remedies
is inextricably linked to theories of predation and antitrust. A sensible way
to deal with accusations of dumping in semiconductors will necessarily be
part of a sensible approach to dealing with questions about anticompetitive
practices and the role governments play in these practices.

From Here to There

In early 1996 the U.S. government and semiconductor industry had
two key objectives as they negotiated with Japan: maintenance of consis-
456 CONCLUSION

tent and reliable data that would permit analytical discussion of market
access issues and some continuing framework that would permit resolu¬
tion of trade and access issues as they arise. Both could contribute to the
quest for an open, competitive world market in semiconductors.
Japan also seemed to have two primary demands in early f996. It
wished to replace a bilateral agreement with a multilateral process and
wanted no numerical targets or benchmarks. These demands also seemed
consistent with construction of an open, competitive trading system.
If Japan and the United States were to simply announce the formation
of a multilateral Global Semiconductor Conference to begin operation
just as the 1991 STA expired, all four objectives could be accomodated.
The GSC would initially include a commitment to collect national data
needed to evaluate market transparency and a framework for timely
consultations on trade, investment, technology, or market access issues
that are raised by GSC members. (As founding members and equal
partners in this enterprise, Japan and the United States would be agreeing
to provide data to each other and react to requests for consultation with
equal alacrity.) The GSC would be open to membership by all nations
that pledged to respect the ground rules established at its foundation.
To tackle an ambitious agenda that might include analyses of pricing
practices, market access, technology programs, and technical standards
clearly will require significant participation and support from industry.
Just as clearly, it will require governments to participate. It is unthinkable
that American semiconductor producers could sit down with foreign
semiconductor producers and discuss pricing, markets, standards, and
cooperative R&D efforts without the U.S. government present. From
the perspective of a user industry, the meeting might seem to be the
perfect cover for the creation of a supplier cartel. At a minimum some
U.S. government oversight would be absolutely necessary to guarantee
that user industry interests are respected (and antitrust suits against
producers avoided).
In addition, most of the issues on the agenda would involve govern¬
ment responsibilities, and companies have no power to cut deals over
these matters. Antitrust rules, dumping, and access to publicly funded
technology programs are areas where governments make policy, not com¬
panies. A framework with industry-government meetings on technical
matters that paralleled government-to-government negotiations over pol¬
icy would need to be constructed.
CONCLUSION 457

Thus if the Global Semiconductor Conference or something like it is

to be the way out of the rhetorical box that the United States and Japan
find themselves in, it will involve both government and industry. It will
not happen if the United States and Japan, whose companies control
80 percent of world semiconductor production, do not take the lead in
building such an institution. It will be much harder to organize it later,
as more countries become established in the industry, and ground rules
become more complicated to negotiate. In 1996, then, an opportunity to
resolve a political impasse and begin to explore solutions to some thorny
policy issues involving global high-technology industries seems to have
presented itself.

Mismanaged Trade?
If one looks only at the Semiconductor Trade Arrangements of 1986
and 1991, simple but contradictory answers to the question posed in
the title to this book spring to mind. Overall, the system of pricing
floors established under the STA probably worked to the net disadvan¬
tage of U.S. interests, and the cartel-like structures fostered by MITI
in implementing what Japan perceived to be its objectives only made
the situation more difficult. There was—and is—no way to set effective
price floors for semiconductors that ultimately would not penalize U.S.
user industries.
The market access the STA opened up has been a different story.
There was a long history of formal and informal barriers that the STA
helped tear down.
At a deeper level, two lessons emerge from trade frictions with Japan
in semiconductors. One is that contradictions between tactical compro¬
mises and strategic, long-run principles in U.S. trade policy ultimately
come home to roost. Having failed to achieve functional access to the
Japanese market in the mid-1970s after formal trade barriers were re¬
moved, the United States essentially decided to use the informal system
of MITI guidance and government collaboration with an industrial inner
circle to achieve an outcome that at least resembled what it thought real
systemic reform might have accomplished. Paradoxically, this decision
probably strengthened what had been waning MITI influence in the Jap¬
anese semiconductor industry. It also enabled Japan to position itself as
458 CONCLUSION

the champion of multilateralism in the international system and claim the

high ground against government intervention and in favor of letting com¬
petitive markets determine economic outcomes. (This latter situation
grated with particular intensity on American nerves, given Japan’s history
of government intervention in this and other high-technology industries,
and official tolerance for collusion and regulation elsewhere in the Japa¬
nese economy.) Rather than supporting a system that would guarantee
real competition in global markets for high-technology products, the
United States has come to wear the mantle of defender of the status quo.
The second lesson from this history is the need to continue working
to guarantee open, competitive markets for high-technology products.
The economics of high-technology industries ensures that the trade re¬
gime in its current form is not going to do the job. Economies of scale
will tempt governments to invent strategic policies that give their firms
advantages against foreign competitors. Collusion that benefits domestic
producers at the expense of foreign producers and consumers will remain
at least theoretically possible. Dumping laws and antidumping pro¬
cedures make it easy to further stack the deck against foreign
manufacturers.
The large share of costs and revenues accounted for by R&D invest¬
ments—the defining characteristic of a high-technology industry—vir¬
tually guarantees continued intervention by government. Even in an
economy cut off from international trade, there is a large economic lit¬
erature that argues that government support for research is desirable
where it corrects for market failures (private investors’ difficulties in
capturing the full fruits of their R&D investments for their exclusive
financial benefit) that cause private returns to investment in technology
to fall short of the full benefit to society. Both case studies and statistical
analysis have confirmed the empirical relevance of this argument.
International trade generates additional complexities. The possibil¬
ity of shifting technology-created profits from foreign producers and
consumers (or avoiding payment of returns to them) raises a second
rationale for government intervention. If national policy can create
situations in which technology-based returns can be captured overseas
for national producers, national income and the standard of living are
increased. This is the strategic trade rationale for intervention in high-
technology industries.
The upshot is that, for both reasons, governments will be deeply in¬
volved in national investment in technology and technology-intensive
CONCLUSION 459

industries. The challenge is to propose some way of neutralizing subsidies

to R&D as a tool of profit-shifting trade competition, yet preserve the
ability of governments to engage in socially beneficial public investment
in R&D. Some move toward reciprocity in R&D—permitting companies
from other countries to join some of one’s subsidized research programs
in exchange for one’s own companies being permitted to join in a com¬
mensurate portion of another country’s R&D projects—would seem a
useful step in that direction.
Over the long haul the only certainty is that an expanded multilateral
framework will be needed to guarantee the open, competitive trade in
high-technology products that serves a common international interest.
One can only hope that the international semiconductor industry, which
pioneered public discussion of problems that are now obvious, will also
lead the way in finding solutions that are not.
 Index

Academic sector, 25, 4I«7 Association of Super Advanced Electronic

Advanced Micro Devices (AMD), 89, 166, Technologies (ASET), 438, 440«21
217 Automobile industry, 2, 166-67
Advanced Telecommunications Research
Institute, 119 Baldrige, Malcolm, 150
Ad van test, 105, 144 Bardeen, John, 30
AEA. See American Electronics Basic Technology Research Promotion
Association Center, 118-19
Akashi Seisaku, 110 Bell Telephone Laboratories: development
Alps, 89 of solid-state electronics, 8, 18, 30-32,
AMD. See Advanced Micro Devices 33; technology transfer, 41, 42, 43, 44,
American Electronics Association (AEA), 49; VLSI technology, 90
216-17 Brattain, Walter, 30
American Telephone and Telegraph “Bubble money.” See Business sector
Company (AT&T): as captive or Bush (George) administration, 2n2, 449
merchant producer, 21«21; patent Busicom, 76
licensing, 44, 49«30 Business sector: “bubble money,” 277-78;
AMI, 72, 89 development programs in
Ampex, 129 semiconductors, 40-41; semiconductor
Ando Electronics, 110 consumption, 34n47, 35t; standards of
Antitrust issues, 150-51, 160, 195, 310, competitive behavior, 311; U.S. company
436; Microelectronics and Computer sales, 286t
Technology Corporation (1983), 148;
Micron, 167; National Cooperative Calculators: “calculator war,” 75-77;
Research Act (1984), 148; nature of, integrated circuit supplies, 72, 74, 75-
448; Union Carbide (1988), 123«254; 79; large-scale integration chips and, 71,
Semiconductor Trade Arrangement and, 72, 73; mass production of, 68—69;
227; U.S. Memories, 217-18; Zenith miniaturization, 67; single-chip, 73;
Radio Corporation (1974), 134«23. See solid state, 60; transistorized, 68n86;
also Monopoly issues U.S. chips in, 431
Apple Computer, 218 Canon: manufacturing processes, 105, 110—
Application-specific integrated circuits 11, 144; purchase of U.S. integrated
(ASICs). See Integrated circuits circuits, 72
ASET. See Association of Super Advanced Cartels. See Economic issues; Ministry of
Electronic Technologies International Trade and Industry
Asia Seisaku, 110 CDL. See Computer Development
ASICs (Application-specific integrated Laboratories
circuits). See Integrated circuits Center for Integrated Systems, 147-48

461
462 INDEX

China, 451 forward pricing and, 309-10, 314, 316-17,

Clinton (Bill) administration, 2-3 318, 320, 321, 322; by Japan, 141-46,
Compaq, 218 149-50, 160, 162, 166, 167, 168-70, 172,
Computer Development Laboratories 305, 383; by Korea, 224-25; litigation,
(CDL), 96, 112-13 162, 168, 169-70, 172, 173, 176, 182, 189;
Computer industry: 1960s, 430; 1980s, 193, prevention of, 187«92, 223-224, 227,
201, 203, 213, 222; 1990s, 283, 434«2; 228, 292, 312-13, 428; sanctions, 187,
commercial, 48, 430; component 188, 211, 225, 230, 266; screening for,
purchases, 24; computer prices, 7; 356, 358; strategic and predatory,
Dendenkosha Information-Processing 140n41, 141-46, 150-51, 306-11, 322-28;
System (DIPS-1), 93; IBM, 59, 60; third-country, 292; by the United States,
integrated circuits in, 34-35, 61; 73-74, 136, 430-31. See also Purchasing
Japanese industry, 39-40, 52, 66, 80-82, and sales; Semiconductor Trade
125, 148, 430; Japanese regulation of, Arrangement
53; Next Generation Basic Computer Dynamic random access memory
Technology program, 113«226; (DRAM): 1988 shortage and crisis, 192-
semiconductors in, 33-34, 48, 49, 63, 201, 203, 207-11, 235, 294-95, 300n84,
203, 220, 442; single-chip 387-96, 442-43; costs, 402; design and
microcomputers, 73; U.S. industry, 442- development, 9-11, 82-83, 93-94, 98-
43; Very High Speed Computer Systems 99, 313—14; dumping cases, 173, 176,
(VHSCS), 61-62, 92, 93 182, 189; forecasting of production, 196-
Computer Technology Research 201, 222-23; product life cycle, 313,
Association, 59n61 354-55; quality, 145-46, 163, 232n9,
Consumer interests, 272-78, 292, 293, 436; role in semiconductor industry, 8-9,
309«9 99, 383; Semiconductor Trade
Contract sales. See Purchasing and sales Arrangement and, 160-226; supply, 256-
Corrigan, Wilfred, 211, 217, 244 66, 443-45; trade issues, 138, 139, 143-
44, 148-49; U.S.firms, 167nl7. See also
Dainippon Printing, 105 Dumping; Indexes; Manufacturing
Dataquest, 21, 236, 404 processes; Production processes;
Dataquest First Monday Report, 244-47 Purchasing and sales
Defense Advanced Research Project Dynamic random access memory
Agency (DARPA), 147-48 (DRAM), prices: below cost and quality
Defense issues: integrated circuits, 34-35; dumping, 141-44, 145, 148-49, 168-69,
national security issues, 130-32, 150, 219-21, 355-58; competition and, 398-
153, 372-82, 425-26, 429; purchases, 405; controls and guidance, 171-72,
24; research and development funding, 175-76, 188, 189-90, 193, 198; cyclical
29-38, 117/z234, 373; transistors, 8, 30- effects, 162-63, 240-41; European
34 markets, 188, 189-90; fair market
Defense Science Board, 373 values, 177n55, 183; Japanese, 184, 187,
Dell Computer, 218 265, 387-88; “pi” rule, 9-11, 240;
Demand. See Purchasing and sales regional differentials, 242-54, 433; two-
Denshinho (Electronics Industry tier, 138-41, 150-52; in the United
Promotion Law). See Law on Temporary States, 203, 210-12, 219-21;
Measures for Promoting the Electronics Semiconductor Trade Arrangement and,
Industry 237-38, 269, 297-301, 445
Digital Equipment Corporation, 217 Dynamic random access memory
Distribution. See Purchasing and sales (DRAM), production and production
DRAM. See Dynamic random access controls; 1980s, 387, 390-91; 1990s, 219-
memory 20; costs, 401-05; forecasting, 196-201;
Dumping: competitive behavior and, 317; investment and, 202-04; learning
constructed cost dumping tests, 312-13, economies, 336-37; memory chip
358; cost structures and, 311-34; supply, 254-66, 356; prices and, 323;
definition, 305, 306«4, 308n5; effects, Semiconductor Trade Arrangement,
211; economic analysis of, 305-71, 428, 180-86, 192-95, 201-08, 269-73; U.S.
445; European charges of, 173, 189, 191; Memories and, 396-97
INDEX 463

Dynamic random access memory (DRAM) value of the dollar, 247nil; value of the
products: 64K: 64K DRAM wars (1981- yen, 203, 393, 436; venture capital, 24-
83), 148-58; controls and guidance, 160; 25; welfare issues, 381, 382, 412-14, 416,
development, 93-94, 99; dumping, 166, 418, 433, 458-49. See also Learning
167, 168-69, 172, 227; pricing and economies; Trade
production, 160, 162, 168-69, 227, 235/, EECA. See European Electronic
247, 249/, 391; production, 196, 262, Component Manufacturers Association
391; Semiconductor Trade Arrangement EEPROMs. See Electronically erasable
and, 297-301; yield, 261/ programmable read-only memory chips
Dynamic random access memory (DRAM) El A. See Electronics Industries
products: 256K: controls and guidance, Association
175, 177, 187, 192, 198, 229-30; Electrical Communications Laboratory
development, 99, 116; dumping, 160, (ECL), 40
168, 169-70, 173, 176; prices, 169-70, Electronically erasable programmable
187, 203, 223, 244, 245-46, 247/, 250-51, read-only memory chips (EEPROMs),
269, 387; production, 183-84, 192, 196, 176
198, 209/7152, 262, 269, 352, 353; Electronic Arrays, 144
Semiconductor Trade Arrangement and, Electronic components: boom of 1988,
160, 177, 269, 297-301; shortages, 203- 201; in Europe, 25; Japanese regulation,
04 53; ranking of industry, 13, 375-76;
Dynamic random access memory (DRAM) sales, 18n20; in the United States, 24,
products: 1M: controls and guidance, 129, 425
182, 192, 193, 198, 215-16; costs, 401- Electronics, consumer: development, 18;
OS; demand, 341-43, 344; development, integrated chips in, 139; in Japan, 47,
96, 99, 116, 192; dumping, 356-58; 49, 52-53, 54, 60, 68, 125, 133, 431; in
learning curves and economies, 337-38, the United States, 53, 129. See also
400-401; prices, 212-16, 221-23, 245, Calculators; Television and radio
246/, 248/, 249/128, 251, 269, 270/, 354- Electronics Industries Association (EIA:
56, 386-87, 398-99; production, 182, U.S.), 128, 129
192, 193, 198, 200, 212-13, 215, 220, Electronics Industries Association of
262, 269, 386-87, 396-97, 398-99; Japan, 87, 135/725 , 390
shortages, 200, 203-04, 208«148, 213; Electronics Industry Council, 53
Strategic Trade Arrangement and, 298- Electronics Industry Promotion Law
301 (Denshinho). See Law on Temporary
Dynamic random access memory (DRAM) Measures for Promoting the Electronics
products: miscellaneous products; 1 Industry
gigabit, 438; IK, 99/7190; 4K, 99/7190; Electrotechnical Laboratory (ETL). See
16K, 98-99, 139-40, 141, 142, 144, Ministry of International Trade and
232/78 , 352/768 ; 4M, 98/7189, 217, 222- Industry
23; 16M, 98/7189; 256M, 314 Emerson, 129
Emitter coupled logic (ECL), 176
ECL. See Electrical Communications Erasable programmable read only memory
Laboratory; Emitter coupled logic (EPROM): 128K, 254, 255/; 256K, 166,
Economic issues: cartels, 382, 384-24, 254, 255/; 512K, 269, 271/; 1M, 269,
405-17, 418, 443; contract pricing, 294- 271/; dumping cases, 160, 162, 170, 173,
301; cost of manufacture, 402; dumping, 174, 176, 189; prices, 166, 174, 210-11,
305-71; economies of scale, 7, 313; 239, 241, 253-55, 265-66, 269, 271,
effects on demand, 342; employment, 278n51; production and production
23/, 24, 47-48; externalities, 380, controls, 182, 198, 201, 239/, 262, 263/,
383/718; forward pricing, 309-10, 314, 264/, 265-66; role in semiconductor
316-17, 318, 320, 321, 322, 327-28; industry, 383; supply and demand, 231,
government policies and support, 3, 256-66, 278-79, 283, 293
380-81; quality control, 145//59; high- Esaki, Leo, 45/724
technology industries, 458; recessions, ETL. See Electrotechnical Laboratory
162, 167, 205; semiconductor and chip Europe: acquisition of solid-state
industry, 7, 17, 38; subsidies, 383-424; electronics, 18; computer industry, 25-
464 INDEX

26, 427; dumping negotiations, 225-26; monitoring system, 195-96;

Korean products, 234nl3, 291; “logic Semiconductor Trade Arrangement, 185,
wars,” 73nl05; products, 288; share of 188, 190, 230; third-country trade, 160,
global chip market, 2, 23-27, 437, 450- 190; World Trade Organization and, 451
51; Semiconductor Trade Arrangement, General Electric (GE): joint ventures, 57;
173, 188-91; tariffs, 26; third-country sale of Great Western Silicon, 122-23;
trade, 160, 243. See also Semiconductor Toshiba and, 129; trade adjustment
industry, Europe assistance, 137; transister production, 33
European Electronic Component Genkyoku, 53
Manufacturers Association (EECA), Geophysics Corporation of America
189 (GCA), 105«202, 111
Everex Systems, 218 Germanium. See Transistors
Export-Import Trade Law of 1952, 133-34 Gidwani, Ramesh, 184
Exports. See Trade Global Semiconductor Conference, 454-55
Goldstar Electron, 225, 291n72
Fairchild Semiconductor, 35-36, 56, 58, Gray market. See Purchasing and sales
76, 88 Great Western Silicon, 122-23
FECL. See Foreign Exchange Control Law
FIL. See Foreign Investment Law
Fisher Ideal price index. See Indexes Hashimoto, Ryutaro, 449-50
FMVs (Foreign market values). See Trade, Hayakawa Electric, 67, 68. See also Sharp
U.S. Hemlock, 122
FONTAC project, 59-60 Hewlett-Packard, 217, 218
Forecasting. See Ministry of International HHI (Hirschman-Herfindahl index). See
Trade and Industry; Semiconductor Indexes
Trade Arragement High Technology Working Group
Foreign Exchange Control Law (FECL; (HTWG), 154-58
1949), 55-56 Hirschman-Herfindahl index (HHI). See
Foreign Investment Law (FIL; 1950), 55 Indexes
Foreign market values (FMVs). See Trade, Hi-Silicon, 122
U.S. Hitachi: Computer Development
Forward pricing. See Economic issues Laboratories, 96; DRAM production
Fransman, Martin, 220 and trade, 116n232, 139, 142«45, 143,
Fuji Electronic Chemicals, 110 166-67, 169, 183, 192, 198, 200, 215,
Fujitsu: 256K EPROM, 166; Computer 291n72, 391, 397, 443n28; dumping,
Development Laboratories, 96; DRAM 170«28, 172; licensing agreements, 42,
production and trade, 116«232, 139, 43-44; market shares, 45; RCA and,
142n45, 143, 192, 277, 291n72, 391; 129; research and development, 41, 62-
dumping, 170n28; joint ventures, 86, 63, 82-83; subsidies, 66, 82-83; supplier
95/7180; subsidies, 66; supplier to NTT, to NTT, 91; transistors, 45, 48; Ultra
91; Ultra Advanced Computer Advanced Computer Development
Development Technology Association, Technology Association, 81, 85, 95/U80;
81, 85; U.S.facilities, 144, 167; Very U.S. facilities, 144, 167; Very High
High Speed Computer Systems, 61-62, Speed Computer Systems, 61-62, 93;
93; very large scale integration ICs, 96, very large scale integration ICs, 96, 107,
107, 110, 116 110, 116
Hitachi America, 166
GATT. See General Agreement on Tariffs Hitachi Software Engineering, 110
and Trade Hoffman Electronics, 57-58
GCA. See Geophysics Corporation of Home Electric Appliances Market
America Stabilization Council, 135n25
GE. See General Electric Honeywell, 129
General Agreement on Tariffs and Trade Hong Kong, 131
(GATT): government subsidies in, 3n4; HTWG. See High Technology Working
Japanese integrated circuit imports, 79- Group
80; Japanese tariffs, 147; MITI Hyundai Electronics, 225, 291«72
INDEX 465
IBM (International Business Machines International Trade Commission (ITC),
Corporation): antitrust case against, 90; 139-40
as captive producer, 20-21; computers ITT Corporation, 41«7, 57, 89, 129
and products of, 59, 60, 61-62, 80,
81nl28, 90, 280; entrance into Japanese
Jacobs, Bernard, 61
market, 58, 282; integrated circuits and,
Japan, 18, 49-52, 56, 272-78. See also
59; local foreign production, 20;
Semiconductor industry, Japan;
memory products, 11, 261/; OS/2
Semicondutor Trade Arrangement;
operating system, 220; subsidiaries, 438; Trade, Japan
U.S. Memories, 217, 218
Japan Economic Journal, 205, 207-08,
Ibuka, Masaru, 43
245n25
ICs. See Integrated circuits
Japan Electrical Manufacturers
IMF. See International Monetary Fund
Association, 119
Imports. See Trade
Japanese Camera Manufacturers
Indexes: aggregate price per bit of memory Association, 390
sold, 10; Fisher Ideal price index, 10, 11,
Japanese Development Bank (JDB):
210-11, 237, 252-53; Hirschman- investments in technological
Herfindahl index (HHI), 254-59, 347,
development, 52n33, 66; low-interest
352,353-54
loans, 53-54, 60, 66, 121, 122
Integrated circuits (ICs): analog circuits, Japan Fair Trade Commission, 390, 448n37
66; application-specific, 173, 176, 192— Japan Radio, 57
93, 243-44, 283-84, 289, 292, 293; Japan Silicon, 122
bipolar, 69n89; competition in, 73-77, JEOL, 105
137; complementary metal oxide silicon Josokuho, 53
(CMOS), 85; development and funding, Junkins, Jerry, 184, 208
8, 34-38, 49, 56, 59-60, 96-98, 390;
diffusion self-aligned (DSA), 62n75;
electronic switching, 92; gate arrays, Kane, Sanford, 217
192-93, 289; hybrid, 60n62; large-scale Kantor, Mickey, 449-50
integration (LSI), 62-63, 67, 71-73, 85; Kawasaki Steel, 122, 123
Mask ROMs, 289; medium-scale Kidenho. See Law on Temporary Measures
integration (MSI), 62«74; metal oxide for the Promotion of Specified
semiconductor (MOS), 62«75, 71; metal Electronics Industries and Specified
oxide semiconductor large-scale Machinery Industries
integration (MOS LSI), 71-72, 75, 82, Kijoho. See Law on Temporary Measures
85, 86-87; negative metal oxide silicon for the Promotion of Specified
Manufacturing Industries
(NMOS), 85, 86; packaging, 105;
Kobe Kogyo (Japan), 42-45, 47
pricing, 138-39, 141-43; quality, 77-78,
Kokusai Electric, 105
92; random access memory (RAM),
Komatsu, 110
62«75; silicon gate metal oxide
Komatsu Denshi, 123
semiconductor large-scale integration,
Komatsu Hoffman, 57-58
85; very large scale integration (VLSI),
Komiya, Ryutaro, 53
62«74, 90, 93-113, 431, 432. See also
Korea: dumping, 224-26, 234«13; exports,
Dynamic random access memory
247«27, 290, 300-301; investment,
Intel Corporation: DRAM production and
435n4; pricing, 162, 234nl3, 300-301;
sales, 98-99, 142n46, 288, 352«68;
products, 288; share of global chip
EPROM production and sales, 166, 170,
market, 2, 23, 161, 219-22, 416, 426,
278-79; import contracts, 76-77;
435, 437, 438, 443, 450. See also
Japanese sales office, 89; production
Samsung
facilities, 167; U.S. Memories, 217
Kyocera, 105
Intellectual property. See Patents and
Kyodo Electronics Laboratories, 61, 86,
licensing
89«160, 95«180
International Monetary Fund (IMF), 130
International Rectifier, 57
International Standard Electric (England), Law on Temporary Measures for
Mril Promoting the Electronics Industry
466 INDEX

(Electronics Industry Promotion Law or and silicon production, 106, 119-24,

Denshinho; 1957): passage and 377-78. See also Production processes
extension of, 52, 53«34, 63; Matsushita, 45, 66, 129
“rationalization” cartels under, 54n39; Matsushita Electronics, 57
research and development subsidies, 66; MCC. See Microelectronics and Computer
support classes, 53-54; trade protection, Technology Corporation
54-55 Merck, 438
Law on Temporary Measures for the Michio Watanabe, 172
Promotion of Specified Electronics Microelectronics and Computer
Industries and Specified Machinery Technology Corporation (MCC), 148
Industries (Kidenho; 1971), 52n33, Micron Technology, 162, 166, 167, 168-69,
53«34, 63 183, 217, 222, 224, 226, 305
Law on Temporary Measures for the Military, U.S. See Defense issues
Promotion of Specified Manufacturing Ministry of International Trade and
Industries (Kijoho; 1956), 52«33 Industry (MITI): administrative
Learning curves and economies: costs and, guidance and control, 149, 155/797, 156—
319, 323«31, 328, 383-84, 414; definition 57, 175-201, 203, 229-30, 391, 395, 418,
and role, 7, 17, 261; model of, 329-33, 431-32, 434-35, 444, 448, 457-58;
336-41, 400-01, 407-09; pricing and, authority and power, 43-44, 55-56, 63,
307, 309, 318n23; semiconductor 70, 119, 124-25, 134, 137; development
production, 295, 312, 313, 314-15, of technology, 40, 70, 83, 106, 119, 430,
318/722 432, 438-41; dumping issues, 162;
Legislation, Japanese, 52-56 Electronics Industry Division, 53;
Letters, side: bilateral agreement on Electrotechnical Laboratory (ETL), 40,
semiconductors of 1983, 155, 157; 48, 60, 61, 82-83, 96, 109; monitoring
Semiconductor Trade Arrangement, 170, and forecasting, 178-79, 192, 195-201,
173-74, 176, 279-92, 433; Texas 202, 205nl39, 209/7152, 210, 216, 222-
Instruments and, 70 23, 224, 228, 229, 260; research and
Liberal Democratic party (Japan). production cartels, 54 , 59 , 74, 80-81,
SeePolitical issues 82/7133, 83-86, 94-95, 111, 132, 134-36,
Licensing. See Patents and licensing 137, 150-51, 171-72; Semiconductor
LSI Logic, 122, 123, 217. See also Nihon Trade Arrangement, 436; Sunshine
LSI Logic Project ^silicon), 119-24; technology
contracts, grants, licenses, and
subsidies, 43, 61-68, 76, 77, 79/7123, 80-
Machinery and Electronics Industry Law 81, 83-86, 95-98, 113-19, 137-38, 438;
(1971), 52-53 Texas Instruments and, 58-59, 69-70.
Machinery and Information Processing See also Semiconductor industry, Japan;
Industries Law (1978), 52-53 Trade, Japan
Machinery industry, 63-64 Mitsubishi: calculator integrated circuits,
Malaysia, 451 69; Computer Development
Manufacturing processes: advances, 8, 39, Laboratories, 96; DRAM production
103-06, 144, 377-78; capital-intensive and sales, 142//45, 215, 391, 397;
nature, 15-16; costs, 16, 113, 232n9, dumping, 172; Intel and, 278-79; MITI
301-02; dynamic random access subsidies, 67-68, 81; Texas Instruments
memory, 8-9, 98-99; electron beam and, 69; Ultra Advanced Computer
production equipment, 62n75, 102; Technology Research Association, 81,
Japanese, 102-06, 125, 383-84, 432; 86, 94; very large scale integration ICs,
metal oxide semiconductor technology, 96, 107, 110; Westinghouse and, 129
71; learning economies, 17, 428, 429; Mitsubishi Electric, 277
local foreign production, 20; Mitsubishi Metals, 123
photolithographic processes, 30/233, Mitsubishi TRW, 57-58
62/?75, 102, 107; “steppers,” 102, 110—11; Models, statistical: Baldwin-Krugman
techniques, 30«33, 377—78; timeframe, (B-K) model of semiconductor industry,
16-17; transistors, 128; very large scale 318-20, 330//39, 332-33, 337/247, 352;
integration technology, 100-13; wafer Bertrand game, 321/727; capacity and
INDEX 467

production choice, 359-71, 407-11, 423- NGK Spark Plug, 105

24; cartels, 406-17; competition, 398- Nihon Business Automation, 110
405, 418-24; Cournot competition, 399, Nihon Kaizai Shimbun (Nikkei), 205, 214-
404; Cournot-Nash equilibrium, 307, 15, 245-47, 250, 404
309, 321, 334, 407, 413-14, 418; “food Nihon LSI Logic, 283-84
chain” theory, 375-77; game theory, Nikon, 62n75, 105, 110-11
380, 407, 418; learning economies, 400- Nippon Columbia, 66
01, 408-09; marginal costs, 401-05; Nippon Electric Corporation (NEC):
semiconductor industry, 312-58, 380, DRAM production and trade, 116«232,
383-424; strategic behavior, 407-11, 139, 142«45, 143, 166-67, 169, 180, 183,
419-24 198, 200, 215, 277, 290-91, 391, 397,
Momota, Tsuneo, 109 443«28; dumping, 170n28, 172;
Monitoring. See Ministry of International Honeywell and, 129; integrated circuits,
Trade and Industry; Semiconductor 60, 78nl21, 92; joint ventures, 56-57,
Trade Arrangement 81-82; market shares, 20-21, 45; New
Monopoly issues: barriers to industries, Computer Series Technology Research
377; dumping and, 140«41; Japanese, Association, 81-82, 85; research and
135, 170n30, 211; Ministry of development, 41, 62-63, 82-83;
International Trade and Industry, 54; subsidies to, 66, 81-83; supplier to NTT,
monopoly power, 211, 319, 321-22, 379- 91; Texas Instruments and, 58, 68, 69;
80, 381, 383wl8, 388; monopoly profits transfer of technology, 56-57, 68;
and rents, 379, 388, 414, 416. See also transistors, 45, 48; U.S. producers and
Antitrust issues manufacturers, 144; Very High Speed
Monsanto Corporation, 392«27 Computer Systems, 61-62, 92, 93; very
Morita, Akio, 43, 70 large scale integration ICs, 96, 107, 110,
Mostek, 167 116
Motorola; Hitachi and, 149«73; Japanese Nippon International Rectifier, 57
sales of 64K DRAM, 149; joint Nippon Kokkan, 122
ventures, 89; research and development Nippon Motorola, 89
funding, 33, 36; Toshiba and, 129, Nippon Peripherals, 81, 95nl80, 111
352rt68, 399n46; U.S. Memories, Nippon Steel, 122
217 Nippon Telegraph and Telephone
Multinational companies, 18, 374 Corporation (NTT): Basic Technology
Multinational companies, U.S.: electronic Research Promotion Center, 118-19;
component sales, 18; foreign affiliates, certification processes, 155-56;
18-19, 28«31; RCA, 42-43; research computers, 93; development of
and development, 18-19; semiconductor semiconductor technology, 40, 45, 90-
assembly, 136-37 94, 95, 106, 115-16; DEX 2 switching
system, 92; patents, 116, 117t;
Nakasone, Yasuhiro, 187 privatization, 117-18; procurement
National Advisory Committee on practices, 137, 116«232; subsidies to
Semiconductors (NACS), 372 semiconductor industry, 113, 117-18,
National Cooperative Research Act of 125-26,137-38
1984, 148 Nittetsu Electronics, 122
National Science Foundation, 12 NKK, 123
National security. See Defense issues NMB Semiconductor, 17nl8, 416
National Semiconductor, 88, 167, 217 Nomura Research Institute (NRI),
NBK, 123 289
NCR, 218 North American Rockwell, 72. See also
NEC. See Nippon Electric Corporation Rockwell International
NEC Anelva, 110 Noyce, Robert N., 56, 138, 140, 142«46,
NEC Toshiba Information Systems 151
(NTIS), 81-82, 96, 111, 112 NTIS. See NEC Toshiba Information
New Energy Development Organization Systems
(NEDO), 120, 123-24 NTT. See Nippon Telegraph and Telephone
New Japan Radio, 57 Corporation
468 INDEX

OCDM. See Office of Civil and Defense 325«32, 334-36, 337n48, 346-47, 356,
Mobilization 359-65, 407-09; pricing and
OECD. See Organization for Economic profitability, 314-17, 318, 320-21, 322-
Cooperation and Development 28, 335, 345, 347, 352, 371; processes
Office of Civil and Defense Mobilization and timing, 13, 213, 231-32, 261-62,
(OCDM), 128 316« 19, 342-43, 409//56; production
Oki Electric, 144, 172, 183, 277, 391 alliances, 446-47; product life cycle,
computer manufacturing, 96; supplier to 313-58; quantities produced, 238/, 239/,
NTT, 91; Ultra Advanced Computer 260-66, 272-73; research and
Technology Research Association, 81, development in, 13, 147, 31-32, 91-92,
86, 94; Very High Speed Computer 200; silicon and memory chips, 62/x75,
Systems, 61-62 200; technological advances and
Okinawa, 88, 131«10 transfers, 9, 91-92, 94, 331-32, 377,
Okura Group, 135 431-32; testing, 232, 236, 295, 333-34,
Organization for Economic Cooperation 345, 402; transistors, 45; very large scale
and Development (OECD), 80 integration equipment manufacturers,
Osaka Titanium Company (OTC), 110, 96, 98/r 189, 431; yields, 198, 200, 258-
120-21, 122, 123 61, 295, 314, 320, 322, 323, 329, 330-31,
OTC. See Osaka Titanium Company 333-34, 339, 341, 343«60, 402. See also
Manufacturing processes
Pacific Semiconductor, 31, 57-58 Purchasing and sales: authorized
Parkinson, Joseph, 181 distributors, 233, 234; contract, spot,
Patents and licensing: licensing and gray market sales, 141-42, 143, 161,
agreements, 129; “patent war” in Japan, 180, 184, 209-10, 221, 230, 233-36, 243,
68-70; royalties, 71; semiconductor 244, 248-52, 269, 292-301, 443-44;
manufacture, 378; technology transfers, demand, 236, 240, 288, 289, 291-92,
42-44, 56-57, 92, 107, 109, 378; LVSI 293, 312, 334, 341-44, 365-71, 387, 388,
projects, 107, 109, 116, 117t. See also 391, 426; domestic versus foreign
Semiconductor Trade Arrangement products, 394—95; original equipment
Pattern Information Processing System manufacturers (OEMs), 232-33, 235-
(PIPS), 82 36, 295-301; pricing, 233, 234, 236-54;
Philco, 33 “pulls,” 233; quantity commitments,
Philips NV, 57, 129 235-36; regional and country
“Pi" rule. See Dynamic random access differentials, 242-54, 259-60, 272, 293;
memory semiconductor chips, 2, 8; supplies,
Political issues: Japanese semiconductor 247n27, 254, 256-66, 272, 295; term of
and computer industry, 27, 79-80, 84- contracts, 294-301; tying arrangements,
85, 94, 167/717, 283, 284; Japanese trade 243-44; volume prices, 244/722;
policies, 130, 144, 166, 172«34, 373n3; warranting, 233. See also Dumping;
Semiconductor Trade Arrangement, Dynamic random access memory
230-31, 283, 284, 356; U.S. (DRAM) prices
semiconductor and computer industry, Purdue University, 30«32
27-30, 283, 284
Polypropylene, 394 R&D. See Research and development
Prices. See Dynamic random access Radar, 29-30
memory; Erasable programmable read Radio. See Television and radio
only memories; Semiconductor Trade Rapoport, Carla, 176
Arrangement; Semiconductor industry; Rashomon (film), 79
Trade Raytheon, 33, 57
Production processes: capacity constraints, RCA: development of solid-state
313, 318, 320, 321, 322, 323, 339; control electronics, 18, 31, 33, 36; technology
of, 195-201, 203, 213, 434-35; cost, 9, transfer, 42-43; trade policies, 129
305, 308-09, 310, 312, 315, 319, 320-21, RCA Victor, 129
322-28, 335, 344-49, 352, 402-05, 427- Reagan (Ronald) administration, 153, 187,
28, 445; labor in, 13; packaging, 296- 188
301; plant capacity, 315, 317, 323/731, Regency Corporation, 45
INDEX 469

Remington Rand, 129 158, 185-87, 223, 232, 254«33, 283, 316,
Research and development (R&D): cost 373-74, 380, 398-405, 441; cyclical
accounting, 311-12; economies of scale, nature, 16-17, 151-52, 157-58, 240-41,
14-17; General Agreement on Tariffs 312, 356, 382-83, 428, 429, 447;
and Trade and, 3n4; government- economic issues and, 427-29, 447-48,
industry collaborations, 59, 80-85, 428- 452-53; foreign sales, 15; High
29, 458; investment and costs, 7, 11-13, Technology Working Group (HTWG),
14/, 25, 313-14, 315, 428-29, 458; 153-58; history, 8, 18, 49, 429;
semiconductor industry, 312 investment in, 7, 376-77, 384-85; market
Research and development, Japan: 3.5 shares, 20-21, 449; memory chips, 8;
generation program, 80-81, 83, 93, 111; models, 307-71, 372-424; production, 7,
Dendenkosha Information-Processing 313; as a strategic industry, 372-82;
System (DIPS), 93; electron-beam (E- technology and industrial structure, 7-
beam) writing systems, 110, 111; 17, 377; subsidies, 413-14, 428-29;
femtosecond technology, 439; funding, supply and demand, 231-36; end users,
92-93, 102, 103/, 111—19; joint research 2. See also Dumping; Dynamic random
labs, 113; as percent of sales, 12-13; access memory; Integrated circuits;
Pattern Information Processing System Manufacturing processes; Research and
project, 82-83; role of government, 45, Development; Semiconductor Trade
112«225; subsidies, funding, and Arrangement; Silicon chips; Trade;
incentives, 54, 60, 61-65, 66-68, 81, 96- Transistors
98, 111-13, 116-17, 125-26, 152, 441; Semiconductor Industry Association
transfer of technology, 102, 103, 109-10, (SIA): formation, 138, 432;
116, 119; VLSI programs, 94-113, 116, investigations and, 139, 140-41;
125, 431 responses to competition, 147, 164, 216-
Research and development, U.S.: funding, 17, 432; section 301 complaint, 158, 165.
27-30, 117//234, 147-48, 429; joint See also Trade, U.S.
ventures, 147-48; tax credits for, 147; Semiconductor industry, Europe. See
technology transfers, 18, 31-32, 41-44, Europe
57. See also Semiconductor Semiconductor industry, Japan: 1950s, 40-
Manufacturing Technology consortium 49, 128; 1960s, 39, 49-72, 124, 125,
Ricoh, 72 382-83, 426, 430; 1970s, 39-40, 72-116,
Rigaku Electronics, 110 124-25, 137-40, 141, 143, 144, 145, 390,
Rogers, T. J., 217-188 435; 1980s, 116-24, 125, 146, 147-58,
162-226, 288, 382-83, 385, 387-98,
Sadao Inoue, 186 426-27, 435, 444; 1990s, 283, 286-87,
Samsung: DRAM production and sales, 288-92, 438-54; competition and, 124-
168, 212, 221, 226, 443n28; dumping 26, 153, 128-58, 383, 431; cooperation
and, 225, 226; growth, 222; Intel and, and collusion within the industry, 6,
352«68; NEC and, 290-91; U.S. 107-13, 142, 163-65, 168, 186-87, 195,
distributors, 300-01 204-08, 214-15, 219-20, 230, 279, 383,
Sanyo, 129 385-86, 387-98, 417, 448-49; “Den-
SBFC. See Small Business Finance Den” suppliers, 91, 92; government
Corporation policies and strategies, 5-6, 27, 39-40,
SEH. See Shin Etsu Handotai 48, 52, 80-81, 113-19, 124, 125, 160-61,
Selete. See Semiconductor Leading Edge 385-86, 418, 430, 448; growth, 39-52,
Technologies Inc. 99, 100, 113, 120-21; investment in, 15-
Sematech. See Semiconductor 16, 39, 57, 65, 80-81, 87, 122-23, 124,
Manufacturing Technology consortium 128, 138, 163, 166, 193-95, 202-08, 215,
Semiconductor industry: 1950s, 128-32; 434-35; joint ventures, 56-57, 60-61,
1960s, 132-36, 382-83, 425; 1970s, 73, 68-70, 122; liberalization of, 73-90, 93,
77, 136-42; 1980s, 142-58, 157-61, 162- 94, 119, 124, 125, 130, 137, 390; market
63, 192-201, 203, 207, 212, 222, 234-35, share, 20/, 21-24, 77-79, 148-49, 280,
234-54, 272, 292, 382-83, 425-26, 432; 451; production in U.S., 22, 144;
1990s, 223, 289«68, 292, 427, 434, 435, products, 23, 45«24, 71, 99, 392; quality
445; competition in, 124, 147, 148, 153, issues, 145; regulations, 53, 125;
470 INDEX

technology transfer, 41-44, 124; 188-91, 224; foreign market values, 177,
transistors, 49-52. See also Dumping; 180, 191, 211, 213-14, 223-24, 228, 306,
Ministry of International Trade and 307-08, 433-34, 450-51; general
Industry; Research and development, description, 1; Japanese views of, 175,
Japan; Semiconductor Trade 436; market shares, 170, 171, 172, 173—
Arrangement; Trade, Japan 74, 175, 188, 206, 229, 279-92, 293,
Semiconductor Industry Research Institute 435-37, 446-49, 450, 452, 457;
of Japan (SIRIJ), 439-41 monitoring and forecasting, 175-201,
Semiconductor industry, U.S.: 1960s, 132, 203, 223-24, 228, 229-30, 243, 252,
136-37, 382-83, 425, 426-27; 1970s, 272, 281-82, 283-84, 292, 306, 356, 388,
136-37, 141-46, 430; 1980s, 146, 147-58, 434, 444-45, 453, 457-58; patents, 228;
162-226, 382-83, 425-26, 442; 1990s, prices, 171-72, 174, 175-80, 189-90,
283, 372, 437-54, 455-59; captive and 241nl9, 242-43, 297-304, 396;
merchant producers, 20-21, 24; profitability, 204-08, 302, 307-08; public
competition, 6, 15, 88-90, 111, 128-58, policy and, 1-6, 38; regional welfare,
373-74, 383, 405-17, 429, 437, 442, 443, 272-78, 293; renewal and renegotiation,
451-52; fabless companies, 23, 444n31; 223-26, 280-81, 306, 434, 449-50; side
foreign sales, 15; foreign subsidiaries letters, 170, 173-74, 176, 178, 229, 279-
and contractors, 136-37; government 92, 433; third-country markets, 160, 174,
policies and strategies, 27-38, 153, 163, 176, 178-79, 180, 183, 185, 188, 190,
373-74, 382-424, 429-30; investment in, 224, 229, 243, 292. See also Dumping
15-16, 144«51; joint ventures, 430, 444; Sharp: calculators, 68rc86, 72, 75, 431;
market share, 20f, 21-24, 279-92, 437- DRAM production and sales, 142«45;
38, 446-49, 451; products, 23, 71, 72- joint ventures, 86, 95nl80; North
73, 278n51, 375«5, 389; quality issues, American Rockwell and, 72; Texas
145-46; research and development, 12- Instruments and, 69
13; sales and trade issues, 2, 18n20, 73- Shiba Electric, 66
74, 132, 145, 216; as strategic industry, Shin Etsu, 110
372-80, 405, 429-30; subsidies for, 380, Shin Etsu Handotai (SEH), 120, 123
405-17, 429; transistors, 49-52; Very Shockley, William, 30
High Speed Integrated Circuit program, SI A. See Semiconductor Industry
37-38. See also Semiconductor Trade Assoication
Arrangement Side letters. See Letters, side
Semiconductor Leading Edge Technologies Siemens, 222
Inc. (Selete), 440 Signetics, 36, 89
Semiconductor Manufacturing Technology Silicon and memory chips: “bi” rule, 240;
consortium (Sematech), 38, 278«52, costs, 9, 16, 50-51, 240, 337n47, 343,
379, 426, 432, 438-39, 440-41 345, 393n29, 402; crossover point, 223;
Semiconductor Research Consortium die shrinks, 260n39, 331; economies of
(SRC), 440 scale, 14-17; flash memory chips, 279;
Semiconductor Research Cooperative, 147 integrated circuits, 8; lifespan, 11;
Semiconductors. See Dynamic random microprocessor performance, 8«7; prices
access memory; Integrated circuits; of, 7, 317; production, 62«75, 200, 241,
Transistors 260-61, 377-78, 402; product lives, 7;
Semiconductor Technology Academia purchase and sales, 392; reduced
Research Center, 440 instruction set computer chips, 36;
Semiconductor Trade Arrangement (STA; research and development, 30, 49-52,
1986, 1991): background, 127, 158-76, 102, 105-06, 260n39, 438-39, 440-41;
227-29, 306, 426, 432; consequences of, role in high-technology industries, 7;
26, 160-61, 185-87, 210-11, 227-304, wafer starts, 315. See also Dynamic
386, 388, 395-98, 405, 433-37, 453-54, random access memory; Erasable
457; cooperation and collusion within programmable read only memory;
the industry, 163-65, 168, 186-87, 195, Manufacturing processes; Production
204-08, 214-15, 219-20, 230, 279, 283, processes
388, 395-405, 453-54; criticisms of, Siltec, 123
450-54, 457; European markets, 173, Singapore, 451
INDEX 471

SIRIJ. See Semiconductor Industry funding, 32n40, 33; role in

Research Institute of Japan semiconductor industry, 56; Sharp and,
Small Business Finance Corporation 69/; Sony and, 58, 68, 69-70;
(SBFC), 52«33 subsidiaries, 438; U.S. Memories, 217
Sony Corporation, 43; market shares, 45; Texas Instruments Japan, 77, 180, 183-84,
Texas Instruments and, 58, 68, 69-70; 185«84
transistor products, 45, 48 Textiles, 132
Specified Manufacturing Industries Law. Thailand, 451
See Law on Temporary Measures for the Third-country exports. See Trade
Promotion of Specified Manufacturing TI. See Texas Instruments
Industries Tilton, Mark, 389-90, 394
Spence, Michael L., 312-13, 400 Toko, 89
Spot sales. See Purchasing and sales Tokyo Oka Kogyo, 110
Sprague Electric, 137 Tokyo Telecommunications Engineering,
SRAMs. See Static random access memory 43. See also Sony Corporation
chips Tokuyama Soda, 120—21, 122
SRC. See Semiconductor Research Toppan Printing, 105
Consortium Toray, 110
STA. See Semiconductor Trade Toshiba: DRAM production and sales,
Arrangement 142//45, 143, 166-67, 192, 214, 215, 277,
Standard Radio, 129 352n68, 397, 399; dumping, 170n28;
Stanford University, 147-48 joint ventures, 57; licensing agreements,
Static random access memory chips 42, 43-44; market shares, 20-21, 45;
(SRAMs), 176 research and development, 41, 82-83,
Steel industry, 394-95 105; silicon wavers, 392-93; subsidies,
Strategic industries, 372-424 66, 81-83; transistors, 45; New
Strauss, Robert, 138//32 Computer Series Technology Research
Sun Microsystems, 218 Association, 81-82, 85; U.S. facilities,
Super Silicon Crystal Research Institute, 167; Very High Speed Computer
438 Systems, 61-62; very large scale
Supply. See Purchasing and sales integration ICs, 96, 107, 110, 116
Sylvania, 33, 129 Toshiba Engineering, 110
Toyo Communication Equipment, 66
Taiwan, 2, 23, 427, 437, 439, 443, 450 Trade: competition, 2, 3-5, 7, 17-27, 51-
Takagi, Noboru, 109 52, 73-77, 83-84, 86-87, 147; European,
Takeda Riken, 144 26; exports in electronics, 49-52;
Tanahashi, 170 frictions and negotiations, 3, 4, 7, 128-
Tanaka, Kakuei, 94 58, 383, 385-86, 391, 449, 451-53;
Tandem Computers, 218 intellectual property protection, 156;
Tandy, 218 Semiconductor Trade Arrangement and,
Tarui, Yasuo, 102, 109 2, 4, 446-49; strategic objectives and
Taxation incentives, 63, 65-66 policies, 5, 21n21, 140, 155, 380-82,
TDK Electronics, 88 426-27, 458-59; tariffs, 26«30, 155, 156;
Technology transfers. See Research and third-country issues, 131, 160, 172;
development World Trade Organization and, 451. See
Tektronix, 218 also Dumping; General Agreement on
Television and radio: cartels, 390; controls Tariffs and Trade; Semiconductor Trade
in the 1960s, 132-36; development, 45, Arrangement; individual countries and
47; Japanese, 49, 66-67, 68, 131; products
licensing issues, 42-43 Trade and Exchange Liberalization
Telex, 90 Guidelines, 130
Texas Instruments (TI): Canon and, 72; Trade, Japan: 1960s, 73, 131—36; 1970s, 73,
entrance into Japanese market, 57, 58- 75-77, 83, 87, 88/, 136-46; 1980s, 120-
59, 60, 68-70, 77, 136; Japanese pricing, 22, 123, 143, 148-58, 159-224, 389, 432;
142-43; Minuteman II guided missile, 1990s, 162, 455-59; controls, 132, 149—
35«50, 36; research and development 50, 166; Council of Nine, 164-65, 391;
472 INDEX

customs issues, 164h3; domestic, 135; United States: cooperation with Japan,
DRAM wars, 148-58; five-company 130, 429-59; semiconductor industry
rule, 133-34; foreign sales, 2, 15-16, output, 1; Soviet Union and, 130; trade
75-76, 133, 262; imports and exports, policies, 4. See also Military, U.S.;
47, 49-52, 54-59, 60, 61, 62/, 70-88, Multinational companies, U.S.;
120-22, 123, 130, 160, 195, 209, 243, Research and development, U.S.;
262, 264, 389, 432; integrated circuits, Semiconductor industry, U.S.;
60, 73, 74-88; price- and quantity-fixing Semiconductor Trade Arrangement;
agreements, 133-46, 149, 150-51, 162, Trade, U.S.
163-64; sales agents, 142«45, 143; Users’ Committee of Foreign
strategic and predadory sales policies, Semiconductors (UCOM), 289
152-53, 165; tariffs and quotas, 87, 131, Ushio Denki, 105
133, 147, 149-50, 152, 156; third-country U.S. Memories, 214, 216-19, 244, 396, 416,
exports, 131, 179; two-tier pricing, 138, 443
140-41, 142, 143, 152; U.S. U.S. Semiconductor, 123
interpretation of policies, 152-53, 216,
385-87, 392; voluntary impact expansion Very High Speed Computer Systems
(VIE) measures, 157. See also Dumping; (VHSCS). See Computers
Semiconductor Trade Arrangement Very High Speed Integrated Circuit
Trade, U.S.: 1950s, 128; 1960s, 130-36; (VHSIC). See Integrated circuits;
1970s, 136-46, 430-31; 1980s, 148-58, Semiconductor industry, U.S.
159-224; 1990s, 455-59; contract, spot, Very Large Scale Integration (VLSI). See
gray market sales, 141; DRAM wars, Integrated circuits
148-58; effects of, 127; foreign market VHSCS (Very High Speed Computer
values (FMV), 177, 180, 191, 211, 213-14, Systems). See Computers
223-24, 266-72, 301-304, 387, 399; VHSIC (Very High Speed Integrated
government involvement in, 149-50, 153, Circuit). See Integrated circuits;
163; imports and exports, 128-29, 130— Semiconductor industry, U.S.
32, 137, 141, 144, 156, 157, 163-64, 266- VLSI (Very Large Scale Integration). See
72; national security and, 130, 132, 150; Integrated circuits
section 301 issues, 158, 160, 164-66, 167, VLSI Technology Research Association,
171, 172, 176, 227, 391, 449; tariffs, 156. 100, 102, 107, 111-13
See also Dumping; Semiconductor Trade von Hippel, Eric, 378
Arrangement
Transistors: consumer versus industrial, Wafers. See Silicon and memory chips
50-51; development, 8, 30-33, 48, 49, Western Electric: research and
52; differences, 128; germanium, 49, 92, development, 31, 33, 736; technology
425; Japanese, 40-42, 45-48, 128; transfers, 41n7, 44
obsolesence, 11; technology transfers, Westinghouse, 129
44; U.S., 48, 128. See also Silicon and World Semiconductor Trade Statistics
memory chips; Semiconductor industry (WSTS), 280
TRW Incorporated, 57-58 World Trade Organization (WTO), 449,
451-52

UCOM. See Users’ Committee of Foreign Yeutter, Clayton K., 172

Semiconductors Yukio Honda, 181, 205
Ulvac, 105
Union Carbide, 123«254 Zenith Radio Corporation, 134n23
Unisys, 218 Zilog, 89
- .
 MISMANAGED TRADE? (■

The semiconductor industry is at the forefront of current tensions over interna¬

tional trade and investment in high-technology industries. In this book Kenneth
Flamm traces the rivalry between U.S. and Japanese semiconductor producers
from its origins in the 1950s to the experiment with “managed trade” embodied
in the U.S./Japan Semiconductor Trade Arrangements of 1986 and 1991 and the
current debate over renegotiating elements of that agreement. Flamm provides a
definitive analysis of this policy experiment and its consequences for U.S. semi¬
conductor producers and consumers, and discusses patterns of competition
within the semiconductor industry.

Using a wealth of new data, the author argues that significant changes to the
trade regime for high-technology industries are needed to escape from the
present impasse. Fie lays out the alternatives, from laissez-faire to managed
trade, and argues strongly for a new set of international ground rules to regu¬
late acceptable behavior by government and high-technology industries.

Flamm’s detailed analysis of competition within the semiconductor industry will

be illuminating for those interested in the organization of high-technology indus¬
tries as well as those concerned with trade and technology policy, international
competition, and Japanese industrial policies.

Kenneth Flamm is an economist and senior fellow in the Foreign Policy

Studies program at the Brookings Institution. Fie has served as Principal
Deputy Assistant Secretary of Defense and Special Assistant to the Deputy
Secretary of Defense.

Brookings Institution Press

Washington, D.C.
Cover design by Stephen Kraft