AIR COOLING

TECHNOLOGY
for ELECTRONIC
EQUIPMENT
About the cover
The inset photograph shows the temperature fields around four
one-inch-square simulated electronic components flush mounted
on a conducting board cooled by air flowing
from left to right. Isotherms are made visible
by using thermochromic liquid crystal
thermography. (Photo courtesy of Professor
Alfonso Ortego of the Heat Transfer
Laboratory at the University of Arizona.)
AIR COOLING
TECHNOLOGY
for ELECTRONIC
EQUIPMENT
Edited by
Sung Jin Kim
SangWoo Lee

CRC PRESS
Boca Raton London New York Washington, D .C.
 Library of Congress Cataloging-in-Publication Data

Air cooling technology for electronic equipment I Sung Jin Kim, Sang Woo Lee, editors
p. em.
Includes bibliographical references and index.
ISBN 0-8493-9447-3 (alk. paper)
1. Electronic apparatus and appliances-Cooling. 2. Electronic packaging-Cooling.
3. Heat-Convection. 4. Air flow I. Kim, Sung Jin. II. Lee, Sang Woo.
TK7870.25.A43 1996
621 .381 '046-dc20 94-45903
CIP

This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the authors and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.

Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microfilming, and recording, or by any information storage or
retrieval system, without prior permission in writing from the publisher.
All rights reserved. Authorization to photocopy items for internal or personal use, or the personal or
internal use of specific clients, may be granted by CRC Press LLC, provided that $.50 per page
photocopied is paid directly to Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923
USA. The fee code for users of the Transactional Reporting Service is ISBN 0-8493-9447-
3/02/$0.00+$.50. The fee is subject to change without notice. For organizations that have been granted
a photocopy license by the CCC, a separate system of payment has been arranged.

The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specific permission mustbe obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation, without intent to infringe.

Visit the CRC Press Web site at www.crcpress.com

© 1996 by CRC Press

No claim to original U.S. Government works

International Standard Book Number 0-8493-9447-3
Library of Congress Card Number 94-45903
Printed in the United States of America 4 5 6 7 8 90
Printed on acid-free paper
 FOREWORD
Cooling technology has been a vital prerequisite for the rapid, if not explo-
sive, growth of the electronic equipment industry. This has been especially true
during the last 15 years with the advent of integrated circuit chips and their ap-
plications in computers and related electronic products. Today, cooling tech-
nology is well recognized as a gating factor for the development of future
electronic equipment. There are many reasons why cooling technology always
will be pivotal in the design of electronic equipment, but the following two are
most fundamental:

• Thermodynamically, all electronic devices are undergoing an irreversible

process with the net result being the generation of heat which must be
removed in order to maintain continuous operation.

• Thermophysically, the reliability and performance of all electronic de-

vices are temperature dependent with various degrees of sensitivity, but
generally the lower the temperature, the better.

With these in mind, the search for new and improved cooling technologies
has been unabated for the last three decades as evidenced by both industrial de-
velopment efforts and university research activities. IBM has been involved ac-
tively with sponsored electronic cooling research at many universities across
the country since 1967 with equal emphasis on both air cooling and liquid cool-
ing. Almost without exception, all research findings sponsored by IBM have
been published, contributing substantially to the advancement of cooling tech-
nology for the industry.
Recent developments since 1992 have resulted in adapting CMOS technol-
ogy for pervasive applications in almost all commercial electronics, including
computers. This development, coupled with the increasingly competitive na-
ture of the industry has made air cooling much more enticing. Therefore, a book
on "Air Cooling Technology for Electronic Equipment" is both natural and log-
ical, and I endorse this timely idea wholeheartedly. Dr. Sung Kim and Dr. Sang
Lee, co-editors of this book, are my colleagues at IBM and all chapter authors
are my professional friends of many years. Ihave every reason to believe that
their joint effort will be a successful one.

Richard C. Chu
IBM Fellow
 PREFACE
Improvements in microelectronics and large scale integration technologies
result in power densities as high as 10 W/cm 2 for semiconductor devices. This
continuous increase in power densities has placed an increased emphasis on the
thermal control of electronic equipment. Convection air cooling is still the most
common method for removing heat from the heat generating elements. This is
because air is readily available and the air moving devices are relatively inex-
pensive. Even though there are a lot of research activities on "exotic" and "ex-
pensive" cooling technologies, more and more emphasis is placed on extending
the limits of air cooling capabilities because of the competitive nature of the
computer industry.
The following questions arise with regard to the thermal design of electronic
systems. What is the optimal spacing between the printed circuit boards? What
is a good estimate of the heat transfer coefficient and the associated pressure
drop for forced convection over package arrays? How are heat transfer and fluid
flow characteristics in the entrance region different from those in the fully de-
veloped region? What is the effect of substrate conduction on convection cool-
ing? How can heat transfer be enhanced to push the limit of air cooling? What
is the eventual upper limit of air cooling? These questions are frequently asked
by thermal engineers and packaging practitioners.
This book addresses all these questions in detail. Six chapters are designated
to answer the above-mentioned questions systematically. They are based on re-
search projects funded primarily by IBM. In addition, we have tried to include
the most recent information on air cooling technologies performed outside of
IBM. Hence, this book will serve as a handy technical source of information
for thermal and packaging engineers who would like to get the most out of air
cooling. The book in its present state reflects the latest development in cooling
techniques and thermal design guides with air as a cooling medium.
It goes without saying that this book would not be possible without the ded-
ication of the chapter authors. We would like to extend our thanks to Bob Stem
and Norm Stanton of CRC Press for their enthusiastic support. Our greatest
thanks go to our wives, Yeon S. Kim and Ann H. Lee, for their continuous
prayers and full support during the last two years. In addition, we would par-
ticularly like to thank Mohinder Grewal and Ron Russell of IBM Corporation
for providing a working environment that makes a fruitful result such as this
possible. Special thanks to Mr. Richard Chu, IBM Fellow, for the insightful
foreword.

SungJ. Kim
SangW. Lee
 CONTRIBUTORS
Yutaka Asako SungJin Kim
Professor of Mechanical Advisory Engineer
Engineering Thermal Engineering Center
Tokyo Metropolitan University IBM Storage Systems Division
Tokyo, Japan Tucson,AZ

KavehAzar Sang Woo Lee

Member of Technical Staff Adjunct Professor
AT&T Bell Laboratories of Mechanical Engineering
North Andover, MA Duke University
Durham,NC
Adrian Bejan
J.A. Jones Professor Majid Molki
of Mechanical Engineering Professor of Mechanical
Duke University Engineering
Durham,NC Esfahan University of Technology
Esfahan, Iran
Mohammad Faghri
Professor of Mechanical Alfonso Ortega
Engineering Associate Professor
University of Rhode Island Department of Aerospace
Kingston, RI & Mechanical Engineering
University of Arizona
Suresh Garimella Tucson,AZ
Cray Research Associate Professor
of Mechanical Engineering Richard Wirtz
University of Professor of Mechanical Engineering
Wisconsin-Milwaukee University of Nevada
Milwaukee, WI Reno, NV
 TABLE OF CONTENTS

Chapter 1
Geometric Optimization of Cooling Techniques 000 00000000000 000 0000ol
Adrian Bejan

Chapter 2
Entrance Design Correlations for Circuit Boards
in Forced-Air Cooling 000 00 000 0000
o o o o o o000 000 0000 0000 00000.47
o o o

M. Faghri, M. Molki, andY. Asako

Chapter 3
Forced Air Cooling of Low-Profile Package Arrays 0 00000 000 00 00 00 00081
R.A. Wirtz

Chapter4
Conjugate Heat Transfer in Forced Air Cooling
of Electronic Components 00000000 000000 0 0 00 000000000 0000 0 000103
o o

Alfonso Ortega

Chapter 5
Enhanced Air Cooling of Electronic Equipment 00 000 000000000 00 0000173
Suresh V. Garimella

Chapter 6
Limits of Air Cooling-A Methodical Approach 000 000000000 000 00 000203
KavehAzar

Index 0000000000000000000000000 0000 000 0 0000 000000000000 000 00245

 Chapter 1

GEOMETRIC OPTIMIZATION
OF COOLING TECHNIQUES

Adrian Bejan

CONTENTS
Introduction ............ . . ......... . ... . . . ... . . .. .. . ... . .... 2

Stack of Vertical Plates Cooledby Natural Convection .. . .. . .. . ... . . . . 2

The Small-D Asymptote ................ ... .. . .. . ... . . . . ... .4
The Large-D Asymptote . . ...... . .. . . .. . . . . . ...... ..... .....5
The Optimal Plate-to-Plate Spacing .. . .. . . ... . .... .. .. .. ...... 5.

Bundle of Horizontal Cylinders (Pin Fins) Cooled

by Natural Convection .. . . . . . ..... .. ..... . . . . ... . .... ... ......6

Stack of Parallel Plates Cooled by Laminar Forced Convection . ........ 8

Stack of Parallel Plates with Flush-Mounted

and Protruding Heat Surfaces .. . . .. ... ..... . .. . .. . . . .. .. ..... . . 12

Stack of Parallel Plates Cooled by Turbulent Forced Convection . . .. . . . 15

Stack of Parallel Plates Immersed in a Free Stream ..... . ... . . . . .. .. 17

Bundle of Cylinders Cooled by Forced Convection ... ......... ... . .25

Heat Generating Plate Cooled Inside a Parallel Plate Channel ...... . .. 27

Board with Large Thermal Conductance
in the Transversal Direction . . .. ..... .. .. . . ... .. . .. . ... . . . . .. 27
Board with Finite Thermal Conductance
in the Transversal Direction .. . .. ......... . .. ... .......... . . .30

Stacks of Plates Shielded by Porous Screens .. .. . . . . . .. .. . .. . . .. . .32

Forced Convection .. ...... .. .. . ..... . . .. .. . . . . .. . .. . . .... .33
Natural Convection . . .. . . .. . . . ... ........... . . . ... . . .... . .34

Plate Fins with Variable Thickness and Height .. . . .. . . .... . .. . ... . .35

Heat Sinks with Pin Fins and Plate Fins . . . . . .. . .......... . .... . .. 37
Pin Fin Arrays . ... . .. . . . . . .. . ... . .. . . . . . ...... .. . .. .. .... 38
Plate Fin Arrays .. . .... . .. .. .. . .. . ... . .... ... . ... .... . ... .40
0-8493-944 7-3/96/\()_()()+$.50
([) I~Q6 by CRC Press, Inc 1
2 Air Cooling Technology

Conclusion ............................................... .44

Acknowledgments ......................................... .45

References ............................................... .45

INTRODUCTION
In the design of packages of electronic components there are strong incen-
tives to mount as much circuity as possible in a given space. This can be achieved
by judiciously selecting the geometry of the package, i.e., the way in which the
components are arranged relative to the coolant and to each other in the fixed
space. An important constraint is that the highest temperature (the "hot spot")
that is registered at a certain point in the package must not exceed a specified
ceiling value. If the temperature rises above the allowable limit, the error-free
operation of the electronic circuit is threatened. Since each component in the
package generates heat, this design objective translates into maximizing the
total rate of heat transfer from the finite space occupied by the package to the
coolant that flows through the package.
In the electronics industry there is a great diversity of components, pack-
ages, and cooling techniques [Moffat and Ortega, 1988; Peterson and Ortega,
1990; Ishizuka, 1992, 1993]. Because of this diversity each optimal cooling
arrangement that emerges out of the design process tends to be specific to a sin-
gle application and lacks general applicability. The challenge is to identify in
this great diversity of configurations those design optimization rules that can
be applied to one or more classes of package configurations. In earlier reviews
[Bejan and Lee, 1994; Bejan, 1995a] we identified several of the fundamental
ways in which the geometric features of entire classes of electronic packages
can be optimized. In this chapter, I systematically review the progress made in
the newly emerging field of Geometric Optimization of Cooling Techniques.

STACK OF VERTICAL PLATES COOLED

BY NATURAL CONVECTION
A sufficiently large number of parallel electronic circuit boards must be
cooled by natural convection in the space of height H, thickness L, and width
W shown in Figure 1. The width is perpendicular to the plane of the figure, and
the board thickness is negligible. Cold fluid (e.g., air) of temperature T=enters
through the bottom of the package, rises through the board-to-board channels,
and exits through the upper opening. The lateral walls of area H x W that con-
fine the package are insulated. The ceiling value of the board temperature, Tmax
is set by electronic operational constraints.
The objective of the design is to maximize the total heat transfer rate re-
moved by the coolant from the H x L x W space. The boards are assumed
equidistant. The only variable is the number of boards
Geometric Optimization of Cooling Techniques 3

~---------L----------iJool

circuit board

t t t t t t t t t
FIGURE 1. Vertical boards cooled by natural convection (Bejan, 1984, 1995a).

(1)

or the board-to-board spacing D. To determine the optimal D that maximizes Q

can be a laborious task, however, we will show that it is possible to solve the
problem almost on the back of an envelope. The intersection of asymptotes
method was outlined first by Bejan ( 1984], and begins with the assumptions that

1. The flow is laminar.

2. The board surfaces are sufficiently smooth to justify the use of heat trans-
fer results for natural convection over vertical smooth walls.
3. The maximum temperature T max is, in an order of magnitude sense, rep-
resentative of the temperature at every point on the board surface.

The method consists of two steps. In the first step, we identify the two ex-
tremes in which the cooling process may function, the smaii-D limit and the
large-D limit. In the second step, the two extreme r~gimes are intersected for
the purpose of locating the D value that maximizes Q.
4 Air Cooling Technology

THE SMALL-D ASYMPTOTE

Consider first the limit of vanishingly small board-to-board spacing,
D ~ 0. In this limit we can use with confidence Q1 = ID1Cp (T max - T ~) for
the heat transfer rate extracted by the coolant from one of the channels of
spacing D. Note that in each channel T~ is the inlet temperature, Tmax is the
outlet temperature, and m1 is the mass flow rate through a single channel
[Bejan, 1984]

U = g~(Tmax- L)D 2 (2)

12v

(3)

The total rate of heat transfer removed from the package is Q =n Q1, or

(4)

In conclusion, in the D ~ 0 limit the total heat transfer rate decreases as D 2 •

This trend is indicated by the small-D asymptote plotted in Figure 2.

natural convection

;"-.. the small-D limit: Q- o2

Q

' \
i
' \
. - l
the large D-limit: Q - D

the actual Q(D)

l
unknown

0
0 D
opt
D
FIGURE 2. The maximization of the total heat transfer rate removed by natural convection from
the vertical stack of Figure I.
Geometric Optimization of Cooling Techniques 5

THE LARGE-D ASYMPTOTE

Consider next the opposite limit in which D is large enough so that it ex-
ceeds the thickness of the thermal boundary layer that forms on each vertical
surface, namely [Bejan, 1995a]

(5)

if the Prandtl number is of order 1 or greater. In this limit the boundary layers are
distinct (i.e., thin compared with D), and the center region of the board-to-board
spacing is occupied by fluid of temperature T ~· The number of distinct bound-
ary layers is 2n = 2L/D, because there are two boundary layers for each D spac-
ing. The heat transfer rate through one boundary layer is h HW (Tmax - T=) for
which his furnished by the correlation for laminar flow only [Bejan, 1993]

hkH =0.517 Ra~ 4 (6)

where RaH = g~ (Tmax- T=) W/(av). The total rate of heat transfer extracted
from the entire package is 2n times larger than h HW (Tmax - T =)

(7)

Equation 7 shows that in the large-D limit the total heat transfer rate decreases
as D- 1 as the board-to-board spacing increases. This second asymptote has also
been plotted in Figure 2.

THE OPTIMAL PLATE-TO-PLATE SPACING

What we have determined so far are the two asymptotes of the actual (un-
known) curve ofQ versus D. Figure 2 shows that the asymptotes intersect above
what woul~ be the peak of the actualQ (D) curve. It is not necess~ to know
the actual Q (D) relation. The optimal spacing Dopt for maximum Q can bees-
timated (approximately) as the D value where Equations 4 and 7 intersect
[Bejan, 1984]:

(8)

This estimate reproduces within 20% the optimal spacing deduced based on
more exact and lengthier methods, such as the maximization of the Q(D) rela-
tion [Bar-Cohen and Rohsenow, 1984] and the finite-difference simulations of
the complete flow and temperature fields in the package [Anand et al., 1992].
An order of magnitude estimate for the maximum heat transfer rate can be
obtained by substituting Dopt in Equations 7 or 4
 Air Cooling Technology

(9)

The unequal sign is a reminder that the peak of the actual Q(D) curve falls under
the intersection of the two asymptotes (Figure 2); however, the right-hand side
of Equation 9 represents the appropriate scale of the maximum heat transfer
rate. This result can be expressed also as the maximum volumetric rate of heat
generation in the H x L x W space

Qmax <
HLW~
0 .45 Hz
k (Tmax-~
T ) RaHl/2 (10)

In conclusion, if the heat transfer mechanism is natural convection, the

maximum density of heat generating electronics (or Q/HLW) is proportional
to (T max- T ~) 312 and H- 112 • The same figure of merit is proportional to the prop-
erty group k(gWav) 112 •

BUNDLE OF HORIZONTAL CYLINDERS (PIN FINS)

COOLED BY NATURAL CONVECTION

An optimal spacing for natural convection cooling exists regardless of the

shape of the heating elements that occupy the H X L cross-section of the fixed
volume of Figure 1. In the preceding section we illustrated the method of in-
tersecting the asymptotes by using the vertical board as the elementary geo-
metric feature. Now we focus on the case where the volume is occupied by an
array of horizontal cylinders of diameter D and length W, for example, by pin
fins on a vertical base surface H x L, Figure 3 [Bejan et al., 1995a].
When the cylinders are arranged in an equilateral triangular array, the inter-
section of the large-Sand small-S asymptotes yields an implicit relation between
the optimal spacing Sopt and the Rayleigh number Ra0 = g[3D 3 (Tw- T=)/(av)

Sopt. 2 + Sopt/D '= 2.75(!:!)1/3 Ra-l/4 (11)

D (1 + sopt /0)2/3 D D

This relation is plotted in Figure 4, which shows that SoptiD is almost propor-
tional to the group (HID) 113 Rao- 114 • In other words, a simpler alternative to the
order of magnitude estimate obtained in Equation 11 is

Sopt ~ (!:!) l/3 Ra -l/4 (12)

D D D

or

Sopt (H} 112 R

H~ D aH
-t/4
(13)
Geometric Optimization of Cooling Techniques 7

f
H

l u
t
I...._ L ----ilia- I

(a) (b)
FIGURE 3. Array of horizontal cylinders cooled by natural convection and detail of one of the
channels traveled by the coolant (Bejan et a!., 1995a).

where RaH = g~H 3 (Tw- T~)/(av). Equation 13 shows that the optimal spacing
is approximately proportional to H 113D- 1112 , which means that it is almost in-
sensitive to changes in the cylinder diameter.
The optimal spacing was verified and calculated more accurately based on
full numerical simulations and experimental measurements (Bejan etal., 1995a).
The numerical results are shown in Figure 4. They are correlated very nicely in
the manner anticipated in Equation 11, namely by using the group (HID) 113
Ran- 114 on the abscissa. The optimal spacings, however, are consistently 2.5
times larger than the values calculated based on Equation 11. The function of
type 12 that fits the numerical data the best (within 1.7% mean error) is

S
;t = 2.72 (H)l/3
0 Rai) 114 + 0.263 (14)

where 0.263 is a small correction terrn. The numerical results for the maximum
heat transfer rate that corresponds to the optimal spacings of Figure 4 are cor-
related within 1.7% by the expression

(15)

In this expression Qmax is the total heat transfer removed by natural con-
vection from the fixed volume H x L x W. Experiments conducted in the
range 300 < Ran < 400, HID = 6.2 and Pr = 0. 72, revealed optimal spacings
8 Air Cooling Technology

10
H/D
Ra 0
Hl 15 20
350 (') A G
Sopt 1000 0
"' 0

D 10,000 + A ill

(1 A G

"' 0
0
1

Eq.(11)

0.1
0.1 1

FIGURE 4. Theoretical and numerical results for the optimal cylinder-to-cylinder spacing in
natural convection (Bejan et al., 1995a).

that agree within 17% with the values determined numerically. It was also
shown that the optimal spacing is relatively insensitive to whether the
cylinders are isothermal or with uniform heat flux. This conclusion widens
the applicability of Equation 14 and agrees with a similar conclusion
reached in Bejan and Sciubba [1992] for stacks of parallel boards cooled by
forced convection.

STACK OF PARALLEL PLATES COOLED

BY LAMINAR FORCED CONVECTION

Consider now the problem of installing the optimal number of heat gener-
ating boards in a space cooled by forced convection [Bejan and Sciubba, 1992].
As shown in Figure 5, the swept length of each board is L, while the transver-
sal dimension of the entire package is H. The width of the stack, W, is perpen-
dicular to the plane of the figure. We continue to rely on the assumptions (a) to
(c) listed under Equation 1. The thickness of the individual board is again neg-
ligible relative to the board-to-board spacing D, so that the number of boards
is approximately
Geometric Optimization of Cooling Techniques

~------------- L ~I

-- j_

--
Tmax D

.. T
.
-
D.p

T~ .
-
H

"

- ..
.
-- f circuit board

FIGURE 5. Package of parallel boards cooled by forced convection (Bejan and Sciubba, 1992).

(16)

It is important that the pressure difference across the package, Ap, is constant
and known. This is a good model for electronic systems in which several pack-
ages and other features (e.g., channels) receive their coolant in parallel from
the same plenum. The plenum pressure is maintained by a fan, which may be
located upstream or downstream of the package.
We begin with the class of applications where the flow is laminar. The ex-
tension to turbulent flow is discussed in a later section. When D becomes suf-
ficiently small, the channel formed between two boards becomes slender
enough for the flow and heat transfer to be in the fully developed regime. The
average longitudinal velocity through the channel is [e.g., Bejan, 1993]

(17)

The total mass flow rate through the package of frontal area H x W is in =
pHWU. In the same limit, the mean outlet temperature of the fluid approaches
10 Air Cooling Technology

the board temperature T max· The total rate of heat transfer removal from the
H X LX w space is Q=mCp (T max- T =) or, in view of Equation 17

(18)

In this way we reach the conclusion that in the D ---7 0 limit the total heat trans-
fer rate decreases as D 2 . This trend is illustrated by the small-D asymptote
in Figure 6.
Distinct thermal boundary layers will cover the board surfaces when the
spacing D becomes sufficiently large. Since the overall pressure drop Ap is
fixed, the immediate question is what free stream velocity U= sweeps these
boundary layers? We answer this question by noting the force balance on the
control volume H x L x W

Ap · HW = 't · 2n L W (19)

in which ::C is the L-averaged wall shear stress provided by the Blasius solution
[Bejan, 1993]. The result for U= is

(20)

forced convection
'I
8
'·, I
\

' /"""- the small-D limit: Q- D2

'' \
I

\
I
•
' I

... I
• -2/3
"'. ... the large-D limit Q - D
I .....

actual
I
unknown
"
0
0 D
opt
D
FIGURE 6. The maximization of the total heat transfer rate removed by forced convection from
the package of Figure 5 (Bejan and Sciubba, 1992).
Geometric Optimization of Cooling Techniques 11

The heat transfer rate through a single board surface is Q1 =fi LW (Tmax-
T =), for which the L-averaged heat transfer coefficient fi is provided by the
Pohlhausen solution for Pr;::: 0.5 [Bejan, 1993]. The total heat transfer rate re-
moved from the entire package is Q = 2n Q 1 or, after using the U= expression
listed above

113
Q. pvL2 DAp)
= 1.208 kHW(Tmax- L) (Pr 2 (21)

In conclusion, in the large-D limit the total heat transfer rate decreases as D- 213
while the board-to-board spacing increases. This second trend is also visible in
Figure 6.
The intersection of the two Q(D) asymptotes, Equations 18 and 21, yields
an approximate estimate for the optimal board-to-board spacing for maximum
heat transfer rate

D
{pt := 2.7 ( A~~L
2)-1/4
(22)

This optimal spacing increases as V 12 , and decreases as Ap- 114 if the applied
pressure difference increases. The Dopt estimate produced by Equation 22 agrees
within 20%. with the more exact value obtained by locating the maximum of
the actual Q(D) curve [Bejan and Sciubba, 1992] and is adequate when the
board surface is modeled as isothermal. For surfaces modeled as uniform flux,
the 2. 7 coefficient is replaced by 3.2 on the right side of Equation 22.
It is instructive to compare Equation 22 with Equation 8 and to notice the
symmetry between the nondimensional optimal spacings for forced convection
cooling and natural convection cooling. Note further that the role in the natural
convection formula is played by RaH, in forced convection is played by the new
dimensionless pressure drop number [Bejan, 1993]:

(23)

This dimensionless group was also identified in problems of external forced

convection [Bhattacharjee and Grosshandler, 1988], entropy generation mini-
mization [Paoletti et al., 1989], and electronic cooling [Knight et al., 1991;
Bejan and Sciubba, 1992] as summarized in Petrescu [1994].
The manner in which the design parameters influence the maximum rate of
heat removal from the package is indicated by

(24)

which is obtained by setting D = Dopt in Equation 18 or 21. Once again, the un-
equal sign is a reminder that the actual Qmax may be smaller by a few tens of
12 Air Cooling Technology

percentage points, because the peak of the Q(D) curve is situated under the
point where the two asymptotes cross in Figure 6. The maximum volumetric
rate of heat generation in the H x L x W space is

Qmax < 0 6!.(T - T )lll/2 (25)

HL W - · L2 max = L

In a subsequent study [Mereu et al., 1993] it was shown that the optimal
board-to-board spacing (Equation 22) also holds when the board thickness (t)
is not negligible in comparison with the spacing (D). This conclusion was based
on the method of the intersection of the asymptotes, as well as on full numer-
ical simulations of the flow field upstream, inside, and downstream of the stack.
The maximum overall thermal conductance between the coolant (T= at inlet)
and the entire stack (hot-spot temperature T max, total heat generation rate Qmax)
is correlated within 5% by the formula

Qmax/[W(Tmax- T=)] 0.18

(26)
Hcp(pilp I PrY 12 1 + 0.36(t/L)ll~ 4

This correlation was tested numerically in the Re range 500-1000, where Re

is based on the board length L and the approach velocity of the free stream. It
is relatively insensitive to whether the surfaces are isothermal or with uniform
flux. Again, the optimization of the D spacing was based on the assumption
that the overall pressure drop across the stack is fixed. Corresponding optimal
D and maximum conductance results have also been developed for stacks
subjected to fixed total mass flow rate and fixed pumping power [Mereu et
al., 1993].

STACK OF PARALLEL PLATES WITH FLUSH~

MOUNTED AND PROTRUDING HEAT SOURCES
The optimization of the spacing between boards with discrete heat sources
was performed numerically [Morega and Bejan, 1994a]. This was an extensive,
systematic study where both flush-mounted and protruding sources were con-
sidered. A sample of the optimization procedure is shown in Figure 7, which
corresponds to a stack where five flush-mounted heat sources are energized on
only one side of each board. The other side of the board is assumed adiabatic.
In addition to the spacing D and swept length L, the geometry of each channel
is complicated by the source length Ls, unheated length Ld, and total heated
length f. The heat generation rate is the same in every heat source. The hot-
spot temperature (Tmax) occurs at the trailing edge of the last heat source. In di-
mensionless form, the maximum overall thermal conductance of the stack (Bmax)
is the group on the left side of Equation 26,

(27)
 Geometric Optimization of Cooling Techniques 13

adiabatic
j_
D
;j Ls
t: 'iTmax

-j 1-- Ld
1- f

1- L

0.099
Ls = 0.117
L
Ld = 0.25
Pr = 0.7 Ls
B
max 0.098

0.097
2 2.4

FIGURE 7. The maximization of the overall thermal conductance when heat is generated by
five equidistant flush-mounted heat sources (Morega and Bejan, 1994).

Another example of the optimization procedure is presented in Figure 8.

When the heat sources are raised, the geometric description is further compli-
cated by the source height Ds. The abscissas of Figures 7 and 8 show that the
scaling law (Equation 22) can be used to predict adequately the optimal spac-
ing. Many geometries of these types (Figure 7 and 8) were optimized, and the
conclusion was that although Equation 22 works, a more accurate correlation
is the one based on the actual heated length (f) instead of L:

Dopt = 2 7TI-l/4
e - . e (28)

The effect of fluid type (Pr) is already built into Equation 28 by using the pres-
sure drop number defined in Equation 23. The overall thermal conductance of
14 Air Cooling Technology

adiabatic
j_
D
q 1
Tmax

I~
,_~ f-- Ld
i
L
-I -1
0.084
Ls =0.088
L
Ld =0.75
0.082
0.006 0.009 Ls

0.080 r-
~·:0.0
I
0.078
2 2.2 2.4 2.6

FIGURE 8. The effect of varying the height of the protruding heat sources in a stack of parallel
boards cooled by forced convection (Morega and Bejan, l994a).

the stack is correlated adequately by Bmax ~ 0.08, which in physical terms

means that

(29)

Although the various changes in geometry and thermal boundary conditions

have some effect, the actual Dopt and Qmaxi(T max - T=) values are relatively in-
sensitive to these details and, for engineering purposes, are described quite ad-
equately by Equations 28 and 29. Furthermore, the peak of the thermal
conductance curve is sufficiently flat so that it is not absolutely necessary to
know the precise value of Dopt· The numerical results also showed that if dis-
crete heat sources are mounted on both sides of each board, the Dopt value is
roughly 30% greater than the estimate based on Equation 28 and that the max-
imum conductance is about 3 times greater than the value furnished by Equation
Geometric Optimization of Cooling Techniques 15

29. Boards with protruding heat sources and laminar forced convection were
analyzed numerically in Kim and Anand [1994a] and numerically and experi-
mentally in Farhanieh et al. [1993].

STACK OF PARALLEL PLATES COOLED

BY TURBULENT FORCED CONVECTION
The optimal spacing of boards with turbulent flow has been determined based
on the method of intersecting the asymptotes [Bejan and Morega, 1994]

D /L
opt = (f C )112 Pr-21J (Pr :S 0.5) (30)
( 1+tiD opt )1/2 r

Cr
In this expression t, f, and are the board thickness, the friction factor for fully
developed flow through a parallel-plate channel, and the average skin friction
coefficient for a plane surface, smooth or rough. The corresponding maximum
value of the overall thermal conductance is

[ q'L J :S (Cr)t/4 Pru6(1 + _t_)-3t4(Ap. Lz)tt2 (31)

k:H(Tmax- T=) max f Dopt jJ.a

(Pr;:;::: 0.5)

where q' = Q!W. The inequality sign is a reminder that if q' is plotted on the
ordinate and D on the abscissa, the peak of the actual q' vs. D curve is lo-
cated under the intersection of the asymptotes (Figure 6). The right side of
Equation 31 represents the correct order of magnitude of the maximum over-
all thermal conductance and can be expected to anticipate within 30% the
exact value.
Beyond this point we must make certain assumptions regarding the values
of the friction factor and skin-friction coefficient If all the board surfaces are
smooth, we can use the standard correlations [e.g., Bejan, 1993]

f = 0.046 Re[i~ 5 (32)

(33)

where Dh = 2D, Renh = DhU/v and ReL = U~Liv. These allowed us to relate U
and U~ to Ap as shown in Bejan and Morega [1994]:
519
U= 5.98Dzt3v-lt9
(
~~ ) (34)
16 Air Cooling Technology

(35)

When combined, Equations 32-35 express f and Cr as functions of the imposed

pressure drop, i.e., functions that can be substituted on the right side of Equation
30. The final expression for the optimal spacing is

D opt 1L = 0.071 Pr-5/11 ( ll.p. L2)-1111

(1 + tiD opt )4/11 jl<X (36)

The geometric meaning of this conclusion becomes clearer if we estimate

the expected order of magnitude of the right side of Equation 36. First, note that
the ReoH range listed in Equation 32 can be rewritten in terms of ll.p by using
Equation 34 and the assumptions that (1 + t1Dopt)4111 = 1 and Pr = 0.72 (air)

ll. L 2 )-1111
0.09 > ( ~~ > 0.032 (37)

Similarly, the ReL range specified in Equation 33 can be rewritten using

Equation 35

1!. L 2 )-1111
0.087 > ( ~~ > 0.038 (38)

Equations 37 and 38 show that the specified ReoH and ReL ranges correspond
to the same range of the pressure drop number ll.p · U/(jl<X). Taken together,
Equations 36-38 show that the slenderness ratio of each board-to-board chan-
nel (DoptiL) takes values between approximately 0.003 and 0.007 and is rela-
tively insensitive to changes in the applied pressure difference.
When the surfaces are smooth, cf. Equations 32 and 33, the maximum over-
all conductance expression (Equation 31) becomes

[ q'L J : :; 0.57Pr4/99(1 + _t_)-67/99(Ap. L2)47/99 (39)

kH(Tmax- T=) max Dopt jla

where q' = QIW. In the case of a fluid with Prandtl number of order 1, Equation
39 is almost the same as the more general Equation 31 with the constant 0.57
in place of (Cr/f) 114 • In conclusion, the maximum overall conductance increases
almost as 1l.p 112 .
Figure 9 shows the optimal spacing calculated by using Equation 36 for tur-
bulent flow. The corresponding DoptiL result for laminar flow has been plotted
to the left while using the coefficient 3.2 (uniform flux) instead of 2.7 (uniform
temperature) on the right side of Equation 22. The figure shows that when the
flow is turbulent DoptiL depends not only on the pressure drop number Ap ·
L 2/(jl<X) but also on Pr and tiL. The optimal spacing in turbulent flow increases
as Pr and tiL increase and is quite sensitive to such changes.
Geometric Optimization of Cooling Techniques 17

0.1
' Pr
.. 0.72
\ laminar flow 1

........
..
0.01 ' turbulent flow

FIGURE 9. The optimal board-to-board spacing for laminar and turbulent forced convection as
a function of the pressure drop number, the Prandtl number, and the board slenderness ratio tiL
(Bejan and Morega, 1994).

The corresponding maximum overall thermal conductance for turbulent flow

(Equation 39) has been plotted in Figure 10 next to the results for laminar flow.
The interesting conclusion made possible by this figure is that the turbulent
portion of each curve is, in an order of magnitude sense, an extension of the
laminar portion. This feature is unlike Figure 9 where there is a sharp change
in the behavior of DoptfL as the flow regime changes.
For boards with rough surfaces, the optimal spacing can be calculated with
Equations 30 and 31 and appropriate correlations for f and Cr. As a start, it can
be noted that when the roughness features are large (e.g., in the fully rough tur-
bulent regime in a duct [Bejan 1993]) the friction factor and the skin friction co-
efficient are relatively constant, i.e., insensitive to the Reynolds number. Boards
with protruding heat sources cooled by turbulent forced convection were stud-
ied numerical! y by Kim and Anand [1994b] and experimentally in Li et al. [ 1993].

STACK OF PARALLEL PLATES

IMMERSED IN A FREE STREAM
Another way of using the information of Figures 9 and 10 is to consider a
cooling arrangement in which specified is not ~p but the coolant velocity well
upstream of the stack, U 0 . In such an arrangement, the scale of ~p across each
channel has the following order of magnitude [Morega and Bejan, 1994a]
18 Air Cooling Technology

1010
1;1
,-----,5 Pr
109

~~
0.72
7
108

c IO'
~
'----'
106

105

104 0 /
...... ......
11211
.,"

103 /:i~~inar flow

102
101

FIGURE 10. The maximum overall thermal conductance for laminar and turbulent forced con-
vection as a function of the pressure drop number, the Prandtl number, and the board slenderness
ratio tiL (Bejan and Morega, 1994).

(40)

This scale can be substituted in Equations 36 and 22 to obtain [Bejan and

Morega, 1994]

(turbulent) l)

Dopt = 3 gp -1/4 R -1/2 (laminar) (42)

L - . r eL

The Reynolds number ReL is based on the specified upstream velocity and the
flow length of the stack
(43)

The optimal spacings recommended by Equations 41 and 42 are displayed in

Figure 11, which shows that in turbulent flow the channel spacing is influenced
not only by ReL but also by Pr and tiL.
The maximum overall thermal conductance (Equation 39) can also be ex-
pressed in terms of ReL by using Equation 40
 Geometric Optimization of Cooling Techniques 19

0.1
Pr = 0.72
laminar flow
Dopt
L
.
·. turbulent flow
0.01

0.001

0.1
Pr = 7

Dopt
L ·••. !.~minar flow

.
0.01
.
turbulent flow

0.001
103

FIGURE 11. The optimal spacing for laminar and turbulent forced convection as a function of
the free-stream Reynolds number ReL = U0 L/v (Bejan and Morega, 1994).
20 Air Cooling Technology

-67199

[ q'L ] :s 0.41 Prsl/99 ( 1 + _t_ ) Ret4199 (44)

lli(~u-L)~ D•

The t!Dopt ratio appearing on the right side is given by Equation 41. The re-
sulting maximum thermal conductance estimate has been plotted in Figure 12
next to the corresponding curves for laminar flow. The analogy between the use
of ReL on the abscissa (Figures 11 and 12) and the use of the pressure drop
number (Figures 9 and 10) is worth noting.
The optimization of the internal geometry of a stack immersed in a free
stream was investigated in great detail by Morega et al. [1995]. As shown in
Figure 13, it was assumed in the beginning that the plates in the stack are not
equidistant. The first objective was to determine the optimal spacing for each
individual pair of plates, i.e., the optimal distribution of spacings across the
stack. This problem was solved numerically by calculating the temperature and
flow fields in a large number of configurations. The plate surfaces were mod-
eled as uniform-flux.
The optimization procedure for a stack with n == 4 nonequidistant plates is
illustrated in Figure 14. The design has only one degree of freedom represented
by the half-spacing d~, or the position of the internal plate. The sum of the two
spacings is fixed, dt + dz = (U2)- 2t. For each flow (e.g., ReL = 200 in Figure
14), the temperature distributions over all the surfaces were calculated. There

107
:3
E
r-----1 Pr ~

~~:s
o.n
~

1 7
1120
c tiL •dl
. . "'
0
1 Q

1120
'-----'
. , "' II
.. 1/211
turbulent flow

..
&

104

"
. "' "' "'
l/20

103

102
103 104 105 106 107
ReL
FIGURE 12. The maximum overall thermal conductance for laminar and turbulent forced con-
vection as a function of the free-stream Reynolds number ReL = U 0 L/v (Bejan and Morega, 1994 ).
 Geometric Optimization of Cooling Techniques 21

free stream I
Uo,To

'b- T
=P

w
H

~"~------------'
~----- L ------1
1

ln- f computational domain

- - - - - - -u-1
=PI
H
t_:_ I
=P l 2 A
I
=
I I
=PI I
q"

----
=PI I
Uo,To ___j- ·n---o ------------PI--n·
0 L Xe

H
2

FIGURE 13. Stack of nonequidistant plates cooled by a free stream (Morega et al., 1995).

are four such distributions, e~, 9z, e3, and 94, which are nondimensionlized as
in Equation 46. The location and value of the maximum temperature was iden-
tified. The four temperature maxima are plotted in dimensionless terms (Stmax,
... , 84max) in Figure 14. These temperature maxima occur close to the trailing
edge of each surface.
The abscissa of Figure 14 accounts for changes in the relative position of
the inner plate. It is clear that this position can be selected such that the peak
temperature of the entire stack is minimized. Of interest then is the minimiza-
tion of the largest of the four temperature maxima, i.e., the minimization of the
dimensionless hot-spot temperature,

(45)
22 Air Cooling Technology

0.1 ReL = 200

n=4

emax,1,2,3,4

e2,max

0.05

e4,max

0
0.5 1
2d /(2d + d)
1 1 2

FIGURE 14. Stack immersed in a free stream: the effect of the position of the inner plates on
the maximum temperatures of the plate surfaces (n =4, ReL =200) (Morega et al., 1995).

This second step of the optimization method is presented in Figure 15. The
symbols that are superimposed on the curve ReL =200 indicate which of the
four temperature maxima is the largest, i.e., the surface on which the hot spot
is located. The hot spot jumps from one surface to another as the position of
the inner plate changes.
Next to the ReL =200 data derived from Figure 14, Figure 15 shows the cor-
responding results developed for ReL = 100 and ReL = 400. The effect of the
inner plate position is clear: the hot-spot temperature ehot is always the lowest
when 2d 1/(2d 1 + d2) is dose to 0.5, i.e., when the plates are positioned equidis-
tantly. The minimum exhibited by ehot is sufficiently flat in the vicinity of
2dtf(2dt + d2) = 0.5 such that we may conclude with confidence that the opti-
mal design for n = 4 and ReL = 100-400 is the one in which the plates are po-
sitioned equidistantly. The ehot minimum becomes flatter as ReL increases; this
means that the relative positioning of the boards becomes less critical as the
Reynolds number increases.
These conclusions were reinforced by repeating these calculations for stacks
with six plates. The geometry-induced changes in the hot-spot temperature be-
come less pronounced when n increases. This means that the fine-tuning of the
Geometric Optimization of Cooling Techniques 23

0.1

ReL = 200, Fig.l4

0.05

ReL = 400

0
e l,max
.. e 2,max
0
e 3,max
.. 9 4,max

0
0 0.5 1
2d /(2d +d)
1 1 2

FIGURE 15. Stack immersed in a free stream: the effect of the Reynolds number and the posi-
tions of the inner plates on the hot spot temperature of the entire stack (n=4) (Morega et al., !995).

position of each plate relative to its two neighbors loses its importance as n in-
creases. Since the best designs for n = 4 and n = 6 are the ones in which the plates
are spaced equidistantly, it is safe to generalize and to recommend the equidis-
tant spacing as an optimal design feature for stacks with more than six plates.
The second design aspect investigated numerically in Morega, et al. [1995]
was the optimal number of plates arranged equidistantly in a stack. The total
heat transfer rate from the stack of fixed volume H x L x W was independent
of the number of boards q' = Q!W. The hot spot temperature

e _That- To (46)
hot - q' /k

was calculated for each number of boards (n) and flow regime (ReL). The re-
sults are presented in Figure 16 for the range 2 :S; n :S; 8 and 100 :S; ReL :S; 1,000.
It is clear that there exists an optimal number of boards and that knowing this
number accurately makes a difference in the effort to maximize the overall ther-
mal conductance (the inverse of ehot). The nopt(ReL) values identified in Figure
16 are recorded in Table l.
24 Air Cooling Technology

~~~~*'•~~~~r~~~
·-··--·-····-·---~----·-··--·-t-·----J.____ j ___-t-··

e hot

10" 3 L-------~--~~~--~~~~~------~~--~

1 10
n
FIGURE Hi. Stack immersed in a free stream: the effect of the number of equidistant plates on
the hot spot temperature (Morega et al., 1995).

A theoretical nopt fonnula based on the method of intersecting the asymp-

totes is [Morega et al., 1995]

_ 0.26~ Pr 114 Re~ 2

nopt = (Pr ~ 0.7, n >> 1) (47)
1 + 0.26[ Pr 114 Re~ 2

The nopt values calculated based on Equation 47 have been added to Table 1.
The agreement between the rounded (integer) values of these order-of-magni-
tude estimates and the numerical data furnished by Figure 16 is good, even
though the number of boards is small, i.e., outside the range of Equation 47.
The relative agreement improves as ReL increases. This trend is supported fur-
ther by the independent nopt data collected from [Nakayama et al., 1988] and
[Matsushima et al., 1992].
The corresponding minimum hot-spot temperatures were correlated with the
theoretical (scaling correct) fonnula derived in [Morega et al., 1995]
Geometric Optimization of Cooling Techniques 25

e . =l&s!:l+3.27(t/L)ReL 2 (Pr :2:: 0.7, n » 1) (48)

hot,mm · H Prl/2 ReL

The ReL range of this correlation was extended from 100 to 10,000 by also
using the data of Nakayama et al. [1988] and Matsushima et al. [1992]. Note
that the dimensionless minimum hot-spot temperature ehot,min is the inverse of
the maximum stack thermal conductance, cf. Equation 46.
When the number of plates is greater than nopt, the flow in each channel ap-
proaches the fully developed regime, and the stack as a block can be modeled
as a saturated porous medium with zero permeability in the transversal direc-
tion [Nield and Bejan, 1992]. It was shown that when the stack porous medium
model is combined with the usual pure-fluid model of the external flow the
computations are faster and permit a study of the effect of heat conduction
through the plate material [Morega et al., 1995].

BUNDLE OF CYLINDERS COOLED

BY FORCED CONVECTION
The optimization of the internal geometry of a fixed volume cooled by forced
convection can be pursued in other configurations, i.e., in applications where
the heating elements are not shaped as parallel plates. An important class is
shown in Figure 17; parallel cylinders (isothermal in theW direction) are a good
model for pin-fin heat sinks with near 100% fin efficiencies.
The optimal spacing S for cylinders arranged in an equilateral triangular
array has been determined based on the method of intersecting the asymptotes
[Bejan, 1995a,b]. The asymptotes were determined by using the large volume
of empirical data accumulated in the literature for single cylinders (large-S
limit) and arrays with many rows (small-S limit) [Zuk:auskas, 1987]. It was as-
sumed that the cylinder diameter (D) and the pressure drop across the bundle
(~p) are specified, and that each cylinder has the same heat transfer rate. In the
range 104 $; P$; 108 , 25 $;HID$; 200 and 0.72 $; Pr $;50, the optimal spacing is
correlated within 5.6% by the expression

Sopt (HI D) 052

D = 1.59 pO.l3 Pro.z4 (49)

where Pis a pressure drop number based on D

- dpD 2
P=-- (50)
uv
The hot spot of the bundle occurs on the cylinders that occupy the last row.
The minimum hot-spot temperature, which corresponds to the optimal spacing
of Equation 49, is correlated within 16% by the expression
26 Air Cooling Technology

-
/
-"'
/

/
/

--
H------1

FIGURE 17. Fixed volume with bundle of parallel cylinders perpendicular to a free stream
(Bejan, 1995a,b).

Tmax - T~ _ 3.33 (51)

QD I (kLW) = p045 Pr0.64

Both Equations 49 and 51 are recommended by the analytical solution obtained

by intersecting the asymptotes. Noteworthy is that the denominator on the right
side of Equation 51 is approximated well by n 1g where the pressure drop num-
ber is based on D

(52)

This allows us to rewrite Equation 51 as the maximum power density that can
be installed on cylinders in the fixed volume H x L x W

(53)
Geometric Optimization of Cooling Techniques 27

It is interesting that since flo is proportional to D 2 the maximum power den-

sity is independent of the cylinder diameter. It is important in thermal design
to selectS in accordance with Equation 49.
For applications where the free stream velocity Uo is specified (instead of
lip), Equations 49 and 51 may be transformed by noting that, approximately
lip- (112) p U~. The results are [Bejan, 1995b]

(54)

Tmax - T= _ 4.5
(55)
QD/(kLW) = Re~9 Pr 0·64

where Reo= UoD/v, or Reo- (2i')u2 • Equations 54 and 55 cover the range 140
< Reo < 14,000, 25 < HID < 200, and 0. 72 < Pr < 50.

HEAT GENERATING PLATE COOLED

INSIDE A PARALLEL PLATE CHANNEL
In certain types of electronic packages, the single-phase coolant flows through
a set of two-dimensional parallel channels formed by a row of printed circuit
boards plugged into a mother board. Each board may be surrounded by a metal
or metal-coated plastic case whose function is to shield the electronic circuitry
from external electromagnetic noise. It is important to know the optimal geom-
etry of each cassette (i.e., the board and its parallel-plate casing) so that the board
operating temperature is minimum [Bejan et al., 1993]. To optimize the geom-
etry of the cassette means to find not only the optimal position for the board in-
side the channel but also the optimal slenderness of the cassette itself (i.e., the
spacing of the channel in which the heat generating board is encased).

BOARD WITH LARGE THERMAL CONDUCTANCE

IN THE TRANSVERSAL DIRECTION
Consider the problem of cooling in the most effective way a board of length
L by positioning it in a stream of coolant that flows through an insulated
parallel-plate channel of the same length. The channel spacing Dis fixed. The
geometry sketched in Figure 18 is two-dimensional, as the board and the
channel are sufficiently wide (width W) in the direction perpendicular to the
figure, W > L.
We assume that the pressure difference across the channel is fixed, .1p, be-
cause the flow is driven by a fan with diameter considerably greater than the
channel spacing D. In an actual application, the fan would blow air through a
stack of ten or more cassettes of profile L x D. One such cassette is presented
in Figure 18.
The total rate of heat transfer Q from the heated plate to the fluid, through
both sides of the plate, is fixed by the electric circuit design. The plate
28 Air Cooling Technology

Coolant,
AP
To

0 X
L

FIGURE 18. Heat generating board cooled by a stream in an insulated parallel-plate channel
(Bejan et al., 1993).

thickness is negligible with respect to the channel spacing D. The only degree
of freedom in choosing the best cooling arrangement is the position of the heated
plate inside the channel. This position is pinpointed by the subchannel spac-
ings above and below the heated plate, Dt and Dz, such that Dt + Dz =D. To
illustrate the solution method in the simplest possible way, we first assume that:

1. The heated plate is isothermal at T w·

2. The flow is fully developed and laminar on both sides of the plate.
3. The NTU on either side of the plate is sufficiently greater than 1 so that
(Tout.l.2- To)/(Tw- To)== l.
4. The surfaces of the plate and the channel walls are smooth.

As shown in Figure 18, Tout.! and Tout.z are the outlet bulk temperatures above
and below the heated plate. The objective is .to determine the best configura-
tion (Dt!D) so that the thermal conductance Q/(Tw- To) is the greatest.
If we label Q1 and Qz the heat transfer rates through the upper side and the
lower side of the heated plate, then the approximation (1) permits us to write

(56)

where, for fully developed laminar flow, [e.g., Bejan, 1993]

(57)

Next, we write y and (1 - y) for the dimensionless spacings of the upper and
lower subchannels

D 1 =yD, D 2 =(1-y)D (58)

and calculate the total heat transfer rate Q = Q1 + Qz

(59)
Geometric Optimization of Cooling Techniques 29

Figure 19 shows that the highest value of they function on the right-hand
side is 1 and that it occurs when y = 0 or 1. The minimum value (114) occurs
when y = 112. We reach the somewhat unexpected conclusion that, from a cool-
ing standpoint, the centerplane (y = 112) is the worst position that the heated
board can have. The best arrangement is the one where the board is attached to
one of the insulated walls of the channel, even though in that case the entire
heat transfer rate Qmust leave the plate through only one of its side sun:aces.
When the board is attached to one of the walls, the thermal conductance Q/(Tw
- To) is four times greater than when the board is positioned in the center of
the channel.
The conclusion that the worst cooling position is y = 112 remains valid even
when some of the simplifying assumptions (1) through (4) are relaxed. For ex-
ample, let us discard (2) and (4) together and assume instead that the flow is
turbulent and fully developed and that the board surfaces are very rough. This
is a good model for cassettes with LID ratios much greater than 10, so that the

""
.---E
.
1~
~
0.5
1~

0
0 0.5 1

y = Dt
D
FIGURE 19. The effect of the board position on the overall thermal conductance of the cassette
of Figure 18: _ _ _ laminar, fully developed flow,-------- turbulent, fully rough, fully devel-
oped flow (Bejan et al., 1993).
30 Air Cooling Technology

entrance region is relatively small, and with boards densely covered with chips
and circuitry that rise as large-scale three-dimensional asperities above the sur-
face. Under these circumstances (i.e., in the fully rough limit of turbulent duct
flow) the friction factor is practically independent of the Reynolds number, and
Equation 57 is replaced by
1/2
rh = ( pAp ) WD3/2 (60)
z fz L 2

The constant friction factors f1 and f2 depend on the dimensions of the rough-
ness elements (assumed the same for both board surfaces) and on the respec-
tive subchannel spacings (DI, D2). When the board is placed in the stream, i.e.,
at y values comparable with 112, the spacings D1 and Dz are also comparable
and, as a first approximation, f1 and f2 may be taken as equal to the same con-
stant f. This is a conservative approximation to which we shall return in the
next paragraph. In the end, Equation 59 is replaced by

Q
· ( -fL ) 112 1 =y3/2+(1-y)3/2 (61)
Tw - T0 pAp c p WD 312

This result shows that the overall thermal conductance Q/(Tw - T0) is once
again minimum if the board is placed in the middle of the channel. The right-
hand side of Equation 61 is plotted as a dashed curve in Figure 19 and is valid
in the vicinity of y = 1/2. The thermal conductance minimum is not as sharp as
for fully developed laminar flow, suggesting that the optimal positioning of the
board is not as critical in the fully rough limit. If one is to repeat the analysis
and take into account the difference between f 1and f2 as the board is positioned
close to one of the walls, one would obtain a curve that falls under the dashed
curve in Figure 19.

BOARD WITH FINITE THERMAL CONDUCTANCE

IN THE TRANSVERSAL DIRECTION
Consider now the more realistic model in which the board of Figure 18 (the
substrate of an electronic circuit board) has a finite thermal conductivity kw and
thickness t. The thickness continues to be negligible with respect to D. The two
surfaces of the board are loaded equally and uniformly with electronics: the
constant heat generation rate per unit board surface is q". It is important to note,
however, that the heat fluxes removed by the two streams are generally not
equal because of the conduction heat transfer across the board.
The temperatures of the two board surfaces (T 1, T 2) increase in the down-
stream direction, and reach their highest levels at the trailing edge, x =L. The
objective is to minimize the larger of these two trailing-edge temperatures by
choosing the optimal board position y. Space limitations permit us to show only
the final results, which are displayed in Figures 20 and 21 in terms of the trans-
versal thermal conductance number
Geometric Optimization of Cooling Techniques 31

(62)

and the dimensionless temperature ceiling

(63)

where T max is the temperature of the hot spot, i.e., the larger of the trailing-edge
temperatures of the two board surfaces.
Figure 20 shows that the best board position Y min depends on B, i.e., on the
degree to which the board substrate is a good thermal conductor:

1. When B is of the order of 1 or larger, the lowest ceiling temperatures are

registered at Ymin = 0 and Ymin = 1, i.e., when the board is positioned close
to one of the insulated walls of the channel. The worst position is in the
middle of the channel, Ymax = 1/2. These conclusions agree completely
with what we learned based on the isothermal board model (i.e., B ~ oo ).
2. When the board is a poor thermal conductor such that B is smaller than the
order of 1, the best position for the board is along the midplane of the D x
L channel. The worst position, Ymax, approaches 0.8 and 0.2 as B decreases.

The transition from conducting boards (1) to poorly conducting boards (2)
occurs when B drops below 0.166. It is fascinating that the best location for
poorly conducting boards, Ymin = 112, happens to be exactly the same as the
worst location for highly conducting boards. This observation stresses the cru-
cial importance of the dimensionless number B. This number must be calcu-
lated early in order to determine the problem type, (1) or (2).
The lowest trailing-edge temperature ceiling that corresponds to the best lo-
cation Ymin is presented as 8min versus Bin Figure 21. The same figure shows
the uppermost trailing-edge temperature that corresponds to the worst position
Ymax, namely 8max· The lowest temperature ceiling (8m;n) is considerably smaller
than the highest temperature ceiling (Smax), regardless of the B value. This shows
the importance of knowing not only the best design (Ymin) but also the worst
design (Ymax).

STACKS OF PLATES SHIELDED

BY POROUS SCREENS
Microelectronics packaging is subject to several conflicting requirements
such as electromagnetic compatibility, acoustic limits, and adequate cooling.
Electronic systems and components are often enclosed completely inside con-
ducting cases to minimize radio frequency interference or electromagnetic in-
terference or for protection against airborne particles. These enclosures have
the added benefit that they reduce the sound pressure level associated with
32 Air Cooling Technology

r ··· · ·····•·······+·.. ·•···i···'··+··•··H···········-~ ·-•··· ·· ··; ...... ;.....;...; ..;..;. ;.. ;.......... ;.... y "• ffil
.

........ i
r················i······+·····i ····i···;..~ •.;.•;..;.....•~~---'·····,y

~~~maxi • j
y ··+··+ +! ++·········+··i·····-~k '·+ H··l-+l·········· j.........; . . . !.....j•..LLU ~
"-< . _
r ·················i········i

Ji min
y
;. +:·HJ:rt:::::::?~4h-l
i
o.s 1---r~+++-H-H--+·+ ·
/~

r··· · · · · · i· · · ·+···i·· · i· ·•··•··i··H··

1-----+--,..,-,...~
...,..::;::'...,:r.Jn tlF· ··-~~~--'~~~!y . . ···ji·.····,i··· !; .. ; ..;..;..;.................••... ..... ;• .....•... ····r'- -'--'--'-~

· · · · · · · ·\ · ··+ ··l-···i···tt"!"t·j········ ·l··+···· ···•· ···:•····!··•··'--i--i--i ····· · ·····l'·····;,_mi'·n - -~­

O L-~~-~-~·~·~
·· ~~
·-J~~:~:~~~,·~~~
· W
O.OI 0.1 10

B
FIGURE 20. The best position (Ym;n) and the worst position (Ymax) of a h eat generating board
with finite transversal thermal conductance (Bejan et al., 1993).

I 5

· ·\·'·\,............. ··+-·..········ •······+· ·+····! ·····; .;... ; .. i . .............. ·······•·············· •····· •....• ····•· ··---~ - i-l

10 ··:.~:::::.~,
. ...._,
.··•···············•·-··-··• 9 max

................:\'\:,
·········+·········+----·+··-·-! ····!···+···'···'··-················
· .....j....... ..L...... ! . . .L i ..L. ~- .L-1
......
5 1-························i········-······i"'-..:··+···---+·: 9

--
' ·.......l min !
,~-,-;·•· ················ · '· ··· ·····...; . ..... ; .......;......;..... ; ;.. ;. ~

• . .....................i.... .... .... ; .........;.....;......; . .. ;....i ., .. ; . ............. ... .. •.. ............ ;........... ; ...... ; .. ; .. ;

0
0.1 I0

B
FIGURE 21. The ol west temperature ceiling (8,.;0 ) and the highest temperature ceiling (Sm.,)
that correspond to t hebest location (ym;n) and the worst location (ymax) (Bejanet al., 1993).
Geometric Optimization of Cooling Techniques 33

acoustic noise generation. There are numerous applications where the thermal
design rules out the use of a complete enclosure around the electronic package.
Openings must be provided in the enclosure because the cooling requirement
of an enclosed packages is greater, and the need to ventilate the electronic com-
ponents becomes critical. This conflict between the need to enclose and the
need to cool poses a significant design challenge as the power density of elec-
tronics is increased.
The importance of the optimal selection of openings for air cooling has been
recognized in the design of stacks of parallel printed circuit boards. The opti-
mal thermal performance of a stack of parallel boards surrounded by perforated
screens formed the subject of a recent study [Bejanet al., 1995]. The study was
structured as a sequence of four distinct problems according to the cooling mode
(forced convection versus natural convection) and whether the board-to-board
spacing is fixed or constitutes a degree of freedom in the design. In every case
a relation was established between the characteristics of the perforated screens
(e.g., porosity) and the overall thermal conductance between the stack and the
coolant. In what follows we illustrate two examples.

FORCED CONVECTION
Consider the L x H stack of heat generating parallel plates (e.g., printed
circuit boards) shown in Figure 22. It is assumed that the stack dimension
perpendicular to the figure is sufficiently greater than the plate-to-plate spac-
ing D such that the flow through each channel is two dimensional. Two elec-
tromagnetic screens (e.g., perforated plates) are placed immediately upstream
and downstream of the stack. The coolant (e.g., air) is forced to flow through
the entire assembly by the pressure difference .t1p, which is fixed. The flow
through the spaces between the parallel plates is laminar. The screen is
described by the porosity <p and contraction pressure loss coefficient Ko, for
which there are sufficient empirical data in the literature (see also Bejan
et al., [1995]).
The optimal spacing was determined by using the method of intersecting the
asymptotes. The solution is presented in Figure 23 in the dimensionless notation

o= Dopt
L
ll"4
, IT=--,
.t1p. Lz
(64)
IHX
The solution accounts for the effect of screen porosity (through cp and Kc), the
effect of screen geometry (through Kc: perforated versus woven screen, sharp
versus rounded edges), and the effect of fluid type (through Pr). We see that when
the screens are absent ( x =0) the optimal spacing parameter approaches()= 2. 73,
i.e., Equation 22. In the opposite limit (x >> 1), the optimal spacing approaches
() = 1.32 x 114 • It is worth noting that in both x limits the() (x) relation is indepen-
dent ofPr; this feature is the result of including Pr in the definition of x, Equation
64. The Prandtl number has the peculiar effect that it shifts the minimum of the
() (x) curve, however, this is a minor effect in the Pr range 0.72-7.
34 Air Cooling Technology

Screen Stack Screen

I
I
I Tw
LlP
I D
=9> I H
I

1
t
U.,., Too
I
I ==i:> U, Uc

I
L

FIGURE 22. Forced convection cooling of a stack of parallel plates with inlet and outlet screens
(Bejan eta!., 1995).

Along the top of Figure 23 we plotted the porosity that corresponds to the
abscissa parameter x when the coolant is air (Pr =0.72) and the screen is a plate
with sharp-edge perforations. For the function Kc(<p) we used the values tabu-
lated in [Blevins, 1992]. The important conclusion made visible in Figure 23 is
that the optimal spacing increases when the screen becomes an increasingly more
significant flow obstruction. The upper abscissa shows that the effect of the
screens on the selection of D begins to be felt when the porosity <p drops below
approximately 0.7. When screens with porosities greater than 0.7 are used, the
optimal plate-to-plate spacing is the same as when the screens are absent.

NATURAL CONVECTION
We now rum our attention to the analogous question for designs where the
parallel plates are vertical and the cooling is by natural convection. As shown in
Figure 24, most of the modeling features described in connection with Figure 22
are repeated. The assembly is immersed in a quiescent fluid of temperature T=·
The number of parallel plates, n =LID, is assumed considerably greater than 1.
The intersection of the small-D and large-D asymptotes produced the opti-
mal spacing reported in Figure 25, where

RaH = g~W(Tw- T~) (65)

a.v
The optimal spacing decreases as the abscissa parameter increases, i.e., as the
screens pose an increasing flow resistance. The decrease in o* versus x* is quite
interesting because it runs against the trend exhibited by the corresponding re-
sult for forced convection, Figure 23. Interesting also are the similarities be-
tween Figures 25 and 23, specifically, the near-same abscissa parameters and
the fact that the effect of the screens becomes important when the abscissa pa-
Geometric Optimization of Cooling Techniques 35

<p
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2

10

II

FIGURE 23. The optimal plate-to-plate spacing as a function of the screen characteristics when
the stack is cooled by forced convection (high screen Reynolds numbers). The upper <p scale refers
to air cooling (Pr =0.72) and screens made out of plates with sharp-edged perforations (Bejan et
al., 1995).

rameter exceeds a critical order of magnitude. In Figure 25 the knee of the o*

(x*) curve is located in the vicinity of x* - 1, which in the case of air cooling
and plate screens with sharp-edged perforations means that the screen effect
becomes important when the porosity falls below approximately 0.8.
Several other optimization problems involving stacks with inlet and outlet
screens in forced or natural convection can be found in Bejan et al. [1995].

PLATE FINS WITH VARIABLE

THICKNESS AND HEIGHT
The hot-spot temperature of a module with finned air heat sink can be re-
duced by allowing the fin thickness and height to vary in the flow direction x,
while the amount of fin material is fixed [Morega and Bejan, 1994b]. This geom-
etry is illustrated in Figure 26. Two models were used. In a two-dimensional
model, the fin conduction was uncoupled from the external convection by as-
36 Air Cooling Technology

r - - - - - - L-------;...,
Vc

---+--- ----+----
D

H
Boundary
Layer

---+--- -~--+----

FIGURE 24. Natural convection cooling of a stack of parallel plates with inlet and outlet screens.
Right side: distinct boundary layers and core temperature in the large-D limit (Bejan eta!., 1995).

suming that the heat transfer coefficient varies as x- 112 , in accordance with bound-
ary layer theory. In the second model the three-dimensional problem of conju-
gate fin conduction and external convection was solved numerically. The main
conclusions are as follows:

1. The maximum temperature on the base of a fin with constant height is re-
duced by approximately 15% if the plate fin is sharpened like a dull knife
such that its thickness increases as X 042 in the downstream direction.
2. The hot-spot temperature at the base of a fin with constant thickness is
reduced by approximately 30% if the crest is inclined to face the flow,
with nearly zero height at the leading edge. The forward inclination of
the crest is the result of having assumed that the base of the fin is heated
with uniform flux. It can be shown that when the top of the module is
conductive enough the fin base is isothermal and the optimal crest incli-
nation has a negative slope, i.e., the crest looks downstream.
3. In addition to lowering the hot spot temperature, each of the design changes
(1) and (2) leads to a considerably more uniform temperature distribution
on the module surface on which the finned heat sink is installed.
4. It is conceivable that an even greater reduction in hot spot temperature
can be achieved by implementing the design features (1) and (2) simul-
Geometric Optimization of Cooling Techniques 37

<p
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2

10
....
-:::.:=
C<:l
~

~~=
t.O
..
II

1 10

FIGURE 25. The optimal plate-to-plate spacing as a function of the screen characteristics when the
stack is cooled by natural convection (high screen Reynolds numbers). The upper <p scale refers to air
cooling (Pr =0.72) and screens made out of plates with sharp-edged perforations (Bejan et al., 1995).

taneously. The determination of the actual magnitude of this combined

reduction would be a good topic for a future three-dimensional numeri-
cal study of the conjugate fin-conduction and convection problem (in-
cluding conduction inside the module) in which the sharpness of the plate
and tilting of the crest both vary.
5. A simpler way to proceed would be, first, to design the plate fin with con-
stant thickness and height by using the classical method (e.g., Bejan,
[ 1993], pp. 65-67) and, second, to reshape the thickness and height dis-
tributions according to conclusions (l) and (2) above, while keeping the
fin volume (weight) fixed. Related air-cooled heat sink geometries are
discussed in [Homentovschi et al., 1996; Ledezma and Bejan, 1996].

HEAT SINKS WITH PIN FINS AND PLATE FINS

The optimization of heat sinks with long finned channels and fully devel-
oped flow in each channel was performed in Tuckerman and Pease [1981],
Knight eta!. [ 1991], and Weisberg et al. [ 1992]. The work summarized in this
 38 Air Cooling Technology

fin

Uoo,Too
9:>
free stream

\ L

\ L

9:>
yt
9:>
viewed
Uoo,Too from
9:> H(x) the side

9:>
0
0 L X

viewed

l
from above

FIGURE 26. Forced convection-cooled electronic package with plate-fin heat sink (top), and
two-dimensional conduction model for a single plate fin (bottom) (Morega and Bejan, 1994b).

section refers to three-dimensional arrays of fins, that is, fins that are inter-
rupted (i.e., not long) in the flow direction [Bejan and Morega, 1993]. The op-
timization was based on an analysis where the space filled by fins and coolant
was treated as a porous medium [Bejan, 1990].

PIN FIN ARRAYS

The heat sink with pin fins cooled by forced convection is illustrated in
Figure 27. Each pin fin has the diameter d and length L. The area covered by
the array is B x X. The generated heat flux is uniform, q" =Q/BX. The array
has been optimized with respect to two degrees of freedom- first, the pin fin
diameter and, second, the array porosity <1> (or hydraulic diameter Db). There-
sults of the first maximization of the thermal conductance are shown in Figure
28, where
 Geometric Optimization of Cooling Techniques 39

0 _ ____,_x X

cover
plate

fin
T
L
coolant
l_
substrate

FIGURE 27. Heat sink with array of pin fins cooled by forced convection (Bejan and Morega,
1993).

() _ dopt L ( k
b ==- 4 Nu-r
)1;2
opt- ""15";:- ' Dh kw
(66)

()112 ( b )
G == 0 + 1 tanh 0112 (67)

and Nu is a constant of order I [Bejan, 1995a]. In the same solution kr and kw

are the fluid and wall (fin) thermal conductivities.
The final results for the second maximization of the thermal conductance
are shown in Figure 29. This time the hydraulic diameter (or its inverse, bopt)
was selected optimally, which is equivalent to selecting the array porosity <l>opt.
cf. porous-medium convection theory [Nield and Bejan, 1992]. The resulting
twice-minimized thermal resistance (Th- Tc)min/(q"Likw) is also reported. Note
that Th and Tc are the base hot-spot temperature and the coolant inlet tempera-
ture. The abscissa of Figure 29 corresponds to the pressure drop number

p- 4 Nu (.!s_)2 ~pL4 (68)

- k, kw f..lUX 2

where k, is a constant of order I 02 [Bejan, 1990] and ~p is the pres_sure differ-

ence maintained across the entire heat sink. It is worth noting that Pis a multi-
40 Air Cooling Technology

0.1

0.01

0.001
0.001 0.01 0.1 1 10
b
FIGURE 28. Optimal pin fin diameter for an array treated as a porous medium (Bejan and
Morega, 1993).

pie of the basic pressure drop number identified in_Equation 23. Figures 29 and
28 must be used together with Figure 29 first and Pas input. To improve the ac-
curacy of these calculations there is a need for information of heat transfer and
pressure drop at low Reynolds numbers [Fowler and Bejan, 1994].

PLATE FIN ARRAYS

The analogous plate-fin geometry is shown in Figure 30. The number of
rows is n. The swept length of each plate fin is ~, the fin thickness is t, and the
fin spacing is D. The overall thermal resistance of the heat sink was first min-
imized with respect tot. The results are shown in Figure 31, where

'topt
topt
= D' <I> = 1.1 (!i.)
kw
112
LI X
(D/X)l/3
(n J!<XX
b.p 2
)
116
(69)

- 'tl/2 [ (1 + 1:)1/6]
F - (1 + 1:)5/6 tanh <I> 1:112 (70)

The results of minimizing the thermal resistance with respect to the spacing
D are shown in Figure 32. The optimal spacing is represented by fl>opr, cf.
Equation 69, or by the array porosity $opt· The abscissa parameter is an appro-
priate pressure drop number
Geometric Optimization of Cooling Techniques 41

10

<j> opt

0.1
1 10 100 1000
p
FIGURE 29. Optimal inverse hydraulic diameter (bopt), minimum overall thermal resistance, and
corresponding optimal porosity (<!>opt) (Bejan and Morega, 1993).

p
*
= Pri13 (.h)2 Llp L 4
kw !10tX2
(71)

that too is a multiple of the pressure drop number defined in Equation 23. The
results for optimal plate fin arrays can be used by starting with P* as input, cal-
culating the optimal spacing (or «Popt. $opt) from Figure 32 and concluding with
the optimal plate-fin thickness from Figure 31. The corresponding minimum
thermal resistance is provided by Figure 32.
An important fundamental development that follows from the geometric op-
timization reviewed in Figures 27-32 is that the flow through a given space filled
with complicated objects can be treated as a flow through a porous medium at
moderate pore Reynolds numbers UK 112/v [Bejan, 1990]. The dimensionless
groups used on the ordinate and abscissa in Figure 33 are used routinely in the
field of convection in porous media [Nield and Bejan, 1992] where U is the vol-
ume averaged velocity. The bundle offour curves corresponds to flow across stag-
gered cylinders (equilateral triangle array) with four (transverse pitch/diameter)
ratios: 1.25, 1.5, 2 and 2.5. The data for these curves were taken from Zukauskas
[1987], or Bejan [1993, p. 488], and the permeability K was modeled as

(72)
42 Air Cooling Technology

I "I

flow q I B
r-----,

1 X

r-------------- ------, _l

---
______ !In
: Tt

L---------------------~

FIGURE 30. Heat sink with array of plate fins cooled by forced convection (Bejan and Morega,
1993).

where kz = 100 and the <jl is the porosity (void fraction) ofthe space. The porous
medium presentation of the cylinder array Llp leads to a much tighter collapse
of the four curves than in the original "heat exchanger handbook" presentation
of the same data.
The porous medium presentation of heat exchanger pressure drop informa-
tion is waiting to be extended to other heat exchanger geometries. In addition
to the tight correlation that this method brings, the fact that all the curves must
approach the Darcy flow limit

<lpKtn = (UKt;z)-t
(73)
XpU 2 v
when UK 112/v < 1, means that the Llp data can be extended (extrapolated) with
confidence into the low Reynolds number range where experimental data are
scarce [Fowler and Bejan, 1994]. Furthermore, flows at low Reynolds numbers
are becoming more common as the miniaturization of classical heat exchanger
configurations continues.
Geometric Optimization of Cooling Techniques 43

10

0.1 Fmax

0.01
0.01 0.1 1 10

FIGURE 31. Optimal plate-fin thickness for an array treated as a porous medium (Bejan and
Morega, 1993).

. . - - - - (Th- Tc)min /(q"L I~), Tuckerman and

,...----,---,--,-r-r..--rrr-I--.--,...-,-..,-,-,.........--....---.-,-,,.,..,... Pease {1981)
10

1111

1 {Th- TJmin
q"L/~

0.1
1 10 100 1000

FIGURE 32. Optimal plate-to-plate spacing and porosity, and minimum overall thermal resis-
tance (Bejan and Morega, 1993).
44 Air Cooling Technology

/'"
Darcy flow
staggered cylinders
in cross-flow

th
UK /v
FIGURE 33. Porous medium representationof the pressure drop data for flow through staggered
cylinders (Bejan and Morega, 1993).

CONCLUSION

In this chapter we reviewed a wide selection of heat transfer configurations.

In each case we minimized the overall thermal resistance between the coolant
and a fixed volume occupied by heat generating components. The heat trans-
fer mechanism varied from natural convection to forced convection. The shape
of the components varied from long plates to parallel cylinders and, in the last
section, to short plates and short cylinders.
The most important conclusion of this work is that the overall thermal resis-
tance can be reduced geometrically by optimally arranging the heat generating
components in the given volume. Quite often, the optimal positions (spacings)
can be determined based on simple analyses such as the method of intersecting
of asymptotes. The results for optimal spacings and corresponding minimum re-
sistance (or maximum conductance) are now available in compact, scaling cor-
rect formulas for key geometries. These results define a newly emerging subfield
of heat transfer, namely, the Geometric Optimization of Cooling Techniques.
Future studies may focus on other applications, namely fixed spaces with
different overall shapes, and components with shapes not considered in this re-
view. In the future work, the body shape correlation method of Refai Ahmed
and Yovanovich [1994] may prove very useful. The effect of self-sustained os-
cillations in the channels [Amon et al., 1992] also deserves to be studied, par-
ticularly with respect to determining the optimal board-to-board spacing. It will
Geometric Optimization of Cooling Techniques 45

be interesting to see how the scaling laws and dimensionless groups identified
in this chapter can be extended (i.e., generalized) to cover the optimization of
future geometries in a field that promises to be very diverse and important.

ACKNOWLEDGMENTS
This work was supported by the IBM Corporation, Research Triangle Park.
The guidance received from Dr. Sang W. Lee is gratefully appreciated. The nu-
merical work was supported by a grant received from the North Carolina
Supercomputing Center.

REFERENCES
Amon, C.H., Majumdar, D., Herman, C.V., Mayinger, F., Mikic, B.B., and Sekulic, D.P. 1992.
Numerical and experimental studies of self-sustaining flows in communicating channels, Int.
J. Heat Mass Transfer, 35:3115-3129.
Anand, N.K., Kim, S.H., and Fletcher, L.S. 1992. The effect of plate spacing on free convection
between heated parallel plates, J. Heat Transfer, 114:515-518.
Bar-Cohen, A. and Rohsenow, W.M. 1984. Thermally optimum spacing of vertical, natural con-
vection cooled, parallel plates, J. Heat Transfer, 106:116-123.
Bejan, A. 1984. Convection Heat Transfer, John Wiley & Sons, New York, prob 11, 157.
Bejan, A. 1990. Theory of heat transfer from a surface covered with hair, J. Heat Transfer,
112:662-667.
Bejan, A. 1993. Heat Transfer, John Wiley & Sons, New York.
Bejan, A. l995a. Convection Heat Transfer, 2nd ed., Wiley, New York.
Bejan, A. l995b. The optimal spacing for cylinders in cross-flow forced convection. J. Heat
Transfer, 117: 767-770.
Bejan, A. and Lee, S.W. 1994. Optimal geometry of convection cooled electronic packages, in
Cooling of Electronic Systems, S. Kakac, H. Yuncu, and K. Hijikata, Eds., Kluwer Academic
Publishers, Dordrecht, The Netherlands, 277-291.
Bejan, A. and Morega, A.M. 1993. Optimal arrays of pin fins and plate fins in laminar forced con-
vection, J. Heat Transfer, 115:75-81.
Bejan, A. and Morega, A.M. 1994. The optimal spacing of a stack of plates cooled by turbulent
forced convection, Int. J. Heat Mass Transfer. 37:1045-1048.
Bejan, A. and Sciubba, E. 1992. The optimal spacing of parallel plates cooled by forced convec-
tion, Int. J. Heat Mass Transfer, 35:3259-3264.
Bejan, A., Fowler, A.J., and Stanescu, G. l995a. The optimal spacing between horizontal cylinders
in a fixed volume cooled by natural convection, Int. J. Heat Mass Transfer, 38:2047-2055.
Bejan, A., Kim, S.J., Morega, A.M., and Lee, S.W. 1995. The cooling of stacks of plates shielded
by porous screens, Int. J. Heat Fluid Flow, 16:16-24.
Bejan, A., Morega, AI., Lee, S.W., and Kim, SJ. 1993. The cooling of a heat generating board in-
side a parallel-plate channel, Int. J. Heat Fluid Flow, 14:170-176.
Bhattacharjee, S. and Grosshandler, W.L. 1988. The formation of a wall jet near a high tempera-
ture wall under microgravity environment, ASME HTD, 96:711-716.
Blevins, RD. 1992. Applied Fluid Dynamics Handbook, Krieger, Malabar, FL.
Farhanieh, B., Herman, C., and Sunden, B. 1993. Numerical and experimental analysis of laminar
fluid flow and forced convection heat transfer in a grooved duct, Int. J. Heat Mass Transfer,
36:1609-1617.
46 Air Cooling Technology

Fowler, A.J. and Bejan, A. 1994. Forced convection in banks of inclined cylinders at low Reynolds
numbers, Int. J. Heat Fluid Flow, 15:90-99.
Homentovschi, D., Stanescu, G. and Bejan, A. 1996. Cooling of a two demensional space with one
or more streams making one or more passes, Int. J. Heat Fluid Flow, 17: in press.
Ishizuka, M. 1992. Application of thermo-fluid analysis to the thermal design of air-cooled elec-
tronic equipment, in Transport Phenomena in Heat and Mass Transfer, J.A. Reizes Ed.,
Elsevier, Amsterdam, 1067-1088.
Ishizuka, M. 1993. Thermal design approach for electronic equipment by use of personal com-
puters, in Computers and Computing in Heat Transfer Science and Engineering, W. Nakayama
and K.T. Yang, Eds., CRC Press, Boca Raton, FL, 391-407.
Kim, S .H. and Anand, N .K. 1994a. Laminar developing flow and heat transfer between a series of
parallel plates with surface mounted discrete heat sources, Int. J. Heat Mass Transfer,
37:2231-2244.
Kim, S.H. and Anand, N.K. l994b. Turbulent heat transfer between a series of parallel plates with
surface-mounted discrete heat sources, J. Heat Transfer, 116:577-587.
Knight, R.W., Goodling, J.S., and Hall, D.J. 1991. Optimal thermal design of forced convection
heat sinks-analytical, J. Electron. Packag., 113:313-321.
Ledezma, G. and Bejan, A. 1996. Heat sinks with sloped plate fins in natural and forced convec-
tion. Int. J. Heat Mass Transfer. 39: in press.
Li, W., Kakac, S., Hatay, F.F., and Oskay, R. 1993. Experimental study of unsteady forced con-
vection in a duct with and without arrays of block-like electronic components, Wiinne und
Stoffiibertragung, 28:69-79.
Matsushima, H., Yanagida, T. and Kondo, Y. 1992. Algorithm for predicting the thermal resistance
offinned LSI packages mounted on a circuit board, Heat Transfer Japan Res., 21 (5):504-517.
Mereu, S., Sciubba, E.,and Bejan, A. 1993. The optimal cooling of a stack of heat generating boards
with fixed pressure drop, flowrateorpumping power, Int. J. Heat Mass Transfer, 36:3677-3686.
Moffat, R.J. and Ortega, A. 1988. Direct air cooling of electronic components, in Advances in
Thermal Modeling of Electronic Components and Systems, Vol. I, A Bar-Cohen and A.D.
Kraus Eds., Hemisphere, New York, 129-282.
Morega, A.M. and Bejan, A. 1994a. Optimal spacing of parallel boards with discrete heat sources
cooled by laminar forced convection. Num. Heat Transfer, Part A, 25:373-392.
Morega, A.M. and Bejan, A. 1994b. Plate fins with variable thickness and height for air-cooled
electronic modules, Int. J. Heat Mass Transfer, 37 (Suppl. 1):433-445.
Morega, A., Bejan, A., and Lee, S.W. 1995. Free stream cooling of a stack of parallel plates, Int.
J. Heat Mass Transfer, 38:519-531.
Nakayama, W., Matsushima, H., and Goel, P. 1988. Forced convective heat transfer from arrays of
finned packages, in Cooling Technology for Electronic Equipment, W. Aung Ed., Hemisphere,
NewYork, 195-210.
Nield, D.A. and Bejan, A. 1992. Convection in Porous Media, Springer Verlag, New York.
Paoletti, S., Rispoli, F., and Sciubba, E. 1989. Calculation of exergetic losses in compact heat ex-
changer passages, ASME AES, 10(2):21-29.
Peterson, G.P. and Ortega, A. 1990. Thermal control of electronic equipment and devices, Adv.
Heat Transfer, 20: 181-314.
Petrescu, S. 1994. Comments on the optimal spacing of parallel plates cooled by forced convec-
tion, Int. J. Heat Mass Transfer, 37:1283.
Refai Ahmed, G. and Yovanovich, M.M. 1994. Approximate solution of forced convection heat
transfer from isothermal simple body shapes. AIAA Paper 94-1971, 6th AIANASME Joint
Thermophys, Heat Transfer Conf., Colorado Springs, June 20-23.
Tuckerman, D.B. and Pease, R.F.W. 1981. High-performance heat sinking for VLS I, IEEE Electron
Device Lett, EDL-2: 126-129.
Weisberg, A., Bau, H.H., and Zemel, J.N. 1992. Analysis of microchannels for integrated cooling,
Int. J. Heat Mass Transfer, 35:2465-2474.
Zukauskas, A. 1987. Heat transfer from tubes in crossflow, Adv. Heat Transfer, 18:87-159.
 Chapter 2

ENTRANCE DESIGN CORRELATIONS

FOR CIRCUIT BOARDS IN FORCED~AIR COOLING

M. Faghri, M. Molki, and Y. Asako

CONTENTS

Introduction ... . . .... .. .. .. .. .. . . ......... . . ... . .. . . . .. . . ..4 7

A New Correlation for Pressure Drop . ... . . ... . . . . . ..... .. ... . . . .49
The Experimental Apparatus and Procedure . . ... ... .... .. . ... . . .49
Experimental Uncertainty for Pressure Data . . . . .. . . . .. . . . .. .. . . .52
Flow Visualization . .... . .. .......... .. .... . . ...... .. . . . . . .53
Pressure Results . . . . . . .. . . . ... . . ..... . . . ... .. .... . .. . .. . . .55

A Correlation for Heat Transfer and Wake Effect . . . .. .. . . ... . . . . . . .63

The Experimental Setup and Procedure . . .. . .. . .. .. ... . .. . .. ...63
Data Reduction . . . .. ... . . . . .... . ... .. .... ..... . . .. . . . .. .. 65
Uncertainty Analysis for Heat Transfer Data .. . .. . . . ... . .... . . . .65
Heat Transfer Results and Discussion . .. ... .. ... .. . . ... . .... .. 66
Application of the Proposed Correlations . . . . . . .. . .. .. . . . . . . .. . .74

A User-Friendly Program for Prediction of Array Temperature .. . .. . ...75

Conclusion . . . . ............. .... ............ . . .... . ..... . . .77

Appendix A: Nomenclature .. . . .... ... .. . . . .. . . . . . ... . .. . . . . . . 78

References ....... . . . ..... . . . . .. .... . .. . .. ... . .. . . . . . . . . . . .79

INTRODUCTION
This chapter describes an experimental study of pressure drop, heat trans-
fer, and wake effect in the entrance region of an array of rectangular blocks in
a duct. The main focus of the work is to present correlations for pressure drop,
convective heat transfer coefficient, and thermal wake effect. Also, qualitative
information will be provided on the nature of the flow field in this region. The
correlations presented here are intended to be used by the practitioners to pre-
dict the temperature of the electronic components that have a similar geomet-
0-8493-'l447-3196/$0.00+$.50
<D 1996 by CRC Press, Inc. 47
48 Air Cooling Technology

ricallayout and especially short entrance lengths. The range of parameters is

close to that often encountered in the computer industry.
A search of literature revealed a number of related studies. Sparrow et al.
[ 1982, 1983] reported heat transfer and pressure drop in arrays of rectangular
modules with barriers and missing modules. The focus of their work was to
study the effect of missing modules and barriers on thermal-hydraulic behav-
ior of rectangular arrays. Their pressure results for no-barrier arrays without
missing modules were obtained for one particular geometry around Reynolds
number of 6900.
In an experimental effort, Lehmann and Wirtz [1985] studied the effect of
streamwise spacing and length on convection from an array of two-dimensional
modules. They also performed visualization tests in the periodically fully de-
veloped region of duct and obtained some information on the nature of flow
field in this region. The pressure drops reported in this reference is limited to
Re = 1000, 2000, and 3000. Some pressure results are also reported by Moffat
et al. [1985] for rather tall and widely spaced cubical elements.
For sparse arrays of elements, the pressure drop is mainly determined by
form drag and thus the pressure coefficient is independent of Reynolds num-
ber. In this connection, the pressure drop coefficients reported in some publi-
cations are independent of Re (e.g., Moffat et al., [ 1985]), while the others (e.g.,
Tai and Lucas, [1985], Souza Mendes and Santos, [1987], and Hollworth and
Fuller, [1987]) showed their pressure results in tenns of Re.
Another issue of importance in flow through array of rectangular blocks is
to determine the flow regime. In a paper by Garimella and Eibeck [1992], the
onset of transition from laminar to turbulent is investigated. They concluded
that, in a three-dimensional situation, transition is not only a function of flow
rate and geometry but also is a function of location in the array. They have fur-
ther stated that the transition Re (based on channel height) varies from 700 to
1900, and by increasing the stream wise spacing between the elements, the tran-
sition occurs at lower Reynolds numbers.
Arvizu and Moffat [1982] presented a superposition method to predict the
temperature distribution in a regular array of cubical elements. Another related
work on the subject is a paper by Wirtz and Dykshoom [1984] in which some
information is provided about heat transfer in the entry region of a sparse array
of elements.
In a series of experimental investigations, Moffat and his group have exten-
sively studied various aspects of thermal-hydraulic behavior of the arrays of
electronic components (Moffat et al. [1985], Moffat and Anderson [1988],
Anderson and Moffat [1990], and Anderson and Moffat [1991]). The overall
objective of their effort has been to develop the techniques and the data bases
needed to predict the operating temperature of the blocks. They have also em-
phasized the very crucial point that the adiabatic heat transfer coefficients, which
are reported in the electronics cooling literature, should be used in reference to
the adiabatic fluid temperature. Other relevant studies are those reported by
Entrance Design Correlations for Circuit Boards 49

Sridhar et al. [1990], Faghri et al. [1995], Wirtz and Weiming [1991], Kang
[1992], and a review article by Peterson and Ortega [1990]. Despite all the use-
ful information and data available in the aforementioned references, a concise
and convenient correlation for temperature prediction is not seen in the open
literature, especially for short channels encountered in electronic equipment.
There are several novel aspects to the results presented here. The focus of
this work is on the entrance pressure drop, heat transfer coefficients, and the
associated thermal wake effects. The data are successfully brought together by
defining modified parameters, and they are presented by simple correlations.
These correlations are subsequently incorporated into a simple algorithm to es-
timate the operating temperature of a circuit board with random heating. Further,
the flow field is visualized by a simple technique, and the results are discussed.
In this study, an array of rectangular blocks (modules) is positioned along
the lower wall of a rectangular duct. These blocks, which represent a model for
the modular electronic components, are arranged in an in-line fashion. The geo-
metric variables, namely the module dimension (L), the module height (B), the
inter-module spacing (S), and the height of the flow passage between the mod-
ule and the opposite wall of the duct (H), are varied in such a manner that BIL
= 0.5, SIL = 0.125, 0.33, 0.5, and HIL = 0.125, 0.25, 0.5, 0.75, 1.0, and 1.5.
These dimensions are close to those often encountered in the computer indus-
try. The working fluid is air, and the Reynolds number based on H and the air
velocity in the bypass channel (i.e., the channel formed between the top sur-
face of modules and the opposite wall of the duct) ranges from 400 to 15000.
Attention is now turned to a full description of the work. First, a new cor-
relation will be presented for pressure drop, followed by the presentation of
correlations for heat transfer and thermal wake effect The chapter will con-
clude with a discussion on how the array temperatures can be predicted a
computer program.

A NEW CORRELATION FOR PRESSURE DROP

THEEXPER~ENTALAPPARATUSANDPROCEDURE
The schematic view ofthe experimental setup is shown in Figure 1. Laboratory
air is drawn into the apparatus through a bell-mouth inlet geometry. The air then
passes through a flow development section (559 mm), flow straightener, test sec-
tion, flow redevelopment section (591 mm), venturimeter, and main valve and
then leaves the flow circuit through the blower operating in suction mode.
The air flow is adjusted by the main valve. There is also available a bypass
valve that allows for a more precise control of the air flow. The flow rate is mea-
sured by a precalibrated venturimeter.
The test section and the corresponding upstream and downstream ducts have
a rectangular cross section. The width of the cross section is equal to W == 178
mm, but its height (H + B in Figure l B and C), depending on the dimensions
of the array, varies from 12.8 to 76.2 mm.
50 Air Cooling Technology

VENTURI METER

AIRFLOW

TEST BYPASS VALVE

SECTION
(A)
FLOW
STRAIGHTENER
L+S

AIRFLOW

A = 14.3 mm for 50.8 mm module

25.4 <B > = 19.1 mm for 25.4 mm module
= 38.1 mm for 12.7 mm module
178

(C)

FIGURE 1. Schematic view of the experimental setup (not to scale). (A) the overall view, (B)
side view of the test section, (C) cross-sectional view of the test section. All dimensions are in
millimeters.

An in-line array of rectangular modules are deployed along the lower wall
of the test section. The dimensions, as noted in Figures lB and C, are S = 6.4
and 8.4 mm; H = 6.4, 12.7, 19.1, and 25.4 mm; L = 12.7, 25.4, and 50.8 mm;
and B = 6.4, 12.7, and 25.4 mm. From the combination of these dimensions
nine test sections are fabricated with dimensions shown in Table l. In all cases
BIL is equal to 0.5. The number of rectangular blocks depends on the test sec-
tion number. For test sections (1 to 3), (4 to 6), and (7 to 9) the number of mod-
ules along and across the test section are, respectively, 5 x 3, 8 x 5, and 15 x 9.
The pressure taps are located at the upper wall. As shown in Figure IB, the
first tap is located a distance A upstream of the second tap, and the other taps
are equally spaced. The distance A between the first and the second tap de-
Entrance Design Correlations for Circuit Boards 51

TABLEt
Geometrical Dimensions of the Test Section

Test BIL S/L HJL B L H s

Section Number [mm] [mm] [mm] [mm]

0.5 0.125 0.125 25.4 50.8 6.4 6.4

2 0.5 0.125 0.25 25.4 50.8 12.7 6.4
3 0.5 0.125 0.5 25.4 50.8 25.4 6.4
4 0.5 0.33 0.25 12.7 25.4 6.4 8.4
5 0.5 0.33 0.5 12.7 25.4 12.7 8.4
6 0.5 0.33 0.75 12.7 25.4 19.1 8.4
7 0.5 0.5 0.5 6.4 12.7 6.4 6.4
8 0.5 0.5 6.4 12.7 12.7 6.4
9 0.5 0.5 1.5 6.4 12.7 19.1 6.4

pends on the test section and is equal to 14.3 (for L = 50.8), 19.1 (for L =
25.4), and 38.1 (for L = 12.7) mm. It should be noted that the location of the
first tap was always fixed. However, to position one tap right above the mod-
ules it was required to have different A values for different module lengths.
In the spanwise direction, the pressure taps are located in the middle of the
cross section and just above the central module (Figure 1C). The location of
all pressure taps in terms of their distance from the first tap is shown in Table
2. It should be noted that the number of pressure taps for test sections 1 to 3
is more than what is shown in Figure lB (see Table 2 for exact location of
pressure taps).
Prior to the onset of a data run, the proper test section is selected and as-
sembled. Then, to prevent air from leaking into the test section, all suspected
joints are sealed by silicon rubber and thoroughly tested for leaks with the aid
of soap solution.
Once the test section is properly assembled, the air flow is activated and
adjusted for a given flow rate. After a warm-up period, the setup is ready for
pressure readings. The pressure signals are transmitted via plastic tubing to a
(l or 10 torr) pressure transducer and an electronic manometer, which are in-
terfaced with an IBM PC. The computer scans the pressure signals for 22 sec-
onds. This arrangement resolves pressure to within I0-5 mm Hg. Each
experiment is repeated twice and the results are averaged to obtain a data point.
A total of 1420 data points are reported here and are used to obtain the pres-
sure drop correlation.
All nine test sections are made entirely of plexiglass so that they can also be
used in flow visualization experiments. The flow visualization method used in
this work is the so-called oil-lampblack technique.
The results of this investigation are presented in terms of Reynolds number
Re with the conventional definition, Re = VH/v. In this equation, Vis the mean
velocity of air in the bypass channel above modules and is related to the vol-
ume flow rate as, V = Q/(WH). With this definition, the flow rate between the
52 Air Cooling Technology

TABLE2
Location of Pressure Taps

Test Sections Test Sections Test Sections

1,2,3 4,5,6 7,8,9
Tap Number X [mm] X [mm] X[mm]

0 0 0
2 14.3 19.1 38.1
3 28.6 52.9 57.2
4 42.9 86.7 76.3
5 57.2 120.5 95.4
6 71.4 154.3 ll4.5
7 85.7 188.1 133.6
8 100.0 221.9 152.7
9 114.3 255.7 171.8
10 128.6 289.5 190.9
II 142.9 210
12 157.2 229.1
13 171.5 248.2
14 185.7 267.3
15 200.0 286.4
16 214.3 305.5
17 228.6 324.6
18 242.9
19 257.2
20 271.5

modules at 0 < y < B is neglected and V is slightly higher than the actual mean
velocity. The pressure drops are nondimensionalized and expressed as

Po-p.
K=---'
0.5pV 2

where K is the pressure drop coefficient. In the periodically fully developed re-
gion, the pressure drops are converted to friction factor defined as

(2)

EXPERIMENTAL UNCERTAINTY FOR PRESSURE DATA

Prior to the onset of experimental runs, the venturimeter was calibrated with
the aid of a lamillfU"-flow element. The parameters recorded during the calibration
process and the respective measurement uncertainties are listed in Table 3. The
precision limits seen in the table are the smallest interval between the scale mark-
ings (least count) of the respective instruments. The bias limit for instruments was
negligible. The sensitivity coefficients were evaluated using the data reduction
FORTRAN program. The results were then combined through the root-sum-square
expression to obtain uncertainties (Kline [ 1985] and Abernethy et al. [ 1985]).
Entrance Design Correlations for Circuit Boards 53

Using these guidelines, the uncertainty of venturimeter coefficient is found

to be ±0.027 with Co= 0.971 ± 2.8%. Other uncertainty values are summarized
in Table 4.

FLOW VISUALIZATION
The flow pattern in the entrance region of the array of modules was revealed
by the application of oil-lampblack technique. The technique is relatively sim-
ple but may not be suitable for any type of flow field. The detailed description
of oil-lampblack visualization technique is described by Sparrow et al. [1981].
The first step is to make a suitable mixture of oil and lampblack. It was found
that the automatic transmission oil gives the best result The mixture is subse-
quently applied to the top surface of the module, which creates a uniform glossy
black surface.
When the airflow is initiated, the mixture moves under the action of shear
stress exerted by the air, and the streaks that form on the surface reveal the pat-
tern of flow field adjacent to the walL The development of the surface patterns
is visually observed during the visualization run. The experience gained from
these visual observations together with the information reflected by the streaks
are used to describe the flow field.

TABLE3
The Precision Limits for the Measured Parameters

Parameter Precision Limit

Barometric pressure ±0.1 mmHg

Venturimeter inlet gage pressure ±lo-s mm Hg
Pressure drop across the venturimeter ±lo-s mm Hg
Pressure drop across the laminar-flow element ±Jo-smm Hg
Inlet air temperature ±0.6°C
Laminar-flow element ±I% of volume flow rate
Duct width, W ±0.1 mm
Distance between top of modules and ±0.1 mm
the opposite wall, H
Module height, B ±0.1 mm
Inlet and throat diameters of venturimeter ±0.1 mm
Pressure gradient, dP/dX Twice the standard deviation
of the slope of P-X plot
test section gage pressure ±lo-s mmHg

TABLE4
Estimated Typical Uncertainties

Parameter Uncertainty(%)

Venturimeter coefficient, Co 2.8

Reynolds number, Re 2.9
Pressure drop coefficient, K 6.1
Friction factor, f2H 15.3
54 Air Cooling Technology

The photographic evidence of the flow field is shown in Figure 2. The pho-
tograph shows the top view of the flow pattern near the top surface of the first
three modules of the array. These patterns are obtained for the modules with
SIL = 0.125 and HIL = 0.125 at Re = 15000. The main flow is from left to right,
and the scale below the photograph shows the distance from the leading edge
of the first module in centimeters.
Special features of the flow are flow separation at the leading edge of the
first module, reattachment of air flow at about 1.8 em from the leading edge
(nearly L/3 ), and the formation of a recirculating bubble in between. The curved
dark line that starts at the lower left comer of the first module and extends to
1.8 em downstream and subsequently returns to the upper left corner is the line
of flow reattachment.
The recirculating bubble is located between the line of reattachment and
the leading edge of the first module. In this region, air moves upstream near
the wall to meet the incoming separated flow. The dark area extending from
the zero of scale to about 0.5 em downstream (see the scale below the photo-
graph in Figure 2) is simply an area of rather weak shear stress. Beyond the
line of reattachment, the air flows downstream toward the other modules
of the array.
The patterns clearly show that the flow field near the first module of the
array is three dimensional. This is also confirmed by the observations of Chou
and Lee [ 1988] who employed a two-module array in their experiments. Farther
downstream the air appears to flow along the direction of main stream and there
is no clear evidence of flow separation.
Another message from these photographs is that, at S/L = 0.125 (i.e., the
value at which the visualization experiments were performed), there is no clear
indication that the flow penetrates the intermodular gaps, as is often observed
for larger values of SIL (e.g., see Chou and Lee, 1988). However, the sharp
edges of the modules are expected to be conducive to flow separation, and pos-
sibly the flow separates behind the modules. This is even more possible at higher
Reynolds numbers. If this is true, the pressure results may reflect a lack of de-
pendance on Re, especially at higher Re.

AIRFLOW
)lor

FIGURE 2. Flow visualization patterns (Re = 15000, L =50.8 mm. SIL = 0.125, HJL =0.125).
Entrance Design Correlations for Circuit Boards 55

PRESSURE RESULTS
Pressure distribution along the flow at Re = 15000, SIL = 0.125, and HIL =
0.125 is shown in Figure 3. The lower diagram in this figure is prepared to show
the location of pressure taps (marked with X) relative to the modules. As shown,
the fist tap is located just upstream of the leading edge of the first module. The
ordinate in Figure 3 indicates the air pressure with respect to the atmosphere
Patm - Pi, so that an increase in the ordinate reflects a drop in pressure.
Examination of the graph indicates that pressure decreases rapidly from the
first to the second taps, while there is a slight pressure recovery for the next
pair of taps. From this point on the air pressure continuously decreases. The
rapid pressure drop between the first and second taps occurs at a point where
flow is separated from the wall and the recirculating zone has reduced the ef-
fective cross section of the airflow. Beyond this point, the flow expands and the
pressure is somewhat recovered.

01)
~
6
a5
a
4
·~

0-.t S/L = 0.125

3
I H/L = 0.125
2 Re = 15000
s1
.......,
(lj
0 5 10 15 20 25
press e ta numb

Flow 1 2 3 4 5 6 7 8 9 10
>X X X X X X X X X X

1st module 2nd module 3rd

FIGURE 3. Pressure distribution versus tap number at Re = 15000, SIL =0.125, and H/L = 0.125.
56 Air Cooling Technology

Pressure distributions along the flow direction are shown in Figures 4 to 6.

The abscissa X/Dh is the axial distance from the first pressure tap divided by
the hydraulic diameter of the duct (Dh = 2W(H + B)/(W + H +B)). The ordi-
nate K is the pressure drop coefficient as defined in Equation 1.
Examination of these figures indicates a fairly rapid drop in air pressure as
flow enters the array of modules. Further downstream, however, the pressures
at similar locations approach a fully developed state and decrease linearly. When
the numerical values of pressure data at respective locations were closely ex-
amined, it was found that a fully developed state was reached after the 2nd, 3rd,
and 4th modules in test sections numbered (1, 2, 3), (4, 5, 6), and (7, 8, 9),
respectively.
Another noteworthy feature in these figures is the independence of K from
Re when Re is greater than 4000. At higher Re, the possible flow separation
behind the modules is intense. In this case, the drag force exerted by the
modules on air flow are determined by form drag and the effect of wall fric-
tion is negligible. It is known that form drag, and thus K, is nearly indepen-
dent of Re.
Fully developed pressure drops are expressed in terms of friction factor,
f2H, in Figure 7. As stated in the preceding paragraph, there are two mecha-
nisms that govern the pressure drop in the array, namely, the form drag and
wall friction. At larger intermodular spacing, the flow is governed by form
drag, while at smaller spacing the wall friction is dominant. The data points
in Figure 7 indicate that as SIL is increased, the f2H values become less de-
pendent on Re.
Also shown in Figure 7 are the laminar flow results of Asako and Faghri
[1988] (the solid line) (at BIL = 0.375 as compared to 0.5 in this investigation),
their turbulent results (Asako and Faghri [1991] at BIL = 0.5), and the laminar
Hagen-Poiseuille flow in parallel plate channel (the dashed line just above the
upper solid line) indicated by 48/Re. There is generally a close agreement be-
tween the data, and the minor differences seen in the laminar region is perhaps
due to differences in BIL values.
As a first step in obtaining a correlation for the pressure data, we combine
the different HIL values with the pressure drop coefficient Kin Figure 8. In ad-
dition to the experimental data, the figure also shows the best fit through the
points. The equation for the solid lines, obtained through a least-squares curve
fitting procedure, are

(3)

(4)
Entrance Design Correlations for Circuit Boards 57

I
S/L ~ 0. d~s HJL = 'o.125
2 0 Re 400 -
1000 ooo
0 oo
.6.
0 4000
0
\1
10000
15000 00
oo o 0 ,6_.6..6.
1 1- oo ,6_6..6-b.t:.
0 6,6_b.b.

1 ggi~~saseeeeaeaeea
0 1- D
I I

-1 0 1 2 3 4 5 6

-1 0 1 2 3 4 5

1.2 1- S/L = 0.125 H/L = 0.50 6 .6..6-t:::.

-
6.t:::.b.b.
0.8 1-
.6.
b. !:lb. -
t:::.b.b. ooooooo
0.4 0&S$e@@@~~~~~~~~~~~ -

0.0 1- D -
I I 1 l

-1 0 1 2 3 4

X/Dh
FIGURE 4. Distribution of pressure drop coefficient in terms of axial distance for SfL = 0.125.
58 Air Cooling Technology

2 I
S/L = 0.33 H/L 1 = 0.25
0 Re = 400
.t:;;,. 1000 oo
0 4000 0
1 :- 0 10000 0
0
.c:,..t::,. -
'i7 1sooo 0 .c:,.
g &@ e a ee
.c:,. .c:,.
0 .t:;;,. @
~
0 - u
I I I I I

-2 0 2 4 6 8 10
2 I

S/L = 0.33 H/L = 0.50 o'o

0 0
0
1 1-
0 .c:,.
.t:;;,..t:;;,.
K 0 .t:;;,. .t:;;,.
0
e e e
A
6.
$ ~e
i ~ 0
0 1- D
I

-1 0 1 2 3 4 5 6 7

2 I
S/L = 0.33 = 0.75 Oo
0
0
0 0 .t:;;,./:;
1 - .c:,. 1::!.

e s a e ee
0 .t:;;,.

e
@ .t:;;,.

8 ~

0 u -

-1 0 1 2 3 4 5 6

X/Dh
FIGURES. Distribution of pressure drop coefficient in terms of axial distance for SIL = 0.33.
Entrance Design Correlations for Circuit Boards 59

3 ~------~-------~,------~------~
S/L = 0.5 H/L 0.5
0 Re = 400
2 £.;;. 1000 oo o -
0
0 tb~~~ oOoo
1 ~
\J 15ooo 0 o0° Lj.ll/:;.~ /:;.

l:l~~BesaM~"Ijl@" e -
g
0 D
I

-5 0 5 10 15

-1
T I
10 0
2 1- S/L = 0.5 H/L = 1.0 ooo -
oo
ooo ~:::..6..~:::>. 6
0
a a- ~:;,.6.

6 ~g88~g~ ..
K 1 1- ~::./:;.
ileeee
0 1- 0 -
'
-2 0 2 4 6 8 10

D.
1.0 S/L = = 1.5 I'::, .b.!::.
~:;,t::./'::,
o<><><> <>
6.~:::..6..
.6.6.aoo<>g~eeT2!Fii 0
0.5 ~s~e ~~e

0.0 a ~
-2 0 2 4 6 8
X/Dh
FIGURE 6. Distribution of pressure drop coefficient in terms of axial distance for S/L = 0.5.
60 Air Cooling Technology

0 S/L = 0.125 H/L = 0.125

l::. S/L = 0.33 H/L = 0.25
Asako & Faghri (1988)
Asako & Faghri ( 199 1)
Hagen-Poiseuille flow, 48/Re

g __ ~-~--
-·-·-·---·-·-·-.

Re
FIGURE 7. Friction factor and comparison with literature.

The standard deviations of data from the correlations 5 to 7 are, respectively,

0.0677, 0.0760, and 0.1187. With these numbers, the respective correlations
are good to within ±17.0%, ±16.9%, and±30.1%.
Further attempts to correlate all data points and to present them by a single
equation requires the definition of modified pressure drop coefficient as

(6)

The upper graph in Figure 9 presents a total of 1420 data points. The figure
indicates that the data points have been brought together relatively well, and
they are represented by the equation

K· = 87.22JfL 0.5!5 (7)

Entrance Design Correlations for Circuit Boards 61

t")
10 1
C\2 S/L = 0.125
l:'-
0
0
I 10°
....--....
t--=1
~
::r:
....__.... 10- 1
~
(') 10 1
m
(')

0
I
............... 10°
....::1
~
:I:
....__....
~ 10- 1

10 1
ro
C\2
C\2
0
I 10°
...............
....::1
~
::r::
....__....
10- 1
~

10- 5 10- 4 10- 3 10- 2 10- 1

FIGURE 8. The first step in correlating pressure data (SIL =0.125, 0.33, and 0.5).

The standard deviation of data from correlation is 0.1899 and the correlation
is good to within ±54.4%. Further, the standard deviation of coefficients 87.223
and 0.515 are, respectively, a 1 = 1.493 and a 2 = 0.6579 x w-2 , corresponding
to ±2a 1/87.223 = ± 3.4% and± 2az/0.6579 x w-2 = ±2.6% uncertainty. This
equation provides a means for estimating the pressure drop in the entrance re-
62 Air Cooling Technology

0 present work
best fit

0 Lehmann & Wirtz (1985)

10 1 ~ Souza Mendes & Santos (1987
best fit

FIGURE 9. Modified pressure drop coefficient for all data points.

gion of an array of rectangular blocks when 400 ~ Re ~ 15000, B!L = 0.5, 0.125
~ S/L ~ 0.5, and 0.125 ~HILs 1.5.
To compare this correlation with the literature, the pressure results of
Lehmann and Wirtz [1985] and those of Souza Mendes and Santos [1987]
are shown in the lower graph. There are a number of differences between
these investigations. Lehmann and Wirtz studied the problem for B/L =0.25,
0 ~ S/L ~ 1, and 0.25 ~ H/L ~ 0.75, which are somewhat different from the
geometric parameters of the present study. In addition, their test section
geometry was two dimensional. The pressure results reported by Souza
Mendes and Santos [ 1987] are for B/L = 3/8, S/L = 0.25, and H/L = 5/8.
Despite all the differences, the level of agreement seen in Figure 9 is
relatively good.
Entrance Design Correlations for Circuit Boards 63

A CORRELATION FOR HEAT

TRANSFER AND WAKE EFFECT

THE EXPERIMENTAL SETUP AND PROCEDURE

The experimental apparatus is very similar to that of pressure drop. It con-
sists of a bell-mouth inlet, a flow development section {559 mm), test section
(343 mm), a flow redevelopment section (591 mm), a venturimeter, a control
valve, and a blower. The test section and the corresponding upstream and down-
stream ducts have a rectangular cross section. The width of the cross section is
178 mm, but the height, i.e., H + B in Figure 1Oa, depending on the dimensions
of the array, varies from 19.1 to 50.8 mm. The laboratory air is drawn into the
apparatus through the bell-mouth inlet, and after flowing through the test sec-
tion (Figure 1) and the related flow passages, it exits the flow circuit through
the blower operating in suction mode.
The air flow is adjusted by the main control valve. There is also available a
bypass valve that is used for fine adjustments. The flow rate is measured by a
precalibrated venturimeter that is located downstream from the test section and
before the main control valve.
The rectangular blocks of the test section are made of copper and are po-
sitioned along the lower wall in an in-line arrangement. The geometry of the
test section, as noted in Figure 10, is identified with S, H, L, and B. In this in-
vestigation, S ranged from 6.4 to 8.4 mm, H from 6.4 to 25.4 mm, L from 25.4
to 49.8 mm, and B from 12.7 to 25.4 mm. From the combination of these di-
mensions, seven test sections were fabricated that could readily be identified
with their respective dimensionless geometric parameters BJL = 0.5; SJL =
0.128 and 0.33; and HIL = 0.128, 0.25, 0.255, 0.50, 0.75, 0.765, and 1.0. With
this arrangement, the number of copper rectangular blocks along and across a
given test section was either 5 x 3 or 8 X 5. In addition, a number of dummy
plexiglass blocks of the same dimensions were positioned downstream of the
last copper block to eliminate possible extraneous effects. The air pressure is
measured by pressure taps along the wall of the test section. Details of the lo-
cation of pressure taps and the pressure drop correlations were reported in the
preceding section.
As shown in Figure 1Oc, a thermofoil heater is attached to the bottom of
the copper block. To minimize the heat losses, a layer of silica powder was
placed at the bottom between the thermofoil heater and the plexiglass lower
wall. This arrangement was repeated for all copper blocks of the array.
Moreover, the outside surfaces of the test section, including the upstream and
downstream ducts, were completely wrapped and insulated by 100 mm of
common glass wool insulation.
Before the onset of a data run, the proper test section is selected and as-
sembled. Then, to prevent air from leaking into the test section, all suspected
joints are sealed by silicon rubber and thoroughly tested for leaks with the aid
of soap solution. Subsequently, the flow and power to thermofoil heater are
64 Air Cooling Technology

INSULATED

AIRFLOW

COPPER BLOCKS

AIRFLOW DO
DO
(b)
COPPER BLOCKS
L
THERMOFOIL HEATER
/
SILICA POWDER
PLEXIGLASS

•vv vvvvvvvvvvvvvvvvv
xxxxxxx:<xxxxxx~'·

(C) INSULATION

FIGURE 10. Schematic of the test section, (a) side view, (b) top view, (c) a typical heated block.

turned on and adjusted for a predetermined Reynolds number. It typically takes

30 minutes to warm up the apparatus.
In a typical data run, only one copper block is heated at a time and the re-
maining blocks are adiabatic. The input power of the thermofoil heater is ad-
justed at such a level that the temperature difference between the block and
ambient remains constant. This temperature difference varies from 30 to 50°C
depending on the run. Once the heated block reaches the steady state, the var-
ious parameters are recorded. These parameters are block surface temperature
(the average temperature of three surface-mounted thermistors), inlet air tem-
perature (typically 20 to 25 ±0.1 °C), air temperature downstream of the test
section (usually I oc higher than the inlet temperature, ±0.1 °C), pressure drop
across the venturimeter (0.05017 to 9.80 ±1 x IQ-5 mm Hg), inlet pressure of
the venturimeter relative to ambient (0.01 to 13 ±1 x I0-5 mm Hg), ambient
barometric pressure (740 to 770 ±0.1 mm Hg), thermofoil heater voltage (5 to
Entrance Design Correlations for Circuit Boards 65

35 ±0.3 V), and resistance of the thermofoil heater (30 to 32 ±0.3 Q). Except
for the barometric pressure, all other pressures were measured by a pressure
transducer and an electronic manometer. All pressure and temperature readings
were performed by an IBM PC. The computer scanned the temperature and
pressure signals at a scan rate of (50/22) per second.

DATA REDUCTION
The convective heat transfer coefficient is evaluated from, h = Q/[A(Th-
Tin)], where Q = V2/R- (conduction losses+ radiation losses). In these equa-
tions, Q is the rate of heat convection from the heated block to air stream, A is
surface area of the block, T h mean surface temperature of the block, Tin inlet
air temperature, V voltage across the thermofoil heater, and R is the therrnofoil
electrical resistance.
Conduction losses were evaluated from a two-D and three-D numerical sim-
ulation of heat transfer in the copper block and the adjacent walls. However,
due to low thermal conductivity of plexiglass, the effect of longitudinal con-
duction through the base board was negligible. Radiation losses were estimated
from a simplified model where the heated block is treated as a small radiating
object surrounded by a large environment. The results suggested that the con-
duction losses were 10% of V 2/R and the radiation losses, with an emissivity
value of0.15 for copper, were always less than 1% ofV2/R. The h values were
subsequently nondimensionalized and expressed in terms of Nusselt number
with the conventional definition, Nu =hL/k.
The heat transfer results will be presented in terms of Reynolds number,
Re =VH/v, based on mean air velocity, V, in the bypass channel above the
blocks and the dimension H (Figure lOa). The velocity V is obtained by di-
viding the air volume flow rate, as measured by the venturimeter, WXH
(W= 178 mm).

UNCERTAINTY ANALYSIS FOR HEAT TRANSFER DATA

The parameters recorded during the calibration process and the respective
uncertainties are: barometric pressure, ±0.1 mm Hg; venturimeter inlet gage
pressure, ±10-5 mm Hg; pressure drop across the venturimeter, ±10-5 mm Hg;
pressure drop across the laminar-flow element, ±1 o-5 mm Hg; and inlet air tem-
perature, ±0.1 oc. In addition to these values, the uncertainty of laminar-flow
element is estimated as ±1% of volume flow rate.
Other variables used to evaluate the final uncertainties are: duct width, ±0.1
mm; dimension H (Figure lOa), ±0.1 mm; block height B, ±0.1 mm; inlet and
throat diameters of venturimeter, ±0.1 mm; test section gage pressure, ±lQ-5
mm Hg; block surface temperature, ±0.1 °C; thermo foil voltage, ±0.3 V, and
thermofoil resistance, ±0.3 n. Using the guidelines of Kline [1985] and
Abernethy et al. [1985], the mean uncertainty ofNu is estimated as 4.8%, with
the maximum uncertainty being 7.9%. The uncertainties of other parameters
are given in the figure captions.
66 Air Cooling Technology

HEAT TRANSFER RESULTS AND DISCUSSION

The entrance region adiabatic heat transfer coefficients are presented in
Figures 11 and 12. The ordinate is nondimensionalized with the periodic fully
developed Nusselt number, Nurd, at the Reynolds number corresponding to the
same set of data. The coordinate X is measured along the flow with X = 0 cor-
responding to the leading edge of the first block. The family of curves in Figure
11 are for S/L = 0.128 and HIL ranges from 0.128 to 0.765, while those in
Figure 12 are for S/L = 0.33 and HIL = 0.25 to 1.0. It is evident from the fig-
ures that the distribution of Nu/Nurd is almost independent of Reynolds num-
ber and HIL. A careful examination of the numerical value of the individual
data points show that the mean deviation of data from the best fit (solid lines)
is 5.3%, and the maximum deviation does not exceed 10.4%.
The family of curves seen in Figures 11 and 12 show a common trend. Starting
with a relatively large value, the Nusselt number decreases rapidly and approaches
the fully developed value. This trend is reminiscent of heat transfer coefficient in
the entrance region of a duct. However, in addition to the high temperature gra-
dients at the entrance, there exists a strong flow separation zone on the first block
of the array (Figure 2 and Molki et al. [1993]). It is well known that the points of
flow reattachment have higher transfer coefficients (e.g., see Molki and Hashemi
[1992]). Therefore, this flow separation and the subsequent flow reattachment
has an enhancing effect on the block-averaged heat transfer coefficients reported
in Figures 11 and 12. It should be noted that the extent of enhancement is even
more pronounced at special locations such as the line of reattachment.
The data reported in Figures 11 and 12, a total of 176 points, are presented
in Figure 13. Also shown in this figure are the results of Anderson and Moffat
[1990] for H/L =0.102-0.735, S/L = 0.273, and BIL = 0.204. Despite all the
differences between the two investigations, it is seen that all data follow a cer-
tain trend. This suggests that a curve fit to all data may be a proper way of rep-
resenting the Nusselt number in the entrance region of the array. The solid line
in Figure 13 is the least squares best fit to the data with the equation

(8)

The mean and maximum deviation of data from this equation are, respectively,
4.8% and 19.8%, while the mean and maximum experimental uncertainty is
estimated as 16.1% and 44.2%. These figures suggest that Equation 8 is a pos-
sible correlation for Nu/Nufd· According to this equation, if the Nu!Nufd = 1.01
marks the beginning of the periodic fully developed region, then the entrance
region of the array is identified with 0::::; X/Dh ::::; 6.53. Therefore, Equation 8 is
particularly useful when X ::::; 6.53 Dh. Otherwise, the fully developed results
can be used.
To facilitate the use of Equation 8 in practical applications, the periodic fully
developed Nusselt numbers had to be written in a convenient way. After ex-
tensive examination of data and several curve-fitting attempts, a modified
Entrance Design Correlations for Circuit Boards 67

2.0
H/L = 0.128
0 Re = 3000
1.6 .6. 5000
0 7000
\] 9000
<) 12000
1.2 15000
*

0.8
0 2 3 4 5 6
2.0

H/L = 0.255
1.6
Nu/Nu1d

1.2

0.8
0 2 3 4 5 6
2.0

H/L = 0.765
1.6

1.2

0.8
0 2 3 4 5 6

X/Dh

FIGURE H. Entrance adiabatic heat transfer coefficients for S/L =0.128 (mean and maximum
uncenainty = 16.1%, 44.2%).

Nusselt number emerged as Nui'd= 2.819 Nurd Re-0 ·607 (S/L)- 0295 , which is pre-
sented in Figure 14. Also seen in this figure are the results from the literature.
The figure suggests that the modified Nusselt number is a suitable parameter
that correlates the data fairly well. The correlation is obtained from a least
squares curve fit to all data with the equation

(9)
68 Air Cooling Technology

2.0

H/L = 0.25
0 Re = 3000
1.6 /::;. 5000
0 7000
\1 9000
1.2
§
E:j

0.8
0 2 3 4 5 6
2.0

H/L = 0.50
1.6

1.2

@
0.8
Nu/Nufd 0 2 3 4 5 6
2.0

H/L = 0.75
1.6

1.2

0.8
0 2 3 4 5 6
2.0

H/L = 1.0
1.6

1.2

0.8
~ ~ ij 0 ! ij ~

0 2 3 4 5 6
X/Dh

FIGURE 12. Entrance adiabatic heat transfer coefficients for S/L =0.33 (mean and maximum
uncertainty= 16.1%, 44.2%).
Entrance Design Correlations for Circuit Boards 69

2.0 .----r------,r----.----,-----.
0 present. work
e Anderson & Moffat, 1990
best fit
1.6

1.2

0.8 t___ ___t_ _ ___L_ _ _~----'-------'

0 2 4 6 8 10

FIGURE 13. Entrance adiabatic heat transfer coefficients for all data and comparison with lit-
erature (mean and maximum uncertainty= 16.1 %, 44.2%).

The mean and maximum deviation of data from Equation 9 are, respectively,
12.8% and 41.1 %.
There are a number of differences between the present results and those of
the literature as seen in Figures 13 and 14. Anderson and Moffat [1990] per-
formed their experiments for H/L =0.102 to 0. 735, SIL = 0.273, and BIL = 0.204,
while those of Sridhar [1990] were H/L =0.5 to 5.5, SIL =0.33 to 0.5, and BIL
=0.5. Sparrow et al. [1982] did not use the electrically heated blocks in their ex-
periments, instead, they employed a mass transfer technique and the heat-mass
analogy to obtain the heat transfer coefficients. Their experiments were con-
ductedforHIL=0.625,SIL=0.25,andBIL=0.375.0ntheotherhan d,Hollworth
and Fuller [1987] results in Figure 14 are measured for a staggered array of
blocks with H/L = 0.25 and 0.5, SIL =1.0, and BIL =0.25. Despite all these dif-
ferences, it is noteworthy that the modified Nusselt number has correlated the
data well. Therefore, Equations 8 and 9 are suggested as a complete set of cor-
relations for the adiabatic heat transfer coefficient in array of heated blocks.
Adiabatic temperature of the first block situated immediately behind the
heated block is presented in Figure 15 as a function ofRe with HILand SILas
parameters. This dimensionless temperature, or the so-called wake effect, is de-
fined as

(10)

In this equation, Tal, T; 0 , and Th are, respectively, the adiabatic (unheated) tem-
perature of the first block, the inlet air temperature, and the temperature of the
70 Air Cooling Technology

0 present work
lD
6. Sparrow et al., 1982
"'
!\.!
0
0 Anderson & Moffat, 1990
I
,--.....
\7 Hollworth & Fuller, 1987
.....:1
10 1 0 Sridhar, 1990
'-....
(/)
best fit
l'-
0
<D
ci ~
I
Q)
p::
:E
:::s
z
,....
0)

oc:
(\]
10°

H/L
FIGURE 14. Distribution of the modified periodic fully developed Nusselt number (mean and
maximum uncertainty= 7.8%, 15.0%).

heated block. It is evident from the figure that the effect of HIL on 81 is negli-
gible. However, the effect of the interblock spacing, SIL, is somewhat impor-
tant. In fact, increasing SIL shifts the location of the first adiabatic block further
away from the thermal wake of the heated block and, therefore, the tempera-
ture is reduced. The results of Arvizu and Moffat [1982] for SIL = 2.0 and 3.0,
Wirtz and Dykshoom [1984] for S/L = 1, and Sridhar et al. [1990] for SIL =
0.33, are also consistent with this observation.
By defining the modified wake effect according to

e~ = 2.680 8 1 Re 0·168 (11)

the results are brought together in Figure 16. In this figure, the solid line is the
curve fit to all data with the equation

8j = 0.998(~)"".{) 540 (12)

The mean and maximum deviation of data from the solid line are, respectively,
6.3% and 34.4%. The uncertainty analysis indicates that the mean and maxi-
Entrance Design Correlations for Circuit Boards 71

I I

0 H/L = 0.255
0.128

I I

Re

present work: Arvizu & Moffat, 1982

0 S/L = 0.125 0 S/L = 3.0
6 S/L = 0 .33 0 S/L = 2.0
f- -

0
6 0
~
0 0 ooooo oo
6. 6 6. 6. 6. 6.!:;.
0
10- 1 - 0 -
0
8 00 0
0

I I

10 3 10 4
Re
FIGURE 15. Thermal wake effect for the first adiabatic block (mean and maximum uncertainty
= 1.4%, 2.8%).
72 Air Cooling Technology

0 present work
10 1 6 Arvizu & Moffat, 1982
best fit
co
....
co
ci
Q)
~

CJ:)

0
co
co
C\1
10°

S/L
FIGURE 16. Modified wake function (mean and maximum uncertainty= 28.3%, 30.1%).

mum experimental uncertainty of fl1 are 28.3% and 30.1 %, respectively. Since
the mean deviation is well within the mean uncertainty and the maximum de-
viation and uncertainty have comparable values, it may be suggested that
Equation 11 be considered for the range 0.125 ::;; SIL::;; 3.0, which extends be-
yond the range of SIL in the present investigation.
The wake effect for other downstream blocks is presented in Figure 17. In
this figure, fiN is defined as eN= (TaN- T;n)/(Th- T;.). Here, the subscript N
refers to the Nth adiabatic block situated downstream of the heated block, with
N =0 corresponding to the heated block itself. During the wake effect experi-
ments, the heated block was located at the centerline and at either the first (lead-
ing) or at the third row of the array. The experimental runs were also carried
out for different geometries. As evidenced from the data in Figure 17, the lo-
cation of the heated block and the geometry does not have a marked effect on
the distribution Of flN/flJ.
There are a number of investigators who have correlated the flN/8 1 data as
1/N (e.g., Arvizu and Moffat [1982], Moffat et al. [1985], and Sridhar et al.
[1990]) or (1/N)m where m is a function of Re (Wirtz and Dykshoorn [1984]).
The major difficulty with this correlation is that when the number of blocks is
large (N -7 oo), 1/N approaches zero. In an insulated duct with a number of
heated blocks, the temperature of the downstream adiabatic blocks approaches
the mean temperature of the air flow, which is somewhat higher than the inlet
temperature. Therefore, a suitable correlation is the one which incorporates this
Entrance Design Correlations for Circuit Boards 73

0 S/L = 0.128, H/L = 0.255

1 .2 1st module heated
S/L = 0.33, H/L = 0.50
3rd module heated
S/L = 0.33, H/L = 0.50
1st module heated
0.8

best fit, eq (13)

0.4

1/N
0.0
0 2 4 6 8
ROW NUMBER
FIGURE 17. Thermal walce effect for other downstream blocks (mean and maximum uncertainty
=4.0%, 44.0%).

limiting case. In the light of the foregoing discussion, the wake effects are cor-
related as

~~ =0.151 + 0.849 N-u 14 (13)

Figure 17 shows that Equation 13 is a better fit to data than liN.

Attention is next turned to the wake effect on the blocks situated in the flank-
ing columns of the heated block. In Figure 18, the ordinate, 8tN/SN is the ratio
of the wake effect for the flanking block at the same row, i.e., SIN= (TfN- Tin)/(Th
-Tin), to eN defined earlier. It is seen from the figure that farther away from the
heated block e!N increases with respect to eN and is independent of Re. In the
range of N = 1 to 4, e!N varies from 5.8% to 27.5% of eN.
The wake effect for the flanking column may be represented by

~: = 0.0575 Nuzs (14)

which is a least-squares fit to all data. The mean and maximum deviation of
data from this equation are, respectively, 7.8% and 22.6%, while the mean and
maximum uncertainty are 13.0% and 16.0%.
74 Air Cooling Technology

0.6

0 Re 3000
~ 5000
0 7000
"V 9000
0.4 best fit
z
CD
--........
z
c:c-

0.2

0.0
0 2 3 4 5
ROW NUMBER
FIGURE 18. Thermal wake effect for the blocks in the flanking columns (mean and maximum
= 13.0%, 16.0%).

Next, attention is turned to the application of the above correlations for pre-
dicting the temperature of the circuit-board elements. The application is lim-
ited to the geometry and the range of parameters considered in this paper.
However, they can also be used as first approximation for a wider range of
geometry and parameters in practical problems.

APPLICATION OF THE PROPOSED CORRELATIONS

Based on the aforementioned correlations, a marching procedure is proposed
for estimating the operating temperature of elements in a circuit-board with
random heating using the superposition approach (Arvizu and Moffat [1982]).
In this procedure, the temperature of each block is considered to be composed
of two components. One is due to self-heating (LlT,h = Th- T.), the other is due
to the thermal wake of the upstream heated blocks (LlT we). It should be noted
that every downstream block is affected by the thermal wake of every upstream
block situated in the same column or in the flanking columns. Therefore, the
Entrance Design Correlations for Circuit Boards 75

thermal wake component will be evaluated from the summation of the wake
effects of all the upstream heated blocks.
The algorithm is started by providing the geometrical parameters (number
of rows and columns, L, B, S, H, W), air inlet velocity, Yin, inlet air tempera-
ture, Tin, and power dissipation in each block. These values are used to obtain
S/L, HIL, X/Dh, and Re. Then, the temperatures are computed according to the
following steps:

1. Adiabatic heat transfer coefficient, h, is obtained from Equations 8 and

9 for all blocks of the array.
2. Operating temperatures of the blocks in the first row are evaluated from,
(Power- heat losses)= hA(Th- T.), where Power is the heat generation
within the block. It should be noted that the adiabatic temperature of the
blocks situated in the first row is equal to the inlet air temperature, i.e.,
Ta==Tin·
3. Wake effect of the heated blocks on all downstream blocks situated in the
same column, i.e., ATwe,s =TaN- Tin, is obtained from Equations 10-13.
In this way, for every downstream block a different ATwe,s value is eval-
uated and assigned to that block.
4. Wake effect of the heated blocks on the flanking columns is evaluated
from Equation 14. Only one column to the left and one to the right of the
respective heated block is considered. The wake effect on other columns
has been found to be negligible. In this manner, two additional wake ef-
fects, namely ATwe,l and ATwe,r, are found and assigned to the block. In
this procedure, for heated blocks in the first row, three AT's are assigned
to each of the blocks situated downstream.
5. Adiabatic temperature of the blocks in the second row are estimated from
T. =AT we+ Tin, where ATwe= LATwei, and i = s,l,r.
6. Consider the second row of heated blocks. Using the adiabatic tempera-
tures of the previous step, repeat steps 2 to 5 to find the operating tem-
peratures of the blocks in the second row and their respective wake effects
on downstream blocks. At the end of every cycle of iteration, the adia-
batic temperature of the blocks in the next row is obtained from all the
previously obtained ATwe); according to step 5.

A USER-FRIENDLY PROGRAM
FOR PREDICTION OF ARRAY TEMPERATURE

The aforementioned procedure is incorporated into a user friendly FOR-

TRAN program that is available from one of the authors (MF) upon request. A
sample output from this computer program is shown in Figure 19. The figure
shows the top view of the array. The cross-hatched blocks are heated while the
remaining blocks are adiabatic. The first of the four numbers on the heated
blocks is the power generation in watts. This number is not seen for the un-
76 Air Cooling Technology

30.46 30.38 29.49

32.72 35.20 32.53

7.4% 15.8% 10.3%

34.29 30.91
36.93 34.03
7.7% 10.1%

POWER
GENERATION
32.48
34.39
5.8%

MEASURED
TEMPERATURE
30.59 35.58
28.29 40.07
8.1% 12.6%
PREDICTED
TEMPERATURE

26.30
27.57
27.57 PERCENTAGE
2.6% 9.5% TEMPERATURE
DIFFERENCE

FIGURE 19. A typical temperature prediction for the elements of a circuit board. The data are
for L =49.8 mm, B =25.4 mm, S =6.4 mm, H =38.1 mm, V;. =!.24 mls, and T;n 27.57°C,=
corresponding to S/L =0.128, HIL =0. 765, BIL =0.5, and Re =5000

heated blocks. The last three numbers on the heated and unheated blocks indi-
cate the measured and computed temperatures in degree C, and their difference,
respectively. Examination of difference percentages indicates that the differ-
ence between experiment and prediction is in the range from 2.6 to 21.8%, and
the mean difference is 11.0%.
Entrance Design Correlations for Circuit Boards 77

CONCLUSION

The research reported here is an experimental investigation of heat trans-

fer and pressure drop in the entrance region of an in-line array of heated
blocks. The study is aimed at obtaining correlations for pressure drop, adia-
batic heat transfer coefficient, and thermal wake effects so that the operating
temperature of the real circuit boards with a similar geometrical layout can
be estimated. However, since the real electronic components are irregular
both in placement and size, the results reported in this paper may not be ap-
plicable to highly irregular layouts. In all of the experiments the working
fluid was air.
Flow visualization indicated the presence of a highly separated flow on top
of the first module followed by flow reattachment and recirculation. It was fur-
ther found that flow reattachment is curved, and thus the flow field is thor-
oughly three dimensional. Farther downstream, the flow visualization could not
present a clear evidence of flow separation. However, the pressure results in-
dicate that at higher values of Re, the flow is possibly separated behind the
downstream modules.
Pressure drop results are presented and correlated through the definition of
a modified pressure drop coefficient. This correlation is good to within ±54.4%
and is a convenient tool for estimating the pressure drop in arrays of rectangu-
lar blocks.
The entrance heat transfer coefficients revealed that Nu!Nurd is a function
of X!Dh, and its dependence on geometrical parameters and Reynolds number
is within the limits of the experimental uncertainty. These coefficients were
successfully correlated by defining a modified Nusselt number. The correlation
is obtained by a curve-fitting procedure that includes the data from other in-
vestigators, and thus is expected to be valid over a wider range than that of the
present work.
Temperature measurements indicated that the thermal wake effect, 81, for
the first adiabatic block behind the heated block is a function of Re and SIL
and is insensitive to HIL. A modified wake effect, Oi, absorbed the effect of Re
and correlated the results as a function of SIL.
The wake effect for other blocks, when presented as SN/8~, are found
to be independent of the position of the heated block and geometry. It was
further found that the commonly employed 1/N function is not a well
representation of 8N/81 distribution. Instead, it was correlated as 0.151 +
0.849 N- 13 ' 4, which gives a more realistic asymptotic value for SN/0,
when N ~ oo. Correlations were also found for the blocks situated in the
flanking columns.
The present correlations are incorporated into a FORTRAN program to pre-
dict the operating temperatures of the circuit boards of similar geometry with
random heating. The results indicate that the temperatures can be predicted with
sufficient accuracy for practical application.
78 Air Cooling Technology

APPENDIX A: NOMENCLATURE

A1 inlet cross-sectional area of venturimeter, (n/4)Df

B module height
Co flow coefficient of venturimeter, Co= Q[(DdD 2) 2 - 1] 0·5/[A 1 (2AP/p 1) 0 ·5]
Dh hydraulic diameter, mm, 2W(H + B)/(W + H +B)
D1 inlet diameter of venturimeter
D2 throat diameter of venturimeter
f2Hfriction factor
h mean heat transfer coefficient for a rectangular block
H distance between top surface of modules and the opposite wall
K pressure drop coefficient
K+ modified pressure drop coefficient
L plane dimension of square module
Nu Nusselt number, hL/k
Pi air pressure at the ith pressure tap
Po air pressure at the first pressure tap
Q volume flow rate
Q rate of heat convection
R thermofoil electrical resistance
Re Reynolds number, VH/v
S intermodule spacing
T temperature, oc
V voltage across the thermofoil heater, air mean velocity in the bypass
channel
W duct width, W = 178 mm
X axial coordinate, X = 0 at the leading edge of the first block

GREEK SYMBOLS
e wake effect
v kinematic viscosity
p density

SUPERSCRIPTS
+ modified parameter

SUBSCRIPTS
a! refers to adiabatic temperature of the first block behind the heated block,
Equation 5
aN refers to adiabatic temperature of the Nth block behind the heated block
fd refers to periodic fully developed, Equation 4
fN refers to flanking column, Equation 9
h refers to heated block, Equation 5
m refers to inlet condition, Equation 1
Entrance Design Correlations for Circuit Boards 79

N refers to the Nth block behind the heated block, Equation 8

sh refers to self heating
we refers to wake effect

REFERENCES
Abernethy, R.B., Benedict, R.P., and Dowdell, R.B. 1985. ASME measurement uncertainty, ASME
J. of Fluids Eng., 107:161-164.
Anderson, A.M., and Moffat, R.J. 1990. A new type of heat transfer correlation for air cooling of
regular arrays of electronic components, in Proc. of ASME Winter Annu. Meet., 27-39.
Anderson, A.M., and Moffat, R.J. 1991. Direct air cooling of electronic components: Reducing
component temperatures by controlled thermal mixing, ASME J. ofHeat Transfer, 113:56--62.
Arvizu, D.E., and Moffat, R.J. 1982. The use of superposition in calculating cooling requirements
for circuit board mounted electronic components, in Proc. of the 32nd Electron. Components
Conf., IEEE, 32:133-144.
Asako, Y., and Faghri, M. 1988. Three-dimensional heat transfer and fluid flow analysis of arrays
of square Blocks encountered in electronic equipment, Numerical Heat Transfer, 13:481-498.
Asako, Y. and Faghri, M. 1991. Parametric study of turbulent three-dimensional heat transfer of
arrays of heated blocks encountered in electronic equipment, Heat Transfer in Electron.
Equipment, HID, 171:135-141.
Chou, J.H. and Lee, J. 1988. Reducing flow nonuniformities in LSI packages by vortex genera-
tors, in Cooling Technology for Electronic Equipment, W. Aung, Ed., Hemisphere,
Washington, D.C., ll3-124.
Faghri, M., Molki, M., Chrupcala, J., and Asako, Y. 1995. Entrance analysis of turbulent flow in
an array of heated rectangular blocks, IEEE Transactions on Components, Packaging, and
Manufacturing, Part A, Vol. 10, No. 3.
Faghri, M., Ray, A., and Sridhar, S. 1991. Entrance heat transfer correlation for air cooling of ar-
rays of rectangular blocks, Heat Transfer Enhancement in Electronics Cooling, ASME HTD
183:19-23.
Garimella, S.V. and Eibeclc, P.A. 1992. Onset of transition in the flow over a three-dimensional
array of rectangular obstacles, J. of Electron. Packag., 114:251-255.
Hollworth, B.R. and Fuller, H.A. 1987. Heat transfer and pressure drop in a staggered array of air-
cooled components, in Proc. of the Int. Symp on Cooling Techno!. for Electron. Equipment,
Honolulu, 732-748.
Kang, S.S. 1992. The thermal wake function for rectangular electronic modules, National Heat
Transfer Conf, Open Forum, San Diego, CA.
Kline, S.J. 1985. The purposes of uncertainty analysis, ASME J. Fluids Eng., 107:153-160.
Lehmann, G.L. and Wirtz, R.A. 1985. The effect of variations in stream-wise spacing and length
on convection from surface mounted rectangular components, ASME HTD, 48:39-47.
Moffat, R.J. and Anderson, A.M. 1988. Applying heat transfer coefficient data to electronics cool-
ing, ASME Winter Annu. Meet, Chicago, IL.
Moffat, R.J., Arvizu, D.E., and Ortega, A. 1985. Cooling electronic components: forced convec-
tion experiments with an air-cooled array, Heat Transfer in Electronic Equipment, ASME
HTD 48:17-27.
Molki, M., Faghri, M., and Ozbay. 0. 1993. A new correlation for pressure drop in arrays of rec-
tangular blocks in air-cooled electronic units, Natural and Forced Convection, ASME J. of
Fluids Eng., 116:856--886.
Molki, M. and Hashemi, A. 1992. Turbulent convective mass transfer downstream of a perforated
baffle blockage. Int. J. Heat and Fluid Flow, 13(2): 116-123.
80 Air Cooling Technology

Peterson, G.P. and Ortega, A. 1990. Thermal control of electronic equipment and devices, in
Advances in Heat Transfer, Vol. 20, J.P. Hartnett and T.F. Irvine, Jr., Eds, Academic Press,
New York.
Souza Mendes, P.R. and Santos, W.F.N. 1987. Heat-transfer and pressure drop experiments in air-
cooled electronic-component arrays, J. Thermophys., 1:373-378.
Sparrow, E.M., Molki, M., and Chastain, S.R. 1981. Turbulent heat transfer coefficients and fluid
flow patterns on the faces of a centrally positioned blockage in a duct, Int. J. Heat Mass
Transfer, 24:507-519.
Sparrow. E.M., Niethammer, J.E., and Chaboki, A. 1982. Heat transfer and pressure drop charac-
teristics of arrays of rectangular modules encountered in electronic equipment, Int. J. Heat
Mass Transfer, 25:961-973.
Sparrow, E.M., Vemuri, S.B., and Kadle, D.S. 1983. Enhanced and local heat transfer, pressure
drop, and flow visualization for arrays of block-like electronic components, Int. J. Heat Mass
Transfer, 26:689-699.
Sridhar, S. 1990. Heat transfer and fluid flow behavior in arrays of rectangular blocks encountered
in electronic equipment, Master's thesis, University of Rhode Island.
Sridhar, S., Faghri, M., and Lessmann, R.C. 1990. Heat transfer behavior including thermal wake
effects in forced air cooling of arrays of rectangular blocks, ASME HTD, 153:15-26.
Tai, C.C. and Lucas, V.T. 1985. Thermal characterization of a card-on-board electronic package,
Heat Transfer in Electronic Equipment, ASME HTD, 48:49-57.
Wirtz, R.A. and Dykshoom, P. 1984. Heat transfer from arrays of flat packs in a channel flow, in
Proc. of4thAnnu. Int. Electron. Packag. Soc., Baltimore, 247-256.
Wirtz, R.A. and Chen, W. 1991. Laminar-transitional convection from repeated ribs in a channel,
Heat Transfer in Electronic Equipment, ASME HTD, 171:89-94.
 Chapter 3

FORCED AIR COOLING

OF LOW-PROFILE PACKAGE ARRAYS
R.A. Wirtz

CONTENTS
Introduction ..... .. . . .. .. ... . ... . . .. . . .. . . ........ . . . .. .. . .82

Array Geometry . .... ... ... . .. . . . . .. . . .. . . . . .. . ...... . . . . . .. 82

Convection Processes ...... . . . . .... . ... . . . . . . . . ..... . ... . . .. .83

Estimation of Package Temperature . . ....... . ........ .. . . .. . .. . .85

Determination of hk and e .. ..... ... .... .. .... .. .... ..... .... .86
Prototype Experiments . .. . . . ... . . . . .. .. . . . ...... . . .... ... . .86
Model Experiments . . . . .. . . .. .... . . . .. .. ......... . . . .... . .87

Package Heat Transfer Uniform In-Line Arrays . . ......... . .. . . . . . .89

Adiabatic Heat Transfer Coefficient . . . . . . .. . . . . . . . . .. . . ... . . . .89
Thermal Wake Function ... . .. .. . .. . . .. . . . . .. . . . .. ... . .. . . .. 92
Example Calculation .. . . . .. .. . . . . .... . .. . . . ...... . ... . .. . .94

Coolant Pressure Drop-Uniform In-Line Arrays . . . . .. . . . . .. . .... . .96

Nonuniform Arrays ...... . . . . .. . .. . ... .. ........ . .. .. . .. . ... 98

Effect of Array Stagger . . . . . .. . . . .. . .... . . . .. . . . ....... . .. .98
Effect of Nonuniform Package Size ... .. . ........ . ... . . ... . . . .98
Effect of Heat Sink Placement . . . . . . .. .. . . . . . . . . . .. .. ..... .. .98

Conclusion ............ . ..... ... .. . . . . . .. ............. . ....99

Appendix A: Nomenclature . . . . . .. ... ... .. . . ... . . .. ... . .. .. ... 99

References . . . ... . . ... .... . .. . ... .. ... . . . . . . . . . . .. . ...... .1 00

For Further Information . . . . . . . . . . . . . . . . . . . . . . . . . . 10

. .1. . .. .. . .

0-8493-9447 -3196/$0.00+$.50
© 1996 by CRC Press. Inc. 81
82 Air Cooling Technology

INTRODUCTION

Air continues to be the preferred coolant for most electronics installations.

Direct air cooling, where fan-driven, conditioned air is passed directly over
board-mounted packages, is a convenient and popular cooling scheme. If cir-
cuit boards are rack mounted, the space between boards acts as a conduit to di-
rect the coolant over and around modules, allowing for convective cooling of
both the modules and the pc-board.
In these applications, heat flows from the device to the ultimate sink via a
complicated path. On the one hand, heat is conducted directly to the package
surface, where it is removed by convection to the coolant and radiation to the
surroundings. The remainder of the energy is conducted through the mounting
substrate and attachment leads to the pc-board where it is either conducted to
the system enclosure, or convection processes on the back of the board or be-
tween modules remove it. A complete thermal analysis to predict device oper-
ating temperatures would require consideration of all of the heat transfer paths
described above. However, if the convection mechanisms are suitably quanti-
fied, the analysis can be completed using a standard conduction analyzer in
conjunction with separate calculations for the convection and radiation com-
ponents of the heat transfer process.
The objective in this chapter is to describe the convection process occurring
in a class of air-cooled packaging configurations where conduction and radia-
tion heat transfer are absent. Heat transfer correlations and calculation proce-
dures that can be used to estimate package surface temperatures and coolant
flow pressure drops are presented.

ARRAY GEOMETRY

Packaging configurations are diverse, usually including a mix of package

sizes on a given pc-board. In addition, some packages may include heat sinks,
and electrical cabling or card separators may partially obscure the flow. In order
to develop a systematic description of convection in electronics, we restrict our
attention to unobstructed flow over uniform arrays of equal-sized elements.
Where information regarding the effect of array nonuniformity is available, it
is pointed out
Consider a uniform, in-line array of modules mounted on a pc-board, as
shown in side view in Figure 1. Let x and z denote the stream-wise and cross-
stream directions, andy is the coordinate perpendicular to the plane of the fig-
ure. Each array element has height, a, length, Lx, and width Ly. The
corresponding spacing between elements is ex and Cy. We can define the array
(area) packaging density as

(1)
Forced Air Cooling of Low-Profile Package Arrays 83

zlJr--------.
,
X

FIGURE 1. Side view of a uniform, in-line array of convectively cooled electronic packages.

where Sx = Lx + Cx and Sy = Ly + Cy are the packaging pitch in x and y direc-

tions, respectively. The packaging density can range from zero to unity. For
square packaging arrangements (Lx = Ly, Cx = Cy) the packaging density is given
by (L!Sf Package lengths range from about 10 mm to 80 mm, and package
aspect ratios (Lia) can range from 2 to 20. Package spacing, c, is typically
4 mm- 16 mm. Array densities typically range from 0.25 to 0.9.
Package dimensions and spacings, in conjunction with the pc-board spac-
ing and array size, comprise the principal length scales that are known to in-
fluence cooling performance. The effect of other geometric features, such as
package edge radii, the presence of lead attachments, etc., are assumed to have
only a small impact on convection.

CONVECTION PROCESSES
As shown in Figure 1, the coolant (air) at temperature, Tc, and average ve-
locity, Vc, flows into a channel of height, H. Such a channel might be formed
by two adjacent pc-boards, or by a board bounded by the system enclosure.
Array elements have surface temperature, Tt<, and the rate of convective heat re-
moval is qk, where k is the row number in the array. Upon encountering the
array, the entry flow divides. Part is diverted to the space above the modules
(the bypass flow), and the remainder is channeled into the spaces between array
elements (the array flow). The extent of interaction between the bypass flow
and the array flow is not generally known. It has been shown for regular, in-line
arrays containing cubical array elements (where the element height, a, is not
negligible) that the fluid velocity in the array is significantly reduced relative to
that in the by -pass region [Arvizuand Moffat, 1982]. Under these circumstances,
convective cooling of the board surfaces between packages probably plays only
a secondary role in package cooling. On the other hand, board-convection is
probably a significant part of the overall heat transfer problem when package
heights are very small, such as in surface-mount applications.
The methods of analysis that follow are directed at geometric situations
where the package height is sufficiently large that the coolant flow is modified
away from the "smooth duct" condition. Situations where a is small can be an-
alyzed using conventional superposition techniques [Kays and Crawford, 1993]
in conjunction with heat transfer correlations suggested by Lehmann and
Pembroke [1991 a].
84 Air Cooling Technology

A good description of the bypass flow is given by the local average veloc-
ity, V, which is the average velocity based on the minimum cross sectional area
of the flow passage.

(2)

Since both a/Hand Lv!Sv are positive and less than unity, Equation 2 shows that
the bypass flow is accelerated relative to the entry flow. This is, of course, due
to the extra blockage that the array elements present.
Consider air cooling pc-boards spaced 20 mm apart with a fan-driven flow
that can range from 200 fpm to 2000 fpm ( 1m/s to 1Ornls ). The package heights
are 4 mm and the packaging density is 60%. Under these conditions the aver-
age bypass velocity is roughly 20% greater than the inlet average velocity, and
the channel Reynolds number for the bypass flow, Re = V2(H- a )lv, will range
from approximately 2,400 to 24,000. Thus, we expect convection process, par-
ticularly on the tops of the array elements, to have mostly turbulent character-
istics.1 Since the bypass flow and the array flow velocities are of different
magnitudes, a shear layer that separates the two regions forrns near the tops of
the array elements. The shear layer is characterized by relatively high levels of
mixing with measured stream-wise turbulence intensities within it approach-
ing 20% or more [Garimella and Eibeck, 1992; Wirtz and Chen, 1992]. This
enhanced mixing along the top of the package results in heat transfer rates that
are greater than would occur in a smooth-walled channel.
Since the bypass and array flows are different, the convection process on
the top of a module is likely to be different from that found on any of the four
sides exposed to coolant As a simplification, we restrict our attention to mod-
ules where convection is mostly influenced by the bypass flow since this quan-
tity is relatively easy to quantify. We define a low profile package as one where
the package top represents at least 50% of the package heat transfer surface
area, i.e.,

LxLy ;;::: O.S (3)

LxLy +2a(Lx + Ly)

For a square planform package, Equation 3 requires that low-profile packages

have aspect ratios, Ua :2:: 4. This restriction covers the majority of packages
found in application.
Wirtz and Mathur [ 1994] studied the local heat transfer rates on the surfaces
of low-profile packages. (They studied square planforrn packages having
L/a = 6). These investigators found that the local heat transfer coefficient dis-

1 Quantative measures of laminar-turbulent transition in tenns of a transition Reynolds number

are not available. An assessment of whether a given flow is laminar, turbulent or transitional is
a judgment call.
Forced Air Cooling of Low-Profile Package Arrays 85

tribution on the top surface of the package is similar to, but greater than, that
expected on a flat plate in a turbulent airstream. They also found that the aver-
age heat transfer coefficient on the top surface of these packages is approxi-
mately equal to the overall heat transfer coefficient for the package. Therefore,
we expect that heat transfer correlations developed for low-profile packages
will be appropriately described by the local average velocity, V.

ESTIMATION OF PACKAGE TEMPERATURE

Two convective effects contribute to the temperature rise of an electronic
package. The first, called the self-heating effect, is due to heat generated within
the package. The second is due to heat releases from other components that
are upstream from the package being considered. These upstream heat releases
raise the mean temperature of the coolant which then washes over the pack-
age in question, raising its temperature. Thus, a completely passive compo-
nent that resides in the wake of a heat-releasing element will experience a
temperature increase. This second heating effect is often called the thermal
wake effect.
For forced convection cooling (no buoyancy effects or radiation heat trans-
fer), these two effects may be combined by simple addition (superposition) be-
cause the energy equation and its boundary conditions are linear under these
conditions. A convenient expression [Arvizu and Moffat, 1982] is

i<k (4)

The first term on the right of Equation 4 expresses the self-heating temper-
ature rise of element k above the entrance temperature, T0 , in terms of its con-
vective heat release rate, qk, heat transfer surface area, Ak, and the adiabatic heat
transfer coefficient, hk. The tenn adiabatic is used in describing hk since it is
the heat transfer coefficient observed in the absence of other (upstream) heat
releases, and this serves as the quantity's definition.

(5)

As a consequence, hk is a function of flow rate, channel and array geometry,

fluid properties, and perhaps position in the array. However, its value is not de-
pendent on upstream thermal boundary conditions. This characteristic of hk al-
lows the two effects (self-heating and upstream heat release) to be separated in
Equation 4. From an experimental viewpoint, this is one of the main advan-
tages of the approach since hk is measured in an array with only the k-th ele-
ment powered-up and all other elements passive. Moffat and Anderson [ 1990]
discuss the relationship of the adiabatic heat transfer coefficient to other heat
transfer coefficient definitions used in heat transfer analysis.
86 Air Cooling Technology

It should also be noted that hk may be related to the case-to-ambient ther-

mal resistance, Rc-a, only where Rc-a is reported for conditions of no upstream
heating. Under these conditions

(6)

The second term on the right of Equation 4 expresses the temperature rise
of element k due to thermal wakes from all elements in array rows upstream
from k. It contains the thermal wake function, f)k-i, which is the fractional tem-
perature rise of element k due to heat release from element i with qk = 0,

(7)

Like hk, £A-i is a function of flow rate, channel and array geometry, fluid prop-
erties, and position in the array. Its experimental determination is relatively
straightforward. Experiments are conducted with one array element heated, and
the temperature increases of elements downstream from it are recorded.
It should be noted that the form of Equation 4 is appropriate to situations
where the upstream heating effect is confined to the same column of the array
as the heat release (i.e., thermal wakes do not spread to adjacent columns).
Experimental observations on in-line arrays of low profile packages support
this assumption [Wirtz and Dykshoom, 1984]. Furthermore, Moffat and
Anderson [ 1990] have developed the thermal wake concept of Equation 4 in
terms of kemal functions, placing the approach on a firm analytical footing.

DETERMINATION OF b" AND e

Due to the inherent complexity of the flow, most correlations of hk, and fA-i
are empirically based. Two approaches have been followed to obtain data used
to construct correlations: prototype experiments and model experiments.

PROTOTYPEEXPER~ENTS
Prototype experiments utilize actual package encapsulations mounted to
pc-boards using standard attachment methods. A small heating element and
temperature sensor replace normal circuitry inside the encapsulation. Package
arrays are arranged as they would be in application. The advantage of this ap-
proach is that the heat transfer data obtained is a direct measure of the cooling
performance of a specific design. The principal disadvantage is that it is diffi-
cult, if not impossible, to separate the effect of the different heat transfer modes
that are active. This is usually because conduction through attachment leads to
the pc-board can not be accurately measured. As a result, heat transfer data ob-
tained in this way is not easily generalized to other, geometrically similar pack-
aging configurations.
Forced Air Cooling of Low-Profile Package Arrays 87

MODEL EXPERIMENTS
Model experiments utilize test articles that simulate actual packages. The
model packages can be carefully designed to minimize measurement errors,
and experimental test rigs can be designed to eliminate unwanted heat transfer
modes. For example, guard heaters may be employed to eliminate conduction
to the mounting board, and package surfaces can be highly polished to mini-
mize radiation heat transfer. Similitude may then be employed to generalize
measurements from a given test rig to other geometrically similar packaging
configurations. The disadvantage of this approach is that the model test article
may neglect a significant geometric feature contained in the prototype, leading
to erroneous conclusions about the convection process.
Copeland [1992] reports prototype measurements that utilized square plan-
form, aluminum/ceramic encapsulated packages (L = 37 mm, L/a = 6.4) that
were pin-grid attached to epoxy-glass interconnect boards. Packages were
arranged in square arrays with the packaging density, D, ranging from 0.25 to
0.94. The boards, with board-to-board spacing varied over the range 2.2 :S: HIa
:S: 5.5, were tested in a special purpose channel designed to simulate flow through
a card rack. For array elements far from the channel entrance, Copeland cor-
relates his heat transfer coefficient data in terms of entrance velocity using a
linear equation. For a packaging density of D = 0.53 and H/a = 2 he suggests
the following equation.Z

h == 20+ 7V0 (8)

Wirtz and Mathur [1994] report heat transfer measurements from a model
experiment that used a packaging configuration similar to the one studied by
Copeland. The model experiments utilized machined aluminum blocks that
were instrumented with small heaters and calibrated thermocouples. External
surfaces were polished to minimize radiation heattransfer. The experiment was
designed so that conduction to the mounting board could be accurately mea-
sured and subtracted from the heater power to give the convection heat trans-
fer rate. The model packages were larger than those used in the prototype
experiment (L = 69.8 mm). However, the package aspect ratio (L/a = 6), pack-
aging density (D = 0.45), and dimensionless channel height (H/a = 2) are al-
most the same as those used to construct Equation 8.
Figure 2 compares Copeland's linear correlation, Equation 8, with the con-
vection heat transfer data obtained in the model experiment. The prototype mea-
surements ofh are much higher than the model experimental results, particularly
at low entrance velocities. Part of this difference is attributable to the longer
package length employed in the model experiments. The remaining difference
is because the prototype measurements mix conduction and convection effects.
As the flow rate increases, the convective component increases (the conduction
component should be essentially independent of flow rate) and the two results
2 Equation 8 is dimensionally inhomogeneous. This practice is not recommended.
88 Air Cooling Technology

100

>:q.(S~HI•·l~

••• ·r. .
10 8
' Wim and Mathur, 1994

0.1 l 10
V,., [m/s)

FIGURE 2. Comparison of package heat transfer measurements: Prototype vs. Model

experiment.

appear to converge. However, Equation 8 implies the wrong dependence of con-

vection on flow rate. The results are plotted in log-log coordinates in the figure,
so the linear equation, Equation 8, plots as a curve that is asymptotic to h - v:;.
On the other hand, the model experimental results (which include convection
only) appear to plot as a straight line over the entire range of velocities investi-
gated. Thus, the model experiment indicates a logarithmic dependence of the
heat transfer rate on coolant velocity, with h - V,:' where n < 1, and this power-
law dependence is what has been observed in aU reported model experiments.
A regression analysis of the model experiment data leads to the conclusion
that h - vg· 8 . A similar regression analysis of the prototype data, assuming a
power-law relation between h and V0 , would give something like h - vg.s. The
lower exponent obtained from the prototype measurements is caused by the rel-
atively high heat transfer coefficients reported at low flow rates. These high
heat transfer rates are caused by the relatively large contribution that conduc-
tion makes to the measured heat transfer at these low flow rates. In general, if
conduction and radiation heat transfer is not adequately accounted for in an ex-
periment, the resulting exponent in a power-law correlation of convection heat
transfer data will give a low estimate of the power-law exponent.
Since there is some uncertainty regarding the role of conduction and radi-
ation in heat transfer results obtained via prototype experiments, the correla-
Forced Air Cooling of Low-Profile Package Arrays 89

tions developed in the following section are limited to results derived from
model experiments.

PACKAGE HEAT TRANSFER

UNIFORM, INs LINE ARRAYS
ADIABATIC HEAT TRANSFER COEFFICIENT
As in any confined flow, fully developed conditions are preceded by a length
of channel where the local shear stress and heat transfer coefficient vary in the
streamwise direction. Most heat transfer observations of convection from ar-
rays of electronic packages suggest that it takes from three to five package rows
for the convection process to become fully developed (in a row periodic sense).
For the channel Reynolds numbers generally encountered, the leading rows ex-
hi bithigher heat transfer coefficients than the fully developed values. Generally,
the leading row his about 20-30% greater than the fully developed value, and
the second-row value is augmented by 5-10%. However, this is only an ap-
proximate "rule of thumb." As the Reynolds number decreases the amount of
augmentation exhibited by leading-row array elements also diminishes so that
at very low Reynolds numbers leading row heat transfer coefficients may be
equal to, or lower than, those measured down stream [Garimella and Eibeck,
1990, Lehmann and Pembroke, 199la]. In the following, we develop a corre-
lation for the fully developed package heat transfer coefficient. Corrections,
such as the above mentioned "rule of thumb," can be applied to the fully de-
veloped coefficients to account for leading row effects.
Table 1 summarizes the geometrical characteristics of model package arrays
considered in some recent heat transfer experiments. The table lists the pack-
age length and aspect ratio, packaging density of the array, and dimensionless
channel height, H/a. All investigations considered uniform arrays and, except
for Anderson and Moffat3 , all considered square planform packages. The pack-
age length is seen to vary from 25.4 mm to almost 70 mm and the package as-
pect ratio varies from 2.67 to 8.75. Packaging density is seen to vary from the
TABLE 1
Geometric Features of Some Model Heat Transfer Experiments

Investigator L, Ua D HIa
[mm]

Wirtz and Dykshoom, 1984 25.4 4.00 0.25 1.5-4.6

Sparrow et a!.. 1982 26.7 2.67 0.64 2.7
Anderson and Moffat, 1992a 37.5 3.95 0.59 1.5-4.6
Wirtz et al., !994b 56.0 8.75 0.49 1.5-10
Wirtz and Mathur, 1994 69.8 6.00 0.45 2.0
Wirtz and Col ban, 1995 69.8 6.00 0.45-0.69 2.0

3 Anderson and Moffat considered rectangular planforrn packages with Lx!Ly = 0.81, Sx/Sy =
0.85
90 Air Cooling Technology

relatively sparse array considered by Wirtz and Dykshoorn (D = 0.25) to very

dense packaging considered by Wirtz and Colban (D = 0.69). Dimensionless
channel heights range from 1.5 to 10.
Figure 3 plots reported heat transfer coefficient measurements of h from the
studies listed in Table 1 in terms of the local average velocity, V (see note4 ).
The following observations can be made:

1. The data lie in a band having a bandwidth that is about ±25% of any mid-
point value.
2. Subsets of the data are generally arranged in parallel straight lines with
a decrease in h corresponding to an increase in package length, L.
3. The heat transfer coefficient is seen to increase logarithmetically with ve-
locity, implying a power-law relation, h- vn, for each subset of the data.
Best-fit values ofn range from 0.6 to 0.8.

100

10
Sparrow d 11., 1982
0
Wirtund Dybboom. 1984
Andersoo md Moffat, 19928
Wimeul,l994b
Wim mel Mathur, 1994
Wim mel Colb111, 1995

0.1 1 10 100
V, (mls)

FIGURE 3. Package adiabatic heat transfer coefficient measurements from the model experi-
ments listed in Table 1.

4 The experimental work of Wirtz et al. [1994]; Wirtz and Mathur [1994]; and Wirtz and Colban
[ 1995] was completed at an altitude of 1300 m M.S.L. In Figure 3, these data are altitude cor-
rected by multiplying reported velocities by 0.86, the average barometric pressure (in bars) at
that altitude.
Forced Air Cooling of Low-Profile Package Arrays 91

4. The use of the local average velocity, V, appears to adequately correlate

the effect of variations in dimensionless channel height, Hla, and pack-
aging density, D.

Observations 2 and 4 in the above list suggest that the appropriate length
scale to unify the data is the package length, L. Figure 4 shows the data ren-
dered in these terrns where the package Nusselt number, NuL= hL! k and pack-
age Reynolds number, ReL = pVLI Jl, are both based on the package length. The
bandwidth of the data is now about ±10% Furthermore, the figure shows a
change in the slope of the Nusselt-Reynolds relation suggestive of laminar-
turbulent transition at ReL "" 5000. Power-law correlations for the data give

(9)

and

NuL = 0.082 Re~72 , Re L > 5000 (10)

where Pr is the coolant Prandtl number. Equation 9, for low Reynolds number
flow, is seen to be a simple 10% downward adjustment of the Blasius solution

100

0 Sp!UYOW et Ill., 1982

[),
Wirtt Md Dyksboom, 1984
- - Lehmann and Pembroke, 199111
v Anderson and Moffat, 199211.
0 Wirtz et at., 1994b
10 <> Wirtz and Mathur, 1994
s Wirtt ~~nd Colbllll, 1995

1000 10000 100000

R~

FIGURE 4. Correlation of data from measurements listed in Table I for package Nusselt num-
ber in terms of package Reynolds number.
92 Air Cooling Technology

for laminar flow over an isothermal plate of length, L. Equation 10, for the
higher Reynolds numbers, is fully empirical. It gives heat transfer coefficients
that are 40 to 75% higher than would be obtained using the Stanton number
correlation for turbulent flow over a smooth, isothermal plate [Kays and
Crawford, 1993]. In both cases, the data is seen to scatter ±10% about the cor-
relations. Since 5-10% (in both Nu andRe) is generally the limit of accuracy
for the experimental results considered, further refinement such as inclusion of
correction factors for variations in packaging density, package height, or chan-
nel height, seems unnecessary.
For low profile packages having small but finite package heights, the cor-
relation of Lehmann and Pembroke [199la], shown as a dashed line in the fig-
ure, is recommended5
(11)

Other correlations for the heat transfer coefficient have been suggested. For ex-
ample, Anderson and Moffat [ 1992b] developed a linear correlation of the heat
transfer coefficient in terms of the coolant turbulence level
(12)

where u'max is the theoretical maximum turbulence level in the shear layer that
separates the bypass-flow from the array-flow. In Equation (12), c1 == 16, c2 ==
30 are empirically determined constants.The quantity u'max is estimated as

u'
max
= 0 •82 [(-.!pdx ]i
dp) VoSx
Lx (13)

where -dp/dx is the coolant flow pressure gradient. Streamwise turbulence mea-
surements by Wirtz and Chen [1992] in a channel containing two-dimensional
ribs having an aspect ratio of 6 show that h(2-D) is nearly linearly proportional to
the measured maximum streamwise turbulence level, u'max (max). Unfortunately,
correlation of the data of Table 1in terms of u'max shows that c 1and c2 in Equation
12 are functions of package length, L, and array density, D. Equation 12 over-
predicts measured values of h for packages having L >50 mm by 20-40%.

THERMAL WAKE FUNCTION

Correlations of the thermal wake function, ek-i, have been proposed
by Arvizu and Moffat [1982], Wirtz and Dykshoom [1984], Anderson and
Moffat [1992), and Kang [1994]. Kang considers a point source of heat in a
turbulent channel flow and finds that e~, the wake function for an array element
immediately downstream from the heat releasing element, is given by the fol-
lowing expressions6 :

5 Lehmann and Pembroke also present correlations for low Reynolds number flows .
6 Kang also gives expressions for the thermal wake function in laminar flow. These expressions
are not presented here since experimental verification of these expressions isnot available.
Forced Air Cooling of Low-Profile Package Arrays 93

I .
(Sx )0.5( LxH )044. (t- E_)
() = 7 19C Pr-{).5 Lx
H
05. Ren-0.94
a,L
(14)

where Reo.L = VoRedV is the package Reynolds number based on inlet average
velocity, V0 • The constant, C, and exponent, n, in Equation 14 are the coeffi-
cient and exponent, respectively, of the power-law that correlates the heat trans-
fer coefficient for the packages

Nu L = C Re"o,L (15)

The thermal wake function for successive packages, Sk-i k > i + 1, is given
in terms of e, by the expression

(16)

where the exponent m has been experimentally determined to range from 0.5
to 1.0. Kang shows that m is a function of the turbulent Pedet number of the
flow. A curve fit of his numerical results gives

=.5 + 0.06e -
Pe1
m 2500 (17)

where Pe, = VLJ!a,Sx is the turbulent Pedet number and a,= 0.006 a.(VH/v)- 88
is an estimate of the turbulent thermal diffusivity. The electronics cooling
applications considered here have Pet ranging from about 250 to l 000, so
=
m 0.55 ± O.ol.
Table 2 compares estimates of the thermal wake function given by Equations
14 and 15 with experimentally determined correlations of Wirtz and Dykshoom
[1984] for two dimensionless channel wall-to-wall spacings. In developing these
expressions, Equation 10 has been used for the package heat transfer coefficient
correlation, thus the coefficient, C, in Equation 15, is given as C =0.082(VN o)0 ·72 ,
where VNo is given by Equation 2. The results show that the fractional temper-
ature increase of the element immediately downstream from the heat-releasing
element decreases with increasing flow rate. The theoretical result predicts a
weaker Reynolds number dependence than was found in the experiments.
However, as shown in Figure 5, the agreement between the theoretical result and
the empirical correlation is very good in both cases, although it is anticipated

TABLE2
Comparison of Thermal Wake Function Estimates of Kang (1994)
with Data Correlations of Wirtz and Dykshoom (1984)

Investigator Hla=2 H/a =5

Kang (1994 ), theoretical a, =0.58 Rea.L-· 22 e, = 0.43 Rea,L-.n
Wirtz and Dykshoom (1984), experiments a, = 1.29 Reo.L-.3 a, =0.82 Reo.L-.}
94 Air Cooling Technology

0.20 - . - - - - - - - - - - - - - - - - - - - - .

- - Kang [1994], Eq. (14)

- - Winz and Dykshoom [1984]

0.15

I Hla=2

0.05

0 10000 20000 30000 40000 50000

FIGURE 5. Comparison of Equation 14 for the thermal wake function with empirical correla-
tion of Wirtz and Dykshoom [1984].

that Equation 14 will overpredict 8 1 at high Reynolds numbers and high di-
mensionless channel heights. The figure shows that an increase in channel wall-
to-wall spacing reduces the thermal wake effect. Furthermore, it is noted that
heat release from an element in row "i" of an array results in a 5-10% fractional
temperature increase for the element immediately down stream in row (i + 1).
Figure 6 compares the predictions of Equation 16 and 17 with the data of
Wirtz and Dykshoom [1984] for the temperature rise of successive elements
further downstream from the heated element. The case shown is for two
Reynolds numbers with HJa = 2. For the higher Reynolds number, Pe = 375
and Equation 17 gives m = 0.55. This results in a good fit of the data, as can be
seen in the upper part of the figure. At the lower Reynolds number (Reo,L =
=
1355), Pe = 285 and Equation 17 gives m 0.55. In this case, the data are bet-
ter fit if m =0. 78, and the theoretical result is seen to overpredict the wake func-
tion by about 20-30%, as shown in the lower part of the figure.

EXAMPLE CALCULATION
We can gain an appreciation of the relative importance of these effects (self
heating and thermal wake) by considering an example. Consider an array of
square planform packages having package length, L = 36 mm, and height, a =
Forced Air Cooling of Low-Profile Package Arrays 95

1.0

anlao.8
1

0.6

0.4
- - Kang [1994], Eqs. (Hi,l7)
1.0 e Wirtz and Dykshoom [1984]

0.8

0 /00.6
II I

0.4

0.2

0.0
1
n=i-k, number of rows down stream
FIGURE 6. Comparison of Equations 16 and 17 for the thermal wake function with empirical
correlation of Wirtz and Dykshoom [1984].

5 mm. The array is five rows long. The packages are spaced c =8 mm apart (D
= 0.67), and the board-to-board spacing is H = 20 mm. Air at 20°C flows to-
ward the array at 800 fpm (Vo =4.lrnls). Under these conditions VNo = 1.26,
the channel Reynolds number is Re = 13,600, and the package Reynolds num-
ber is ReL = 12,300. Equation 10 gives a fully developed flow heat transfer co-
efficient, h = 52W/m2°C. Equations 14 through 17 givdh, k= 1, .. , 4as (0.110,
0.075, 0.060, 0.051). Table 3 summarizes the resulting operating temperatures,
T k - To, when the first row, or all rows, are powered at 5W per element.
The first row of the Table shows package temperature increases (relative to
the entrance flow temperature) for the case where only the element in the first
row is powered and no correction is made for entrance effects. Self heating re-
sults in a 4 7. 7°C temperature increase of the first package, and the temperature
increase of successive (unpowered) elements is relatively small.
96 Air Cooling Technology

TABLE3
Estimated Package Temperature Increase Due to Convection Only

Package Temperature Increase, T.-T., oc

Row number, r 1 2 3 4 5

only q 1 =5W, no entrance correction 47.7 5.3 3.6 2.9 2.5

all q's =5W, no entrance correction 47.7 52.9 56.5 59.4 61.8
all q's =5W, with entrance correction 39.7 48.6 56.5 59.4 61.8

Note: Cooling conditions are defined in the accompanying text. Temperature increases are rounded
to one digit of significance.

The second row of the table demonstrates how these seemingly small ther-
mal wake effects are compounded by multiple heat sources. The case consid-
ered is for all rows powered at 5W per element with no correction for entrance
effects. Successive element temperature increases are obtained via Equation 4
by adding the thermal wake temperature increase contribution of each upstream
heat-releasing element in the same array column to the self-heated temperature
rise of the element under consideration. For example, the temperature increase
of the third-row element is 47.7 + 5.3 + 3.6 = 56.6 (intermediate calculations
rounded to 1-digit). In this case, heat release from the first four rows results in
a 30% increase over the selfheated temperature rise in the surface temperature
of the fifth-row element
Finally, the third row of the Table demonstrates how entrance effects affect
package operating temperatures. In this case, we assume the first and second
row heat transfer coefficient is augmented by 20% and 10%, respectively, rel-
ative to the fully developed value of h. The results show 17% and 8% reductions
relative to uncorrected estimates in the operating temperatures of the first- and
second-row elements. Furthermore, the entrance effect does not extrapolate to
successive element rows past the first two rows. Since the heat release rate from
the first and second row elements is fixed (at 5W each), the sturcture of the
thermal wake downstream from these elements is not changed by the assumed
augmentation in the element heat transfer coefficient.

COOLANT PRESSURE DROP-

UNIFORM, IN-LINE ARRAYS
Upon encountering a package array, the coolant is accelerated because of
the reduced cross-sectional area available to the flow, and the shape of the ve-
locity distribution changes. As a result, there is an increase in the streamwise
pressure gradient, -dp/dx. As the entrance effect diminishes (three to five pack-
age rows) the local pressure gradient reduces to a constant (in a row-periodic
sense), fully developed value. The fully developed pressure gradient can be
expressed in dimensionless terms as a friction factor based on the bypass
flow, f, as
Forced Air Cooling of Low-Profile Package Arrays 97

(18)

Dimensional analysis shows that f is a function of Reynolds number and the

geometric lengths that describe the array. Figure 7 shows some typical data
[Wirtz et al., 1994b, Wirtz and Colban, 1995] where the friction factor is plot-
ted in terms of the channel Reynolds number, Re. The upper part of the figure
compares two arrays having different packaging densities, and the lower part
compares two channel heights. 7 Also shown in the figure are laminar and low-
Reynolds number turbulent [Beavers et al, 1971] correlations for flow in a
smooth-wall channel having a height equal to (H-a). The smooth-wall correla-

0.1 Wirtz and Colban [1995]

0
0 0

D HIa Ua
0 0.45 2 6
0.01
D 0.69 2 6
6. 0.49 2 8.75
0.1 "V' 0.49 s !l1S

Wirtz et al. [1994b]

6
* ~ '\l"V' * l:i'ffJ4Jt::J&w

0.01 -+----,----.--,---.--.--,--.--,..------..----..- -
2000 3000 5000 7000 10000 20000 30000

Re
FIGURE 7. Friction factor measurements for flow over uniform, in-line arrays.

7 Since Re is defined in terms of local average velocity, V, a comparison off for different geo-
metrical configurations does not imply a comparison at equal entrance flow rates.
98 Air Cooling Technology

tions can be interpreted as applicable to arrays having maximum packaging

density (D ~ 1).
The data exhibit turbulent-like behavior, with the friction factor slowly de-
creasing with increasing Re. The slope of the plot of data in log-log coordi-
nates suggests f- Re-0· 2 • This implies that the pressure drop across an array,
tlp, is proportional to Vo 1.s so that a doubling in flow rate results in a factor-3.5
increase in tlp. The upper part of the figure shows that f decreases with an in-
crease in packaging density, presumably approaching the smooth-wall corre-
lation as D increases. The lower part of the figure shows that the present friction
factor definition, Equation 18, appears to adequately correlate variations in
channel height. As a result, for a given array we expect Ap - (H-at 1•
Furthermore, a comparison of the two plots suggests that f decreases with in-
crease in the package aspect ratio, Lla. It must be pointed out that there is only
a limited pressure-drop data set available, therefore a comprehensive correla-
tion of these effects is not feasible.

NONUNIFORM ARRAYS

EFFECT OF ARRAY STAGGER.

The effect of staggering alternate rows of packages has been studied by
Hollworth and Fuller [1987], Garimella and Eibeck [1990], and Wirtz and
Colban [1995]. Bazydola andTaslim [1993] have looked at entrance region
effects on a staggered array. At fixed coolant flow rate, stagger increases both
the package heat transfer coefficient (by up to 50%) and the friction factor
(by up to a factor of two). However, Wirtz and Col ban found that there is no
cooling performance advantage resulting from row stagger on the basis of
equal coolant pumping power such as exists in a fan-driven system.
Furthermore, if cooling performance is expressed in terms of heat removal
per unit system volume, maximum capacity is expected with the most dense
packaging possible.

EFFECT OF NONUNIFORM PACKAGE SIZE

Sparrow et al. [1984], and Torikoshi et al. [1988] considered how changes
in array element height affected the package heat transfer coefficient. In gen-
eral, they found that package heat transfer is slightly enhanced in the vicinity
of the change in geometry.

EFFECT OF HEAT SINK PLACEMENT

Lehmann and Kosteva [1990], Lehmann and Pembroke [199lb], and Gavali
et al. [1993] studied how placement of a heat sink on an array element affects
the thermal performance of other array elements in the array. In general, ele-
ments immediately adjacent to the heat-sinked element (in the adjacent
columns) experience a small amount of augmentation in their heat transfer co-
Forced Air Cooling of Low-Profile Package Arrays 99

efficient (about 5-10%) due to local acceleration of the coolant around the
blockage caused by the heat sink. Array elements downstream from the heat-
sinked element may experience either an increase or a decrease in their heat
transfer rate due to the heat sink's presence. Wirtz et al. [l994a] explain how
this effect influences operating temperatures. Wirtz et al. [1994b] present cor-
relations for computing longitudinal fin heat sink performance in arrays of pack-
ages, and Lee [1995] presents a calculation methodology for estimating heat
sink performance.

CONCLUSION
Convection in uniform arrays containing low profile packages is strongly
influenced by the bypass flow. The primary scaling quantities that control the
magnitude of the heat transfer coefficient are the local average velocity, V (an
estimate of the bypass velocity), and the package length, Lx. Other geometric
parameters that influence h are the packaging density, D, package height, a,
and the channel height, H. Equations 9 and 10, heat transfer correlations based
on modeling experiments, appear to adequately incorporate these effects in
situations where the height of the low profile package is sufficient to modify
the coolant flow. These correlations estimate the package heat transfer coef-
ficient with an uncertainty of about ± 10%. If the package height is very small,
Equation 11 is recommended. In a similar manner, Equations 14-17 allow
for estimation of the thermal wake function. These two effects, self heating
and thermal wake, may then be superimposed using Equation 4 to estimate
package surface temperatures under conditions where radiation and conduc-
tion are neglected.

APPENDIX A: NOMENCLATURE
a Package height
A Package surface area
c Package spacing
D Packaging density
f Friction factor
h Adiabatic heat transfer coefficient
H Board-to-board spacing
k Thermal conductivity
L Package length
Nu Nusselt number
p Pressure
q Convective heat transfer rate
Rc-a Case-to-ambient thermal resistance
Re Channel Reynolds number, Vl(H-a)lv
ReL Package Reynolds number, VUv
 100 Air Cooling Technology

S Packaging pitch
T Temperature
V Local average velocity
Vo Inlet average velocity
x Streamwise coordinate
y Cross-stream coordinate
z Spanwise coordinate
a Thermal diffusivity
ll Viscosity
v Kinematic viscosity
ek-i Thermal wake function, (TJ,-To)/(T;-To)
p Density

REFERENCES

Anderson, A.M. and Moffat, RJ. 1992a. The adiabatic heat transfer coefficient and the superposi-
tion kernel function: Part 1-Data for arrays of flatpacks for different flow conditions. J.
Electron. Packag., 12:14-21.
Anderson, A.M. and Moffat, R.J. 1992b. The adiabatic heat transfer coefficient and the superposi-
tion kernel function: Part II-Modeling flatpack data as a function of channel turbulence, J.
Electron. Packag., 112:14-21.
Arvizu, D.E. and Moffat, R.J. 1982. The use of superposition in calculating cooling requirements
for circuit board mounted electronic components, IEEE paper CH 1781-4/48-0133.
Beavers, G.S., Sparrow, E.M., and Lloyd, J.R. 1971. Low reynolds number turbulent flow in large
aspect ratio rectangular ducts, ASME J. Basic Eng., 285-289.
Bazydola, S.M. and Taslim, M.E. 1993. An experimental investigation of a staggered array of heat sinks
in the hydrodynamic and thermal entrance regions of a duct. J. Electron. Packag., 115:106-111.
Copeland, D. 1992. Effects of channel height and planar spacing on air cooling of electronic com-
ponents. J. Electron. Packag., 114:420--424.
Garimella, S. V. and Eibeck, P.A. 1990. Heat transfer characteristics of an array of protruding
elements in single phase forced convection, Int. J. Heat Mass Transfer, 33(12):
2659-2669.
Garimella, S.V. and Eibeck, P.A. 1992. Onset of transition in the flow over a three-dimensional
array of rectangular obstacles, J. Electron. Packag., 114:251-255.
Gavali, S., Kailish, C., Patankar, S.V., and Miura, K. 1993. Effect of heat sink on forced convec-
tion cooling of electronic components, Adv. in Electron. Packag., 1993, ASME EEP,
4(2):801-808.
Hollworth, B.R. and Fuller, H.A. 1987. Heat transfer and pressure drop in a staggered array of air
cooled components, in Proc. Int. Symp. on Cooling Techno!. for Electron. Equipment, Honolulu,
732-748.
Kang, S.S. 1994. The thermal wake function for rectangular electronic modules, J. Electron.
Packag., 116:55-59.
Kays, W.M. and Crawford, M.E. 1993. Convective Heat and Mass Transfer, 3rd. Ed., McGraw-
Hill, New York.
Lee, S. 1995. Optimum design and selection of heat sinks, in Proc. Ilth IEEE SEMJ-THERM
Symposium, San Jose, 48-54.
Forced Air Cooling of Low-Profile Package Arrays 101

Lehmann, G.L. and Kostiva, S.J. 1990. A study of forced convection direct air cooling in the down-
stream vicinity of heat sinks, J. Electron. Packag., 112:234-240.
Lehmann, G.L. and Pembroke, J. 1991a. Forced convection air cooling of simulated low profile
electronic components. I. Base case, J. Electron. Packag., 113:21-26.
Lehmann, G.L. and Pembroke, 1. 1991b. Forced convection air cooling of simulated low profile
electronic components. II. Heat sink effects, J. Electron. Packag., 113:27-32.
Moffat, R.J. and Anderson, A.M. 1990. Applying heat transfer data to electronics cooling, J. Heat
Transfer, 112:882-890.
Sparrow, E.M., Niethammer, J.E., and Chaboki, A. 1982. Heat transfer and pressure drop charac-
teristics of arrays of rectangular modules encountered in electronic equipment, Int. J. Heat
Mass Transfer, 25(7):961-973.
Sparrow, E.M., Yanezmoreno, A.A., and Otis, D.R. 1984. Convective heat transfer response to
height differences in an array of block-like electronic components, Int. J. Heat Mass Transfer,
27:469-473.
Torikoshi, K., Kawaxoe, M., and Kurihara, T. 1988. Convective heat transfer characteristics of ar-
rays of rectangular blocks affixed to one wall of a channel, Natural and Mixed Convection
in Electronic Equipment Cooling, ASME HTD, 100:59-66.
Wirtz, R.A. and Chen, W. 1992. Laminar-transitional convection from repeated ribs in a channel
flow, J. Electron. Packag., 114:29-34.
Wirtz, R.A., Chen, W., and Colban, D.M. 1994a. Convection in arrays of electronic packages con-
taining longitudinal fin heat sinks, in Cooling ofElectronic Systems, S. Kakac, H. Yuncu, and
K. Hijikata, Eds., Kluwer Academic, Dordrecht, The Netherlands, 145- 164.
Wirtz, R.A., Chen, W., and Zhou, R. 1994b. Effect of flow bypass on the performance of longitu-
dinal fin heat sinks, J. Electron. Packag., 116:206-211.
Wirtz, R.A. and Colban, D.M. 1995. Comparison of the cooling performance of staggered and
in-line arrays of electronic packages, in Proc. ASMEIJSME Thermophys. Conf, Maui,
4:215-221.
Wirtz, R.A. and Dykshoom, P. 1984. Heat transfer from arrays of flat packs in a channel flow, in
Proc. 4th IEPS conf, Baltimore, 318- 326.
Wirtz, R.A. and Mathur, A. 1994. Convective heat transfer distribution on the surface of an elec-
tronic package. J. Electron. Packag. , 116:49-54.

FOR FURTHER INFORMATION

Basic information on thermal control in electronics is found in Thermal

Analysis and Control of Electronic Equipment, A.D. Kraus and A. Bar-Cohen
(McGraw Hill, 1983). More advanced summaries on heat transfer in electron-
ics are found in Advances in Thermal Modeling of Electronic Components and
Systems, Vols. 1, 2 and 3, A. Bar-Cohen and A.D. Kraus, Eds. (ASME Press).
The Electrical and Electronics Packaging Division (EEPD) of ASME acts
as a coordinator and disseminator of information on electronic packaging. The
division's Journal of Electronic Packaging contains technical papers on recent
developments in heat transfer in electronics. The K-16 Committee on Heat
Transfer in Electronics (a committee of EEPD and the Heat Transfer Division
of ASME) acts as a forum for exchange of technical information.
 Chapter4

CONJUGATE HEAT TRANSFER IN FORCED AIR

COOLING OF ELECTRONIC COMPONENTS
Alfonso Ortega

CONTENTS
Introduction .... . .... . .................................. . .104

Background ... ..... . ... . . . .. . ....... ....... ... . .. ..... . .. 104

Classification of Problems ... .... ..................... . . . .... 114

Two-Dimensional Situations: Strip Source of Heat

in Boundary Layer and Channel Flow . ... . . ........ .. .. . . .... .. 117
Experimental Observations .. . . . . . . ... . ....... . .. .... . . . .. . 117
Conjugate Models ..... . . ... .... ..... . . . ...... . . ... ..... . 125
Laminar Channel Flow-Finite Difference/Control
Volume Method .... . . . ...... .... . ... . ... .. . ...... . . ... 125
Laminar Channel Flow-Boundary Element Method .. . . .. . .. . 133
Solid Side Model .... . . . . . . ... ..... . ...... ... .... . .. 133
Fluid Side Model . .. ... . . ..... .. . .. . . ... ... ... . . .. . . 136
Solid-Fluid Coupling . . . .. . ... .... . ..... . . . .. . . . . . .. . 138
Laminar Boundary Layer Flow . ... . .. . . .. . . .... . . . .. .... .141

Three-Dimensional Situations: Rectangular Sources of Heat . . . . . . . .. . 145

Boundary Layer Flow ...... ... .. . . . ..... . ....... ....... .. 145
Uniform Flow .... . .. . .. .. . .. . ... . .. .. . . . .. . . . . .... . .... 153

Conclusion . ... . . . . .... .. .. . . . .... . . . .. . .... . .. . . . . . ..... . 166

Acknowledgments . ... ..... ............ .. .... . .. . . . .... . .. . 169

References .. . .. . . . .. . .. ... . ... ... . .. . . . . .. . .... . ... .. ... . 169

For Further Information .. . ... . .. . ..... . ..................... 171

0-849J-9447-J/96J$0.(Xl+$.50
© 1996 by CRC Press, Inc 103
104 Air Cooling Technology

INTRODUCTION

In air cooling of electronics, the heat transfer process from single chip pack-
ages mounted on circuit packs or boards is a combined mode process involv-
ing conduction within the chip carrier and the board and convection from the
package and board surfaces. There are numerous approaches for analyzing the
thermal behavior of the package/board system including traditional uncoupled
approaches that apply convective heat transfer coefficient boundary conditions
to the solid conducting domains and more rigorous coupled approaches that
acknowledge the conjugate nature of the heat transfer process through the mul-
tiple paths from the chip to the package to the board and eventually to the con-
vective air flow. In assessing modeling strategies for electronics thermal
management, the need for several approaches, ranked in a hierarchy according
to application, ease of use, computational requirements, and accuracy, is fre-
quently overlooked. A traditional uncoupled conduction analysis of a packag-
ing structure, such as a board with convective heat transfer coefficients imposed
as surface boundary conditions, is approximate at best, but it has a low com-
putational overhead, is relatively easy to develop, and may lead to satisfactory
results for a system analysis. On the other hand, a full conjugate approach re-
quiring simultaneous solution of equations describing diffusion of heat in the
solid and diffusion and advection of heat in the fluid may carry a higher com-
putational penalty but will generally be more accurate when the fluid flow can
be modeled well. As computational power has become increasingly available
to the thermal analyst, the ability to perform reasonably accurate flow simula-
tions coupled directly to the thermal analysis has rapidly evolved.
The present chapter is devoted to a description of some selected conjugate
problems in air cooling of electronic components. There are several objectives.
The first is to introduce nonspecialists to the subject by discussion of key fea-
tures of conjugate problems involving electronic components on boards. It is
hoped that the discussion will provide a starting point for understanding the
differences between coupled and noncoupled approaches and in particular point
out difficulties in the use of the heat transfer coefficient when conjugate con-
duction mechanisms are important. For specialists, the aim of the chapter is to
illustrate the use of various approaches for conjugate analysis and to present
physical insight that has recently been gained by their application in the au-
thor's research group. The experienced analyst may wish to skip the background
material as it is intended primarily to engage the less experienced reader who
may be encountering the subject for the first time.

BACKGROUND

In principle, convective heat transfer from a solid surface to a flow is al-

ways a conjugate process involving the combined and simultaneous action of
conduction in the solid wall and advection/diffusion in the fluid. In other words,
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 105

all practical heat transfer problems, including those that arise in air cooling of
electronics, are conjugate in nature. The fact that the word "conjugate" is so
frequently associated with the electronics cooling problem, when in fact all
practical convective heat transfer problems are conjugate, tells more about the
historical evolution of convective heat transfer than the nature of the elec-
tronics cooling problem. The electronic package structures examined in this
chapter are illustrated in Figure 1, where it is understood that many single chip
components may reside on a single circuit pack. What is so special about this
problem, and why does it carry the "conjugate" descriptor? To elucidate, ex-
amine the two problems depicted in Figure 2, each with identical flow over
upper and lower board surfaces, with boards of identical thickness but with
different board thermal conductivity and heat dissipation conditions on the
surfaces. In Figure 2(a) the surfaces of the board are entirely given up to uni-
formly dissipating sources of identical strength, q. Because the board is non-
conducting, the board has no effect on the temperature distribution and the
surfaces may just as well be represented with a uniform heat flux thermal
boundary condition, i.e.

(1)

g ' SUBSTRATE
CHIP ..,--~~
~---~·:
I I I
I I

~
PACKAGE

I
I I
I I
\ I BOARD

·~ t;;;} t;;;} {;;} ~

{;;} {;;;; t;;;;J {;;} {;;}
{;;} {;;;} {d ~ {;;}
{;;} {d {d {;;;} {;;}

FIGURE 1. A hierarchical representation of chip, chip carrier. and board packaging levels.
106 Air Cooling Technology

where k1 is the fluid thennal conductivity, Tis fluid temperature, and q is the
heat dissipation per unit area. Contrast the just described situation to that of
Figure 2(b) in which a single source with total heat dissipation lztotal is located
on a board of finite thickness and thennal conductivity. In this case, the heat
dissipated by the source may be conducted into the board and released from ei-
ther surface, resulting in a nonuniform distribution of heat flux from the board
to the fluid on the board surfaces. To the thermal engineer's distress, the heat
flux distribution can no longer be specified on the board surfaces a priori as in
the first problem because it is unknown. Equally disturbing is the fact that the
distribution of convective heat transfer coefficient, so nicely catalogued in
countless heat transfer textbooks for cases such as in Figure 2(a), is completely
different for case 2(b ), despite the fact that the flow field is identical.
One approach to the thennal problem is to adopt an uncoupled strategy in
which all information about the convective part of the problem is embodied in
a convective conductance, or heat transfer coefficient, h, specified on the solid
surfaces. For the problem of Figure 2(b ), the Laplace equation on the board
may be solved using any of a number of approaches, most commonly by ap-
plying a finite-difference or finite element approach, subject to convective heat
transfer coefficient boundary conditions on the board surface. The art of choos-
ing a correct representation for the convective heat transfer coefficient h on the
board surfaces will be the key to a successful model. In the application at hand,
the local heat transfer coefficient is sometimes defined in terms of the undis-
turbed fluid temperature as

h=-q-
T,- T,,
where T, is the local surface temperature and To is the air temperature far from
the plate surface.
Here, q is interpreted as the local heat flux entering the fluid from the sur-
face. A typical solution, for example, for laminar boundary layer flow over a
smooth surface with conditions of constant heat flux on the wall, such as in
Figure 2a., is commonly found in heat transfer monographs. For example, Kays
and Crawford [1993] give the solution as

Nux = ~x = 0.453Re~ 12 Pr 113 (3)

f

where Nux is local Nusselt number, xis local streamwise position, Rex is local
Reynolds number, Re, = U.x/v, and Pr is fluid Prandtl number. It is seen as in
Figure 2(c) that h declines as x- 112 on the surface, a well-behaved monotoni-
cally decreasing behavior, and that it gets infinitely large at x = 0. This latter
behavior is unrealistic and simply reflects the failure of the boundary layer ap-
proximations at small x. In the theory of heat exchangers, which forms the basis
for so much of the material found in heat transfer textbooks, these types of
boundary conditions are frequently used with great success on convective sur-
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 107

faces despite the fact that they may be uniform neither in temperature nor in
heat flux. In general, only the surface averaged heat transfer coefficient is of
interest. The effects of wall conduction, especially in the stream wise direction,
are generally unimportant in such problems, and thus it is frequently forgotten
that solutions such as Equation 3 are not only dependent on the geometry under
consideration but also on the thermal boundary conditions and the temperature
used as the local reference temperature, in this case, T0 • It would be erroneous
to specify a heat transfer coefficient derived from Equation 3 to the surfaces of
the board depicted in Figure 2(b) for the purpose of calculating the surface tem-
perature. To illustrate, when problem 2(b) is solved using a true coupled con-
jugate approach, which did not require a priori specification of h on the surfaces,
the resulting distribution of heat transfer coefficient defined by Equation 2
would appear as in Figure 2(d). In this case, the board conduction produced a
highly nonuniform heat flux distribution on the surface that is dependent on the
thickness and conductivity of the board. The distribution of h reflects the nonuni-
formity in the distribution of the heat flux. The heat transfer coefficient h is
badly behaved and is so dependent on the conjugate board conduction that it
loses its primary value, which is its generality, and hence is not very useful for
predicting the board and component temperatures. Proper solution of a conju-
gate problem requires an approach that simultaneously considers fluid con-
vection and substrate conduction.
To illustrate conjugate, combined mode heat transfer in an actual air-cooled
board, consider the temperatures that result from a single PQFP located on a
test board in a low speed parallel air flow. Two such test boards are illustrated
in Figure 3. The two boards are at this writing under evaluation by the JEDEC
JC15.1 subcommittee on thermal measurements [JEDEC, 1994] with the aim
of establishing industry standard procedures and hardware for thennal char-
acterization of single chip packages. The traces on the board of Figure 3(a)
are designed to accommodate a single 64 lead PQFP and are not nested. The
lead out is conventional and "radial" in design. The board of Figure 3(b) is de-
signed to accommodate packages of different size footprints, and this is ac-
complished with nested traces that emanate in the four principle axis directions
on the board surface. In both cases the copper traces are 2.5 mil thick. The
copper traces produce a highly anisotropic thennal conductivity in the boards,
with in-plane conductivity much higher than out-of-plane conductivity and
with significantly different spatial distribution. Figure 4 presents a series of
infrared thermographs made at three air flow speeds for the two boards. The
thermographs for the unnested "radial" trace pattern, Figure 4(a) show that the
traces conduct heat radially from the package, more or less equally, and that
the convection to the air distorts the radial symmetry because of the thermal
wake effect. The results for the nested trace pattern, Figure 4(b), were surely
anticipated-the trace spatial density is highest in the direction of the princi-
pal axes and the resulting preferential conduction via these routes is apparent.
The convective wake that develops due to the spatial evolution of the air-side
108 Air Cooling Technology

---
---------
~--------------k~s_=_o ______~lt
~----------------L----------------~

2 (a)

2 (c)

FIGURE 2. Comparison of a nonconjugate and conjugate problem in air cooling of a PCB; (a)
board with zero conductivity and uniform dissipation of heat on its surface, (b) board with nonzero
conductivity and localized dissipation of heat on its surface, (c) local heat transfer coefficient dis-
tribution for case (2a), the nonconjugate problem (d) local heat transfer coefficient distribution for
case (2b ), the conjugate problem.
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 109

---
--------- qll

~--------t_______ks~~-o ______ ~lt

~------------------L------------------~

2 (b)

k /k
s f

0.1
1.0
10.0
100.0
h

I . . . ,..
\
.. ·/..
I

---~------~------

2 (d)
FIGURE 2. Continued.
110 Air Cooling Technology

"
cc

(a)

(b)

FIGURE 3. PCB test coupons for 0;_, measurements of a single Motorola™ 64 lead PQFP in
low speed wind tunnel; (a) nonnested copper traces (2.8 mil Cu), (b) nested copper traces (2.8 mil
Cu). [Courtesy of Darvin Edwards, Texas Instruments].
 g
<2!.
Single i
Pattern (a.) ~
~
~
§
~
II>

"'
s·
61
f:l
~
100 fpm 300 fpm 500 fpm ;J:.
::;·
g
~
;:s
<><>
<Q.,
~
(b.) a
(3
;:s
Nested r;·
g
~<:::>
~~
FIGURE 4. Infrared thermographs of the surface temperature of a single Motorola TM 64 lead PQFP in low-speed wind tunnel flow from left to right on (a)
nonnested PCB, and (b) nested PCB [Courtesy of Darvin Edwards, Texas Instruments].
............
112 Air Cooling Technology

thermal boundary layer is still apparent in the flow direction. In both cases,
the distribution of heat flux on the surface of the package and the board is
highly nonuniform and is dependent on the trace pattern and board thickness
and conductivity. A second illustration of conduction and convection acting
simultaneously is shown in Figure 5, which shows two infrared photographs
of the top surface of a horizontally oriented PQFP mounted on a board. Figure
5(a) is an infrared thermograph of the top surface of the PQFP cooled by nat-
ural convection to air, and Figure 5(b) is the same package cooled by air flow-
ing from left to right at 1 m/s. The heat spreading by conduction from the die
through the plastic to the PQFP top surface is readily observed from the
isotherms. In the case of natural convection, Figure 5(a), the isotherms are
nearly circular, indicating that the surface convective resistance to heat flow
is more or less uniform over the surface, or at least circumferentially uniform.
In the forced cooling case, Figure 5(b), the convective thermal wake effect is
again observable by the elongation of isotherms in the flow direction. Despite
the fact that the heat transfer from the heat-dissipating die to the air flow is a
complicated conjugate process, for practical reasons it is common in the ther-
mal characterization of single chip packages to characterize the package ther-
mal resistance with simple metrics such as junction to ambient thermal
resistance, ej-a, defined as

(4)

where P d is total power dissipation from the die and Tj is the average die tem-
perature. It is not surprising that the use of these simple metrics is fraught with
inconsistency and misconception. One reason is the general unwillingness to
reject the notion that surface heat transfer coefficient h depends solely on the
geometry and the flow descriptors such as velocity. There appears to be a gen-
eral lack of appreciation of the fact that in most problems in air cooling of elec-
tronics, h depends simultaneously on the characteristics of the flow and on the
characteristics of the conducting solids of the package structure, i.e., most of
the problems are "conjugate" in nature.
Returning to the example of Figure 2(b), the degree of coupling between
the convective heat transfer from the source and conduction in the board can
be parameterized by the Biot number based on the board thickness. Table 1
compares the thermal conductivity and the inverse Biot number (Bi-' = k/ht)
of a 1-mm-thick layer of various packaging materials with a convectively
cooled surface, where h may be a nominal or average value of the heat trans-
fer coefficient on the surface. If the heat transfer coefficient is high, or if the
substrate conductivity is low, coupled conduction effects in the substrate are
negligible, as, for example, is the case for a substrate composed of balsa wood
with Bi-' order 1.0. Conversely, as substrate thermal conductivity increases, or
if the convective heat transfer from the surface is low, conjugate conduction
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 113

FIGURE 5. Infrared thermographs of the upper surface of a horizontally oriented PQFP under
conditions of natural convection air cooling (top), and forced convection air cooling at a velocity
of I m/s (bottom). [Courtesy of Prof. Zhu Dezhong, Mechanical Engineering Department, Tsinhua
University, Beijing. China].
114 Air Cooling Technology

TABLE 1
Comparison of Thermal Conductivity of Common Packaging Materials
and Their Inverse Biot Number for a Typical Comparison Case

Material Thermal Conductivity k Biot Nurnber'(klht)

{W/m-K) h =50 W/rn'-K
t=lmm

Balsa Wood 0.05 1.0

Epoxy-Glass 0.18 (out of plane) 3.6
Plastic Molding Compound' 0.7 14.0
Epoxy-Glass 1.1 (in plane) 21.7
Alumina 19.7 384.6
Al-Nitride 170 3333.3

Source: Properties abstracted from Jensen et al. [1989], except for 'Tummala and Rymaszewski
[1989], with permission.

mechanisms may be important. Thus, from Table 1, lateral conduction and

heat spreading from a component to an epoxy-glass board may be important
to consider in a low speed, air-cooled application and will certainly be im-
portant for the ceramic substrates. Although the length scale used in Table 1
is somewhat arbitrary, the point to be made is that the magnitude of Bi-' is
indicative of the strength of the coupling between the convection and the
solid conduction.

CLASSIFICATION OF PROBLEMS

The class of problems that are of interest here are illustrated in Figure 6,
which shows a series of subproblems of the general printed circuit board
(PCB) cooling problem [Ortega, et al., 1994]. Starting from the top, the prob-
lems are identifiable by geometry, thermal boundary conditions, and type of
flow phenomena induced. They also increase in physical feature and compu-
tational difficulty but follow a tractable path to the most general practical
problem, namely the problem of Figure 6(f). In each case the flow may be
further organized according to whether it is an unconstrained boundary layer
flow over the surface or a channel flow between the surface and its opposing
neighbor board, and whether the flow is laminar, turbulent, or transitionaL
The situations of Figure 6(a) and 6(b) introduce the discrete, conjugate na-
ture of the heat dissipation in a one-dimensional and two-dimensional sense
without any flow complexities that may arise in the more difficult geome-
tries. The situations of Figures 6(c) and 6(d) introduce predominately two-
dimensional, separated, unsteady, mean flow both on the rib surfaces and on
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 115

the board. The heat transfer from the two-dimensional rib-like components
may be highly coupled to the board conduction and may require a conjugate
treatment. The low magnitude of span wise velocity components may allow a
two-dimensional computation, and results for this class of conjugate prob-
lem have been reported by numerous investigators in the laminar regime. The
protruding components introduce premature transition and high levels of near-
wall turbulence even at low Reynolds numbers, and hence the applicable range
of laminar computations is limited.
The extension to three dimensions results in the situations of Figures 6(e)
and 6(f). The mean flow in these situations is far more complex, involving sec-
ondary flows such as the horseshoe vortex that wraps around the base of the
upwind components and three-dimensional, unsteady, reversing flow in the re-
gion between components. In addition, the air flow in the array, at entrance
velocities ranging from 1 to 4 rn/s, is in a highly unstable, intermittent, tran-
sitional regime. In the fully turbulent regions, the turbulent motions are ex-
traordinarily high, anisotropic, and obviously not in local equilibrium.
Numerous studies have experimentally extracted average heat transfer coeffi-
cients on the surfaces of such three-dimensional components under conditions
that suppress conjugate mechanisms. Most of the data (see for example
Anderson and Moffat [ 1990a,b]) are for components fabricated from conduc-
tive metal such as aluminum or copper, and the components are mounted on
thermally insulating boards. Hence, under typical low-speed air cooling con-
ditions, such components are nearly isothermal and board conduction is small.
The ensemble of heat transfer coefficients thus measured are weakly depen-
dent on the conduction in the component and in the board and thus depend
marily on geometry and fluid flow parameters. The behavior of the heat transfer
coefficient in this type of "uncoupled" problem is discussed in the reviews
of Moffat and Ortega [1988] and Peterson and Ortega [1990] among
other sources. The present chapter aims to broaden the scope of these previ-
ous treatments by pointing out the importance of conjugate phenomena in
similar problems"
The remaining discussion will focus primarily on the geometries of Figure
6(a), the two-dimensional strip source of heat on a conducting board, and
6(b), the rectangular source of heat on a conducting board. By limiting the
discussion to flows that are well understood, the basic features of conjugate
mechanisms can be better observed and generalized and solution techniques
can be compared. For completeness and in order to point out the ability to
analyze more realistic geometries, a recent investigation of a more specific
geometry resembling Figures 6(c) and 6(d) will be briefly discussed" Since
there are many possible approaches to the analysis of the conjugate prob-
lem, Table 2 lists various types of flows, flow models, and solid models that
have been found to be useful in addressing the PCB cooling problem; some
but not all of these combinations of flow models and solid models will
be discussed.
116 Air Cooling Technology

(c)

(f)

FIGURE 6. Some subproblems of the printed circuit board cooling problem organized accord-
ing to thermal boundary conditions and geometry [Ortega et al., 1994].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 117

TABLE2
Types of Flows and Models Useful for Conjugate Analysis
of Printed Circuit Boards

Type of Flow Flow Thermal Model Solid Thermal Model

Uniform flow Analytical kernel Finite element method

Laminar boundary layer flow Analytical surface Boundary element method
conductance
Laminar channel flow CFD- finite volume Finite difference method

TWO-DIMENSIONAL SITUATIONS:
STRIP SOURCE OF HEAT IN BOUNDARY
LAYER AND CHANNEL FLOW
EXPERIMENTAL OBSERVATIONS
To introduce the relevant physics, results from an experiment specifically
designed to examine conjugate heat transfer for the geometry of Figure 6(a) in
a relatively simple air flow are first presented [Wirth, 1994; Ortega et al., 1994].
The geometry and thermal conditions of interest are shown in Figure 7, and the
experimental realization of the problem is depicted in Figure 8. The apparatus
consisted of a low-speed wind tunnel and associated flow conditioning appa-
ratus, a flow development section, and a heat transfer measurement section.
Experiments were performed with a single heat transfer section but with vari-
ous flow development sections. The flow development surface and heat-
transfer surface formed one wall of the main flow channel, and the opposite
wall was formed by a 0.5 in. plexiglass wall. Smooth plates were used to gen-
erate normal flat plate laminar and turbulent boundary layer profiles at the en-
trance to the heat transfer section. Plates with regular spanwise grooves were
used to generate disturbed boundary layers with amplified near-wall turbulence.
As shown in Figure 9, the latter plates were grooved such that the rib length B
was a constant 1.0 in., the groove depth D was a constant 0.25 in., and the groove
length S was 0.25, 0.5, or 1.0 inches.
The heat-transfer surface shown in Figures 8 and 9 was designed to mea-
sure conjugate heat transfer from a single 1.0 in. surface-mounted heat source.
The panel was 0.5-in.-thick plexiglass (nominal thermal conductivity between
0.18 and0.2 W/m-K) with a stream wise length of 12.0 in. and a spanwise length
of 24.0 in. to match the wind tunnel spanwise dimension. The 1.0-in. heater
was 20.0 in. long in the spanwise direction, thus approaching a one-dimen-
sional strip source of heat. It was located 3.0 in. downstream from the begin-
ning of the instrumented heat transfer surface, as shown in Figure 9. The buildup
118 Air Cooling Technology

-u------------------------------------------------------------------- -:
.
~

.~-
''
t=:T

FIGURE 7. Problem domain for a one-dimensional heat source on a conducting substrate [Ortega
et al., 1994].

[ "-E>OGCASO ( 0.0 '"'c"m I

IT
1 rLs H
1
(1.0) (7.362)

r-3.01
r~l'---~f_ __JITO.S)
lLOW DEVELOPMENT FOAM INSULATION
PLATE

DIMENSIONS IN INCHES

FIGURE 8. Experimental apparatus for measurements of conjugate heat transfer from a one-di-
mensional heat source on a conducting substrate [Ortega et a!., 1994].

of the heater section is shown in Figure 9. The 1.0 in. x 20.0 in. heat source
was a 1-mil-thick etched Nichrome foil with a serpentine pattern that provided
95% areal coverage. The foil was encapsulated by 2 mil Kapton film on both
sides and was attached to a custom heat flow sensor. The heat flow sensor con-
sisted of a 12-mil-PVDF (trade name KYNAR) spacer with five pairs of evenly
spaced 1 mil Type K thermocouples adhered to both sides. Surface tempera-
ture measurements were made by embedding 1-mil-diameter Type K thermo-
couples in a 5-mil-thick adhesive layer adhered to both the flow side and the
back side of the 0.5 in. plexiglass panel. The junctions, with diameter of roughly
3 mil, were placed at the spanwise centerline of the heat transfer surface, and
the thermocouple wires were laid out along spanwise lines in order to follow
isothermal lines on the surface. Relative to the discussion of Table 1, the ther-
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 119

DIMENSIONS IN INCHES

xs-- _._-~3-.0-==-:1.10~

.003 KAPTON
.005 3M ADHESIVE
i~~~~~~~~~~~::::::-002 KAPTON
.001 NICHROME FOIL
.002 KAPTON
.003 B-STAGE EPOXY
.012 PVDF
.003 B-STAGE EPOXY
.005 3M ADHESIVE
PLEXIGLASS

FIGURE 9. Details of development plate and heater section. [Ortega et al., 1994].

mal conductivity of the plexiglass substrate was measured to be 0.19 W/m 2-K,
similar to the out-of-plane conductivity of epoxy fiberglass.
Figures 10 and 11 show the mean velocity, the RMS fluctuating velocity,
and the percent turbulence level (RMS velocity normalized on mean velocity)
for nominally laminar (2 m/s) and turbulent (10 m/s) flows over the smooth
plate. The zero pressure gradient Blasius profile is shown for comparison in
Figure 10. The laminar flow compares well with the expected theoretical be-
havior, but the maximum turbulence level of about 1% is rather high and leads
to early transition. Similar data for flow developed over the grooved surfaces
show significantly increased turbulence levels, up to 55% for a mean flow speed
of just 2.0 m/s. The large surface grooves introduce additional modes by which
turbulence is generated, especially when they introduce mean flow separation
and reattachment. Furthermore, the grooves induce premature transition to tur-
bulence. The presentation of Figure 12, showing the magnitude of the maxi-
mum fluctuating velocity as it depends on Reynolds number at a fixed position
x, unambiguously shows an inverse monotonic dependence of the transition ve-
locity on groove geometry, as characterized by the length scale B. The most im-
portant aspect to be noted is that at the low air velocities commonly encountered
in cooling of boards, the flows are invariably unstable transitional flows and
the true laminar regime is limited to air velocities of 1 m/s or lower. Furthermore,
the transitional velocity is strongly dependent on the surface geometry.
The moderately high plate thermal conductivity and the low rate of con-
vective heat transfer from the surface at these low air velocities combine to pro-
120 Air Cooling Technology

2.2 5.5
2 5
1.8 4.5
1.6 4
mean velocity
~ 1.4 3.5 ~
(/)
iii
l1.2 rms turb. level (%) >
--<>-
3 ..9.l
~ Q)
'6 0
_Q rms fluctuating velocity 2.5 c:
Q)
Q)
3
> 0.8
Blasius' solution 2 -e
2
0.6 1.5
0.4

0.2 0.5
0 +=~~~~~~~~~~~~~~~~~~4o

0 0.5 1.5
position (em)

FIGURE Ill. Streamwise mean velocity, RMS fluctuating velocity, and % turbulence for flow
developed over smooth plate, x = 0.762 m, U = 2 m/s, [Ortega et al., 1994].

10
30

8
l::::1
i mean velocity :::-
::>
20 ai
6 >
~ rms turb. level (%) ..9.l
'6 Q)
0 0
iii c:
> 4 rms fluctuating velocity Q)
3
10 2
-e
2

0 0
0 0.5 1.5 2
position (em)

FIGURE ll. Streamwise mean velocity, RMS fluctuating velocity, and % turbulence, for flow
developed over smooth plate, x = 0.762 m, U = 10 m/s, [Ortega et al., 1994].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 121

1.8

1.6 • flat plates

..... 1/4 inch grooves
1.4 ..... 1/2 inch grooves
-... 1 inch grooves
1.2

~ 1
E
-:::J
0.8

0.6
0.4

1.0 1.2

FIGURE 12. Maximum RMS velocity fluctuation dependence on Reynolds number for devel-
opment plates with increasing groove width, x =0.762 m, [Ortega eta!., 1994].

duce a coupled conjugate heat transfer situation. Defining a nondimensional

surface temperature as

(5)

where qtotal is the total source power dissipation per unit area of source, surface
temperatures may be compared under varying conditions of total source power.
Figures 13 and 14 show the normalized temperature on the flow side and on
the backside, respectively, at freestream speeds varying from 2 to 20 m/s for a
smooth development surface. Flow is from left to right, and the heat source is
located from position 3 to 4. The effects of the substrate conduction manifest
themselves in the significant upwind temperature increase, easily discernible
even two source lengths upstream. The temperature peaks on the source down-
stream from its center and drops steeply downstream in the thermal wake re-
gion. The temperature on the rear insulated surface is somewhat more
symmetrical because of heat spreading, and the maximum temperature is nat-
urally less than on the heated side. Figure 13 points out the sensitivity of the
temperature distribution to transition from laminar to turbulent behavior. There
are three cases indicated at a velocity of nominally 10 m/s (Re = 778400); two
of the cases, denoted by "weak trip," cluster at a higher temperature, and the
122 Air Cooling Technology

0.045

Re=155,700, 0=581 W/m', ZZ trip

Re=155,700, 0=775 W/m', ZZ trip

Re=389,200 0=775 W/m', ZZ trip

0.03
Re=778,400, 0=1550 W/m', weak trip
--o-
Re=778,400, 0=775 W/m', weak trip
---«--
Re=778,400, 0=1550 W/m', ZZ trip

0.015 Re=1,167,700, 0=1550 W/m', ZZ trip

-4--

Re=1,556,900, 0=1550 W/m', weak trip

0 2 3 4 5 6 7 8 9 10 11 12 13
position

FIGURE 13. Dimensionless surface temperature on flow side with smooth development plate,
x, =0.838 m, [Ortega et al., 1994].

0.03 ~-------------------------,

Re::155,700 0=581 WJml, ZZ trip

f\8::155,700, 0=775 W/rrta, ZZ trip

Rs=389,200, 0=775 Wfm2, ZZ trip

0.02
Re=nB.400, 0=1550 Wfm2, weak trip

Re=nB,BOO, O=~SSOW/W, ZZ trip

0.01 Ae=1, 167,700, 0=1550 W/ffil, ZZ trip

-----
Re=1,556,900, 0=1550 Wlrril, weak trip

Re=1,556,900, 0=3100 W/m2, weak trip

2 3 4 5 6 7 8 9 10 11 12 13
position

FIGURE 14. Dimensionless surface temperature on back side with smooth development plate,
x, =0.838 m. [Ortega et al.. 1994].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 123

third, denoted by "ZZ trip," is significantly lower. The weak trip cases corre-
spond to smooth plates with a single 0.5-in. span wise strip of sandpaper at the
beginning of the smooth plates; this boundary layer trip was not effective in
tripping the boundary layer and was replaced by a strip of sandpaper that was
placed in a zig-zag pattern on the plate, with a peak-to-peak amplitude of 2.0
in. The ZZ trip was more effective in tripping the boundary layer, and thus, at
the same velocity produced lower temperatures.
A typical nonnormalized surface temperature distribution is shown in Figure
15, for a smooth development section and freestream speed of 5 m/s. The LO-
in. source again is located 3.0 in. from the origin. In the downstream thermal
wake region, it is noteworthy that within approximately one heater length scale,
the heated side and the back side temperatures become indistinguishable as
does conduction in the substrate, implying that the normal temperature gradi-
ent in the substrate vanishes. The continued temperature decrease further down-
stream, therefore, is caused solely by diffusion in the fluid. This peculiar situation
arises because the back surface is insulated in this case. Close examination of
Figures 13 and 14 will reveal that the thermal wake decays more rapidly for the
turbulent flow cases, i.e., for velocities less than about 5 m/s (Re = 389200) for
the smooth plate cases. It may also be seen that the onset of turbulent convec-
tion confines the region where substrate conduction is significant to no more
than one length scale upstream from the source. Despite the conjugate nature of
the problem, it is nevertheless tempting to define a local convective heat trans-
fer coefficient as

(6)

and a local Nusselt number as

Nu = hL, == (qtnta/ - IJcond - IJrad) L, (7)

k.r (T.,.- ~,)kf

Figure 16 shows that typically Nu is well behaved, decreases monotonically

in the streamwise direction, and decreases monotonically with flow velocity,
all consistent with expectations. Figure 17 demonstrates the dependence of the
source average Nusselt number (averaged on the source area) on Reynolds num-
ber for a fixed source position of x, = 0.838 m and for both smooth and grooved
development surfaces. The results are surprisingly well behaved and readily in-
terpreted. Regardless of the nature of the development surface, the data collect
reasonably well on two paths: the upper path corresponds to the laminar regime,
and the lower path corresponds to the turbulent regime. As seen previously in
Figure 12, the transition from the laminar to turbulent behavior is dependent
on the nature of the disturbances introduced by the surfaces. The best fit cor-
relations through the data of Figure 17 are shown for the sole purpose of as-
sessing the dependence on velocity.
124 Air Cooling Technology

25 250

-----
20
.
flow side 200

back side
~ --o- N"
5l 15 150 E
!
heat flow
·;::
~ $:
::>
-e~ 10 ,g
100 'lil
(!)
E .<::
!!

5 50

0
0 2 3 4 5 6 7 8 9 10 11 12 13
position (inch)

FIGURE 15. Surface temperature elevation above ambient on flow side and back side with smooth
development plate, x, = 0.838 m, U = 5 m/s, Q,oml =0.5 W/in.', [Ortega et al., 1994].

140

120 Re=1,557,400, 0=3100 W/m2

-<>-
Re=1,557,400, 0=1550W/m2
100 ----
Re=1,168,000, 0=775 W/m>
(i; ___.._
.a
E 80 Re=778,700 0=1550 W/m2, ZZ.
::>
<: ---&-

Qi Re=778,700, 0=1550 W/m2

<ll
<ll 60 -A-
::>
z Re=778,700, 0=775 W/rrf'
-o-
40
Re=389,400, 0=775 W/m2

20 Re=155,700, 0=775 W/m2

Re=155,700. 0=581 W/m2

0
0 0.2 0.4 0.6 0.8
position on source

FIGURE 16. Local Nusselt number on source based on T" and L.,, smooth development plate,
x,=0.838m, [Ortegaetal..l994].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 125

turbulent regime
o flat plates
• flat plates Nu=0.042 Re 0·54 il(""
100 f- • 1/4 inch grooves X

• 1/4 inch grooves

" 1/2 inch grooves
• 1/2 inch grooves
• 1 inch grooves
• 1 inch grooves

laminar regime

Nu =0.320 Re 0·35

100 1000
Re "10-a
X

FIGURE 17. Average Nusselt number on source, hL.Jk1, dependence on Re." x =x, =0.838 m,
[Ortega et al., 1994].

CONJUGATE MODELS
Laminar Channel Flow-Finite Difference/Control Volume Method
Numerous approaches for modeling the conjugate problem described by
Figure 7 are possible, and many have been reported in the literature. Seminal
work on the conjugate analysis of forced convection heat transfer from small
isothermal heat sources embedded in large substrates for hydrodynamically
fully developed laminar channel flow was performed by Ramadhyani et al.
[ 1985] followed by the study oflncropera et al. [ 1986] on flush mounted isother-
mal heat sources embedded in one wall of the horizontal channel with hydro-
dynamically fully developed laminar or turbulent flow. Their efforts were
concentrated on liquid cooling but their numerical approach was a general
control-volume-based method in both the fluid and the solid. They conducted
experimental studies and compared their results with a numerical investigation
using a finite volume CFD approach in the fluid domain and a finite volume
approach in the solid domains. Krause et al. [ 1989] repeated their study using
a finite element method and investigated the effects of buoyancy. Sugavanam
et al. [ 1995] performed a similar study for air and extracted many pertinent
physical issues using the method. The approach of Sugavanam et al. is dis-
cussed next in some detail as it is representative of a full numerical formula-
tion of the problem.
The domain of interest is shown in Figure 18 in which the strip source of
heat is deployed on a conductive wall with laminar flow of air entering with
126 Air Cooling Technology

H kf
.~ks
Xs
f
4;
: : :
: : :
:
~
L f

Lx
H kf

///////////////////////////////////////////////////////. /////////////////////////////////////

FIGURE 18. The general two-dimensional conjugate problem [Sugavanam et al., 1995].

uniform velocity at the inlet to the channels formed by the substrate plate and
the insulated walls above and below it. The exit conditions are that the stream-
wise velocity and temperature gradients in the flow direction vanish at the exit
plane. Unlike the experiment previously discussed in which the back side of
the substrate surface was insulated, in this numerical experiment both the in-
sulated backside and the convectively cooled backside cases were considered.
For brevity, only results for the insulated backside conditions are discussed.
The mathematical formulation for the case of the insulated backside is pre-
sented in brief. Steady, two-dimensional, Newtonian, incompressible, and lam-
inar flow is assumed, and the properties of the material of the substrate and the
fluid are considered to be uniform, isotropic, and constant. The primary vari-
ables are normalized as
T-T0
X=_!_ Y=L U=.!:!__ V=~ 8=-- p = ___!!__, (8)
Dh Dh Uo Uo qDhlk1 pUo

Re = UoDh Pr= ~ Pe = RePr (9)

v ()(

where Dh =2H is the channel hydraulic diameter and q is total power dissipa-
tion of the source per unit area.
With the above commonly applied assumptions, the continuity, x-momen-
tum, y-momentum, and energy equations for the fluid side are given as

au+ av = 0 ( lO)
ax ar
au au aP l
u ax + v ar = -ax+ Re ax" +arz
[azu azul (11)
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 127

(12)

(13)

and the corresponding boundary conditions at the inlet and exit are

u =1, V=O, 8=0 for X=O, and _!_ < y < (H + t) (14)
D"- - Dh

au =O V=O, ae =0 for X=!::__ and _!_<Y<H+t (15)

ax ' ax D" D" D"
Zero velocity, no-slip conditions are enforced at the interface and the upper
wall, and the upper wall is assumed to be adiabatic, leading to

t Y= H+t (16)
U=O, V=O for 0:::;; X:::;; L/D11 and
Y= D"'
D"

ae =O
ay for O<X<!::_
- - D", and Y= H+t (17)
D"
The Laplace equation describes the steady conduction in the solid side

(18)

and the ends of the substrate are taken to be adiabatic, implying

ae = 0 for X=O, X=D, L

and O<Y<_!_ (19)
ax h
- - D"

ae = 0 for o:::; x:::; ~", and Y=O (20)

ar
Finally and most critically, it is necessary to match heat flux and temperature
at the interface between the fluid and solid domains

k,[ae] = [ae]
kr ()y , ar f for 0:::;; X:::;; X, and

k, [ae] _ [ae] _ 1 for <X<X

X ,_ L,
- ,+fi (22)
kf ar ' ()y f - h

er = e, for and (23)

128 Air Cooling Technology

A numerical solution was found using the well documented simpler scheme
described by Patankar [ 1980], and the details are available in Sugavanam [1994).
Unless otherwise specified, the results presented next are for the following nom-
inal conditions: Reynolds number based on hydraulic diameter, ReDh = 1,260,
corresponding to an average inlet velocity of 1.0 m/s, channel height H = l em,
heat source streamwise length L, = 1 em, and substrate thickness t = 0.5 em.
Variations of temperature, heat flux into the fluid, and Nusselt number along
the interface, for a fully developed flow maintained at a constant Reynolds num-
ber and kslkr varying between 0.1 and 100, are shown in Figures 19, 20, and 21.
The source is located at a position 2.5::; X::; 3.0. The normalized temperature
along the interface, Figure 19, remains undisturbed until the upstream conduc-
tion becomes nonnegligible. It then rises until it reaches a maximum on the heat
source and decreases thereafter. In the wake region downstream of the source,
the nondimensional temperatures for all conductivity ratios merge. It is espe-
cially instructive to learn that the downwind thermal wake achieves the adia-
batic board behavior within only one length scale downstream, for ksfkf::; 10,
and within two length scales for greater conductivities. This is partly explained
by the variation of the heat flux into the fluid side shown in Figure 20. For in-
creasing kslkr the spreading of the heat flux due to the board conduction is read-
ily apparent, but the conduction is preferential to the upstream surface since the
near wall fluid is cooler in this region. In the far downstream there is no heat
flux into the fluid from the board, and hence the upstream distribution of the
heat flux from the board into the fluid due to board conduction has little effect

0.06
-IU
---1
0.05 --10
······100

0.04

..,. 0.03

0.02

0.01
'
/ '
3 4 5 6 7 8
X

FIGURE 19. Interface temperature varying with position for fully developed !low at Re = 1260
for insulated board backside [Sugavanam, et al.. 1995 ].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 129

1 ~--------~==----------------------,
-0.1
k/k --l
f --- 10
0.8 s ····100

0.6

0.4

0.2

oL-~~~~~~~~~~~~~~~~~

0 1 2 3 4 5 6 7 8
X

FIGURE 20. Heat flux into the fluid varying with position for fully developed flow at Re =!260
for insulated board backside [Sugavanam et al., I 995].

80~--------------------------------,

70
60
50
z 40
,,
30 ..
~~~-~~,<~';\

20 --------------------------- '' \

10
0
0 1 2 3 4 5 6 7 8
X

FIGURE 21. Nusselt number on the board-fluid interface as a function of position for fully de-
veloped flow at Re =1260.0, insulated board backside [Sugavanam eta!., 1995].
130 Air Cooling Technology

on the wall temperature. The wall approaches the true adiabatic wall tempera-
ture, for all kJk1 ratios. A local Nusselt number may be arbitrarily defined as

Nu = -[elf aae? 1 (24)

where qr is the local convective heat flux into the fluid. The length scale used
is arbitrarily chosen as the channel hydraulic diameter. As shown in Figure 21,
the local Nusselt number is highly dependent on the conjugate board conduc-
tion. In the downstream wake region the heat flux approaches zero but the in-
terface temperature elevation over the inlet temperature does not due to the
convective wake effect, thus Nu approaches zero. The Nusselt number in the
upstream region is a ratio of two small numbers, the heat flux from the board
into the fluid and the temperature rise of the interface over the inlet. In the up-
stream region the ratio is nearly constant, but its value is dependent on the con-
ductivity ratio. The difference in the behavior of Nu in the upstream region of
the source compared to the downstream region is an excellent example of the
difficulty in predicting Nu in heat transfer problems that have significant con-
jugate coupling and highly nonuniform surface thermal boundary conditions.
Nu is lowest for the most conductive case as a result of heating of the near wall
fluid upstream from the source through upwind conduction through the board.
Increasing board conductivity results in a monotonic decrease of source aver-
aged Nusselt number from its adiabatic board value as illustrated in Figure 22
for fully developed flow. This result is somewhat misleading in that the degra-

35 .----------------------------------------.

30 1- 01260.0
~-126.0
I

20 1-

li
15 1-

10 ~------ -o-------<>-- -O- -o----- <>---

--o-_-o-o.,_
--o- __
---0
5 1-

0 L-----~~~~~--~--~--~~~--~~~~~~
0.1 1 10 100
k
s
I k
f

FIGURE 22. Source averaged Nusselt number dependence on conductivity ratio for fully de-
veloped flow, insulated board backside [Sugavanam et al., 1995].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 131

dation of Nusselt number, when defined in terms of a fixed ambient tempera-

ture reference, appears to degrade the heat transfer from the heat source. In fact,
increasing board conduction also increases the effective board area available
for heat transfer and thereby decreases the heat source temperatures, a fact that
is best observed in Figure 19. That fact notwithstanding, the source averaged
Nusselt number may also be used to show the effects of the developing hydro-
dynamics, as illustrated by Figure 23. At a fixed board conductivity, the Nusselt
number on the source decreases from its highest value at the inlet of the chan-
nel, where the boundary layer is thinnest, to its constant value under fully de-
veloped flow conditions. The asymptotic fully developed limits are indicated
in Figure 23. Correlations were developed for the asymptotic fully developed
flow Nusselt numbers as a function of Peclet number and board to fluid con-
ductivity ratio. If Nuad is the Nusselt number for the adiabatic board case (k..lkr
= 0), the correlational fit may be forced to yield the adiabatic board solution
when kiktgoes to zero. After a series of trials the following correlation emerged:

k )0.586
(
Nu _ { 1.186 }- !/; (25)
Nuad - Pe0.013

where Pe = RePr = UoDh/a.. Equation 25 compared with the actual solution to

within 1.2% for 0::; kslkr s 10. Alternatively, a simpler form valid for O.l s kslk1
s 10.0 can be stated as

45 ~------------------------------------~

k/k -o--0.1
40 ' ( -<>- 10.0

Re = 1260
35

I:Z 30

25

20

15
0 1 2 3 4 5 6 7 8
X

FIGURE 23. Source averaged Nusselt number dependence on position of source, laminar de-
veloping flow, Re = 1260. insulated board backside [Sugavanam et al., 1995].
132 Air Cooling Technology

-{).0754

Nu = 1.833 Pe 0 ·37
(
~; )
(26)

and this form collapsed the computed data to within 10%. Both Equations 25
and 26 are for the single substrate thickness, tiL, = 0.5.
An important virtue of the single source solutions is that they may be used
to find the solution for a situation of multiple sources on a conducting substrate,
interacting both convectively and conductively through the board by linearly
superposing the single source kernel solutions. The energy equations both on
the solid side and the fluid side are linear with respect to temperature, thus al-
lowing the superposition. This fact is verified numerically for two discrete, iden-
tical, flush-mounted, constant-powered heat sources embedded in a conductive
substrate for the case where the backside of the board is insulated. Figure 24
demonstrates the comparison between the superposed interface temperature so-
lution and the actual temperature solution for two discrete, flush-mounted
sources under fully developed flow conditions and ksfk1 of 10. The length of the
channel is 17 em with the first source located at 5 em from the inlet and the
second source spaced 5 em from the first source. Perfect agreement is observed
as expected. It can be concluded that the solution for a single isolated source
of heat on a conducting substrate may be used as a kernel solution thus reduc-
ing the computational time required for a thermal design tool.

0.045
---lsi source
0.04 -2nd source
-both sources
0.035 o superposition
0.03 k/k =10
• f

= 5 em
... 0.025

0.02

0.015

0.01
0.005 I
/······ ....
--- /

3 4 5 6 7 8 9
X

FIGURE 24. Interface temperature varying with position for fully developed flow at Re =1260,
for two sources with distance between sources, La = 5 em, k,/k1 = 10, insulated board backside
[Sugavanam et al., 1995].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 133

Laminar Channel Flow-Boundary Element Method

The finite difference/volume approach discussed in the previous section is
by far the most commonly used method in commercially available CFD codes
that allow solution of conjugate problems. The accuracy of the approach may
suffer in regions of high gradients because of improper grid resolution. In par-
ticular for the problems of the type described in this chapter, proper computa-
tion of the heat flux at the fluid-solid boundary, especially in regions of rapid
changes such as in the vicinity of heat sources, is difficult using a finite vol-
ume approach. A powerful alternative approach, using the boundary element
method (BEM) in the solid domain, is computationally superior for the two-
dimensional problem because it allows the computation of both temperature
and heat flux as primary variables [Kabir et al., 1995] and it can be readily im-
plemented with a variety of flow solvers. Because the approach is not well-
known in the electronics thermal management community, a brief description
of its implementation follows, and results for the two-dimensional problem of
Figure 25 for fully developed channel flow are discussed and compared to the
results of the finite-volume approach.

Solid Side Model

In the boundary element method, only the boundaries of the domain of in-
terest are discretized. This is accomplished by the transformation of the gov-
erning differential equation into boundary integral equations. As stated earlier,
the accuracy of the BEM solutions is greater than those obtained using finite
element or equivalent methods because the temperature (potential) and heat
flux density (gradient) are directly calculated, thus retaining the same accuracy

FIGURE 25. Problem domain for BEM analysis of two-dimensional conjugate problem [Kabir
et al., 1995].
134 Air Cooling Technology

for both. In contrast, finite element results are accurate for the primary variable
(potential) of the field problems. The gradients are obtained by numerically dif-
ferentiating the potentials. In doing so, the heat flux densities are much less ac-
curate and are usually discontinuous between elements [Lee and Palisoc, 1990].
Referring to the general domain of Figure 26, the steady state heat conduction
equation for an isotropic solid domain Q bounded by the surface r is

(27)

subject to the boundary conditions

T =To on ro (28a)

(28b)

aT + bq = c on rM (28c)

where Tis the temperature and q is the normal heat flux density. Subscript 0
denotes the prescribed values and subscripts D, N, and M denote boundaries
with Dirichlet, Neumann, and mixed boundary condition types respectively;

0
c
cr
II
cr
z

M: aT+ bq =c
FIGURE 26. BEM elements along the boundary of a typical domain indicating three types of
possible boundary conditions [Kabir et al., 1995].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 135

n is the outward normal to the boundary. The constants a, b, and c are known
quantities (a and b both are nonzero).
Multiplying Equation 27 by a weighting function wand integrating by parts
results in

(29)

in which subscript n denotes the normal derivative. Now w is chosen as the free
space Green's function T* satisfying

"iPT* (:X) = 8 (:X) (30)

which in two dimensions has the solution

T* = __!__In
2n
(l)R (31)

R being the distance between the field point and the source point. Merging of
Equation 30 into Equation 29 and taking the limit as x tends to any point p on
the boundary results in

c(p)T(p) = fr(qT* -Tq*)dr (32)

The coefficient c is a function of the internal angle the boundary makes at point
p (local geometry). If the boundary is locally smooth c has the value of 112.
Otherwise, it may be calculated directly or indirectly [Brebbia et al., 1984].
Now Equation 32 is discretized by dividing the boundary into N number of el-
ements as shown in Figure 26, while assuming that the value of T and q at a
point within an element is related to its values at the element nodal points by
some interpolation function. If linear interpolation functions are chosen for both
temperature and normal flux densities, Equation 32 becomes

IJ ('JI,7; + 'J12T2)q* dr
N N
c,T; = :~:J('JI,q, + 'J12q 2)T* dr- (33)
;=I J j=i 1

where

and 11 is a dimensionless coordinate equal to 2yll, l being the element length

andy a local coordinate. T,, T2, q,, q2 are the nodal quantities of the jth ele-
ment along the boundary. Now defining

and adding the contributions from the two neighboring elements, (j-l) and (j),
into one term, Equation 33 can be rewritten in the form
136 Air Cooling Technology

N N
2.Hij1j = LGiJqj (34)
j=l J=l

where N is the total number of boundary nodes and each nodal coefficient Gii
is equal to the term gi/ of element (j-1) plus the term gi/ of element (j), for a
counterclockwise numbering system. The same applies for Hii· For a well-posed
problem, either Tor q or some relation between the two is known for each node.
Rearranging Equation 34, a set of Nlinear equations with N unknowns are found
as described by the following equation, which can be readily solved for nodal
unknown quantities.
N N
LAiJlJ(u) = LBiJYY) (35)
J=l J=l

The superscripts u and k mean unknown and known quantities, respectively.

Fluid Side Model

As in any conjugate problem, the interface temperature distribution or heat
flux uniquely determines the temperature field in the fluid; however, neither can
be specified a priori. In principle, any model of the fluid side energy equation
that can be posed in terms of the interfacial temperature or heat flux can be cou-
pled to the solid side formulation. To illustrate, an analytic approach proposed
in this context by Culham et al. [199la] is used, both to contrast it with the full
CFD approach of Sugavanam et al. [ 1995] and to illustrate the power of ana-
lytical approaches for the practical problems in air cooling of electronics.
It is not necessary to solve the fluid flow problem explicitly to formulate the
conjugate heat transfer if the response of the flow in question due to a step
change in either wall heat flux or wall temperature is available in analytical
form. The reason is that any arbitrary distribution of surface temperature or heat
flux may be represented by the superposition of a series of step changes, (see,
for example, the discussion of Eckert and Drake [1972], pp. 321-327). In the
conjugate problem, when the solid side is discretized, we may consider that the
objective is to develop an expression implicitly relating the fluid temperature
at the wall to the wall heat flux in each discrete interval, where neither is known
a priori. For the present two-dimensional problem it is straightforward to il-
lustrate. For a surface in which the boundary condition specified is some spa-
tially varying temperature, Tw(x), the local heat flux qj(x) can be evaluated using
the Duhamel superposition technique, first proposed by Rubes in [ 1951]

(36)

where h(~,x) is the unit thermal conductance, which may be found from the so-
lution of the relevant energy equation in the presence of a step function in tem-
perature. In this case, x is the stream wise spatial position, 0 ::; x ::; Lb, and ~ is
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 137

a local dummy variable, 0 :::;; ~ :::;; x accounting for the evolution of the temper-
ature in the flow up to the local position x. Solutions for h (~, x) are documented
for many types of flows. For example, integral formulations of the laminar
boundary layer solutions were collected by Rubes in [ 1951] and Tribus and Klein
[ 1952], and these are discussed in a subsequent section. In the present case, for
fully developed laminar channel flow, the solutions of McCuen [1962] for h(~,x)
are perfectly suitable. The unit thermal conductance is

(37)

where Cn are eigenconstants, A,. are the eigenvalues, and Y'n are the derivatives
of eigenfunctions evaluated at the wall where the step in temperature is applied.
Equation 36 is discretized along the solid-fluid interface in the streamwise di-
rection by dividing the interface along the streamwise direction into (k-1) ele-
ments, as shown in Figure 27

All the coefficients of Ti are evaluated numerically using standard quadrature

except for the case when Xi =xi, for which a dosed form solution exists. In ma-
trix form Equation 38 can be rewritten as
j

iJJ(x) = l:,aJiTw(x;)- cJ j = 2,3, k (39)

j=;2

The c/s are related to i}j(x1), and are due to Tw(x;).

Fluid
f
Interface

k-1 k
ks

FIGURE 27. Discrete along the fluid-solid interface when applying BEM scheme [Kabir et al..
!995].
138 Air Cooling Technology

Solid-Fluid Coupling
The coupling between the solid and the fluid region is accomplished by en-
suring the energy balance and matching the temperatures at the interface. That
is, at any node j, along the interface
(40)

(41)

For nonsource nodes, Qis are zero. The BEM model, Equation 35, can be in-
terpreted as a relation between the unknown }j'") and the known (specified) con-
ditions. The change in the heat flux density at the jth node, due to a change in
temperature at the /th node, can be obtained by taking the derivative of Equation
35 with respect to T;
N dq
LBu a-/.= Ail. (42)
j=l 1[

From Equation 42 it is evident that all the derivatives are independent of

temperature and flux densities, hence, an exact linear relation can be derived
to calculate the correct flux densities
(Jq
= LC1t- ~(O))_j
k
q - q(O) j = 2,3, k (43)
J J 1=2 CJI;

Here qj")) is the solution of Equation 35 based on an arbitrary prescribed Tt)). It

should be pointed out that, depending on the desired heat flux condition, the
correct temperature T1 can be determined exactly from Equation 43 without any
iteration. Multiplying Equation 43 by k, and adding the resulting equation to
Equation 39, Equation 41 can be used to arrive at the following:
k
'I.Au1J = D;
j=2
(44)

where

k a
D = ~) _!lj_ T(O) + Q + c - k q(O)
I 1=2 I di/ I I I ·' I

Equation 44 now can be solved readily to obtain the interface temperature

distribution and either Equation 35 or Equation 39 to calculate the heat flux
distribution.
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 139

For comparison, the geometry (Figure 25) was chosen to match that of
Sugavanam et al. [1995]; in the notation of Figure 25, 2HIL = 1, 2Hit = 2, L, =
1 em, t = 0.5 em, and Lb = 16 em, and the nondimensional temperature was
defined as

e = (T- T0 ) k
Q4H f·
(45)
Figure 28 compares the BEM computed results with those of Sugavanam for a
nominal velocity of 1 m/s, Re = 1260. There is nearly perfect agreement be-
tween the two techniques. The computed convective heat flux, given in Figure
29, also agrees reasonably well with the previous results shown in Figure 20.
The BEM method is probably more accurate in the vicinity of the source lead-
ing edge since it shows continuous, smooth behavior, whereas the finite volume
method, Figure 20, shows discontinuous behavior, especially at low conduc-
tivity ratios. When comparing local Nu at the interface, Figure 30, more obvi-
ous differences can be seen. Whereas the finite volume approach predicts the
Nu to approach a constant value upstream of the source, different for each con-
ductivity ratio, the BEM approach predicts this behavior only at relatively high
conductivity ratio; otherwise, the Nu starts at zero and rises to its peak value at
the leading edge of the source, but with an unusual intermediate inflection point.
We believe that the BEM approach captures the true behavior of Nu because

0.05

0.04

~ 0.03

0.02

0.01

0 ~~~~~~~~~~~~~~~~~~~~

0 2 3 4 5 6 7 8
X

FIGURE 28. Interface for fully developed channel flow; comparison of BEM result to result of
Sugavanam eta!., [ 1995] using finite volume method [Kabir eta!., 1995].
140 Air Cooling Technology

1
/ \
0.9 t- k!l/kj
\
O.B 1-
0.1
1.0
0.7 t-

-
10.0
100.0
0.6 1- ···.
0
0.5 r--
"'cr
0.4 r--
0.3 1-

0.2 1-

0.1 li-
[\..---·-..
0
0 1 2 3 4 5 6 7 8
X

FIGURE 29. Convective heat flux into fluid q" computed with BEM approach, normalized by
total heat source power per unit area, Q, fully developed channel flow, Re =1260 [Kabir et al., 1995].

80

70 k

80 0.1
1.0
10.0
50 100.0

30

20
I
_i___ -~--·"·,·· .
10
·. --- ·-·-·-·-·-·-·-·-------·
0
0 1 2 3 4 5 6 7 8
X

FIGURE 30. Nusselt number distribution along interface computed with BEM approach, fully
developed channel flow, Re = 1260 [Kabir et al.. 1995].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 141

both wall heat flux and temperature are computed directly. The finite volume
method requires that the wall heat flux be determined by numerically evaluat-
ing the normal gradient, and therefore its accuracy degrades when both heat flux
and temperature rise approach zero. On the other hand, the BEM method is not
as straightforward to extend to three dimensions as either the finite-differ-
ence/volume or the finite element method, and this is a major drawback.

Laminar Boundary Layer Flow

Given the previous discussion on the flow regimes in typical PCB applica-
tions, an equally important flow to consider is that of laminar boundary layer
flow. The domain of interest is represented in Figure 31. Again, various ap-
proaches for the conjugate problem can be taken and several are briefly illus-
trated here.
Using a finite difference, control volume approach similar to that of
Sugavanam et al. [1995], Gorski and Plumb [1990, 1992] analyzed the conju-
gate behavior of the two-dimensional strip heat source problem [ 1990] and the
three-dimensional rectangular source problem [1992] with the flow described
by the analytical solution of Blasius for the laminar boundary layer. They con-
sidered only the limiting case of infinitely thick substrate. In the domain illus-
trated by Figure 25, they solved the energy equation for the flow side, Equation

FIGURE 31. Problem domain for BEM analysis of two-dimensional conjugate problem with
laminar boundary layer !low [Kabir et al., 1995].
142 Air Cooling Technology

13, using a finite-difference method with the velocity field given by the Blasius
solution at each streamwise position. In the solid side, the energy equation,
Equation 18, was discretized also with a finite-difference method, and the en-
tire system of discretized equations in both domains was simultaneously solved
using a line-by-line algorithm with temperatures along each line evaluated using
matrix inversion. The boundary and matching conditions were the same as those
previously formulated, and the discontinuity in the thermal conductivity at the
fluid/solid interface was handled using the harmonic mean technique suggested
by Patankar [1980]. Their results were very similar to those of Sugavanam et
al. [ 1995] but cannot strictly be compared because the developing channel flow
adds freestream acceleration to the boundary layer. Gorski and Plumb con-
ducted a systematic variation of parameters, varying Peclet number, Pe =
( U 0 X,) a between 1 x 103 and 1 x 105 , conductivity ratio, kslkr between 0.1 and
10.0, and stream wise position to source-length, xJ L, between 5 and 100. Their
source averaged Nusselt number, Nu = hLJkr was correlated to within 5% by
the expression

071( )-o.057
Nu = 0.486Pe 0 · 53 (~:). :; (46)

The Nusselt number was defined in terms of the source length, L,, and Peclet
number was defined in terms of streamwise position of the source, x,, in con-
trast to the correlations of Sugavanam, Equations 25, 26 in which the channel
hydraulic diameter was used for both Nu and Pe. As in the previous case for
fully developed channel flow, increase in substrate conductivity decreased the
source Nusselt number. In addition, the boundary layer growth decreased the
average Nusselt number as shown by its inverse dependence on source position,
varying approximately as x-"". This is consistent with the channel flow prob-
lem, Figure 23, in which Nu decreases in the channel entry region to its fully
developed value. In some instances it might be more useful to average the heat
transfer coefficient or Nusselt number over the entire substrate area rather than
just over the source. For example, if one were to perform a conjugate analysis
of the heat transfer in the plastic monochip package shown in Figure 5, it might
be more meaningful to determine the average heat transfer coefficient over the
entire cooled top surface area. In the present cases, an average heat transfer co-
efficient is extracted for the source area rather than for the entire board.
Alternatively, the BEM approach, Equations 27 to 44, can be used with the
laminar boundary layer flow provided a suitable expression is used for the unit
surface conductance, Equation 36. The exact solution for boundary layer flow
is cited in Eckert and Drake [ 1972] as

~
h('>,x)=-~
0.332kf
X
I
Pr 3 Rei
I
lJ1- {~~}~ )-j (47)
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 143

Culham eta!. [199la] showed that a general form of Equation 47 can be writ-
ten as

(48)

where the parameters, C, y, ~' a, and b have been derived for various types of
flows, including the present laminar boundary layer flow. Table 3, abstracted
from their paper, summarizes some of the available solutions.
Using Equation 4 7, results were obtained by Kabir et al. [ 1995] using the
BEM approach in the solid, for kslkr between 0.1 and 100 and for free stream
velocity of 1.0 m/sec, equivalent to ReLb = 10,000. Results were very similar to
those for fully developed channel flow. Figure 32 shows that local Nusselt num-
ber behaves in much the same way, with the intermediate inflection point in the
upstream region occurring at low board conductivity. By comparison with
Figure 23, for the fully developed channel flow, and Figure 33, for the bound-
ary layer flow, one may note that for both conductivity ratios, Nu for the chan-
nel problem is significantly higher, presumably because the growth rate of the
unconstrained boundary layer is greater than the growth rate of the channel
flow. In addition, the source Nusselt number does reach a fully developed as-
ymptote. For low to moderate conductivity ratios, the thermal boundary layer
thickness over the source is less than half the channel height so that the ther-
mal boundary conditions on the opposing wall have no influence on the source
heat transfer.

TABLE3
Constants To Be Used in Equation 48 for Boundary Flow

Solution c y ~ a b Reference

laminar

exact 0.332 1/3 1/2 3/4 1/3 Eckert [ 1950]

approx 0.304 1/3 l/2 3/4 1/3 Rubesin [ 1945]
approx 0.323 1/3 1/2 3/4 1/3 Kays [1993]

turbulent

approx 0.0288 1/3 4/5 39/40 7/39 Rubesin [ 1951]

approx 0.0289 1/9 4/5 9/10 1/9 Seban [ 1950]

Source: Culham et al.. 1991a. With permission.

144 Air Cooling Technology

70

60 k /k
s f
0.1
50 1.0
10.0
100.0
40
::I
z
30

20
·- ... ,
__

I
'·-1·-.
10
·.:-:-·- -·-.
0
0 1 2 3 4 5 6 7 a
X

FIGURE 32. Nusselt number distribution along interface computed with BEM approach, lam-
inar boundary layer flow, ReLb =!0,000, source position xiLh =5/16 [Kabir et al., 1995].

k /k
s f

~
~
Re=1260.0

'<>-
--<>--o_
---<>-----
--<>- - - <> - - - <>- - -

10
0 1 2 3 4 5 6 7
X

FIGURE 33. Average Nusselt number on source computed with BEM approach, dependence
on stream wise position of source, laminar boundary layer flow, ReL, = I 0,000 [Kabir et al., 1995].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 145

THREE-DIMENSIONAL SITUATIONS:
RECTANGULAR SOURCES OF HEAT
We next consider selected approaches and results for the problem of Figure
6(b ), the cooling of a rectangular, flush-mounted source of heat on a conduct-
ing substrate or board. Figure 34 indicates the geometry with the pertinent geo-
metric parameters. Although the system ignores the difficulties associated with
more difficult flow, as for example in Figure 6(f), the fact that the flow is well-
known facilitates the understanding of the conjugate phenomena. In addition,
as demonstrated by Culham et al. [199lb], at low laminar Reynolds numbers,
the flush-source model may give reasonable agreement with measured data
on PCBs.

BOUNDARY LAYER FLOW

In the coordinate system given in Figure 34, the boundary layer equations
for mass and momentum are commonly given as

(49)

(50)

and the energy equation, retaining molecular diffusion terms in all three direc-
tions is

FLOW

~ '

/1=''·"[/
~-'~----~------------------·-·
~~ X /

FIGURE 34. Geometry for boundary layer flow over a rectangular heat source on a conducting
substrate or board.
146 Air Cooling Technology

(51)

In general only the last term on the right, representing diffusion normal to
the substrate surface, is retained for the low speed laminar boundary layer, but
all three terms are retained here for initial discussion. The energy equation in
the solid is in general

(52)

where S is a volumetric source term that allows specification of heat dissipa-

tion at the location of the source.
Gorski [ 1992] and Gorski and Plumb [ 1992] extended their finite-difference,
control-volume-based approach discussed in the previous section to the cool-
ing of a rectangular source on an infinitely thick substrate in a laminar bound-
ary layer flow. Rather than solving the laminar boundary layer equations 49
and 50, they adopted the classical solution of Blasius for the velocity compo-
nents. The energy equations 51 and 52 were discretized using a finite-
difference, control-volume approach with a staggered grid. The discontinuity
in thermal conductivity at the fluid/solid interface was handled using the har-
monic mean suggested by Patankar [1980]. The edges of the board were as-
sumed insulated and the substrate thickness was chosen to be large enough such
that infinitely thick substrate behavior was achieved. The substrate surface area
was assumed to be large enough to allow the fluid temperature at the leading
edge and at the lateral edges to be taken as the undisturbed temperature Tn. At
the outflow boundary, the fluid temperature gradient in the streamwise direc-
tion was assumed to be zero. Results were obtained for 0.1 ::;; kJk1 ::;; 10, 5 ::;;
(x,.c~)/2b::;; 150, 0.2::;; lib::;; 5.0, and 1 x 10'::;; Pe::;; 1 x 105 , where Peclet num-
ber for this case is Pe = UoX.vtfa. Representative results are shown in Figures
35 and 36. Figure 35 demonstrates the dependence of temperature on the span-
wise centerline of the source on board conductivity at a fixed Pe = 5 x 103 • At
the lowest conductivity, in which the substrate is nearly adiabatic, the temper-
ature of the square source nearly matches that of the two-dimensional strip
source, indicating that span wise conduction either through the substrate or fluid
is negligible compared to stream wise advection. As conductivity ratio increases,
the square and strip source temperatures substantially diverge, even for kikr =
1.0, in which the solid conductivity would be that of air. This divergence would
decrease at higher Pe in which streamwise advection would dominate the heat
transfer. For example, Figure 36 shows that for a conductivity ratio of 1.0, but
at Pe = 7.5 x l 04 , the square source, lib = 1.0, more nearly approaches the two-
dimensional strip source behavior. For lib:;:::: 5 .0, the square source temperatures
substantially match the two-dimensional solution. Based on the data generated
using their model, the authors developed the following correlations for average
source Nusselt number, Nu = h2blkr
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 147

30
- - - -2-D
----3-D

Cll
g 20

--
CD
I..
CD

=a
CD
I..
2CIJ
; 10
ll..
E
Cll
1-

o~_.~J_~~~~~~~~-L~~~-7

0.11 0.2 0.4

Non-dimensional distance surfaee,

FIGURE 35. Effect of the board conductivity on the span wise centerline temperature for a square
source compared to strip source of same streamwise length in a laminar boundary layer, Pe =
U,x,."!a =5 x 10', Xs.c/2b = 25, 1/b = 1.0 [from Gorski, 1992].

Nu = 0.60Pe 0 .48 (-
2b )0.63( -
b )0.18 (53)
x,,cl A/P

Nu = 0.43Pe 0·52 ( -
2b )0.70( b )0.07 (54)
- k,fkf ::= 10.0
Xs,ci A/ P

where AlP is the source surface area to perimeter ratio. In order to correlate the
source geometry, both the aspect ratio lib and the source surface area to perime-
ter ratio normalized on source length (AIP)Ib were tried with the latter giving
slightly better fit to the data. Since in practice the total heat dissipated Q is gen-
erally known, it is useful for practical calculations to correlate the total heat
transferred to the fluid Qj. The authors found the following best fits of their nu-
merical data

Qt = 0.36{Pe(J:.!!._)I.s(AIP)I.s}o.!9 (55)
Q X,ct b
148 Air Cooling Technology

za

u 16. 3D, Ub .. 5.1!

!...
30, lib .. 2.0
Ill
()
<: 30, Ub . 1.0
--
Ill
i 12
=a
Ill
30, Ub .. 0.5
;a
L.
8
"'...Ill
1:1.. 30, Ub .. 0.2
e
Ill
!-
4

llt..-..~.--!........1-..l-J-..l--l..-!--!........!.-I......L-'-_._...._..~.._1-,....!_...........I

!I.CI 0.2 0.4 11.6 0.8 Ul

Non-dimensional distance along heater surface,(:

FIGURE 36. Effect of aspect ratio 1/b on the span wise centerline temperature of a rectangular source
in a laminar boundary layer, Pe =U,x,_d(:J. =7.5 x HY, x,,,i2b = 100, k.Jfv= 1.0 [from Gorski, 1992].

Qf = 0.063{Pe(J!?__)I.s(A/ p)l. 8
}
036
k,fkr =10.0 (56)
Q x,.cJ b

In contrast to the full numerical approach of Gorski and Plumb, Culham et

al. [ 1991 a, 1991 b] developed an analytical/numerical model forthe problem of
Figure 34 for a board with finite thickness. The board was assumed to be thin
enough so that normal {z-direction) temperature gradients could be neglected.
This "thin board" model was justified for board Biot number Bi =htlks < 0.1,
which includes most practical applications of air-cooled PCBs. Neglecting nor-
mal temperature gradients allows integration of Equation 52 across the thick-
ness of the board to yield the thin board result

(57)

where q, is local power dissipation per unit surface area, lJJ is the convective
heat flux into the fluid, and q, is the radiative heat flux from the board surface.
Both qr and q, were assumed to be the sum of the fluxes from the upper and
lower surfaces. The lateral edges of the board were assumed to be insulated.
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 149

The solid-side solver was built on a finite-element based approach. Culham et

al. [199la] introduced the solution to the flow-side energy equation by adopt-
ing the unit surface conductance method discussed previously in the context of
the BEM. If diffusion in the fluid is neglected in the streamwise (x) and span-
wise {y) directions, Equation 51 becomes

(58)

which is the common boundary layer form of the energy equation. Again, under
conditions of a temperature-specified wall, the local heat flux is given by
Equation 36, where the form of the unit thermal conductance h (~,x) is depen-
dent on the type of flow. For boundary layer flow, Culham et al. [199la] pro-
posed the general form given by Equation 48. Substitution of Equation 48 into
Equation 36 gives

(59)

where the constants were previously tabulated in Table 3. It was further demon-
strated that Equation 59 can be inverted so that the wall temperature elevation,
8w(x) = Tw(x)- To, can be expressed explicitly for conditions of a heat flux spec-
ified boundary as

(60)

where r is a gamma function. This form can be simplified if qris constant over
a discrete interval, which is the form adopted by Culham et al. in coupling to
the board FEM equations.
It is worth emphasizing that these solutions neglected both streamwise and
spanwise conduction in the fluid resulting therefore in a solution that is func-
tionally dependent only on the streamwise position. In the absence of conju-
gate coupling to the board, this solution could not be used in the
three-dimensional problem, because no mechanism would exist for spanwise
diffusion of the thermal energy. However, when coupled to the solid-side so-
lution that allows conduction in the board in both the stream wise and span wise
directions, the two-dimensional solution represented by Equations 59 or60 may
be used, with the implicit assumption that the magnitude of span wise or lateral
conduction in the board renders the spanwise diffusion in the fluid negligible.
Obviously, this assumption can be true only for kJk1 > 1.0, but this includes all
the cases of practical importance. Note however that this fluid side solution
would not be useful in the case of a poorly conducting or adiabatic board.
The finite-element model in the board allows discretizing of the board into
finite areas sufficiently small to allow the surface heat flux calculated at the mid-
150 Air Cooling Technology

point to be representative of the heat flux over its area. The unit step function
models for the fluid side and the FEM models on the solid side were coupled by
iterating between each solution until the calculated wall heat flux distribution
forced the wall temperature calculated using the fluid flow equations to match
the wall temperature calculated with the solid side model. The iterative proce-
dure was instituted in a general purpose code for the PCB problem [Culham et
al., 199la], and a detailed parametric study was performed [Culham eta!., 199lb]
on the effects of board emissivity, board thermal conductivity, and for the two
source model illustrated in Figure 37, effects of upstream source heating and in-
tersource spacing on the downstream source. Nomenclature is shown in Figure
37. The dimensionless temperature on the span wise centerline for the two source
test case studied is presented in Figure 38, where

(61)

and the average heat flux over the front and back surface of the board is de-
fined as

(62)

As board conductivity gets very large, heat spreading renders the board
nearly isothermal. The board conductivity influences the peak temperature on

y
L

L/2

2 ~ l x t heat sources
(flu.sh mounted}

BASE CONDITIONS
L x Wxt =0.2m x0.1m x0.0016m
k, =2Wim -K T~=20C

l X I =0.02m X 0.02m QI=Qz=2W

FIGURE 37. Geometry and nomenclature for two source PCB [from Culham et al.. 1991 b ].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 151

2.4r---------------------------------------,
k. (W/mK) = 0.2

1.6

...8
X 1. 2

.8

•4

.2 •4 .6 .8

FIGURE 38. Spanwise centerline temperature dependence on board thermal conductivity, Q1 =

Q2 = 2.0 W, [from Culham eta!., l991b].

both sources, and the convective thermal wake mechanism increases the tem-
perature of the downstream source above the upstream source for all conduc-
tivities, the relative difference depending on the intersource spacing. In a
separate study [Lemczyk et al., 1991 ], it was found that an accurate method for
determining the effective thermal conductivity of a multilayer printed circuit
board is by use of the harmonic mean of the thermal conductivity of individ-
ual layers calculated using appropriate weighting of the series and parallel re-
sistances. Figure 39 shows the effect of the strength of the upstream source on
the temperature of the downstream source. For the intersource spacing shown,
about 90% of the temperature increase of the downwind source is due to its
own heating, while I 0% is due to the influence of the upwind source. The up-
wind source temperature increases linearly with its own power dissipation.
Effect of intersource spacing is shown more generally in Figure 40, where the
temperature increase of the second source !1T2 above its base value for the iso-
lated case, 1'1ho. is shown. The ratio <j> is

<j> = !1T2 - 1'1T"·" (63)

f1T2.n

and d = dll is the center-to-center spacing normalized on source streamwise

length. The cased= 0 indicates that the two sources overlap and the tempera-
ture is that of a single source at twice the dissipation. Interestingly. since the
problem is linear, the temperature increase is twice !'1T2.o· The effect of the up-
152 Air Cooling Technology

2. 4

4.0
!.6
3.0
§ 2.0
)( 1.2 1.0
.,:: 0.0

.8

•4

0
0 .2 '4 .6 .8

FIGURE 39. Span wise centerline temperature for two sources, parametric on power dissipa-
tion of upstream source Q 1 for downstream source power Q2 =2.0 W, k., =2.0 W/m-K [from Culham
et al., 199lb] .

•B
54
18
.5
6 = k.(W/mK)
2
~ 0.2

'4

'2

0
-7.5 -5 -2.5 0 2.5
d
FIGURE 40. Effect on maximum temperature of downstream heat source due to location of up-
stream heat source, [from Culham et al., 1991 b].
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 153

stream heat source is felt on the second source even when it is downstream
strictly because of upwind board conduction, as can be seen by its increase with
increasing conductivity.

UNIFORM FLOW
Ortega et al. [ 1993] proposed that the simplest model of the convective flow,
one that assumes that the flow is a uniform parallel flow over the board, may
be considered to be the most basic flow for the PCB cooling problem of Figure
34, and they used it to analyze both the heat transfer from the source and the
behavior of the thermal wake downstream of the source. Sugavanam et al.
[1995], for the two-dimensional problem, showed that the uniform flow solu-
tion provides the upper asymptotic limit for the board Nusselt number, since
for a given inlet Reynolds number, it gives the highest near-wall streamwise
velocity compared to either a boundary layer flow or a developed channel flow.
Although the uniform flow model is not realistic in a strict sense, not satisfy-
ing, for example, the no-slip condition at the wall, it successfully predicts the
behavior of the downstream thermal wake [Ortega et al., 1993] that has been
experimentally observed in the turbulent flow over arrays of 3-D wall-mounted
blocks simulating electronic components. The approach is discussed here pri-
marily because it illustrates the use of full three-dimensional analytic kernel
solutions that admit diffusion in the fluid in all directions.
The flow field is assumed to be a simple uniform flow parallel to the board
with magnitude U leading to a degenerate form of the energy Equation 51

(64)

where £ is used as the symbol for the fluid effective thermal diffusivity so as
not to be confused with the molecular diffusivity, ar. If Equation 64 is used as
a primitive model for a highly turbulent flow, the effective diffusivity £ may be
interpreted as the sum of the molecular and the eddy diffusivities for heat [Kays
and Crawford, 1993]

(65)

leading to a definition for the "effective fluid conductivity"

(66)

Henceforth, the model is referred to as the uniform flow effective diffusivity

(UFED) model. Equation 64 may be nondimensionalized in the following way:

(67)

where
154 Air Cooling Technology

X*= Ux; Y* = Uy; Z* = Uz (68)

£ £ £

Brer is a characteristic temperature for the problem. The fluid can be visualized
as a bulk flow with homogeneous properties moving in the direction x over a
plane surface; heat transfer occurs by diffusion in each principle direction and
advection by the bulk flow. The kernel solution for a point source of strength
Q moving at constant velocity over an infinite plane surface is available [Carslaw
and Jaeger, 1959] and has previously been used in problems of moving sources
of heat on the surface of conducting solids [Rosenthal, 1946] and in the analy-
sis of film cooling [Eckert and Drake, 1972] by attaching the reference frame
to the moving source. The solution is given by

(69a)

where R = -.Jx2 + y 2 + z2 • In dimensionless form

8* = - 1-exp{-!(R* -X*)} (69b)

2nR* 2

where the temperature reference is given by

e - Q (69c)
ref - ke f..jU

The temperature field at z = 0 for the PCB cooling problem represents the so-
lution for a point source on an adiabatic board with a uniform isotropic flow
moving over it with a constant velocity U in the x direction. The point source
solution is discontinuous at the origin but nevertheless may be used to gener-
ate solutions for sources of finite dimension [Ortega eta!., 1993] of which the
rectangular source is the most relevant for the printed circuit board cooling
problem. Integrating Equation 69(a) over a rectangular source of dimensions
2b x 21 in the streamwise and spanwise directions, respectively, and using the
nomenclature of Figure 34 yields

*- 1 I.~{ Y*+Pe 1 Y*-Pe 1}

8 - 2 (21!:)1/2 o erf 2 (2u)l/2 - erf 2 (2u)lf2 X

{ X* +Peh- 2u X* -Peh- 2u} du (70)

erf 2(2u)'f2 - erf 2(2u)'l2 u'l2

where

Ub Ul
Peb =- , Pe 1 = - (71)
£ £
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 155

The source strength is now more appropriately indicated by q, heat flux on the
surface of the source, assumed to be uniform.
Sample solutions are shown in Figure 41 for low Peclet number and in Figure
42 for high Peclet number. For illustration, the effective diffusivity has a value
of 0.002 m'ls corresponding to the eddy diffusivity of turbulent flow of air over
a flat plate. The surface plots clearly show the three-dimensional nature of the
thermal wake brought about by diffusion in the x, y, and z directions and its re-
sponse to increased Peclet number. At elevated Peclet number, the temperature
over the source is nearly two-dimensional and one would expect the two-
dimensional strip source solution to be accurate. Spanwise diffusion in the ther-
mal wake is obvious at low Pe but relatively minor at the higher Pe as can be
seen by comparison of Figures 41 (b) and 42(b ).
The UFED model was used to formulate the conjugate problem of Figure
34 by coupling it to a thin board model that was discretized with a Galerkin fi-
nite element method [Ramanathan, 1994 and Ramanathan and Ortega, 1996].
The analytical UFED model for the fluid domain provides a particularly sim-
ple coupling to the board FEM model allowing a noniterative procedure for ob-
taining the steady state temperature distribution over a rectangular board with
a rectangular source. The analytical treatment of the board side follows that of
Culham et al. [199la] and is described by the thin board model given by
Equation 57. The lateral edges and the backside of the board are assumed to be
insulated. The domain of the board was discretized into bilinear rectangular el-
ements and the Galerkin formulation for finite element analysis was applied to
Equation 57. Bilinear shape functions were used for temperature within each
element. The process resulted in the following matrix equation:

[KG][9]- [FG] + [i~}qr]- [~~] = 0 (72)

where [KG] is the global stiffness matrix, [FG] is the matrix containing the
edge boundary conditions (in this case a null matrix because of the insulated
edge conditions), [qG!k,t] is the global coefficient matrix for the convective
heat flux, and [QG/k,t] is the global matrix for the source term. [llf] is the sin-
gle column matrix of unknown surface heat flux.
The point source solution for the flow, Equation 69(a), explicitly relates e
to llf allowing a direct substitution into the board equations. In each discrete
board element qr is assumed to vary bilinearly, and the kernel solution is inte-
grated over each element to compute the temperature distribution over the en-
tire board due to the heat flux over each element. The temperature at every nodal
point has contributions from all the elements on the surface. Discretizing of
Equation 69(a) results in a matrix relation between 8 and qr

[8] = [A][qf] (73)

where [A] is the coefficient matrix for qr, and inverting

156 Air Cooling Technology

0.1
Pe = 5.0
b

[q 1 ] = [B][9] (74)

where

[B] = [Ar 1 (75)

Then substituting Equation 74 in Equation 72 and grouping the coefficients of

the unknown temperatures into one matrix results in a system of equations in
e given by
(76)

The Gauss-Seidel iteration method was used to obtain the results presented here.
In order to observe the relevant dimensionless parameters, the board con-
duction equation, Equation 57, can be normalized using the definitions given
in Equation 68 to give

(77a)
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 157

Pe
b
= 5.0
6
..0
.......... 5
>.
A
4
Q
.9
..... 3
()

....
~
0
1-<
2
Q
C'd
~
til 1

0
0 1 2 3 4 5 6 7 8 9 10
(b) Flow direction --- > x/b

=
FIGURE 41. Dimensionless temperature rise of a square source on adiabatic surface, Pe. 5.0
2b x 21 = 2 em x 2 em, U = 1.0 m/s, E =0.002 m'/s, k, = 2.362 W/m-K; (a) surface plot, (b) sur-
face isotherms.

where the temperature reference is

e q, (77b)
ref = k, Uje
The source stream wiselength scale b was introduced arbitrarily, hence Peb could
just as well be Pe,. Comparison of the flow side equation, Equation 67, and the
solid side equation, (77a), reveals finally that the conjugate solution is ex-
pressible as

(78)

For discussion of representative results, the following base case was used:
t = 1 mm and 2b x 21 = 2 em x 2 em. The board dimensions were Lx x Lv = 10
em x 6 em for low Peb and 9 em x 4 em for high Peb. Nondimensional temper-
ature fields are plotted for low Peb of 5.0 and high Peb of 500.0 each for con-
ductivity ratios kslk, of 1.0 and 10.0. The surface plots along with the
corresponding contour plots are shown in Figures 43 and 44. Comparing the
results for Peb of 5.0 and ksfk, of 1.0 with the adiabatic solution in Figure 41, it
is clear that when board conduction is negligible, the spreading of the thermal
wake is essentially due to the diffusion in the fluid. At higher Peh the convec-
158 Air Cooling Technology

o.os
o.o1 Pe == 500.0
I>

tion becomes dominant rendering the transport through conduction in both the
solid and the fluid negligible. Comparison of the temperature fields on an adi-
abatic board for the same Peb (Figures 41 and 42) with those for kslke of 1.0
shows that at higher Peh the conduction in the solid is insignificant, and for the
range of Peb considered, conductivity ratio of 1.0 does not introduce percepti-
ble conduction effects. A tenfold increase in conductivity ratio enhanced the
conduction heat transfer from the source at all Peh. Figures 43b, d and 44b, d
show pronounced conduction and heat spreading in all directions from the
source. At low Peb the conduction in the solid dominates, and the heat is spread
more effectively on the board thus reducing the peak temperature as kJke in-
creases. At higher Peb the convection dominates, and thus even a kJke of IO.O
does not yield dramatic heat spreading by board conduction as is evident in the
contour plot, Figure 44(d). In the context of air cooling of PCBs, at high Peh
the conduction effects are not significant enough to affect laterally displaced
electronic components in the downstream wake; however, the strength of the
thermal wake does emphasize the importance of thermal interaction of com-
ponents along the flow direction by both board conduction and flow convec-
tion. Proper prediction of the thermal wake is critical.
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 159

Pe = 500.0
b

-
4
..c
; .,
3 f-
A
I
....__,
--.::: ----.::
ag=
I
2 - I
<.J I
....
<!)
I
;a I .../ _.---/ ...-.../

"'=
1 -
~
"'
0
0 1 2 3 4 5 7 8 9
{b) Flow direction - > xfb

FIGURE 42. Dimensionless temperature rise of a square source on adiabatic surface, Peh =
500.0, 2b x 21 = 2 em x 2 em, E = 0.002 m'ls, k, = 2.362 W/m-K; (a) surface plot, (b) surface
isotherms.

A comparison of the thermal wake along the flow direction can be best made
by considering the centerline temperatures. Figures 45 and 46 show the effect
of conduction in the solid for Peh of 5.0 and 500.0, respectively. Each is also
compared to the corresponding adiabatic board solution from the UFED model.
It can be clearly seen from these plots that with increase in board conduction
the thermal wake spreads preferably upstream since the convective wake down-
stream reduces the ability to remove heat downstream, hence reducing the con-
duction within the solid in that direction. Moreover, the plots show that for kJke
ratio of 0.1 the board essentially behaves as adiabatic for all the Peh range an-
alyzed. At higher Peh deviation of the thermal wake with board conduction, ef-
fects from the adiabatic wake becomes negligible. It should be noted that Peh
of 5.0 are below any practical electronics cooling applications and that the con-
ductivity of the fluid ke is the effective conductivity that is defined with respect
to the effective diffusivity (Equation 66). With respect to the molecular con-
ductivity of the fluid, the conductivity ratio of 10.0 would yield solid conduc-
tivity close to that of epoxy boards used in electronic packaging. The plots of
the centerline temperatures indicate that the thermal wake sufficiently down-
stream of the source is self-similar and as a first approximation may be com-
puted with the adiabatic board results. The advantages of using such an
analytical solution for the prediction of the thermal wake downstream of the
source has been explored in detail by Ortega et a!. [ 1993 ].
The average heat transfer coefficient or Nusselt number is a useful quanti-
tative measure of the effectiveness of heat transfer from the source despite the
fact that it is a defined quantity. In addition, it may be used in an uncoupled
 o.o7
P e "" soo.o; k /k
o.o6 b
• e
"" 1.0

o.os e· == B /( q b
/k )
"
o.o4
.... o.o3

FIGURE 43.
'" ''
N m 'l i" " " '' '"
" '' ' ' " ' ('
\ Pe. • 5 O.
'' "
m P " '' " "
k ik · • \. 0.
''~due
(b ) Pe. • 5.
<o • " I® " be
o< 'M"'
Peh = 50 0.0, 0. kJk, • \0 oo c o o d u "' ''
kJ k, = I 0. .0 . (c ) lh
• sQO.O, kJ
'' " " ·
0. k, • \. 0, (d )
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 161

0.4

o.35 Pe = 5.0; k /k = 10.0

b • e

0.3 9. = 9 /(qb/k )
c
o.z5
....

(b)

o.o7
Pe = 500.0; k• /k c = 10.0
o.06 b

9. = 9 /(qb/k )
e

o.04
.... o.03
o.UJ.
0.ol

(d)

FIGURE 43. Continued

162 Air Cooling Technology

Pe = 5.0; k /k = 1.0
b s e
6
.t::J
.......... 5
>.
1\
4
c:l
......
0
..... 3
(J

.....
~
1-<
"0 2
c:l
cd
~ 1
0 ~~~~~~~~~~~~~~~~~~~

0 1 2 3 4 5 6 7 8 9 10
(a) Flow direction --- > xjb

Pe = 500.0; k /k = 1.0
b s e

o ~~~~~~~~~~~~~~~~--~~~

0 1 2 3 4 5 6 7 8 9
(c) Flow direction --- > xfb

FIGURE 44. Nondimensional temperature rise due to a square heat source on conducting plate,
surface isotherms: (a) Peb =5.0, k,/k, = 1.0, (b) Peb =5.0, k,/k, = 10.0, (c) Peb =500.0, k,!k, = 1.0,
(d) Peh = 500.0, k.,lk, = 10.0
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 163

Pe = 5.0; k / k = 10.0
b s e
6
..0
.......... 5
>.
A
4
d
......
....0 3
y

....
"0
<1.}
1-<
2
!=:
~
0..
<l:l 1

0
0 1 2 3 4 5 6 7 8 9 10
(b) Flow direction --- > xjb

b
Pe
s e
= 500.0; k /k = 10.0
4 .-~----~~---------------------------,

o ~~~~~~~~~--~~~~~~~~~

0 2 3 4 5 6 7 8 9
(d) Flow direction --> x/b

FIGURE 44. Continued

164 Air Cooling Technology

0.7
,. = fJ /(qb/k ) k/k :
e s e
0.6 -adiabatic
-- 0.1
.... 1.0
0.5 -- 10.0

0.4 Pe "'5.0
b
*q, ... - ... ,
''
0.3 ' I

' I
\
\
\\
0.2 '·,
·'.....
................

0.1

0
0 2 4 6 8 10
xfb
FIGURE 45. Nondimensional span wise centerline temperature rise, lib= 1.0, Pen= 5.0.

analysis in such a way as to account for the degradation of convective heat

transfer due to upwind heating through the board. Average Nusselt number is
defined as

(79)

where h is the average heat transfer coefficient over the source.

The Nub were computed for various Peb ranging from 5.0 to 5000 and for
lib= 0.5, 1.0, and oo. The case lib= oo represents the infinitely wide 2-D strip
of heat; the analytic solution for the adiabatic case is given in Ortega et al.
[1993]. Nub plotted against the Peh is shown in Figure 47, logarithmic on both
the axes. Curve fits were performed on the data for each aspect ratio. Above
Peh = 100.0 there is no appreciable difference between the three aspect ratios,
thus for the range of Peb encountered in practice, the Nub for a rectangular source
of any aspect ratio lib greater than 0.5 can be approximated as that for a square
source. At low Peh the spanwise diffusion is significant, and hence the shape of
the source affects the heat transfer from it. The magnitude of Nub varies in-
versely with lib, being lowest for the 2-D case and highest for lid= 0.5. Are-
gression analysis on the square source data resulted in the following linear fit:
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 165

0.08 ,. = B/(qb/k) k/k :

e s e
0.07
-adiabatic
1.0
0.06 -·- 10.0

0.05
Pe = 500.0
.
<ll» 0.04
b

0.03

0.02 -·-·-·-·-·-·--=-:::-::---
--·-·---

0.01

0
0 1 2 3 4 5 6 7 8 9
x/b
FIGURE 46. Nondimensional spanwise centerline temperature rise, lib= 1.0, Peb =500.0.

(80)

The standard deviation of the regression was 0.16. The error in using the square
source correlation in predicting the strip source Nub at Peh of 5.0 and 5000.0
was 5.0% and 0.07%, respectively, and 2.7% and 0.8% for a rectangular source
of aspect ratio 0.5 at Peb of 5.0. An interesting note about the exponent on the
Peclet number in Equation 80 is that an analysis of the two-dimensional prob-
lem with negligible streamwise diffusion shows the exponent on Peclet num-
ber to be exactly 0.5.
To determine the response of the heat transfer over the source to variations
in Peb at various board conductivities, Nub for various Peb ranging from 5.0 to
1000.0 and for k..lke ofO.l, 1.0 and 10.0 were computed for the square source.
After extensive analysis, two correlations were developed [Ramanathan, 1994].
In the first case, a correlation was developed in such a way that average Nusselt
number tended to the adiabatic board limit as conductivity ratio kslk1approached
zero. The result was

(Js.)
0.334489 Peg· 0368
Nub = 1- 0.203 Pe-;D 174 (81)
Nub(ad) ke
166 Air Cooling Technology

100
- - 2cm x lcm source
- 2cm x 2cm source
--- strip source

i
.0

10

/
/
/

1
1 10 100 1000 10000
Pe
h

FIGURE 47. Average source Nub dependence on PeJ, comparison of three somce aspect ratios
on an adiabatic board.

and is useful for 0.0::; k,Jk1 ::; 10.0. An expression that does not tend to the cor-
rect adiabatic limit but is simpler to use is

Nub = 0.943349(k,/ker-{)·05 Peg- 53 (82)

applicable for Peb = 5.0 to 1000.0 and kJke = 0.1 to 10.0. The goodness of the
correlation is tested in Figure 48 for all the parameter variations attempted.
These correlations offer a way to characterize the conjugate heat transfer from
a square source of heat, and in particular, to account for the effects of board
conduction in an uncoupled analysis.

CONCLUSION

This chapter has presented analysis of and results from a limited class of
conjugate subproblems of the printed circuit board cooling problem, problems
in which the flow is sufficiently simple so that the conjugate mechanisms can
be emphasized without being obscured by difficult fluid flow physics or nu-
merical issues that invariably arise in CFD analysis of more difficult geome-
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 167

100
k/k :
s e

0 0.1
• 1.0
Ill 10.0
regression

---
~
:;..
~ .. 10

z""

1
1 10 100 1000
Pe
b

FIGURE 48. Correlation of source average Nusselt number for various board conductivities.

tries. It has become increasingly possible to apply commercially available CFD

codes for the solution of fully conjugate problems of the type discussed in this
chapter but with greater complexity in both the geometry and in the thermal
boundary conditions. Readers are referred to the study ofChoi eta!. [1994] and
the references reviewed in that study in which the geometry of Figure 49 has
been extensively analyzed using a commercially available CFD tooL The geom-
etry is a two-dimensional representation of a printed circuit board with a some-
what detailed model of the electronic package modules mounted on the board
surfaces. The solution approach is nearly identical to that of Sugavanam et a!.
[ 1995] described in this chapter. Because such problems introduce a very large
number of parameters, it is difficult to extract broadly applicable results from
such studies. Nevertheless, the ability to successfully compute such geometries
is by now readily available in commercial form and is limited only by com-
puting resources and computer resourcefulness.
It is worth remembering that a thermal analyst's toolbox should contain many
vertical compartments; we earlier referred to them as hierarchies. The upper-
most, most accessible compartment should contain simple, easy to use tools
such as design rules of thumb that can be developed on the back of an envelope.
 ,_.
~
00

t-L, I L. l lr-j
Printed Circuit
(a) T Board (PCB)
====:J

llL
INLET OUTLET

(b)
~
Hp ::;·
Printed Circuit Board (PCB) ~
£..
~-
FIGURE 49. Electronic components mounted on a conducting printed circuit board, two-dimensional representation [after Choi et al., I 994]. ~
~
0

g
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 169

When the problem resists solution with the tools in the uppennost compartment,
the tools in the next lower compartments should be pulled out. At the lower-
most compartment, the analyst should keep the biggest hammers and most so-
phisticated tools. The successful analyst will probably not have to resort to that
lowest compartment very often. It is the author's hope that this chapter, having
demonstrated various ways to approach the PCB cooling problem by use of
tools from different compartment levels (but primarily from the mid-level ones),
will prompt users to question their choice of tools and expand their toolbox.

ACKNOWLEDGMENTS

Much of the work presented would not have been possible without the sup-
port of my sponsors. With gratitude I acknowledge the support over the past
five years of the National Science Foundation, the Semiconductor Research
Corporation, the IBM Corporation, the Intel Corporation, and Advanced Micro
Devices. My colleagues Drs. Chris Choi of the University of Arizona, Sung Jin
Kim of the IBM Corporation, and Cho-lik Chan of the University of Arizona
have taught me much about numerical techniques in conjugate heat transfer,
and I thank them for their friendship and technical contributions. Darvin
Edwards of Texas Instruments and Tom Tarter of AMD have been invaluable
mentors through the sponsorship of the SRC. It has been a pleasure to explore
these issues with my assistants Humayun Kabir, Shankar Ramanathan, Ramesh
Sugavanam, and Uli Wirth who have contributed immeasurably to the techni-
cal and not so technical aspects of my work.

REFERENCES
Anderson, A.M. and Moffat, R.J. 1990a. Convective heat transferfrom arrays of modules with non-
uniform heating: experiments and models, Report HMT-43, Thennosciences Div., Mech.
Eng. Dept., Stanford University, Stanford, CA.
Anderson, A.M. and Moffat, R.J. 1990b. A new type of heat transfer correlation for air cooling of
regular arrays of electronic components,ASME HTD- 153. Thermal Modeling ofElectronic
Systems and Devices, ASME, New York, 27-40.
Brebbia, C.A., Telles, J.C.F., and Wrobel, L.C. 1984. Boundary Element Techniques: Theory and
Application in Engineering, Springer-Verlag, New York.
Carslaw, H.S. andJaegar, J.C. 1959. Conduction ofHeat in Solids, 2nded, Clarendon Press, Oxford.
Choi, C. Y., Kim, S.J., and Ortega, A. 1994. Effects of Substrate conductivity on convective cool-
ing of electronic components, ASME J. Elec. Pakag., 116:198-245.
Culham, J.R., Lemcyzk, T.F., Lee, S., and Yovanovich, M.M. 1991 a. META-a conjugate heat trans-
fer model for air cooling of circuit boards with arbitrarily located heat sources, AS ME HTD-
171, Heat Transfer in Electronic Equipment, A. Ortega, D. Agonafer, and B. Webb, Eds.,
ASME, New York, 117-126.
Culham, J.R. and Yovanovich, M .M. 1991 b. The effect of common design parameters on the ther-
mal performance of microelectronic equipment: II. Forced convection, ASME HTD-171,
170 Air Cooling Technology

Heat Transfer in Electronic Equipment, A. Ortega, D. Agonafer, and B. Webb, Eds., ASME,
New York, 55-62.
Eckert, E.R.G. 1950. Introduction to the Transfer of Heat and Mass, McGraw-Hill, New York.
Eckert, E.R.G. and Drake, R.M. 1972. Analysis of Heat and Mass Transfer, McGraw-Hill, New
York.
Gorski, M.A. and Plumb, O.A. 1990. Conjugate heat transfer from a finite strip heat source in a
plane wall. ASME HTD-129, AIAA-ASME Thermophysics and Heat Transfer Cmiference,
Seattle, ASME, New York, 47-53.
Gorski, M.A. and Plumb, O.A. 1992. Conjugate heat transfer from an isolated heat source in a
plane wall. ASME HTD-210, Fundamentals of Forced Convection Heat Transfer, M.A.
Ebadian and P.H. Oosthuizen, ASME, New York, 99-105.
Gorski, M.A., 1992. On the Heat Transfer from a Small, Discrete Heat Source, Ph.D. thesis,
Washington State University, Pullman.
Incropera, F.P., Kerby, J.S., Moffat, D.F., and Ramadhyani, S. 1986. Convection heat transfer
from discrete heat sources in a rectangular channel, Int. J. Heat Mass Transfer,
29:1051-1058.
JEDEC. 1994. Methodology for the thermal measurement of component packages (single semi-
conductor devices), JCI5.1 subcommittee test proposal, JEDEC.
Jensen, R.H., Andrejack, G.A., Button, D.P., and Bydal, B.A. 1989. Comparative thermal per-
formance of various substrate materials in a simple packaging application: actual versus
predicted in Proc. 39th IEEE Electronic Components Conference, IEEE, New York,
572-576.
Kabir, H., Ortega, A., and Chan, C.L. 1995. A boundary element formulation of the conjugate heat
transfer from a convectively cooled discrete heat source mounted on a conductive substrate.
IEEE Transfer Camp., Packag. Mfg. Technol-Part A., I 8(1 ): 108-116.
Kays, W.M. and Crawford, M.E. 1993. Convective Heat and Mass Transfer, 3rd ed., McGraw-Hill,
New York.
Krause, W.B., Klasi, M.L. andAddala, S.R. 1989. Conjugate heat transfer from isothermal surface
mounted heat sources embedded in a large substrate. ASME HTD, Ill, Heat Transfer in
E/ectronics-1989, 1989 Heat Transfer Conference, 161-168.
Lee, C. C. and Palisoc, A.L. 1990. Topics in Boundary Element Research: Electrical Engineering
Application, C. A. Brebbia, Ed. Vol. 7, Springer-Verlag, New York.
Lemczyk, T.F.. Mack, B.F., Culham, J.R., and Yovanovich, M.M. 1991. PCB trace thermal analy-
sis and effective conductivty in Pmc. 7th Annual IEEE Semiconductor Thermal Meas. and
Manag. Symp. (SEMITHERM), IEEE, New York, 15-22.
McCuen, P.A. 1962. Heat transfer with laminar and turbulent flow between parallel planes with
constant and variable wall temperature and heat flux, Ph.D. thesis, Stanford University,
Stanford.
Moffat, R.J. and Ortega, A. 1988. Direct air cooling of electronic components, in Advances in
Thermal Modeling of Electronic Components and Systems, Vol I, A. Bar-Cohen and A.D.
Kraus, Eds., Hemisphere, New York.
Ortega, A., Ramanathan, S., Chicci, J.D., and Prince, J.L. 1993. Thermal wake models for forced
air cooling of electronic components in Pmc. 9th Annu. IEEE Semiconductor Thermal
Meas. and Manag. Symp. (SEMITHERM). IEEE 93CH3226-8, New York, 63-74.
Ortega, A., Wirth, U., and Kim, S.J. 1994. Conjugate forced convection from a discrete heat source
on a plane conducting surface: a benchmark experiment, ASME HTD-292, Heat Transfer
in Electron. Sys., ASME, New York, 25-36.
Patankar, S. V. 1980. Numerical Heat Transfer and Fluid Flow, Hemisphere, New York.
Peterson, G.P. and Ortega, A. 1990. Thermal control of electronic equipment and devices, in
Advances in Heat Transfer, Vol. 20, J.P. Hartnett and T.F. Irvine, Jr., Eds, Academic Press,
New York.
Ramadhyani. S., Moffatt, D.F., and lncropera F.P. 1985. Conjugate heat transfer from small isother-
mal heat sources embedded in a large substrate, Int. J. Heat Mass Transfer, 28: 1945-1952.
Conjugate Heat Transfer in Forced Air Cooling of Electronic Components 171

Ramanathan, S. 1994. Uniform Flow Effective Diffusivity Model for Convective Cooling of
Electronic Components, Master's thesis, University of Arizona, Tucson.
Ramanathan, S. and Ortega, A. 1995. A uniform flow effective diffusivity approach for conjugate
forced convection from a discrete rectangular source on a plane conducting surface, Procs.
1996 IEEE ITHERM Symposium.
Rosenthal, D. 1946. The theory of moving sources of heat and its application to metal treatment,
ASME J. Heat Transfer, 849-866.
Rubesin, M. W. 1945. An Analytical Investigation of the Heat Transfer Between a Fluid and a Flat
Plate Parallel to the Direction of Flow Having a Stepwise Discontinuous Surface
Temperature, Master's thesis, University of California at Berkeley, Berkeley.
Rubesin, M. W. 1951. The effect of an arbitrary surface temperature variation along a flat plate on
the convective heat transfer in an incompressible turbulent boundary layer, NACA TN, 2345,
NACA.
Seban, R.A. 1950. Calculation method for two-dimensional boundary layers with arbitrary free
stream velocity variation and arbitrary wall temperature variation, Inst. ofEngrg. Research,
Univ. of California at Berkeley, Berkeley.
Tribus, M. and Klein, J. 1952. Forced convection from non-isothermal surfaces, in Proc. Heat
Tran~fer Symp., University of Michigan, 211-235.
Sugavanam, R. 1994. Numerical Investigation of Conjugate Heat Transfer from Flush Heat Sources
on a Conductive Board in Laminar Channel Flow, Master's thesis, University of Arizona,
Tucson.
Sugavanam, R., Ortega, A. and Choi, C. Y. 1995. Anumerical investigation of conjugate heat trans-
fer from a flush heat source on a conductive board in laminar channel flow, Int. J. Heat and
Mass Transfer, 1995, 38:2969-2984.
Tummala, R.R. and E.J. Rymaszewski, Eds. 1989. Microelectronics Packaging Handbook. Van
Nostrand Reinhold, New York.
Wirth, U. 1994. Experimental Investigation of Conjugate Heat Transfer from a Flush Mounted Heat
Source on a Conductive Substrate in Laminar and Turbulent F lo ws, Master's thesis,
University of Arizona, Tucson.

FOR FURTHER INFORMATION

In addition to the references cited in the text, readers are referred to these
additional references for further information about conjugate heat transfer as
related to electronics cooling issues.
Brosh, A., Degani, D. and Zalmanovich, S. 1992. Conjugated heat transfer in a laminar boundary
layer with heat source at the wall, ASME J. Heat Transfer, II :709-724.
Davalath, J. and Bayazitoglu, Y. 1987. Forced convection cooling across rectangular blocks, ASME
J. Heat Transfer, 109:321-328.
Kim, S.H. and Anand, N.K. 1994. Laminar developing flow and heat transfer between a series of
parallel plates with surface mounted discrete heat sources, Int. J. Heat Mass Transf er,
37:2231 - 2244.
Kim, S.H. and Anand, N.K. 1994. Turbulent heat transfer between a series of parallel plates with
surface-mounted discrete heat sources, ASME J. Heat Tramfer, 116:577- 587.
Nigen, J.S. and Amon, C.H. 1994. Time-dependent characteristics of conjugate heat transfer char-
acteristics of self-sustained oscillatory flows in a grooved channel, ASME J. Fluids Eng.,
in press.
Zebib, A. and Wo, Y.K. 1989. A two dimensional conjugate heat transfer model for forced air cool-
ing of an e lectronic device, ASME J. Electron. Packag., 9:11 - 16.
 Chapter 5

ENHANCED AIR COOLING

OF ELECTRONIC EQUIPMENT

Suresh V. Garimella

CONTENTS

Introduction . . . .. . . . . .... .... .. ... . ..... . .. .. . ...... .. . . . .174

Enhancement Strategies . .... . . . .... . . ...... .. . . . .. .. .. . . .. . .174

Enhancement Techniques .. . .. ................ . ....... . ..... .176

Inherent Enhancement at Electronic Component Surfaces ... . .. ... 176
Enhancement by Controlling Geometric Layout ..... . .. . . . .. . . . . 176
Enhanced Surfaces . . . . ..... . . . .. . . . . . .... . ... . ..... . . . . .. 179
"Large" Roughness Elements (2-D Ribs and 3-D Protrusions) ...... 179
Classical Roughness .. . ........... . .. . .... .. ..... . ... . . . . 182
Vortex Generators and Barriers .. . . . . . . ..... .. ....... ... . . . . 183
Displaced Promoters and Flow Modulation . . . . .... . .. . ....... .187
Impinging Jets . . ... .. . . . ... ... . .... ... . .. ... .. . . . . . .. . . . 189
Compact Heat Sinks . .. . . .. . .. ... .. . . . . .. . . . ... . .. .. . ... .193
Porous Metallic Matrices .... . .. . . . .. . . ... . . .... . . ..... . ... 195
Improved Substrate Conduction ..... . ....... .. ...... . . ..... . 195
Hybrid Techniques ... . . . . .. . . ....... . ... . . .. ... .. . . ...... 195

Performance Evaluation Criteria . . . .. . . . . . . . .... . ... . . ... . . . . . .195

Conclusion ... . .. . . .. ... .. . . .. . ...... ... ...... ...... .. . ... 196

Acknowledgments . . .. ................. . . . . .... .. .. . .. .. . .. 196

Appendix A: Nomenclature ...... .. .. ... . . . . .. . . . . . . . .. .. . .. .196

References . . . .. . .. ... . ..... . . .. .... .. . . . .. . ... . .. .. .. . . .. 197

0- ~493 -944 7-3/96/$1l.OO+S.50

© 1996 by CRC Press, Inc. 173
174 Air Cooling Technology

INTRODUCTION

In view of the simplicity of implementation and cost effectiveness of using

air as the cooling medium for electronic equipment, it is highly desirable to
explore means for enhancing the cooling capabilities of air before recourse
is made to more aggressive cooling techniques such as using liquids in free
or forced convection, in a duct-flow, an impingement, or a spray-cooling
arrangement. Since unassisted (unenhanced) air cooling is not adequate for
many applications, a thorough investigation of enhancement techniques is
crucial. This chapter covers various enhancement techniques including the
use of geometric adjustments such as staggering chip-arrays; surface en-
hancements in the form of a large variety of fins; and passive flow-modula-
tion devices such as turbulators, vortex generators, barriers, displaced
promoters, two-dimensional ribs, and three-dimensional roughness
elements. Subsequently, jet impingement from single and multiple arrays
of round and slot jets will be discussed as well as microchannel and heat-
pipe heat sinks, porous metallic matrix inserts, improved substrates, and
hybrid techniques.

ENHANCEMENT STRATEGIES
Air-cooling techniques may be broadly classified, based on driving force,
into free-convection and forced-convection techniques. This chapter is primar-
ily concerned with forced convection. Detailed discussions of free convection
air cooling and its enhancement are beyond the scope of this chapter. For in-
formation on these topics, the reader is referred to reviews of the literature avail-
able in Aihara [1991] and Moffat and Ortega [1989].
A large number of choices exist for the enhancement of forced-convection
cooling of electronic equipment The following strategies may be identified for
the design and implementation of any enhancement technique:

Improvement of convective heat transfer coefficient: Improvement can

be accomplished by the choice of a fluid with superior thermophysical
properties or by increasing flow velocities. Since this chapter deals with
air cooling, the former option cannot be considered further. The latter op-
tion, on the other hand, may come at a disproportionate cost in terms of
the required pumping power and is often not a preferred option. It is worth
noting in this connection that the so-called Mouromtseff number may be
used to evaluate the thermal performance of a coolant in a given config-
uration [Chu et al., 1970]; this number simply recasts the appropriate
forced-convection correlation to reflect the dependence of the Nusselt
number on the thermophysical properties included in the nondimensional
parameters of the correlation. The Mouromtseff number for different flu-
ids for turbulent flow through smooth tubes is shown in Figure 1 as a
Enhanced Air Cooling of Electronic Equipment 175

FC75

...
0
u
<JJ ....
0 0
~ ::t Coolanol-45
"'0Q.
II
0
~ 0.1

r-----.. . . . . . - . . . . . . . . . . . . Air

Temperature-"F

FIGURE 1. Mouromtseff number for fully developed turbulent flow in smooth tubes (Chuet al.
[ 1970], by permission).

function of temperature; the higher the value of this number, the better is
the heat transfer performance of the fluid.
.. Increase of surface area available for heat transfer: This has traditionally
been accomplished by the use of enhanced surfaces (fins), and a
body of literature exists on their design and implementation.
Flow modulation: This strategy aims to increase mixing in the flow using
devices such as vortex generators, turbulators, and swirl flow devices.
Increases due to flow modulation in the heat transfer coefficient of as
much as 100% have been reported for turbulent flows.

These three strategies may often not be entirely distinct; for instance, flow
modulation is essentially a means for increasing heat transfer coefficients by
selectively increasing the local velocity or shear in the flow near the surface to
be cooled. Similarly, arrangements such as louvered fins increase both the heat
transfer coefficient as well as the surface area [Webb, 1987].
Different strategies are typically effective in the different flow regimes-
laminar, transitional, and turbulent. Flow-regime transition has been well char-
acterized in the literature for smooth ducts and pipes. However, when flow
176 Air Cooling Technology

channels carrying electronic components are considered, the definition of lam-

inar to turbulent transition is less obvious. This transition was investigated using
flow visualizations and turbulence measurements in water by Garimella and
Eibeck [ 1992]. The onset of transition was shown to be a function of channel
height as well as the chip layout and geometry. For instance, as channel height
increased from 1 to 3.6 chip heights, the channel height-based Reynolds num-
ber (UH/u) for the onset of transition in interior (fully developed flow) regions
of the chip array was found to increase from 700 to 1900. This compares to the
smooth-pipe value of 2400 for transition Reynolds number based on hydraulic
diameter, or roughly 1200 for Reynolds number based on channel height.
Garimella and Eibeck [ 1992] also suggested that though the chip height was not
a variable in their study, it appeared that a Reynolds number based on chip height
appeared to be the correct predictor for transition, with transition occurring for
UB/u > 550 at a chip spacing of about 2 chip heights; this threshold was low-
ered to 400 at larger spacings of 6 chip heights. Igarashi and Takasaki [1992]
proposed a similar threshold of UB/u > 900 for transition in air. Lower limits of
transition in a rectangular duct are tabulated in Bhatti and Shah [1987], where
the critical Reynolds number is listed as a function of duct aspect ratio and varies
from 3100 for a parallel-plates channel to 2200 for a square duct. The critical
Reynolds number is also affected by the conditions at the entrance to the chan-
nel-an abrupt entrance is more susceptible to transition than a smooth entrance.
Individual techniques to achieve these enhancement objectives are discussed
in the rest of this chapter.

ENHANCEMENT TECHNIQUES

INHERENT ENHANCEMENT AT ELECTRONIC

COMPONENT SURFACES
It is important to recognize that electronic components have naturally en-
hanced heat transfer (due to enhanced hA products) when compared to predic-
tions based on thermally ordered conditions (smooth surface, vibration-free
analyses). In addition to the beneficial flow-field disruption caused by virtue of
their protrusion, this natural enhancement results from the surface roughness of
the circuit boards, vibrations in electrical equipment, and electric fields in power
equipment [Kraus and Bar-Cohen, 1983]. The surface roughness on a printed
circuit board has been shown to lead to an increase in the heat transfer coeffi-
cient relative to a smooth surface by up to a factor of 10 [Wenthen, 1977].

ENHANCEMENT BY CONTROLLING GEOMETRIC LAYOUT

Significant enhancements in heat transfer can be realized by controlling the
geometric layout of the chips and the assembly of the boards. Staggering the
chips on a board has been shown in several studies to increase the heat trans-
fer coefficient relative to an inline configuration for a given inlet velocity and
channel height. In air-cooling experiments, Hollworth and Fuller [1987] ob-
Enhanced Air Cooling of Electronic Equipment 177

tained as much as a 50% increase in heat transfer; in water, a 10 to 40% en-

hancement was seen to be accompanied by a 20 to 110% increase in pressure
drop [Garimella and Eibeck, 1991c]. Figure 2 shows the heat transfer coeffi-
cients for inline and staggered arrangements of an array of chips with water as
coolant [Garimella, 1991]. The flow visualizations in Figure 3 (water with
hydrogen-bubble tracers and laser-sheet illumination) demonstrate the superior
mixing obtained by staggering the chips in an array. With fan power held con-
stant, Ashiwake et al. [1983] obtained a 70% drop in cooling-air temperature
by staggering the chips of an inline array.
Even in inline arrays, the heat transfer coefficient can be increased by in-
creasing either the stream wise spacing between chips (35-40%) or the span-
wise spacing (15%) as shown in Garimella andEibeck [1990, 199lb]. General
recommendations about component orientation and the use of mixed compo-

~ H/B = 1.2
A H/B:::: l.9
!Ill H/B =2.7
+ H/B =3.6
,-..
~
INLINE ARRAY
N

.E.
~
.c:
1000

0
0 1000 2000 3000 4000 5000 6000
0
;.§
.5 50
:a>
•..
···•···•···•·······®
0
40
~0
eo
30
!f Ill .. -~
..
"'
1:5 20 .. .~;, .. ·A···A···A···A
~
0
<.,)
c 10
•··.
-'<. :.,_4(· ·-~-- ..... --11!··· .m- .. :::::::::::::::::::::
«l
.c: ····-llll·' , ......... ··
c
v
~ 0
0 1000 2000 3000 4000 5000 6000
Reynolds number

FIGURE 2. Heat transfer coefficients in water for an inline array of chips, and the percent en-
hancement obtained by fully staggering the chips (by one chip width); the Reynolds number (ReH)
is based on the channel height, and the chip spacing is equal to the chip width and length.
178 Air Cooling Technology

FIGURE 3. Flow patterns visualized with hydrogen bubbles in water for inline and staggered
arrays of chips; flow is from right to left (H/B =3.6, ReH =3450).

nent sizes were drawn by Azar and Russell [1991] based on flow patterns vi-
sualized in water. There are also indications that chips in the horizontal ori-
entation have lower thermal resistances than those in the vertical orientation,
even in forced convection due to buoyancy effects; the vertical orientation also
results in greater asymmetry in component temperature distribution [Azar et
al., 1989]. This effect would, however, be swamped at sufficiently large
Reynolds numbers.
 Enhanced Air Cooling of Electronic Equipment 179

When possible, chips in a horizontal channel should be mounted on the

bottom plate (facing upwards), so as to benefit from the enhancement effect
from free convection; in contrast, a heated top-plate configuration displays a
thermally stratified behavior that suppresses the effect of buoyancy. The ori-
entation of the chips is important at the lower-laminar and weakly turbu-
lent-range of Reynolds numbers. For instance, Osborne and Incropera [1985]
showed that the heat transfer coefficients for the heated bottom plate in a hor-
izontal parallel-plates duct could be as much as 50% larger than for the
top plate at a weakly turbulent Reynolds number based on channel height
of 5200.
Another system parameter of control may be high free-stream turbulence.
This information is included under controlling geometric layout, since it may
be possible to enhance the turbulence level of the incoming coolant stream ei-
ther by appropriate choices in manifold design, or by placing turbulence-pro-
moting peripheral devices upstream from the chips. Simonich and Bradshaw
[1978] reported significant increases in Stanton number in response to high
free-stream turbulence. Pedisius et al. [ 1979] found increases in heat transfer
of 10 to 15% with turbulence intensities up to 8%; upon further increasing the
turbulence intensity to 14%, however, only a 20% increase in heat transfer was
observed. Blair [1983a, b] reported heat transfer enhancements of as much as
20% for a grid-generated free stream turbulence of 7%. While there have been
different explanations in the literature of the precise effect of free-stream tur-
bulence on heat transfer, it appears that free-stream turbulence not only causes
transition to occur earlier (at lower Reynolds numbers) but also has an effect
on the fully turbulent boundary layer [Blair, 1983a, b]. Maciejewski and Moffat
[1989] reported increases in heat transfer of 1.8 to 4 times relative to that pre-
dicted using accepted correlations for turbulent boundary layers, due to very
high free-stream turbulence intensities of 20 to 60%.

ENHANCED SURFACES
A vast amount of literature exists on the use of extended surfaces for heat
transfer enhancement. Reported studies have ranged from analytical solutions
for the simpler fin configurations to detailed conduction and conjugate heat
transfer computations and to experimental databases for a large variety of ex-
tended surfaces. Detailed discussions are widely available (for instance, see
Bergles et al. [1983, 1991]; Webb [1987, 1994]) and this material will not be
covered in the present chapter.

"LARGE" ROUGHNESS ELEMENTS

(2-D RIBS AND 3-D PROTRUSIONS)
As a general rule, obstructions are less effective in laminar flow than in tur-
bulent flow. The laminar sublayer in turbulent flow is a lot more fragile and
thus more amenable to being disrupted. However, obstructions may help to trig-
ger transition at lower Reynolds numbers in an otherwise smooth channel. Thus
180 Air Cooling Technology

in a narrow range of flow rates at the upper end of the laminar regime, ob-
structions can cause sharp increases in heat transfer.
Electronic components themselves often act as protrusions that enhance mix-
ing in the flow [Garimella and Eibeck, 199lc]. While the layout of the com-
ponents on the board is mostly governed by nonthermal hardware
considerations, the strategic placement of additional barriers could still be used
to significant advantage.
Numerous studies of two-dimensional rectangular ribs have appeared in the
literature. Han et al. [1978] studied the effect of rib geometry (shape, angle of
attack, and pitch-to-height ratio) on friction factor and Stanton number for fully
developed turbulent air flow. Ribs were placed on both walls with the rib height
being less than 15% of the channel height. Both the ribs and the interrib spaces
were heated. The ribs were treated as classical roughness and a maximum in
both friction factor and Stanton number was found to occur at a rib spacing to
height ratio of9. This spacing was reported to correspond roughly to the situ-
ation where separated flow from a rib reattaches just prior to the next rib.
Interestingly the cross-sectional shape of the rib had only a modest effect on
heat transfer, while the friction factor was affected markedly.
Velocity and turbulence characteristics in the vicinity of a two-dimensional
rib were presented by Crabb et al. [1977]. The reattachment point was found
to be approximately 12-rib-heights downstream of the leading edge. The height
and streamwise length of the rib were shown to affect the trajectory of the
streamline dividing forward and backward flow (zero-velocity streamline), and
hence, the reattachment length. This was proposed as a reason for the observed
difference in reattachment lengths behind a rib, a fence, and a backward-
facing step. Significant fluctuating energy at a discrete frequency was found to
be present in the vicinity of the shear layer above the rib.
The reattachment length behind a two-dimensional obstacle (backward-
facing step) increases markedly with Reynolds number in the laminar regime,
with a sudden drop at transition, and a gradual increase with Reynolds number
in the turbulent regime [Durst et al., 1986; Eaton and Johnston, 1981; Tropea
and Gackstatter, 1985].
Three-dimensional obstacles behave very differently from their two-
dimensional counterparts. The primary difference comes from the behavior of
the flow field in terms of the three-dimensional separation and reattachment.
For instance the reattachment length is on the order of 1 to 4 obstacle heights
in the three-dimensional case as opposed to 11 to 15 rib heights for two-
dimensional ribs, although the trends of variation with Reynolds number are
similar for the two kinds of obstructions. The three-dimensional separation and
reattachment process is more vigorous due to the reattachment being from three
sides (as opposed to solely from the top for a rib) and has a greater impact on
heat transfer. Maximum turbulent stresses occur near the top of the obstacle.
The wake behind both cylindrical and cubical obstructions disappears almost
completely by a distance of about five heights downstream, and no velocity de-
Enhanced Air Cooling of Electronic Equipment 181

feet remains [Castro and Robins, 1977; Lim and Lewkowicz, 1986]. As
Reynolds number increases, the lateral spread of the fluid dynamic and ther-
mal wake decreases [Lim and Lewkowicz, 1986; Morris et al., 1995]. Greater
turbulence intensities in the upstream flow have also been found to reduce the
size of the recirculation region downstream of an obstruction. The horseshoe
vortex shed from a three-dimensional obstacle mounted on a wall may also have
an impact on heat transfer enhancement; this is discussed in a later section.
Garimella and Schlitz [1995] recently studied heat transfer enhancement in
a rectangular duct with very small height to width ratio (0.0 17) simulating
stacked printed circuit boards carrying computer chips using two- and three-
dimensional mixing devices. Their experiments were performed over a wide
range of transitional and turbulent Reynolds numbers in water and FC-77 (a
perfluorinated dielectric fluid). Heat transfer and pressure drop results were
presented for a discrete, flush heat source, a protruding heat source, and for an
array of protrusions. Heat transfer enhancement obtained by these means was
measured, as well as that obtained by the introduction of transverse ribs on the
opposite wall. The greatest enhancement of 100% relative to the flush heat
source was obtained for the array of roughness elements used in conjunction
with a rib on the wall opposite the heated element. Results from this work are
shown in Figure 4, where the enhancement is defined with respect to a flush-
mounted chip.

--~
y --~-
/
/ -~-- """'0-- _
- -a- -element array with I rib
20 / - 9- -element array with 4 ribs - ~- - ~
1 - -111- - flush source with I rib
~ - "'?" - flush source with 4 ribs
- ~ -element arr y
Qb-k-~-L~~--~~~~~--L-~~-L~~~~~~~~--~

0 10000 20000 30000 40000

FIGURE 4. Extent of heat transfer enhancement (FC-77, relative to flush chip) of the protrud-
ing chips and with two-dimensional ribs placed on the opposite wall [Garimella and Schlitz, 1995].
182 Air Cooling Technology

CLASSICAL ROUGHNESS
Roughness elements, either random in nature or in a repeated-rib configu-
ration, have been extensively studied as a technique for heat transfer enhance-
ment in channel flows. The roughness elements considered are typically much
smaller than the ribs and obstructions of the preceding section-at least an order
of magnitude less than the boundary layer thickness, and in tubes, one or two
orders of magnitude less than the tube diameter. Nakayama [1982] reviewed
the influence of roughness Reynolds number, Prandtl number, and geometri-
cal parameters on the momentum and heat transfer roughness functions, for
both granular, three-dimensional surface roughness and for repeated-rib rough-
ness. It was concluded from comparing a large number of studies in the litera-
ture that the relative merits of roughening a surface are large when the roughness
Reynolds number is small; granular (3-D) roughness provided more favorable
results than rib (2-D) roughness. The heat transfer enhancement is also larger
for higher Prandtl-number fluids. It should be pointed out that in implement-
ing roughness as a heat transfer enhancement technique variable physical prop-
erties have a more pronounced effect on heat transfer in rough passages than
in smooth passages [Wassel and Mills, 1979]. Since fairly large temperature
variations are experienced in electronic applications, care should be taken to
account for variable-property effects.
Heat transfer and friction in tubes with rib-roughness (0.01 to 0.04 times
tube diameter in height) were modeled as classical roughness by Webb et al.
[1971]. The correlation for friction factor was based on the law-of-the-wall sim-
ilarity used by Nikuradse [1933] for sand-grain roughness, while the heat trans-
fer correlation was based on a heat and momentum transfer analogy (also used
by Dipprey and Sabersky [1963]). The authors argued that this method of cor-
relation could be used for geometrically similar fonns of arbitrary roughness.
Perry et al. [1969] identified two types of two-dimensional roughness ele-
ments: d-type associated with narrow cavities where the roughness function
(which describes the deviation of the log-law velocity profile from the smooth-
wall behavior) depends on duct diameter, and k-type with larger spacing be-
tween roughness elements where the roughness function depends on the
roughness height. It is this latter k -type roughness that, when installed upstream
from electronic components, could be used for enhancing heat transfer. This
type of roughness is characterized by an interelement spacing greater than about
three times the roughness height, resulting in the flow between roughness ele-
ments interacting strongly with the core flow, with eddies shed into the core.
Townes and Sabersky [ 1966] also observed a periodic instability when the cav-
ities between roughness elements were shallow (small element height, large
spacing); in this case, the vortex fonned in the cavity broke down and inter-
acted with the core flow. These instabilities result in higher heat transfer rates.
Kader and Yaglom [ 1977] made the first clear distinction between two- and
three-dimensional roughness. The heat transfer dependence on roughness
Reynolds number derived in their model was different for the two kinds of
Enhanced Air Cooling of Electronic Equipment 183

roughness. A similar difference in heat transfer dependence was found by

Garratt and Hicks [ 1973] and Webb et al. [ 1971]. The influence of the rough-
ness-element geometry on the heat transfer and friction characteristics has been
investigated by several researchers [Baumann and Rehme, 1975; Daile Donne
and Meyer, 1977; Hall, 1962; Maubach, 1972; Meyer, 1980].

VORTEX GENERATORS AND BARRIERS

Significant work has been done for decades on the use of vortex generators
to enhance heat transfer and to control flows in different applications. Figure 5
illustrates the principle of operation of a half-delta-wing vortex generator and
the longitudinal vortex formed; a pair of counter-rotating vortex generators are
also shown.
Fundamental studies of longitudinal vortices imbedded in turbulent bound-
ary layers were carried out by Eibeck and Eaton [1987] and Wroblewski and
Eibeck [ 1991] among others. Wroblewski and Eibeck showed that the vortex in-
teraction with the turbulent boundary layer enhanced the heat transfer to a greater
extent than did the momentum transport. Heat transfer coefficients and friction
factors were obtained as a function of Reynolds number and angle of attack using
rectangular, delta-wing, and half-delta-wing vortex generators by Fiebig et al.
[ 1986]. Delta wings were found to yield the greatest enhancement, with the ratio
of Colburn factor to friction factor being the highest at small angles of attack
(10 deg). Russell etal. [1982] showed that a value for this ratio of more than 0.5
could be achieved with counter-rotating vortex-generator pairs produced by rec-
tangular elements oriented at an angle of attack to the flow. The most persistent
vortices were produced with an angle of attack of 20 to 30 deg. Edwards and
Alker [1974] measured local heat transfer coefficients downstream from a row
of cubes, and a row each of corotating and counterrotating vortex generators.
The cubes produced the highest local increases in heat transfer; the counterro-
tating vortex pairs were more effective than the corotating vortices, with a max-
imum increase in heat transfer of 60% over flat-plate values. Mantle [1966] and
Zhang et al. [ 1989] studied vortex generators in the form of delta wings and cu-

FIGURE 5. Schematic of the single and counter-rotating pair of longitudinal vortices produced
by half-delta-wing vortex generators.
184 Air Cooling Technology

bical and cylindrical obstructions. Pearcey [1961] provided design criteria for
optimum arrays of vortices including corotating and counterrotating pairs.
Garimella and Eibeck [199la] reported a study in which half-delta wings
were placed upstream from, and on the same wall as, each streamwise column
of chips; heat transfer enhancement with water as coolant was studied as a func-
tion of streamwise position and Reynolds number. Vortex generators of two
heights (one and two times the chip height) were studied. As with other kinds
of barriers, the greatest enhancement was observed at the second row of chips
downstream and at a Reynolds number in the transitional flow regime, as shown
in Figure 6. In a succeeding study by Garimella and Schlitz [1993], half-delta
wings were placed on the wall opposite a chip, singly and in counterrotating
pairs, and localized heat transfer enhancement on that chip was investigated
with the height of the delta wing and its position with respect to the chip as
variables. A dielectric liquid (FC-77, with a Prandtl number of 25.3) was used
as the coolant. A single, tall vortex generator, twice the chip in height, yielded
the best results. A peak enhancement of 17% was obtained at transitional
Reynolds numbers. Optimum placement locations for the vortex generators
were also identified.
Enhancement techniques may be deduced from the study of the effects of
missing elements, height differences between chips on a board, and implanted
barriers, on heat transfer and pressure drop in arrays of chips by Sparrow et al.
[ 1982, 1983] and Sparrow and Otis [1985]. The mass (heat) transfer coefficient
just downstream from a missing element was found to be 40 to 50% higher as

FIGURE 6. Heat transfer enhancement with vortex generators twice the chip in height placed
upstream from Row I, HfB =3.6 [Garimella and Eibeck, !99la].
Enhanced Air Cooling of Electronic Equipment 185

a result. The second row downstream from a barrier implanted in a chip array
was found to experience the greatest enhancement (about a factor of 2); how-
ever, the pressure drop increased due to the barrier by 10 to 150 times. The lo-
cation of the second row downstream coincides with the location where the
flow reattaches. This is a recurring result that has been found in studies with a
variety of obstacles and barriers. The parametric effects of installing multiple
barriers and the enhancement on the wall opposite from the chip-carrying wall
of a parallel-plates channel have also been studied. The effect of tall elements
on heat transfer from the remaining shorter elements in an array was investi-
gated by Souza Mendes and Santos [1987] and Torikoshi et al. [1988]. Chou
and Lee [ 1987] attempted to reduce temperature nonuniforrnities in air flow
over a heated chip using a rectangular barrier mounted at the front face of the
downstream chip. The maximum temperature of the chip was reduced as a re-
sult, as were the temperature variations over the chip surface. The barrier was
found to have an optimum effectiveness for a ratio of barrier to chip height of
2/3, beyond which no increase was found.
Anderson and Moffat [1991] suggested the introduction of scoops (Figure
7a) in the low-velocity recirculation region downstream of each chip to enhance
thermal mixing and thus reduce overall temperature rise. A decrease in the com-
ponent temperature rise of up to 19% was observed as a result of the scoops in
the row just downstream (see Figure 7b); the associated increase in pressure
drop over the eight streamwise rows of chips with scoops introduced behind
one row was found to be 11%. Similarly, Matsushima and Yanagida [1992]
measured a 25% increase in heat transfer coefficient in the last of five rows of
chips due to cylindrical obstacles placed on either side of each chip in a row,
that is, in the flow paths between chips in each spanwise row. With this increase,
the heat transfer coefficient was nearly the same as that predicted under the as-
sumption of complete thermal mixing of the cooling air. This arrangement with
a total of 10 cylinders placed in the five rows, caused the pressure drop to in-
crease by 1.6 times.
Heat transfer effects due to horseshoe vortices (see Figure 8) along the wall
downstream from a wall-mounted cylinder and streamlined cylinder were stud-
ied by Fisher and Eibeck [ 1990], who observed local increases in heat transfer
of 20 to 30%. The vortices from a streamlined cylinder were found to be more
concentrated (with greater heat transfer augmentation) while those from a cir-
cular cylinder were not as strong but were spread over a wider area due to a
large region of separation. Local heat and mass transfer effects of horseshoe
vortices produced at a cylinder/wall junction were studied by Ireland and Jones
[ 1986] in channel flow and by Goldstein et a!. [ 1985] in an external boundary
layer. Ireland and Jones found that the maximum heat transfer coefficient along
the channel wall occurs at the stagnation point upstream from the cylinder, and
a double peak in Nusselt number occurs downstream. The mass transfer ex-
periments of Goldstein et al. showed strongly enhanced mass transfer imme-
diately upstream (4.5 times the undisturbed value at the leading edge) and along
186 Air Cooling Technology

100
I I I I I

80 -
~ ~
~ -
• Q
0
~ i
•
0
0
0
G 60 - liD -
~
@
~- "
,_"
'
40 I-

0 No Scoops
e Scoops
20 I- -

0
I I I I I
0 2 4 6 8 !0

Row Number

FIGURE 7. (a) Schematic of the scoops installed between two rows of chips; (b) Comparison
of the component temperature rise with and without scoops; the scoops are placed between rows
land 2 [Anderson and Moffat, 1991].
Enhanced Air Cooling of Electronic Equipment 187

FIGURE !1. Simplified schematic of the horseshoe vortex formed at a wall-mounted cylinder
(Goldstein and Kami [1984], by permission).

the sides of the cylinder. In their experiments, a short cylinder with a height
to diameter aspect ratio of 1 yielded average mass transfer rates about 8%
higher than a taller cylinder (aspect ratio 12) but with markedly different local
distributions.

DISPLACED PROMOTERS AND FLOW MODULATION

One technique investigated in the literature to effect heat transfer en-
hancement has been to mount devices above the chips, either suspended or as
an attachment on the opposing walL These devices serve to locally increase
the flow velocity, shed vortices, shift reattachment points, and/or increase tur-
bulent mixing. For instance, Karniadakis et al. [1988] and Thomas [1965,
1966] showed that circular wires positioned above a smooth surface in lami-
nar flow result in a premature transition to turbulence, causing increased heat
transfer rates. Myrum et al. [ 1993] investigated the heat transfer enhancement
obtained by placing cylindrical rods immediately above heated ribs in a duct.
They defined enhancement in terms of the increase of the entropy generation
per unit duct length, and found increases of up to 27% in this quantity due to
the introduction of the cylindrical rods. Chou and Chiang [1990] also tested
the enhancement due to unsteady vortex shedding from cylindrical rods held
above chips, with the flow Reynolds number and the rod location as parame-
ters. A temperature reduction of about 15% was measured at the chip imme-
diately downstream from the chip above which the rod was installed for
Reynolds numbers at the high end of the laminar range. At low Reynolds num-
188 Air Cooling Technology

bers (~ 100 based on rod diameter), however, the rods had no discernible en-
hancing effect. Gan et al. [ 1990] reported results of experiments performed
with a half-cylinder and half of a streamlined body mounted on the wall op-
posite the chips. With two such displaced promoters in tandem, they obtained
a peak enhancement of 51%.
Another class of forced convective heat transfer enhancement techniques is
based on using flow destabilization mechanisms in regions of separated flow.
They promote lateral, large-scale convective mixing and hence increase heat
transport normal to the heat transfer surface. Flow unsteadiness is induced by
active, passive, or supercritical flow destabilization. It appears that above a crit-
ical Reynolds number, flows in electronic chip geometries exhibit laminar, self-
sustained oscillations at a frequency that corresponds to that of the most unstable
Tollmien-Schlichting channel modes compatible with the periodicity of the
geometry.
Active destabilization techniques have been proposed by Sobey [1980],
Ghaddar et al. [1986], Greiner [1991], and Azar [1992]. In this approach, the
external flow is actively modulated at an appropriate frequency to excite flow
instabilities, and thus to disrupt the shear layer and increase communication
between stagnant flow regions between chips in an array and the bypass chan-
nel flow. Flow destabilization may also be accomplished in a passive mode by
using obstacles such as cylinders (displaced promoters) that periodically shed
vortices and again serve to disrupt the confined nature of the interchip flow and
hence to improve mixing [Kamiadakis et al., 1988; Suzuki et al., 1991]. Ratts
et al. [1987] studied the flow modulation induced by vortex shedding from
cylinders in cross flow; enhancements of up to 82% in heat transfer coefficient
were obtained with cylinders placed periodically above the back edge of each
row of chips. The cylinder position, diameter, length, and number of cylinders
were investigated as parameters.
Active, passive, and supercritical flow destabilization techniques were com-
pared on an equal pumping-power basis by Amon [1992] via numerical simu-
lations. While the passive techniques were found to be the best with respect to
minimum power dissipation at low Nusselt numbers, supercritical destabiliza-
tion was found to become competitive as higher Nusselt numbers became nec-
essary. These results are shown in Figure 9.
Another approach involving the modification of the incoming flow has been
to incline the inlet of air in order to intensify the cooling experienced. Inclined
inflow could be naturally occurring, for instance, as in the cooling of electric
motors where the air enters obliquely into the stators. This approach was ex-
plored by Jicha and Horsky [ 1990] who investigated the effect of the variation
of the angle of attack from 0 (axial entrance) to 70 deg for a Reynolds number
of 42,520. For longer channels with a length of the order of 15 channel diam-
eters, there was a modest increase in heat transfer of about 15% due to the in-
clined inlet of air. For short channels (about five diameters long), the increase
was a much more dramatic 50% (angle of attack 70 deg).
Enhanced Air Cooling of Electronic Equipment 189

107 ..-------..--1-----..--,--------.,---- --.

b. Active Flow Modulation
0 Passive Flow Modulation
8 Supercritica! Flow Destabilization

0
115 0
b. liD
b.
0
e
0

loS
•
f- -
0

0
0

I
JifL-~~-L~--~~~-4~--~~~~~~~~~~--~ I I
0 2 4 6

Nusselt Number

FIGURE 9. Dimensionless pumping power as a function of Nusselt number for various flow
modulation techniques [Amon, 1992].

IMPINGING JETS
Impinging jets may be classified as free-surface or submerged, and the spent
flow can be in a confined or unconfined state. Also, the nozzles producing the
jets could be round orifices or slots, singly or in arrays, based on the applica-
tion (Figure 10). Free-surface jets are those where the fluid issuing from the jet
is different from the ambient fluid, resulting in a distinct free surface separat-
ing the two fluids; an example is a jet of water issuing into surrounding air. In
submerged jet impingement, on the other hand, the ambient fluid is the same
as the impinging jet; this is the case for air-in-air or water-in-water jets.
Four main variables can be identified that affect the local heat transfer from
a chip to an impinging jet jet Reynolds number, fluid thermophysical proper-
ties (Prandtl number), nozzle-to-chip spacing, and distance from the stagnation
point. A host of other parameters may also be identified including nozzle geom-
etry, thermal and hydrodynamic boundary conditions, turbulence level in the
jet, and number and configuration of nozzles. The stagnation-point heat trans-
fer coefficient has been observed to remain constant or to increase slightly as
the nozzle-to-chip spacing is increased to about four nozzle diameters and then
to decrease with further increases in the spacing. This behavior can be explained
based on the fluid mechanics of the jet. As the jet issues from the nozzle, a mix-
ing layer forms over its periphery and penetrates towards the center of the jet
with distance traversed away from the nozzle. The so-called potential core of
190 Air Cooling Technology

FIGURE 10. Various configurations of round and slot jets, singly and in arrays.

a jet is the region that is unaffected by the mixing layer, and the fluid therein
remains at the nozzle exit velocity. For well-formed nozzles, the potential core
has been shown to extend over 6 to 8 nozzle diameters, while for square-edged
nozzles, the core persists up to only 2 or 3 diameters away. When the spacing
between the nozzle and chip is less than the length of the potential core, the
core strikes the impingement surface; as long as this condition is met, the stag-
nation point heat transfer remains roughly constant For larger spacings, the
axial velocity of the jet decreases with increasing distance, and the stagnation
heat transfer progressively decreases. However, due to the large-scale struc-
tures in the mixing layer of the jet, the level of turbulence in the jet simultane-
ously increases and compensates for some of the drop in heat transfer due to
the drop off in axial velocity.
Much of the experimentation in jet impingement heat transfer has involved
air jets, and by default, in the submerged mode, since the experimentally easiest
configuration is that of an air jet issuing into air. Air cooling of heated surfaces
with impinging jets has been studied extensively [Baughn et al., 1991; Gardon
and Akfrrat, 1965; Goldstein et al., 1986; Hollworth and Gero, 1985]. Reviews
of the literature including design correlations for the use of impinging jets are
available in Martin [1977], Obot et al. [1980], and Downs and James [1987].
While a large number of air-jet studies have been reported in the litera-
ture, it is important in electronics applications to consider the effect of the
confining wall on the heat transfer from a surface. The nozzles used in cool-
ing computer chips would be located in the board facing the chips, and thus
the spent flow from the jet would be confined to a parallel-plates channeL
The recirculating flows created by the confining wall (in an axisymmetric
toroidal shape around the jet centerline for a round jet) have a significant ef-
fect on the heat transfer distribution on the target surface as shown by
Garimella and Rice [1995] and Rice and Garimella [1994]. Thus care should
be employed in extrapolating results from unconfined jets to electronics ap-
plications. Figure 11 shows a photograph of the flow patterns in an axisym-
Enhanced Air Cooling of Electronic Equipment 191

Zld=4

r/d
FIGURE 11. Photograph and sketches of the flow patterns in a confined and submerged im-
pinging jet of FC-77 for a nozzle diameter of 3.18 mm and a nozzle to target spacing of 2 (sketch
only) and 4 diameters; the photograph and sketches are not to the same scale.

metric, confined, and submerged jet of FC-77 visualized with pliolite parti-
cles and laser illumination; also included (with a different scale) are sketches
of the flow patterns visualized.
For round jets, the local heat transfer coefficient on the target surface has a
bell-shaped distribution with respect to radial distance from the stagnation point
The maximum value occurs at the stagnation point and decreases symmetrically
with radial distance. Under some conditions, secondary maxima are observed in
these curves corresponding to the location where the wall jet becomes turbulent
[den Ouden and Hoogendoom, 1974; Martin, 1977] and where the recirculating
flow reattaches [Garimella and Rice, 1994]. The local heat transfer distribution
for a confined and submerged round jet ofFC-77 illustrating the secondary peaks
for different nozzle to target spacings is shown in Figure 12; similar results have
been obtained in air (for instance, Huber and Viskanta [1994]).
One of the implications of using multiple jets to cool a single chip, or sin-
gle jets to cool an array of chips on a board, especially in a confined config-
uration, is the resulting cross flow. As the spent fluid from each jet drains
away, the accumulated drainage adds a cross-flow component to the existing
impinging-jet flow field. The effect of this accumulated cross flow from neigh-
boring jets (or with cross flow imposed) has been studied by Florschuetz et
al. [ 1981], Goldstein and Behbahani [ 1982], Obot and Trabold [ 1987], and
Whidden eta!. [1992]. According to Obot and Trabold, the flow field for these
 192 Air Cooling Technology

multiple jet configurations is complicated by interference between neigh-

boring jets prior to impingement, as well as by collision of the developing
wall flows from adjacent nozzles (cross flow). Jiji and Dagan [1987] per-
formed experiments with single (free) jets of water and FC-77 impinging on
a single heat source and an array, with the heat sources oriented vertically.
The effect of cross flow due to the drainage of spent flow from neighboring
jets had only a minor effect on average surface temperature and its unifor-
mity. As in other studies of this type, the recommendation was to increase the
number of jets and decrease the jet diameter for improved heat transfer per-
formance. The interactions between neighboring jets for confined and sub-
merged air jets was recently studied by Huber and Viskanta [1994]. A
nozzle-to-target spacing of 0.25 diameters was shown to result in minimal
preimpingement interaction, thus leading to much improved heat transfer rel-
ative to a larger spacing of 6 diameters. The secondary maxima were also
much more pronounced at the smaller spacings (as also shown in Figure 12).
Lytle and Webb [ 1994] obtained similar results for low nozzle-to-target spac-
ings but in an unconfined configuration.
The literature on jet impingement heat transfer has utilized various fluids
ranging from water to freons in addition to air. It would be advantageous to cast
all these results with a unified perspective so that they could be used irrespec-
tive of the fluid used in an application. However, the scaling of results between

15000

13000

11000
.......
~
9000
e
<'!

E??
,_., 7000
..=

5000
OZ/d=l
"Z/d=2
3000 XZJd=J
0 Z/d=4
+ ZJd=5
1000
-5.0 -2.5 0 2.5 5.0
r/d
FIGURE 12. Local heat transfer coefficient distribution for a confined. round jet of FC-77 with
a diameter of 1.59 mm, at different nozzle to target spacings and a Reynolds number of 13,000
[Garimella and Rice, 1994].
Enhanced Air Cooling of Electronic Equipment 193

fluids with different thermophysical properties is not yet well understood, and
most studies have simply used assumed Prandtl-number dependencies in their
correlations. One study by Metzger eta!. [1974] performed experiments with
water and oil jets (free-surface) to determine the effect of Prandtl number on
heat transfer; a Prandtl number range between 3 and 150 was tested by vary-
ing the fluid used and its inlet temperature.
The sizeable literature on the enhancement of jet impingement heat trans-
fer by employing surface roughness, protuberances and fins on the target sur-
face is beyond the scope of this chapter.

COMPACT HEAT SINKS

One of the most compact and high-performance heat dissipation devices is
the microchannel heat sink, first proposed by Tuckerman and Pease [1981,
1982] as a technique to lower the convection resistance between the substrate
and coolant. The microscopic flow channels incorporated in this heat sink are
based on the fact that for laminar flow in confined channels, the convection co-
efficient scales inversely with channel width. A prototype water-cooled inte-
gral heat sink with 50 11m wide and 300 11m deep channels machined into the
silicon substrate was shown to maintain a maximum substrate-to-coolant tem-
perature difference of 71 oc for a power density of 790 W/cm2 • Clogging of the
channels due to entrained debris and the possibility of water leakage were two
of the problems that were identified with the implementation of such devices.
To address the former, Tuckerman and Pease [1982] suggested that the longi-
tudinal "microfin" structure of the microchannels could be replaced by "mi-
cropillars" that would be less prone to clogging by particulate contamination.
It was also demonstrated that using FC-77 instead of water as the coolant would
lead to only minor degradation of the heat transfer. Phillips et al. [1987] pro-
posed models for evaluating the thermal and fluid performance of such heat
sinks. Their numerical predictions showed that turbulent-flow designs could
have equivalent or better performance than comparable laminar flow designs.
Recently, Bowers and Mudawar [ 1993] suggested that based upon practical
considerations such as pressure drop, erosion, choking, clogging, and manu-
facturing ease, a minichannel geometry (with channel widths of a few mm) of-
fered inherent advantages over the microchannel geometry, except when the
heat dissipation required is beyond the capability of the minichannels, and when
minimizing weight and liquid inventory is a must.
Another example of compact heat sinks is the heat pipe. The advantages of
the high heat transfer rates obtained through phase-change heat transfer can be
coupled with forced convection using a heat pipe, which dissipates heat by a
cyclic process involving a coolant that undergoes evaporation and condensa-
tion in a contained unit (for instance, see Dunn and Reay [1994]). In a heat
pipe, the inherent problems associated with direct liquid cooling and boiling at
the chip surface are avoided while high heat fluxes are still achieved. The ad-
vantage of using a heat pipe lies in its ability to carry heat from densely popu-
194 Air Cooling Technology

lated areas through a very narrow conduit to a location where a larger heat sink
can be installed. Thus a fin array can be used at a location remote from the chip,
which would otherwise have been too bulky to install. Heat pipes may also be
used as heat spreaders for controlling temperature nonuniformities on the sur-
face of semiconductor devices, as in the "micro" heat pipes ofWu eta!. [1990].
At the normal operating temperatures of chips, typically less than l00°C,
Babin and Peterson [1990] achieved heat dissipation rates of over 125 W/cm 2
with a maximum total power of 65 W. Unlike other enhancement techniques
where the heat flux can be used to calculate the power dissipated, the maxi-
mum power that can be carried away by a heat pipe is governed by the capa-
bilities of the phase change process. A heat-pipe design for cooling high
flux/high power chips was recently explored by North and Avedisian [1993].
The design involved a series of holes drilled into a manifold base plate lined
with sintered copper powder that served as the wick. With an air-cooled con-
denser section, a maximum heat flux of 4 7 W /cm2 and a total power of 900 W
were achieved with surface temperatures under l00°C.
While a heat pipe represents a very high thermal-conductivity path between
the chip and the ultimate heat sink such as a cold plate or a fin array, it may
often be limited by the capability of this heat sink to dissipate the heat that is
wicked to it. Thus, while large amounts of heat can be carried through the heat
pipe, impractically large condenser surface areas may be necessary to manage
this heat at the sink. The main impediments to the widespread use of heat pipes
has been their high manufacturing cost and uncertainties about their long-term
reliability. With stepped-up research and development efforts, improved man-
ufacturing methods, and increased heat-dissipation needs they are fast becom-
ing a viable design choice.

POROUS METALLIC MATRICES

Porous metallic inserts have been proposed for service as high-efficiency
fins. The microchannel cooling schemes discussed above could be considered
an extreme application of this technique. The microstructures have been mod-
eled as a porous medium with Darcy's law being used to describe the flow
[Koh and Colony, 1986]. However, as pointed out in Vafai and Tien [1981],
heat convection problems may not accurately be modeled using Darcy's law.
Using a modified Darcy equation for the flow and volume-averaged energy
equations for the solid and fluid, Tien and Kuo [1987] proposed a model that
could be applied not only to microchannels and dense porous inserts but also
to other complex nonhomogeneous systems such as pin-fin structures. A foam-
filled duct was shown to yield two to four times the heat transfer relative to
laminar slug flow in a parallel-plates duct. Hadim and Bethancourt [ 1993]
showed that, instead of completely filling the duct with porous material, dis-
crete porous inserts mounted only above each heat source resulted in signifi-
cantly lower pressure drops, while the heat transfer enhancement effects did
not degrade noticeably.
Enhanced Air Cooling of Electronic Equipment 195

IMPROVED SUBSTRATE CONDUCTION

One approach, which does not fall into the traditional classification of en-
hancement techniques but is of great interest due to its passivity, is improved
substrate conduction. Since the choice of board materials, die-attach and bond-
ing techniques are not primarily governed by thermal management consider-
ations, this approach may not always be practicable. However, if thermal
concerns were to be addressed at an early design stage, the improvement of
substrate conduction could be considered an enhancement parameter.
Antonetti [1990] reviewed the progress made in the development of predic-
tive theory, recent experimental studies, and examples of thermal interface
management in current electronic equipment. The theory and the applications
of constriction and spreading resistance concepts as applied to microelectronic
thermal management were reviewed by Yovanovich [1987]. Through their
models Culham and Yovanovich [1987] and Negus and Yovanovich [1986] il-
lustrated that the thermal conductivity is the single most important design pa-
rameter for reducing board temperatures. A marginal increase in the relative
percentage of copper in the board was shown to result in an increased effec-
tive thermal conductivity (for board conductivity in the range of 1 to 10 W/mK)
and significantly lower board temperatures. The effect on board temperature
of the small alterations in circuit-board thickness was shown to be not suffi-
cient to warrant changing the design of the board. It was also demonstrated
that a decrease in thermal resistance of the board due to increased chip spac-
ing occurs only when accompanied by a similar increase in the board thick-
ness (which is often not feasible). Thus increasing the chip spacing to a distance
greater than twice the chip width would produce little benefit in terms of de-
creased thermal resistance.

HYBRID TECHNIQUES
Some of the enhancement techniques discussed in this chapter may be used in
combination with each other. Air cooling is often supplemented by liquid cool-
ing as the ultimate heat sink. For instance, water-cooled finned-tube heat ex-
changers are incorporated between successive rows of circuit boards. In the IBM
Liquid Encapsulated Module (LEM), the substrate carrying the integrated circuit
chips is mounted within a sealed module-cooling assembly containing a fluoro-
carbon liquid coolant. Heat from the chips is transferred to the fluorocarbon using
internal fins and then to water flowing through an externally attached cold plate.
One of the more successful designs for indirect liquid cooling was the Thermal
Conduction Module (TCM), developed at IBM and used extensively in their com-
puters. These and other such techniques are discussed in Chu and Simons [1990].

PERFORMANCE EVALUATION CRITERIA

In a given application, the performance evaluation criteria may include in-
crease in heat duty, increase in surface area, pumping power, initial cost, main-
196 Air Cooling Technology

tenance cost, safety, and reliability. These criteria are discussed in Bergles et
al. [1974] and can be variously applied to optimize heat transfer based on the
particular application in question. For instance, Bejan and Morega [1994] pro-
posed methods of calculating the optimal spacing of circuit boards in turbulent
forced convection for a stack of fixed volume by maximizing the overall ther-
mal conductance. Mildc et al. [1990] addressed the optimization problem with
respect to the selection of the channel hydraulic diameter and flow Reynolds
number to yield the minimum pumping power for a given convective cooling
configuration. Although this is a very important part of any study of enhance-
ment techniques, it is beyond the scope of this chapter; the reader is referred to
the review articles on enhancement techniques cited earlier as well as Chapter
1 in this book for further information.

CONCLUSION
All of the techniques discussed in this chapter have focussed on enhancing
air cooling. The focus has been on those methods that are particularly attrac-
tive in the electronics-packaging field, and thus, techniques such as the appli-
cation of vibration (either in the fluid or at the substrate) or of electric fields
have not been included. Also, while air cooling is mechanically the simplest al-
ternative, the continuing advances in miniaturization and packaging have al-
ready necessitated non-air-based cooling technologies in several applications;
such technologies are likely to see more widespread use in the near future.

ACKNOWLEDGMENTS
The author's early work in this area at the University of California at Berkeley
(with P. A. Eibeck) was supported by IBM Corporation and the National Science
Foundation; ongoing work at the University of Wisconsin-Milwaukee is sup-
ported primarily by Cray Research, Inc. This support is gratefully acknowledged.

APPENDIX A: NOMENCLATURE
A exposed (active) surface area of chip
B chip height
d, D nozzle diameter
h heat transfer coefficient
H channel height
Mo Mouromtseff number
Nu Nusselt number
ReH Reynolds number based on channel height
ReL Reynolds number based on chip length
Te element (chip) temperature
Enhanced Air Cooling of Electronic Equipment 197

channel mean-inlet temperature

channel inlet velocity
nozzle to target spacing ( 3 d Vh 4 )
d~mensi?nl~ss p~mping power <!> = 4 d~ --y:r
kmemat1C VISCOSity p

REFERENCES
Aihara, T. 1991. Air cooling techniques by natural convection, in Cooling Techniques for
Computers, W. Aung, Ed., Hemisphere, New York, 1-45.
Amon, C.H. 1992. Heat transfer enhancement by flow destabilization in electronic chip configu-
rations, J. Electron. Packag., 114:35-40.
Anderson, A.M. and Moffat, R.J. 1991. Direct air cooling of electronic components: reducing com-
ponent temperatures by controlled thermal mixing, J. Heat Transfer, 113:56-62.
Antonetti, V.W. 1990. Thermal Contact Resistance in Electronic Equipment, in Heat Transfer in
Electronic and Microelectronic Equipment,A.E. Bergles, Ed., Hemisphere, Washington D.C.,
663-673.
Ashiwake, N., Nakayama, W., Daikoku, T., and Kobayashi, F. 1983. Forced convective heat trans-
fer from LSI packages in an air-cooled wiring card array, Heat Transfer in Electronic
Equipment, ASME HTD, 28:35-42.
Azar, K. 1992. Enhanced cooling of electronic components by flow oscillation, J. Thermophys.
and Heat Transfer, 6(4):700-706.
Azar, K., Develle, S.E., and Manno, V.P. 1989. Sensitivity of circuit pack thermal performance to
convective and geometric variation, IEEE Trans. Components, Hybrids, and Mfg. Techno!.,
12:732-740.
Azar, K. and Russell, E.T. 1991. Effect of component layout and geometry on the flow distribu-
tion in electronics circuit packs, J. Electron. Packag., 113:50-57.
Babin, B.R. and Peterson, G.P. 1990. Experimental investigation of a flexible bellows heat pipe for
cooling discrete heat sources, J. Heat Transfer, 112:602-607.
Baughn, J.W., Hechanova, A.E., and Yan, X. 1991. An experimental study of entrainment effects
on the heat transfer from a flat surface to a heated circular impinging jet, J. Heat Transfer,
3:143-149.
Baumann, W. and Rehme, K. 1975. Friction correlations for rectangular roughness, Int. J. Heat
Mass Trans., 18:1189-1197.
Bejan, A. and Morega, A.M. 1994. The optimal spacing of a stack of plates cooled by turbulent
forced convection, Int. J. Heat Mass Transfer, 37:1045-1048.
Bergles, A.E., Blumenkrantz, A.R, and Taborek, J. 1974. Performance evaluation criteria for en-
hanced heat transfer surfaces, Japan Soc. Mech. Eng., 2:239-243.
Bergles, A.E., Jensen, M.K., Somerscales, E.F.C., and Manglik, R.M. 1991. Literature review of
heat transfer enhancement technology for heat exchangers in gas fired applications, Report
GRI 91-0146, Gas Research Institute, Chicago, IL.
Bergles, A.E., Nirmalan, V., Junkhan, G.H., and Webb, R.L. 1983. Bibliography on augmentation
of convective heat and mass transfer II. Heat Transfer Laboratory Report HTL-31, ISU-ERI-
Ames-84221, Iowa State University, December.
Bhatti, M.S. and Shah, R.K. 1987. Turbulent and transition flow convective heat transfer in ducts,
in Single Phase Convective Heat Transfer Handbook, Wiley-lnterscience, New York.
Blair, M.F. 1983a. Influence of free-stream turbulence on turbulent boundary layer heat transfer
and mean profile development. I. Experimental data, J. Heat Transfer, 105:33-40.
198 Air Cooling Technology

Blair, M.F. l983b. Influence of free-stream turbulence on turbulent boundary layer heat transfer
and mean profile development. II. Analysis of results, J. Heat Transfer, 105:41.
Bowers, M.B. and Mudawar, I. 1993. Two-phase electronic cooling using mini-channel and micro-
channel heat sinks. H. Flow rate and pressure drop constraints, Adv. in Electron. Packag.,
ASME EEP. 4, 2:703-712.
Castro, I.P. and Robins, A.G. 1977. The flow around a surface-mounted cube in uniform and tur-
bulent streams, J. Fluid Mech., 79:307-335.
Chou, 1.H. and Chiang, K.F. 1990. Thermal enhancement for simulated LSI packages by external
rods, in Heat Transfer in Electronic and Microelectronic Equipment, A. E. Bergles, ed.,
Hemisphere, Washington, D.C., 393-404.
Chou, J.H. and Lee, J. 1987. Reducing flow non-uniformities in LSI packages by vortex genera-
tors, in Proc. Int. Symp. Cooling Technology for Electron. Equipment, Hawaii, 583-594.
Chu, R.C., Seely, J.H., Antonetti, V.W., and Pascuzzo, A.L. 1970. Thermal design optimization in
large digital systems, IBM Technical Report, TR 00.2039, June.
Chu, R.C. and Simons, R.E. 1990. Evolution of cooling technology in medium and large scale
computers-an IBM perspective, in Heat Transfer in Electronic and Microelectronic
Equipment, A. E. Bergles, ed., Hemisphere, Washington D.C., 41-60.
Crabb, D., Durao, D.F.G., and Whitelaw, J.H. 1977. Velocity characteristics in the vicinity of a two-
dimensional rib, in Proc. 4th Brazilian Congress of Mech. Eng., Paper B-3, 415-429.
Culham, J.R. and Yovanovich, M.M. 1987. Non-iterative technique for computing temperature dis-
tributions in flat plates with distributed heat sources and convective cooling, in Proc.
ASMEIJSME Heat Transfer Conf., Hawaii, 403-409, March.
Daile Donne, M. and Meyer, L. 1977. Turbulent convective heat transfer from rough surfaces with
two-dimensional rectangular ribs, Int. J. Heat Mass Transfer, 20:583-620.
den Ouden, C. and Hoogendoorn, C.J. 1974. Local convective heat transfer coefficient for jets im-
pinging on a plate: Experiments using a liquid crystal technique, in Proc. 5th Int. Heat Transfer
Conf, Tokyo, Japan, September 3-7.
Dipprey, D.F. and Sabersky, R.H. 1963. Heat and momentum transfer in smooth and rough tubes
at various Prandtl numbers, Int. J. Heat Mass Transfer, 6:329-353.
Downs, S.J. and James, E. H. 1987. Jet impingement heat transfer-a literature survey,ASME Paper
87-HT-35.
Dunn, P.D. and Reay, D.A. 1994. Heat Pipes, 4th ed., Pergamon Press, New York.
Durst, F., Founti, M., and Obi, S. 1986. Experimental and computational investigation of the two-
dimensional channel flow over two fences in tandem, in 3rd Int. Symp. on Applications of
Laser Anemometry to Fluid Mechanics, Paper 2.5, Lisbon, July.
Eaton, J.K. and Johnston, J.P. 1981. A review of research on subsonic turbulent flow reattachment,
AIAA J, 19:1093-1100.
Edwards, F.J. and Alker, C.J.R. 1974. The improvement of forced convection surface heat transfer
using surface protrusions in the form of (A) cubes and (B) vortex generators, in Proc. 5th Int.
Heat Transfer Conf, Tokyo, 244--248.
Eibeck, P.A. and Eaton, J.K. 1987. Heat transfer effects of a longitudinal vortex embedded in a tur-
bulent boundary layer. J. Heat Transfer, 109:16--24.
Fiebig, M., Kallweit, P., and Mitra, N.K. 1986. Wing type vortex generators for heat transfer en-
hancement, in Proc. 8th Int. Heat Transfer Conf, San Francisco, 6:2909-2913.
Fisher, E.M. and Eibeck, P.A. 1990. The influence of a horseshoe vortex on local convective heat
transfer, J. Heat Transfer, I J 2:329-335.
Florschuetz, L.W., Truman, C.R., Metzger, D.E. 1981. Stream wise flow and heat transfer distrib-
utions for jet array impingement with crossflow, J. Heat Transfer, 103:337-342.
Gan, Y.P., Wang, S. Lei, D.H. and Ma, C.F. 1990. Enhancement of forced convection air cooling
of block-like electronic components in in-line arrays, in Heat Transfer in Electronic and
Microelectronic Equipment, A. E. Bergles, Ed., Hemisphere, Washington, D.C., 223-233.
Gardon, R. and Akfirat, J.C. 1965. The role of turbulence in determining the heat transfer charac-
teristics of impinging jets, Int. J. Heat Mass Transfer, 8:1261-1272.
Enhanced Air Cooling of Electronic Equipment 199

Garimella, S.V. 1991. Physical mechanisms for the local heat transfer enhancement caused by fin-
!ike obstacles in heat exchanger flow passages, SAE 1991 Transactions, J. ofMater. and Mfg.,
100(5):208-217.
Garimella, S. V. and Eibeck, P.A. 1990. Heat transfer characteristics of an array of protruding ele-
ments in single phase forced convection, Int. J. Heat Mass Transfer, 33(12):2659-2669.
Garimella, S.V. and Eibeck, P.A. 199la. Enhancement of single phase convective heat transfer
from protruding elements using vortex generators, Int. J. Heat Mass Transfer,
34(9):2427-2430.
Garimella, S.V. and Eibeck, P.A. 1991b. Effect of spanwise spacing on the heat transfer from an
array of protruding elements in forced convection, Int. J. Heat Mass Transfer,
34(9):2431-2433.
Garimella, S.V. and Eibeck, P.A. 199lc. Fluid dynamic characteristics of the flow over an array of
large roughness elements, J. Electron. Packag., 113(4):367-373.
Garimella, S.V. and Eibeck, P.A. 1992. Onset of transition in the flow over a three-dimensional
array of rectangular obstacles, J. Electron. Packag., 114(2):251-255.
Garimella, S.V. and Rice, R.A. 1995. Confined and submerged liquid jet impingement heat trans-
fer,]. Heat Transfer, 117(4): 871-877.
Garimella, S.V. and Schlitz, D.J. 1993. Reducing inter-chip temperature differences in computers
using vortex generators in forced convection, J. Electron. Packag., 115(4):410-415.
Garimella, S.V. and Schlitz, D.J. 1995. Heat transfer enhancement in narrow channels using two
and three-dimensional mixing devices, J. Heat Transfer, 117(3): 590--596.
Garratt, J.R. and Hicks, B.B. 1973. Momentum, heat, and water vapor transfer to and from natural
and artificial surfaces, Quar. J. Roy. Met. Soc., 99:680-687.
Ghaddar, N.K., Korczak, K.Z., Mikic, B.B., and Patera, A.T. 1986. Numerical investigation of in-
compressible flow in grooved channels. II. Resonance and oscillatory heat-transfer en-
hancement, J. Fluid Mech., 168:541-567.
Goldstein, R.J. and Behbahani, A. I. 1982. Impingement of a circular jet with and without cross
flow, Int. J. Heat Mass Transfer, 25:1377-1382.
Goldstein, R.J. and Karni, 1. 1984. The effect of a wall boundary layer on local mass transfer from
a cylinder in crossflow, J. Heat Transfer, 106:260--267.
Goldstein, R.J., Chyu, M.K., and Hain, R.C. 1985. Measurement of local mass transfer on a sur-
face in the region of the base of a protruding cylinder with a computer-controlled data ac-
quisition system, Int. J. Heat Mass Transfer, 28(5):977-985.
Goldstein, R.J ., Behbahani, A .I., and Kieger Heppelmann, K. 1986. Streamwise distribution of the
recovery factor and the local heat transfer coefficient to an impinging circular air jet, Int. J.
Heat Mass Transfer, 29:1227-1235.
Greiner, M. 1991. An experimental investigation of resonant heat transfer enhancement in grooved
channels, Int. J. Heat Mass Transfer, 34(6):1383-1391.
Hadim, A. and Bethancourt, A. 1993. Numerical study of forced convection in a partially porous
channel with localized heat sources, Adv. in Electron. Packag., ASME, EEP-4-2:649-657.
Hall, W.B. 1962. Heat transfer in channels having rough and smooth surfaces, J. Mech. Eng. Sci.,
4:287-291.
Han, J.C., Glicksman, L.R., and Rohsenow, W.M. 1978. An investigation of heat transfer and fric-
tion characteristics for rib-roughened surfaces, Int. J. Heat Mass Transfer, 21:1143-1156.
Hollworth, B.R. and Fuller, H.A. 1987. Heat transfer and pressure drop in a staggered array of air
cooled components, in Proc. Int. Symp. Cooling Techno! for Electron Equipment, Honolulu,
Hawaii, 732-748.
Hollworth, B.R. and Gero, L.R. 1985. Entrainment effects on impingement heat transfer: Part I l -
Local heat transfer measurements, J. Heat Transfer, 107:910--915.
Huber, A.M. and Viskanta, R. 1994. Effect of jet-jet spacing on convective heat transfer to con-
fined, impinging arrays of axisymmetric air jets, Int. J. Heat Mass Transfer, 37:2859-2869.
Igarashi, T. and Takasaki, H. 1992. Fluid flow around three rectangular blocks in a flat-plate lam-
inar boundary layer, Experimental Heat Transfer, 5:17-31.
200 Air Cooling Technology

Ireland, P.T. and Jones, T.V. 1986. Detailed measurements of heat transfer on and around a pedestal
in fully developed passage flow, in Procs. 8th Int. Heat Transfer Conf, C. L. Tien et al. Eds.,
3:975-980.
Jicha, M. and Horsky, J. 1990.lntensification of air cooling by inclined inlet of air, in Heat Transfer
in Electronic and Microelectronic Equipment, A. E. Bergles, Ed., Hemisphere, Washington,
D.C., 365-380.
Jiji, L.M. and Dagan, Z. 1987. Experimental investigation of single phase multi-jet impingement
cooling of an array of microelectronic heat sources, in Procs. Int. Symp. Cooling Techno/. for
Electron. Equipment, Honolulu, Hawaii, 265-283.
Kader, B.A. and Yaglom, A.M. 1977. Turbulent heat and mass transfer from a wall with parallel
roughness ridges, Int. J. Heat Mass Transfer, 20:345-357.
Kamiadakis, G.E., Mikic, B.B., and Patera, A.T. 1988. Minimum dissipation transport enhance-
ment by tlow destabilization: Reynolds' analogy revisited, J. Fluid Mech., 192:365-391.
Koh, J.C. Y and Colony, R. 1986. Heat transfer microstructures for integrated circuits, Int. Comm.
Heal Mass Transfer, 13:89-98.
Kraus, A.D. and Bar-Cohen, A. 1983. Thermal Analysis and Control of Electronic Equipment,
McGraw-Hill, New York, chap. 19.
Lim, F.K. and Lewkowicz, A.K. 1986. Investigation by laser Doppler anemometry of the turbulent
flow around cylindrical obstacles on a rough surface, Int. J. Heat Fluid Flow, 7:102-108.
Lytle, D. and Webb, B.W. 1994. Air jet impingement heat transfer at low nozzle-plate spacings,
Int. J. Heat Mass Transfer, 37:1687-1697.
Maciejewski, P.K. and Moffat, R.J. 1989. Heat transfer with very high free stream turbulence,
Report HMT-42, Stanford University, Stanford, CA.
Mantle, P.L. 1966. A new type of roughened heat transfer surface selected by flow visualization
techniques, in Proc. 3rd Int. Heat Transfer Conf, 1:45-55.
Martin, H. 1977. Heat and mass transfer between impinging gas jets and solid surfaces, Adv. Heat
Transfer, 13:1-60.
Matsushima, H. and Yanagida, T. 1992. Heat transfer from finned LSI packages in a channel be-
tween circuit boards, Heat Transfer-Japanese Res., 21(2): 165-176.
Maubach, K. 1972. Rough annulus pressure drop-interpretation of experiments and recalculation
for square ribs, Int. J. Heat Mass Transfer, 15:2489-2498.
Metzger, D.E., Cummings, K.N., and Ruby, W.A. 1974. Effects ofPrandtl number on heat trans-
fer characteristics of impinging liquid jets, in Proc. 5th Int. Heal Transfer Conf, H:20-24.
Meyer, L. 1980. Turbulent tlow in a plane channel having one or two rough walls, Int. J. Heat Mass
Transfer, 23:591-608.
Mikic, B. B., Kozlu, H., and Patera, A.T. 1990. A methodology for optimization of convective cool-
ing systems for electronic devices, in Heat Transfer in Electronic and Microelectronic
Equipment, A. E. Bergles, Ed., Hemisphere, Washington, D.C., 269-286.
Moffat, R.J. and Ortega, A. 1989. Direct air-cooling of electronic components, in Advances in
Thermal Modeling of Electronic Components and Systems, Vol. I, A. Bar-Cohen and A. D.
Kraus, Eds., Hemisphere, New York, chap. 3.
Morris, G.K., Garimella, S. V., and Hlavac, D.L. 1995. Measurements of the wake behind a heated
obstacle in forced convection, to be presented at the National Heat Transfer Conference,
Portland, Oregon.
Myrum, T.A., Qiu, X., and Acharya, S. 1993. Heat transfer enhancement in a ribbed duct using
vortex generators, Int. J. Heat Mass Transfer, 36:3497-3508.
Nakayama, W. 1982. Enhancement of heat transfer, heat transfer-1982, in Proc. 7th Int. Heat
Transfer Conf, Munich, Germany, I :223-240.
Negus, K.J. and Yovanovich, M.M. 1986. Thermal analysis and optimization of convectively cooled
microelectronic circuit boards, HeatTransferinElectron. Equipment,ASME HTD, 57: 167-176.
Nikuradse, J. 1933. Laws for flow in rough pipes, NACA TM, 1292, (English translation, 1950).
North, M.T. and Avedisian, C.T. 1993. Heat pipes for cooling high flux/high power semiconduc-
tor chips, J. Electron. Packag., 115:112-117.
Enhanced Air Cooling of Electronic Equipment 201

Obot, N.T., Mujumdar, A.S., and Douglas, W.J.M. 1980. Design correlations for heat and mass
transfer under various turbulent impinging jet configurations, Drying, 388-402.
Obot, N.T. and Trabold, T.A. 1987. Impingement heat transfer within arrays of circular jets. Part
I--effects of minimum, intermediate, and complete crossflow for small and large spacings,
1. Heat Transfer, 109:872-879.
Osborne, D.G. and Incropera, F.P. 1985. Experimental study of mixed convection heat transfer for
transitional and turbulent flow between horizontal, parallel plates, Int. 1. Heat Mass Transfer,
28(7): 1337-1344.
Pearcey, H.H. 1961. Shock-induced separation and its prevention by design and boundary layer
control. IV. Boundary Layer and Flow Control, Its principles and application. G. V.
Lachmann, Ed., Pergamon Press, New York.
Pedisius, A.A., Kazimekas, P.V.A., and Slanciauskas, A.A. 1979. Heat transfer from a plate to a
high-turbulence air flow, Soviet Res., 11(5):125-.
Perry, A.E., Schofield, W.H., and Joubert, P.H. 1969. Rough wall turbulent boundary layers, 1.
Fluid Mech., 37:193-211.
Phillips, R.J., Glicksman, L.R., and Larson, R. 1987. Forced-convection, liquid-cooled, mi-
crochannel heat sinks for high-power-density microelectronics, in Proc. Int. Symp. Cooling
Technol.for Electron. Equipment, Honolulu, Hawaii, 227-248.
Ratts, E., Amon, C.H., Mikic, B.B., Patera, A.T. 1987. Cooling enhancement of forced convection
air cooled chip array through flow modulation induced by vortex-shedding cylinders in cross-
flow, in Proc.lnt. Symp. Cooling Technol.for Electron. Equipment, Honolulu, Hawaii, 651--662.
Rice, R.A. and Garimella, S.V. 1994. Heat transfer from discrete heat sources using an axisym-
metric, submerged and confined liquid jet, in Proc. Int. lOth Heat Transfer Conf, Brighton,
UK3:89-94.
Russell, C.M.B., Jones, T.V., and Lee, G.H. 1982. Heat transfer enhancement using vortex gener-
ators, in Proc. 7th Int. Heat Transfer Conf., Munich, 3:283-288.
Simonich, J.C. and Bradshaw, P. 1978. Effect of free-stream turbulence on heat transfer through a
turbulent boundary layer, J. Heat Transfer, 100:673.
Sobey, I.J. 1980. On flow through furrowed channels. I. Calculated flow patterns, J. Fluid Mech.,
96:1-26.
Souza Mendes, P.R. and Santos, W.F.N. 1987. Heat transfer and pressure drop experiments in air-
cooled electronic-component arrays, J. Thermophys. and Heat Transfer, 1(4):373-378.
Sparrow, E.M., Niethammer, J.E., and Chaboki, A. 1982. Heat transfer and pressure drop charac-
teristics of arrays of rectangular modules encountered in electronic equipment, Int. 1. Heat
Mass Transfer, 25:961-973.
Sparrow, E.M. and Otis, D.R. 1985. Ductflow heat transfer at a smooth wall which faces a wall
covered by protuberances, Int. 1. Heat Mass Transfer, 28: 1317-1326.
Sparrow, E.M., Vemuri, S.B., and Kadle, D.S. 1983. Enhanced and local heat transfer, pressure
drop, and flow visualization for arrays of block-like electronic components, Int. 1. Heat Mass
Transfer, 26:689-699.
Suzuki, H., Kida, S., Nakamae, T., and Suzuki, K. 1991. Flow and heat transfer over a backward-
facing step with acylindermounted near its top comer, In/. 1. Heat Fluid Flow, 12(4 ):353-359.
Thomas, D.G. 1965. Forced convection mass transfer. Part II. effect of wires located near the edge
of the laminar boundary layer on the rate of forced convection from a flat plate, A/ChE 1.,
II :848-852.
Thomas, D.G. 1966. Forced convection mass transfer. Part III. Increased mass transfer from a flat
plate caused by the wake from cylinders located near the edge of the boundary layer, AIChE
1., 12:124-130.
Tien, C.L. and Kuo, S.M. 1987. Analysis of forced convection in microstructures for electronic system
cooling, in Proc./nt. Symp. Cooling Techno/.jor EleCiron. Equipment, Honolulu, Hawaii, 217-226.
Torikoshi. K., Kawazoe, M., and Kurihara, T. 1988. Convective heat transfer characteristics or ar-
rays of rectangular blocks affixed to one wall of a channel, Natural and Mixed Convection
in Electronic Equipment Cooling, ASME HTD, I 00:59-65.
202 Air Cooling Technology

Townes, H.W. and Sabersky, R.H. 1966. Experiments on the flow over a rough surface, Int. J. Heat
Mass Transfer, 9:729-738.
Tropea, C.D. and Gackstatter, R. 1985. The flow over two-dimensional surface-mounted obstacles
at low Reynolds numbers, J. Fluids Eng., 107:489-494.
Tuckerman, D.B. and Pease, R.F.W. 1981. High-performance heat sinking for VLSI, IEEE Electron
Device Lett., EDL-2(5): 126-129.
Tuckerman, D.B. and Pease, R.F. 1982. Ultrahigh thermal conductance microstructures for cooling
integrated circuits, in Proc. 32nd Electron. Components Conf, IEEE, EIA, CHMT, 145-149.
Vafai, K. and Tien, C.L. 1981. Boundary and inertia effects on flow and heat transfer in porous
media, Int. J. Heat Mass Transfer, 24:195-203.
Wassel, A.T. and Mills, A.F. 1979. Calculation of variable property turbulent friction and heat trans-
fer in rough pipes, J. Heat Transfer, 101:469-474.
Webb, R.L. 1987. Enhancement of Single-Phase Heat Transfer, in Single Phase Convective Heat
Transfer Handbook, Wiley-Interscience, New York, chap. 17.
Webb, R.L. 1994. Principles of Enhanced Heat Transfer, John Wiley and Sons, New York.
Webb, R.L., Eckert, E.R.G., and Goldstein, R.J. 1971. Heat transfer and friction in tubes with re-
peated-rib roughness, Int. J. Heat Mass Transfer, 14:601-617.
Wenthen, F.T. 1977. Experimental verification, presented at NSF Research Workshop on Directions
of Heat Transfer in Electronic Equipment, Atlanta, Georgia, October.
Whidden, G.L., Stevens, J., and Webb, B.W. 1992. Heat transfer and flow characteristics of two-
dimensional jets impinging on heated protrusions with crossflow of the spent air, J. Electron.
Packag., I 14:81-87.
Wroblewski, D.E. and Eibeck, P.A. 1991. Measurements of the turbulent heat transport in a bound-
ary layer with an embedded streamwise vortex, Int. J. Heat Mass Transfer, 34:1617-1631.
Wu, D., Peterson, G.P., and Chang, W. 1990. Experimental investigation of the transient behavior
of micro heat pipes, AIAA Paper 90, 1791.
Yovanovich, M .M. 1987. Theory and applications of constriction and spreading resistance con-
cepts for microelectronic thermal management, in Proc. Int. Symp. Cooling Techno/. for
Electron. Equipment, Honolulu, Hawaii, 1-80.
Zhang, Fiebig, M., and Mitra, N .K. 1989. Vortex breakdown on heat transfer enhancement in flows
between parallel plates, Eurotherm Seminar91 , 97- 104, Heat Transfer in Single Phase Flows,
Proceeding from seminar in Bochum, Germany.
 Chapter 6

LIMITS OF AIR COOLING-

A METHODICAL APPROACH

Kaveh Azar

CONTENTS

Introduction . . . . . .. . . . .. . . . .. ... . .. . ... . . . . ... . . . . ... . .. . .204

Thermal Phenomenon in Electronic Enclosures . . . . . .... . .. .. .. . . .204

Heat Transfer Mechanisms .... . .. .. ..... . . . .. . . . . . . . .......205
Fluid Flow in Circuit Packs . .. . . .. .. . .... . . . . . . .... . . . . ....208
Heat Transfer in Electronic Components (Modules) .. . . .. . .. .. . . .210
Heat Transfer in Circuit Packs and Its Effect on Components ... . .. .211

Thermal Coupling in Electronic Enclosures . . . . . . . . ......... . . .. .212

Component . .... .. . ........ .. . . .. .. . .. . ......... . .. .. . .212
Board/Shelf . . . . . . .. ... .. . . . . .. . .... . . . . . .... . . . ... .. . . .213
Frame (Enclosure) . .... .. . . . .. ...... .. . .. . . . . ... .. .. . . . . .214
Environment . . .... .. .. ... . . . .. . . .. . . .. . .. . . .. . .. .. . . . . .214
System-Level Approach to Thermal Management . ... . . . . . . .... . 2 15

Understanding the Concept of Cooling Limit . . . . . . ... . . . . . . . .. . . .215

Parameters Impacting Heat Transfer . . . . . . . . .. . .. . .... . . . . . . . .. .218

Heat Transfer in a Channel .. ... .. .... .. .. . . . . ... . ... .. . . . . .219
Heat Transferfrom a Component . . . .. . . . ........... .. . . . . . . .221
Significant Parameters ... . . . . ... . . . . . . . . .. . .. . .. . . . .. .. .. .224

Potential Limits of Natural and Forced Convection Cooling . . ... . ... .226
Forced Convection Limit Analysis . . .. . . . . . . . ... . . . . . . . .. ... .226
Natural Convection Optimization . . . .. . . . . . . . .. . . . ... . . . . .. . .233
Methodical Procedure-An Integral Approach . ... ... .. . . . .. ... .239

Conclusion ... . .. .... . ... . . .. . . ...... . .. . .. . .. . .. . ...... ..240

Appendix A: Nomenclature . . .. .. . . .. .. .. . .... . . . .. . . . .. . .. . .241

References .... .. ..... . .. . . . .. ... .... . . . . ..... .... ...... . .243

0-8493-9447-3/961$0.00+$ .50
© 1996 by CRC Press, Inc. 203
204 Air Cooling Technology

INTRODUCTION

Most electronic systems are cooled by air because of its availability and ease
of implementing. Cooling is done either by natural or forced convection de-
pending upon the system's requirements and application. Cooling system designs
vary from simple natural convection to high-capacity forced convection. In nat-
ural convection cooled systems, the circulation of air is accommodated by place-
ment of vent ports, e.g., a VCR, along with proper conduction heat transfer paths
to disperse the heat from the critical areas. In forced convection systems, fan is
the most common fluid mover used. In high-capacity forced convection cooled
systems, jet impingement along with surface enhancement (heat sink) is used
[Bar-Cohen, 1987]. Typically, these systems have a great deal of customization
that is the inherent nature of them; thus, they are not as generic as placing fans
in the system.
In a typical design process, one confronts the question of which cooling sys-
tem or method is most suitable. Typically, a design criterion relating to junction
temperature or industry standards is considered, e.g., NEBS for telecommuni-
cation equipment [NEBS, 1993]. The desire to cool all systems with natural con-
vection continues to persist, but adherence to specifications and limited cooling
capacity is a point of contention. Likewise, cooling with forced convection may
not be feasible and high-capacity cooling, i.e., liquid cooling, may be considered
for a given system. The question is: What are the limits of air cooling and at what
point must higher-end cooling be used? In this chapter, I attempt to answer this
question and provide mechanisms for predictions. In addition, I will attempt to
highlight the important parameters that contribute to the performance of cooling
with air.
To achieve the goal of the chapter, I first review the thermal phenomenon in
electronics systems. The intent of this section is to highlight the importance of
starting thermal management at system level versus individual component or
circuit board. It further elevates the need to look at the component junction tem-
perature instead of total heat dissipation of the system or circuit pack. Hence,
thermal coupling and transport is a point of interest and needs to be understood.
Then the chapter focuses on the concept of limit, and why it is important to look
at the junction temperature instead of the heat flux as the limiting factor. The
two subsequent sections show a first-order model that shows heat transfer from
a circuit board and a component. These models allow us to highlight the im-
portant parameters that impact heat transfer from a component and must be con-
sidered in the design process. The last two sections deal with the potential limits
of cooling with natural and forced convection when air is the cooling fluid.

THERMAL PHENOMENON
IN ELECTRONIC ENCLOSURES

Thermal phenomenon or process defines the procedure for removal of heat

from the components. Thermal process is defined as the merger of heat trans-
Limits of Air Cooling-A Methodical Approach 205

fer and fluid flow to transport the energy. The very nature of the thermal
phenomenon is then a function of the mode of heat transfer and the fluid flow
regime in a system. Thus, proper definition of thermal phenomenon requires
understanding of transport mechanisms-heat transfer and fluid flow.
In this section, I start by defining the basic principles of heat transfer.
Then heat transfer in electronic components is discussed. Here, I discuss
how the power, being dissipated in the chips, is eventually transferred to
the cooling fluid. Since circuit packs by definition contain several compo-
nents and play an important role in thermal response, thermal transport in
circuit packs (PCB) are then discussed. The last two sections talk about the
thermal coupling (communication) between elements that form an electronic
enclosure (system).

HEAT TRANSFER MECHANISMS

There are three modes of heat transfer: conduction, convection, and radia-
tion. Conduction heat transfer is when the heat is transferred by molecular vi-
bration-solids or stagnant fluids. An example of a solid is the molding
compound or the substrate in a component, and stagnant fluid is the air trapped
between the component and circuit board. The conduction heat transfer is gov-
erned by Fourier cooling law defined as

(1)

In this equation, k is the thermal conductivity and is a property of the material.

Table 1 shows the therrnal conductivity of typical materials used in components.
Convection heat transfer is when the transport of heat takes place by fluid
motion. Three types of convection heat transfer are recognized: natural (free),
forced, and mixed. Natural convection occurs as a result of fluid (air) being
in contact with a heated surface. The density of the fluid decreases, causing
the fluid to rise and thus creating a natural circulation. Forced convection
occurs when the fluid motion is induced by external sources. These external
sources include fans, pumps, blowers and other fluid movers. Mixed con-
vection occurs when natural and forced convection are both present. This
is typically observed in low-velocity flows in the presence of high-
powered components.
Convection heat transfer is governed by Newton's cooling law defined as

(2)

Irrespective of the type of convective heat transfer, Equation 2 is used for

its solution. What sets the three types of convection heat transfer apart is h
(heat transfer coefficient). The coefficient h is obtained from empirical data.
Table 2 gives ranges of h encountered in cooling of electronic components.
Also, the literature is rich with research papers that show different correlations
for a heat transfer coefficient. References [Kays, et al., 1980; Wirtz, et al., 1989
206 Air Cooling Technology

TABLE!
Material Property Typically Used in Components

Material Maximum Use K (W/cm 0 C) Application

Metals

Aluminum 660 2.1 Chip conductor and wire bonds

Gold 1,063 3.4 Hybrid conductor and wire
bonds
Copper 1,083 3.8 Lead frame and hybrid, PWB,
and conductor
Lead 327 0.3 Solder attach
Molybdenum 2,610 1.3 Cofired on ceramic conductor
Tungsten 3,380 !.5 Cofired on ceramic conductor

Organics

Epoxy (70% Si02) 170 0.002 Packaging

Epoxy glass (FR-4) 120 0.02 Multilayer board substrate
Adv. epoxy (resin only) 180 0.02 Multilayer board substrate
Triazine 250 0.002 Hybrid dielectric
BT resin (laminate) 290 0.005 Flexible substrate
Polymide 400 0.0007 Flexible substrate
Polymide 310 0.0007 Interlayer dielectric

Inm:ganics

Alumina (ceramic) 1,600 0.3 Hybrid substrates/chip carriers

Silica (fused) 1,100 0.02 Filler for molding epoxies
Silicon nitride 2,000 0.3 Candidate substrates
Aluminum nitride 1,800 3.2 Candidate substrates
Silicon carbide 2,100 2.7 Candidate substrates
Silicon 1,400 1.5 Candidate substrates
Diamond >3,500 20.0 Candidate encapsulation
Glass-ceramic >1,000 0.05 Candidate substrates
Beryllia 1,500 2.6 Chip carriers

Source: Azur, 1992.

and 1984; Sparrow, et al., 1982; Sridar, et al., 1990] are examples of some of
these articles that may become useful to the reader.
A word of caution seems merited at this point with respect to h and its re-
spective correlations. As mentioned, h is a coefficient and its value is obtained
from empirical data. To use a given correlation, its constraints must match your
specific problem. Otherwise, that particular correlation is not suited for your
analysis. The second point is that convective heat transfer analysis is an itera-
tive process. The value of h obtained from a correlation is not absolute.
Therefore, the results should be verified in the overall scope of the problem.
Radiative heat transfer occurs when heat is transported by photons or elec-
tromagnetic waves. What sets radiation heat transfer apart from conduction and
Limits of Air Cooling-A Methodical Approach 207

TABLE2
Heat Transfer Coefficient Value
for Different Coolants

Coolant h, W/m'·°C

Natural Convection (plate)

Air 5
Engine oil 37
Water 440
Forced Convection (Uo = 10 m/s)
Air-plate length = 0.1 m 39
Air-plate length= 0.5 m 17
Boiling
Water at I atm, in a container 3,000
Water at l atm, at peak heat nux 35,000
Film 300
Condensation
Steam at l atm-on horizontal plate 9,000--25,000
Steam at I atm-on vertical plate 4,000--ll ,000
Drop wise 60,000--120,000

convection is the medium for transport and its nonlinear dependence on tem-
perature. Radiation heat transfer requires no medium to transport the energy,
and it is always present regardless of the application. However, its magnitude
and degree of contribution to overall thermal transport, similar to other modes
of heat transfer, is a function of temperature difference. A typical belief that ra-
diation heat transfer can be ignored is purely a misconception. This is more so
in natural convection problems, where in excess of 20% of thermal transport is
attributed to radiation.
Radiation heat transfer is governed by Equation 3

(3)

Values for Fhc (view or shape factor) are tabulated in any heat transfer text
[Incropera, 1990; White, 1984]. A point that should be noted is that T hand Tc are
in absolute units of temperature. For more in-depth review of the fundamentals
ofheattransfer, the reader is referred to references [Incropera, 1990; White, 1984].
Equations 1 through 3 define the basic concepts of heat transfer, and more
importantly, parameters impacting it. The common denominator in the three
equations is the temperature difference, stemming from the definition of heat
transfer. Second, area, either cross-sectional in the case of conduction, or sur-
face, in the cases of radiation and convection, plays a significant role in the
magnitude of heat transfer. Likewise, it can be a limiting factor in the transport
of heat. In the case of convection heat transfer, heat sinks are implemented to
increase surface area. Similar surface enhancements are also used for heat trans-
fer improvement by radiation.
208 Air Cooling Technology

Material properties also play a significant role in heat transfer. Thermal con-
ductivity and surface emissivity are such examples. Heat transfer coefficient h
in Equation 2, is another parameter affecting heat transfer. Although h is not a
property, it is a function of air velocity and cooling fluid properties. Depending
upon a given application, fluid velocity can be the dominating parameter in in-
fluencing the magnitude of h. Hence, knowledge of the value of velocity is es-
sential in heat transfer calculation. This is the subject of the next section.

FLUID FLOW IN CIRCUIT PACKS (PCBs)

As described in the previous section, air velocity can have a major impact
on the overall heat transfer. In the following sections, the dominant role that
the velocity plays becomes more apparent as I describe temperature rises of the
board and component. Here, I show a methodology for determining air veloc-
ity in a circuit pack based on the integral form of the momentum equation
[Murray, 1990]. The focus of this section is at the circuit pack and component
levels; however, its expansion to the system level is obvious as terms are not
confined to these levels.
We would like to develop an expression from which air pressure drop and,
subsequently, velocity can be obtained. Consider the local momentum integral
balance under the assumption of approximately equal inlet and outlet momen-
tum fluxes:

L, (AP)dAcr - A,L (Kp V /2)- L"CwdA,

2 (4)
= Momentum flux;" - Momentum fluxout +Buoyancy

Integrating by using the first order model yields

Estimating the shear stress by the following:

Where () is the momentum boundary layer thickness. The above equation for
pressure drop becomes

AP = (Djorm + DfriJR 2 (7)

where

Drorm = pK/2A,; (8)

and

Dr,;c = (!liA;V 2 )I:{v;A,,/oi) (9)

Limits of Air Cooling-A Methodical Approach 209

The form drag (Drorm) is a function of the loss factor (K), which is based on
contraction and expansion in ducts. The contraction (cc) and expansion (ec) co-
efficients are obtained from the following equations:

(10)

and

(11)

The above equations are accurate as long as the inlet to outlet area ratio equals
0.76, beyond this point Kcc = Kec.
The inlet area for the contractive loss is the channel height multiplied by the
control volume width, while the exit area is the inlet area minus the height times
the width of the component. This is reversed for the expansion loss because the
inlet area is around the component while the outlet is the area of the control
volume. We should note that because of typical low air velocities observed in
electronic systems, frictional losses can be ignored without jeopardizing the
accuracy of the solution.
Having developed an expression for pressure drop, now we need to calcu-
late the velocity. For a forced convection system, where a fan is used, the pro-
cedure is straight forward. To obtain the velocity, we have to match the system
curve to the fan curve to obtain the volumetric flow rate. Calculation of veloc-
ity for natural convection requires the following steps.
Initially, the volumetric flow rate through the system based on the ambient
temperature is estimated by Equation 9:

02)

Here the word "system" applies to the problem at hand. This flow estimate is
based on the equality

(13)

where density (p) is obtained from the ideal gas law and

Dsystem = Drorm + Drric (14)

An estimate for T can be made on the basis of

Q = P Rsystem Cp (T - Tamb) (15)

After calculating the temperature changes for each board in the rack, the flow
per board is calculated based on the following:

Rsystem = (gHp,1.T /(2 Tin Dsystem )) 05 (16)

210 Air Cooling Technology

The preceeding process is iterative in nature, and these calculations are repeated
until a converged solution is obtained.
Equations 8 and 16 show how the air velocity in a system or board can be
calculated and what the pertinent parameters are that impact flow and sub-
sequently heat transfer. These equations clearly highlight the important role
which physical geometry plays in air flow through electronic circuit packs
and enclosures.

HEAT TRANSFER IN ELECTRONIC COMPONENTS (MODULES)

This section is intended to familiarize the reader with generation, spread-
ing, and eventual departure of heat from a component. It should provide the
reader with sufficient understanding to be warned of simplistic solution ap-
proaches to a rather complex problem.
Electronic components (modules) are made of an aggregate of materials
with different physical geometry. The central core of the component is the
die containing resistors, capacitors, inductors, transistors, etc., typically
bonded to a silicon. The combination of these parts form the chip where the
electrical functions are performed. The chip is mounted by an epoxy or other
material onto the substrate. The chips and substrate assembly is, in some
cases, molded to protect the chip from potential environmental or processing
hazards. Some multichip modules (MCM), or single-chip modules (SCM)
continue to be made without molding, but the plastic molded packaging is
becoming an industry standard.
The electrical signals are brought to the component via the leads and then
to the chip(s) through the wire bonds. The leads are either press fitted or sol-
dered to the substrate. The component is connected to the board through its
leads either by surface-mounting or through-hole methods. This connection
creates a package with potentially multiple heat sources in thermal communi-
cation with each other via the substrate and its ambient.
The generated heat at the chips seeks the "least resistant" path to reach the
sinks. The sinks where the heat is eventually transferred are the cooling fluid
and the board. The paths available for heat flow are through the molding ma-
terial and the leads after it is first conducted to the substrate. The flow of heat
is impeded by each material, regardless of its thickness, as it travels from the
source( s) to the sinks. The wire bond also provides another avenue for the trans-
port of heat. This can become a major path if the cross-sectional area is pro-
portionally large.
As the heat reaches the leads, part of it is conducted to the board and the rest
is either radiated and/or convected to the ambient. Also, a similar process oc-
curs as the heat reaches the physical boundaries of the component. The heat-
transfer path through the component is not necessarily one-dimensional and it
tends to flow in any direction that poses the smallest resistance. Combination
of multiple heat sources and different possible avenues for heat flow has cre-
ated a rather complex and nonuniform temperature field in SCMs and MCMs.
Limits of Air Cooling-A Methodical Approach 211

HEAT TRANSFER IN CIRCUIT PACKS

AND ITS EFFECT ON COMPONENTS
The circuit board acts as a channel not only for electrical communication
but a thermal one as well. Similar to electrical signals transmitted through the
routings on a board, the thermal signals are transmitted through the solid. The
thermal signals go through any material, even material with very low thermal
conductivity. The rate of heat transmission is a function of the material con-
ductivity and cross-sectional area. Of course, this transmission rate is different
between electrical and thermal, but it occurs nevertheless. Based on this dis-
cussion, two questions are raised: what are the modes of thermal communica-
tion and how does it affect thermal design?
The heat transfer in a circuit pack takes place by fluid motion (convection)
and conduction heat transfer in the board. For the fluid motion, imagine a cir-
cuit pack inserted vertically into a frame and air used as the cooling fluid. The
circuit pack is populated with a number of components and there are two crit-
ical components placed on the bottom and top of the board. Remember to make
these components functional; a number of other components are necessary.
These components may be moderate to high in power dissipation. As the air
enters the channel formed by the circuit pack and its neighboring circuit packs,
its temperature rises as the result of convective heat transfer. By the time air
gets to the top of the board, where the other critical component resides, its tem-
perature can be significantly higher than the inlet. Part of the heat that was dis-
sipated by the bottom critical component was carried to the other critical
component by air. Although these components are sufficiently apart, they are
in thermal contact via the fluid motion-thus, thermally coupled.
Another mode of thermal communication is through the board itself. The cir-
cuit board material can vary from pure glass-epoxy to glass-epoxy with multi-
ple layers of copper. Because of electrical functionality, most components require
several layers of copper in the board. The copper increases thermal conductiv-
ity of the board, hence enhancing conduction thermal coupling between com-
ponents. The effect or radius of influence of conduction coupling is a direct
function of board thermal conductivity. Let's reconsider the vertically inserted
circuit pack and the two critical components. For the sake of discussion consider
the board to have four layers of copper and assume we can tum these compo-
nents on and off. Tum off the component in the bottom and switch on the top
one while monitoring the temperature of the bottom one. We will see that the
temperature (junction or case) of the one that is off varies as we switch (after
steady state) the other component. If these components were placed in close prox-
imity, the response would have been faster. Also, if the board was an eight layer
instead of four then the component temperature would have been higher. If the
component is low in power but surrounded by higher power components on a
board, the impact of neighboring components must be considered in the thermal
design [Manno, 1992]. This simplistic example illustrates the importance of ther-
mal coupling through the circuit pack and its potential impact on thermal design.
212 Air Cooling Technology

THERMAL COUPLING
IN ELECTRONIC ENCLOSURES

To appreciate the impact of system (enclosure) on thermal performance of

components, it is necessary to review the thermal phenomenon in an enclosure.
The thermal phenomenon is described by coupling between components and
system parts. The term "coupling" implies interdependency. Coupling means
that the thermal performance of the component is directly dependent on other
components and systems. As I previously discussed, the heat-transport mech-
anism, either by the fluid or through solid, can become very complex. To gain
a better understanding of thermal coupling, we must focus on each system part.
We first start from the component and go up to the frame and see how the heat
is transported within an electronic system. This should assist us with the de-
sign process by helping to generate the necessary information or at least in ask-
ing the right questions.
An electronic enclosure is composed of the following:

1. Environment
2. System
3. Shelf (cage)
4. Circuit pack
5. Component
6. Die parts

Figure 1 is a schematic view of the above. What follows is an elaboration of

the above list.

COMPONENT
The thermal process in a component was described in the section on Heat
Transfer in Electronic Components. Since MCMs have gained more attention
and are appearing in many products, a brief discussion about MCMs merits the
effort. MCMs by their definition contain more than one chip. These chips are
residing on one or both sides of a substrate. If we look at an MCM closely, we
see that it resembles a miniature circuit pack. Many of the thermal coupling is-
sues that we discussed in the section 2 on Heat Transfer in Circuit Packs and
Its Effect as Components are also prevalent within MCMs. Since the chips re-
side on a substrate and substrates typically have high thermal conductivity, the
potential for conduction thermal coupling is even greater than the circuit pack.
Much thermal spreading takes place as a direct result of substrate thermal con-
ductivity. This suggests that the critical chip(s) is not necessarily the one that
is the largest or has the highest power dissipation. Because of the thermal spread-
ing, smaller chips with potentially lower junction temperature tolerance may
become the critical ones. Therefore, in the limit analysis, it is essential to look
Limits of Air Cooling-A Methodical Approach 213

Environment
Circuit
I Component/Die
ack

Syste

FIGURE 1. Schematic of system configuration.

at the junction temperature of each chip and ensure that it is within its design
specification.

BOARD/SHELF
The shelf or card holder (cage) is where the circuit packs reside in the sys-
tem. The shelf basically acts as a housing and facilitates electrical connection
of the boards through the backplane. Boards are normally inserted into the
shelves through card guides. Except in some specialized cases where a latch-
ing mechanism is used to rigidly attach the board to the shelf, the boards are
loosely fitted inside the shelf (e.g., a PC motherboard). In these cases, the pri-
mary holder of the circuit packs is the backplane or the motherboard. Therefore,
the necessary contact to facilitate conduction heat transfer from the board to
the shelf does not exist.
Heat generated at the component is conducted to the board through the leads
and the gap. Heat is then transported away from the board by all three modes
of heat transfer. Since poor contact exists between the board and the card guides,
the conduction heat transfer is significantly less than that of convection and ra-
diation, respectively. This may not be the case, however, if the boards are in
good thermal contact with the shelf. We can generalize, based on observation,
that in most system designs conduction coupling between the board and the
shelf is very weak, e.g., telecommunication rack or PC motherboard.
The backplane, e.g., motherboard in a PC, is another avenue for the heat
to be transported to the ambient or the shelf. If the thermal conductivity of
the board is very large, i.e., multilayered boards with several layers of cop-
per, conduction heat transfer to the backplane can be significant. Depending
upon the magnitude of heat dissipation and the nature of contact between the
214 Air Cooling Technology

board(s) and the backplane, boards can also be thermally coupled via the
backplane. The thermal coupling by convection and radiation heat transfers
is significantly larger than conduction heat transfer. Therefore, board/shelf
combination provides another avenue for the heat to be coupled with the rest
of the system.

FRAME (ENCLOSURE)
Frames that house single or multiple shelves are generally designed to be
isolated from the shelves. The heat that is generated within the system normally
finds its way out through the vent holes. Although this constitutes the bulk of
heat flow, there exists significant thermal coupling between the boards/shelves
and the frame. The thermal coupling, in the order of significance, is by radia-
tion, convection, and conduction heat transfer. Since the frame is in contact
with the system ambient, it can act both as a sink and source of heat for the sys-
tem. In general, the coupling cannot be categorized as an insignificant part of
the thermal response of the system.
The magnitude of conduction heat transfer is system-design dependent. A
general statement cannot be made that heat transferred by conduction is small
compared to other modes of heat transfer. But, because of the contact resis-
tance, the conduction heat transfer between the shelves and the frame is usu-
ally weak. In addition, convective heating of the frame is also design dependent.
If the flow of the coolant is in dose contact with the frame, the convection heat-
ing will then be appreciably more. The radiation heat transfer, however, is gen-
erally the predominant mode of thermal coupling between the shelves and the
frame. The radiation heat transfer tends to be even more significant if the sys-
tem is cooled by natural convection.

ENVIRONMENT
The frame is coupled to the surrounding ambient via radiation and con-
vection heat transfer. The system ambient can act both as a source and a sink.
Convection and radiation cooling and heating is possible depending upon the
nature of the system ambient, i.e., open atmosphere or climatically controlled
buildings. The magnitude of the heat transfer can vary significantly with the
changes in the system surroundings. This can constitute a major portion of
the total energy transport to or from the system. Thus, frame-to-ambient ther-
mal coupling must be an integral partof the thermal design consideration and
limit analysis.
We can conclude that the thermal transport process in electronic systems is
quite involved and can become complex. Because of many different thermal
processes and strong coupling at various system levels, thermal bookkeeping
is necessary for accurate analysis. In addition, it should be clear that we can-
not only focus on a component (module) without considering the system, en-
vironment, and other parameters affecting thermal transport.
Limits of Air Cooling-A Methodical Approach 215

SYSTEM-LEVEL APPROACH TO THERMAL MANAGEMENT

I have described how system environment and parameters affecting it can
impact MCM thermal performance. To remind ourselves, the objective of ther-
mal design or analysis is to ensure component junction temperature meets de-
sign specifications. It is apparent from our discussion up to now that, in thermal
design of any component, the component alone is not sufficient. Parameters af-
fecting junction temperature vary from system ambient to the neighboring com-
ponents and must be included in the design and limit analysis.
Two procedures are recommended for thermal design or analysis. First,
thermal design simply requires system down approach. It implies that we have
to look at the system ambient and work our way down to the component of
interest. All parameters affecting the process should be considered. The sec-
ond is a methodical approach to thermal design or analysis. In the previous
section, I referred to this as "thermal bookkeeping." What it implies is that a
successful analysis requires keeping track of all the parameters influencing
design. Superficial or casual treatment of these parameters (starting from the
environment to the component) will yield less than desirable results, [Okutani
et al., 1984].

UNDERSTANDING THE CONCEPT

OF COOLING LIMIT

We now explore why the question of air cooling limit is important to ther-
mal engineers. Perhaps we can shed some light on this question by restating
the objective of electronics cooling. In thermal design of electronic systems,
our goal is to ensure that component junction temperature, either an MCM or
SCM, will be retained below a certain limit. This limit by default is l25°C. I
say by default because there are no references that argue for this or any limit.
However, the reason for the low temperature is rather obvious. The activation
energy attributing to the expected life of the component is exponentially de-
pendent on temperature [Klinger et. al., 1990]. Further, the silicon tends to be
temperature sensitive and at elevated temperatures detectable performance
degradations are observed. Therefore, to ensure that the design is within the
specifications, one needs to know how much heat can be removed by the cool-
ing system of an electronic enclosure-lending itself to the concept of limit.
If we consider the discussion of the previous sections and the goal of elec-
tronics cooling, the focus of the limit becomes obvious. Typically, as will be
seen throughout this chapter, any discussion with respect to a limit has been fo-
cused on the total heat removal from the channel formed by the circuit packs or
the entire system. But, as I attempted to highlight, the point of contention is the
component and how well it can be kept below its limit of 125°C. The maximum
heat removal from the channel becomes a secondary issue. However, the process
of limit analysis is more meaningful if we look at the temperature rise instead
of the absolute temperature.
216 Air Cooling Technology

The earlier focus on the channel or the system as the limiting prospect
has diminished the potential of air cooling for some systems. If we look at
the excellent pioneering work [Kraus and Bar-Cohen, 1983], Figure (2)
shows the potential component case to ambient temperature rise as a func-
tion of heat flux.
For the sake of understanding the limit consider the following: a plastic molded
package with a maximum junction temperature (Tj) of 125°C, an ambient tem-
perature of 50°C, and a junction to ambient temperature rise of 75°C. If we as-
sume a 10° to l5°C temperature drop from junction to case, typical of most
plastic molded components, case to ambient temperature rise will be 60°C. If
we were to cool this component with air in natural or forced convection, the ex-
pected heat flux from Figure 2, will be 0.05 and 0.15 W/cm\ respectively. Note
that in this exercise we did not include any board or system level coupling and
effectively looked at the component when it would reside on a glass-epoxy board.

I01!13r---------------------~r---~r---~----~--------~
•
4

u
0
4,)
(.J
c
...
4,)

~
~
'0

- ...:::»
41)

...0
41)
a. G
e 4
{!

FIGURE 2. Temperature differences attainable as a function of heat fluxes for various heat trans-
fer modes and coolant fluids [Kraus and Bar-Cohen, 1983].
Limits of Air Cooling-A Methodical Approach 217

Let's also consider Figure 3 [Simons, 1983], which shows the heat transfer
coefficient for different fluids. Figure 3 suggests that the highest attainable value
for heat transfer coefficient is 0.0025 and 0.022 for natural and forced convec-
tion using air, respectively.
If we use Equation 2 and the case-to-ambient temperature rise (approxi-
mately 60°C, used above), we get 0.125 and 1.1 W /cm 2 for natural and forced
convection, respectively. Comparison of the heat fluxes obtained from Figures
2 and 3 suggests a range of heat flux capability and depicts a very limiting
prospect for air cooling of electronic systems. When more than 80% of systems
are cooled by air, one begins to wonder how it is done.
Consider the more recent work by the author [Azar eta!., 1992] and Hilbert
[Hilbertet a!., 1990]. Both systems consisted of narrow channel heat sinks along
with a Muffin fan as the air mover. The heat flux in the case of [Azar et al.,
1992] was 20 W /cm 2 with components placed on pure glass epoxy board. This

Air 1·3 atm--11

--•
Fluorochemlcal
Vapor
Silicone 011 Natural
Convection
Transformer 011

Fluorochemlcal
Liquids
Air 1·3 atm

Fluorochemlcal
Vapor
Forced
Transformer 011 Convection
Fluorochemlcal
liquids
Wat~r

Fluorochemlcal
liquids } Boiling
Water

.0001
I I II!".001 f I tpf!l t I I tppl I II I"1.0
f

FIGURE 3. Range of heat transfer coefficients for various coolants [Simons, 1983].
218 Air Cooling Technology

implied that there was minimal conduction heat transfer from the components
to the board and the primary path of heat loss was by convection. The tem-
perature rise in this case was 28°C, which was significantly below the limit
we had discussed before. This suggests that by perhaps increasing the volu-
metric flow rate and convective surface area we can still increase the poten-
tial heat flux from the components. The reader can appreciate the fact that if
the board was multilayer (copper ground layers), the thermal performance
would have been even better because the board would have acted as a heat sink
for the components.
There are a few salient points in this exercise. Foremost, the component junc-
tion temperature rise above ambient is a more realistic criterion for gauging
cooling limit than the channel heat flux. Use of the heat flux, either from the
component or board, does not provide a realistic picture of the capability of
cooling with air. The temperature is a much better gauge which is in line with
the design process. Secondly, there is limited data on the potential range of heat
transfer coefficients or heat removal capability when extended surfaces (heat
sinks) are used. Third, there is a need to be focused on component thermal trans-
port when designing the cooling system. Considering the level of thermal cou-
pling taking place in a system, as discussed in Section 3, the system-level-down
approach is essential for proper gauging or design of the cooling system.
However, one should note that the data from Figures 2 and 3 provide a good
starting point in the design process.

PARAMETERS IMPACTING HEAT TRANSFER

In designing the cooling system, be it natural or forced convection, one
typically starts with the calculation of component junction temperature
for different cooling modes. The hierarchy of cooling methods considered
starts with natural convection and extends into high velocity forced con-
vection. The junction temperature is calculated for the worst-case ambient
with natural convection cooling first and then extends into high-velocity
forced convection. The criterion for determining suitability of a cooling mode
is the following:

11 = Tj.calculated /125 :::; 0. 9 (17)

that allows a 10% margin. If 11 is not satisfied for natural convection, en-
hancements, such as improving radiation heattransfer or increasing convective
surfaces, and others are examined. If the criterion is still not satisfied, higher
convection heat transfer modes should be considered. This unsatisfied criterion
suggests an unavoidable iterative process to seek a desirable cooling mode while
barring the tremendous market pressures to adhere to the simplest cooling mode
(i.e., natural convection).
To reduce the iterative process it is desirable for a thermal engineer to
have the important parameters and their impacts to heat transfer a priori.
Limits of Air Cooling-A Methodical Approach 219

Furthermore, the order of impact or the significance of these parameters is

equally desirable. Hence, we attempt to shed some light on this so-called iter-
ative process by looking at the heat transfer from a component. To further ex-
plore the contribution or significance of each parameter, we will use order of
magnitude analysis to assist us in this task.

HEAT TRANSFER IN A CHANNEL

In this section, I attempt to develop a first-order model of heat transfer in a
channel formed by adjacent circuit packs. The final objective is to develop the
parameters affecting heat transfer. In addition, we would like to develop an ex-
pression for the air and board temperature rise in the channel.
In most applications, circuit packs are placed parallel with each other to
form a channel. Telecommunication racks and PCs are examples of these chan-
nels where both walls have a specified heat flux. In the case of even a single
circuit pack in an enclosure, a channel is still formed containing a single heated
wall. Consider a typical case of a channel formed by four circuit packs in a
rack shown in Figure 4. We are going to focus on the channel formed by cir-
cuit packs 2 and 3 for the purpose of this development. It is apparent to the
reader that the development is independent of the channel considered. Begin
by stating some assumptions:

tt 0
t tt

FIGURE 4. Channel formed by adjacent circuit packs placed in a rack (shelf).

220 Air Cooling Technology

1. The heat flux is uniform over the circuit pack-this ignores the local hot
spot and assumes heat is uniformly distributed. In many cases this is a
valid assumption, especially when the circuit board contains multiple
copper layers and components are fairly uniform in power dissipation.
2. All the heat generated leaves the board only on one side-in reality,
both sides of the board dissipate the generated power. However, since
we are considering a rack of boards of equal power dissipation, the over-
all power dissipation in the channel is the same. If the board power dis-
sipation is not the same then the channel heat transfer can be estimated
[Azar et aL, 1994].
3. Radiation heat transfer between the boards is negligible since both boards
are of equal temperatures.
4. Inlet channel temperature T; and board power dissipation, Q, are given.

The heat transfer from board 3 is obtained from Equation 2

(18)

where T m is the mean air temperature in the channel give by

(19)

Since To is not known we can obtain it from the following equation:

(20)

This yields the air temperature rise inside the channel

(21)

solving for To

Substituting into Equation (19) forT m, we get

(23)

Substitute equation 23 into 18 forT m

Q = hA[Tb- T;- Q/2mCp] (24)

Solving for the board temperature:

(25)

Solving for the board temperature rise above the channel inlet:
Limits of Air Cooling-A Methodical Approach 221

(26)

The mass flow rate, m, is equal top VA, where Vis the air velocity. Equations
18 and 26 are applicable to natural and forced convection, and both equations
display strong dependency on the air velocity. Hence, effective determination
of the temperature rise requires calculation of the air velocity as shown in the
section on Fluid Flow in Circuit Packs. In addition, Equation 26 also requires
a value for h where it is obtained from a number of correlations in the litera-
ture [Azar, 1994].

HEAT TRANSFER FROM A COMPONENT

In this section, we will develop a first-order expression showing the heat
transfer from the component. This process will yield an expression for the junc-
tion temperature as a function of parameters impacting it. Consider Figure 5
where a half of a component is shown. We will apply a control volume to dif-
ferent segments of the component and apply conservation of energy to each
control volume. The component is a plastic molded package that is mounted
on a printed wiring board-a case most commonly encountered.
We start with the molding that houses the die, wire bonds, and the chip-
carrier. Apply a control volume around the molding as shown in Figure 6.
The heat transfer leaving the control volume is characterized by Q1, Q2, and
Qc. Q 1 and Q 2 are heat transfer from the top and bottom of the molding, re-
spectively, and are composed of radiation (r) and convection (h). Note that Q2
can also include the conduction heat transfer through the air gap. The impor-
tant point is that there is heat transfer from the bottom side of the component.
Qc signifies the heat conduction through the lead.

(27)

Applying the conservation of energy to the control volume and assuming steady-
state condition, we get the following:

A Tc

Tj

Tb

FIGURE 5. Half of a component with plastic molding.

222 Air Cooling Technology

FIGURE 6. Energy balance on the molding part.

Qin + Qdissipation = Qout (28)

Since there is no heat entering the control volume, the conservation equation
becomes

Qdissipation = q n = Qout (29)

where q" = Q1 + Q2 + Qc solving for Qc

Solve for Tc a reference temperature on the component

(31)

The value of (kAIL)etr can be obtained and is subject to further discussion

[Azar, 1994]. For the purpose of this discussion we should note that the phys-
ical geometry plays a significant role in the thermal performance. Performing
an energy balance on the lead:
Applying conservation of energy to the lead, Figure 7, Equation 17 reduces to

(32)

The heat transfer into the control volume is by conduction; Qc and leaving the
control volume is by convection from lead surface (Qh.L) and the board (Qh,b).
Hence, the above equation reduces to

(33)

where Qh,b is the convection heat transfer from top and bottom of the board and
is equal to conduction heat transfer in the board
(34)
Limits of Air Cooling-A Methodical Approach 223

- ~-

Qh,b
FIGURE 7. Energy balance on the lead.

Here, kb is the board conductivity and AT is the temperature difference

between Tb and another isotherm typically at the centerline between two adja-
cent components. A and L refer to the cross-sectional area and the length of the
board associated with the conduction heat transfer, respectively. Hence, kb, an
important parameter, is implicitly introduced in Qh,b· Estimating lead surface
temperature as

(35)

then, solving for Qc in terms of temperatures

Solving forT b

Solving for Tj, assuming h's are equal (if we have the value for h around com-
ponent, we do not need to make this assumption)

Tj = Tc + RenfTb(hAL /2 + 2hAb) + hALTJ2

- h(AL +Ab)Tm -hAbTm.bl (38)

The above expression requires Tc and Tb to be known. Considering that leads

tend to be highly conductive in nature, one would not expect a large temperature
drop across the lead. Hence, assuming Tc to be equal to Tb will not significantly
jeopardize the accuracy of the solution, thus eliminating one variable.
Equation 23 requires T m.b. which is the mean air temperature of the channel
formed by boards 3 and 4 in Figure 4.

(39)
224 Air Cooling Technology

Substituting the equivalent of flow rate yields

(40)

Substituting Equation 40 into equation 38, and setting Tc = Tb, we get an ex-
pression for Tj as a function of parameters that are readily measurable.

Ti = Tb + Ren {Tb(hAL /2 + 2hAb) +hAL Tb /2

- h (AL + Ab)Tm- hAb (Q4/(2pV3.4A3,4Cp)) + T;) (41)

Substitution of the values ofT mand T m,b in terms of the flow rate and power
dissipation results in the following:

Ti - T; = (Tb - T;) + RerrfhAdTb - T;)- hAL Q 3 / m 2,3CP (42)

+ 2hAb I Cp[(Q3/mh,3 + (Q4/ mh,4]}

Equation 41 or 42 represents the junction temperature of a component re-

siding on a board. This model is for the case when the power dissipation of the
component being modeled is higher than its neighbors. The model explicitly
does not include heat transfer by conduction through the board from the adja-
cent components. If this is not the case, the model shown here can be expanded
to include the conduction heat transfer between the component and its neigh-
bor. This is done at the lead level, Figure 7, where additional heat transfer en-
ters the control volume at the board.
The reader also can appreciate the sensitivity of the model to Rerr-eompo-
nent internal thermal resistance. An accurate estimation of this parameter can
improve in the level of accuracy in Tj.
Equation 41 can be used irrespective of the mode of convection heat trans-
fer. The junction temperature can be obtained if the value of heat transfer co-
efficient (h) is known. The reader may note that the process for obtaining the
junction temperature is indeed iterative. Solution of a set of Equations 7, 23,
25 and 42 is required in order to obtain Tj.

SIGNIFICANT PARAMETERS
The junction temperature, Equation 41, is the parameter of utmost interest
to thermal engineers in electronics cooling. We would like to know what the
parameters are that have the most impact on the value of Ti- We developed a
series of equations that relate junction temperature to other parameters affect-
ing it Our objective in thermal management is to ensure that Tj is as small as
possible within given constraints. By examining equation 41, maybe we can
shed some light on the important parameters and gain some insight as to how
we can reduce Ti·
We can reformulate equation 36 by the following:

(43)
Limits of Air Cooling-A Methodical Approach 225

Or expressed in general terms:

Tj = f (geometry, properties, velocity) (44)

Equation 27 effectively shows the parameters impacting Tj. For Tj to be as

small as possible (our goal in thermal management), the terms in the braces
must be minimized.
The parameter that has a dominant presence in Equation 41 is the air ve-
locity. This is seen in the convection terms associated with the lead surface and
board areas. Heat transfer coefficient (h) is directly proportional to velocity and
to change of temperature (density) in forced and natural convection systems,
respectively. This is somewhat intuitive and obvious since the ultimate sink for
heat is the air stream. It is interesting to see that convection heat transfer from
the head plays such a role in the magnitude of Tj. This is often overlooked in
modeling and design of electronic components and circuit packs.
In addition, we see the presence ofTi-inlet temperature to the channel. This
stems directly from our earlier discussion, Section 3, that in thermal analysis
system down approach is the only way to obtain an accurate solution. This is
evident in Equation 41, since the value for Ti can be obtained by system level
analysis or direct measurement
We also see the impact of neighboring boards in the value of Tj. The adja-
cent boards impact the mean air temperature (T m) in the channels (front and
back) subsequently affecting heat transfer from the board. Further, the adjacent
board geometry (layout) also impacts heat transfer. This is observed in the hAb
(Q,J(2p VJ.4A3.4Cp) term of Equation 41 by the presence of velocity in the de-
nominator. The resistance to flow, stemming from component layout, will di-
rectly impact flow inside the channel, Equations 7-9.
The role that a board plays in thermal performance of the component can-
not be emphasized enough. Boards with higher thermal conductivity Equation
34 can be instrumental in removing heat from the component. Higher board
conductivity can increase Ab, resulting in higher convection heat transfer from
the board. Again, this is demonstrated by equation 41, even though board ther-
mal conductivity does not directly appear in the equation.
Irrespective of mode of convection heat transfer, I mentioned the important
role velocity plays in the magnitude ofTj. An often overlooked point in circuit
board design is the board layout. Board layout plays a pivotal role in compo-
nent flow exposure. Poorly laid-out boards can create zones of stagnation that
hamper convection heat transfer from the board and component. Azar and
Russell ( 1991) provide many flow visualizations of such effects. This can also
be deduced from Equations 7-ll. The pressure drop is the direct function of
the expansions and contractions created by the components on the board, hence,
affecting air velocity and flow distribution in the rack. The pressure drop in the
system can be reduced by proper placement of components on the board. This
is done such that the critical components and/or boards have the least resistance
to air flow.
226 Air Cooling Technology

POTENTIAL LIMITS OF NATURAL

AND FORCED CONVECTION COOLING

The word limit conveys a message that there is a number associated with
power dissipation beyond which cooling with natural and/or forced convec-
tion is not possible. In the foregoing discussions I have attempted to replace
this notion with the need to look at each case explicitly. I also have advanced
the argument of reshaping our thought by looking at the junction temperature
limit instead of heat transfer when dealing with the issue of limit. For exam-
ple, a common question we as thermal engineers face is "If this board will be
dissipating x Watts, can I cool it with natural convection or do I have to use a
fan?" It is convenient to have a number, say 25W per board, to compare against
and respond accordingly. If we look at the question more closely, the issue is
really the junction temperature limit and not the total power dissipation.
Further, as I indicated in the previous section, there are numerous parameters
that can impact this evaluation, e.g., channel dimensions, board material, com-
ponent layout, system configuration, environment, etc. Hence, we need to look
at the concept of limit from temperature perspective instead of the heat flux
number that may not be applicable and is a function of several system-
dependent parameters.
I need to side step and add that by no means am I suggesting that we cool
all systems with natural and forced convection. The consideration for a cool-
ing system is driven not only by physical and thermal limits but by market lim-
its as well. A case in point is the implementation of jet impingement for cooling
of the processors in a PC. The fan noise is difticult enough to bear. Imagine
contending with a whistling noise coming out of your PC if it contained jet im-
pingement. Likewise, a consumer electronics product, e.g., CD player, has its
market-defined limits or a super computer, e.g., Cray, cannot have a jet turbine
to provide enough flow for air cooling. Hence, in dealing with the issue of lim-
its, other parameters in addition to junction temperature should be considered.
To address the issue of limit, I will review the work available in the litera-
ture and then focus on the temperature limit versus heat flux limit that I have
advocated here. Because of the diversity in electronic systems and the above
discussion on system and application dependency, I have purposefully avoided
generating numbers that can stand as a limit for comparison. However, the tools
for calculating the potential limits for your specific problem are provided that
readers can use for their analysis.

FORCED CONVECTION LIMIT ANALYSIS

Based on the issues we raised, the literature is very limited on convection
limits. The works, by Hannemann ( 1990), Jacobs ( 1989, 1990), and (Chung,
1987) are noteworthy for review.
Hannemann looked at the limits of forced convection as a function of num-
ber of chips per unit area versus number of pins per chip for the mini- and
Limits of Air Cooling-A Methodical Approach 227

microcomputer. The analysis was done for the case when the chips were at-
tached to a heat sink as shown in Figure 8.
The chip temperature rise was described by

(45)

Considering that the convective resistance is the dominant factor, the above
equation reduces to

AT = ~ (46)

The number of chips per unit area is given by

N ! U =l!Ab (47)

substituting into Equation 27, we obtain the following:

AT = qAc
hyAb (4 8)

FIGURE 8. Extended surface chip cooli ng. (Hanneman, 1990].

228 Air Cooling Technology

where

y =A I Ab = 1 + 2y Is, (49)

and
(50)

Extracting the chip area that may be cooled with an extended surface area,
Ab and chip heat flux q is given by
h!J.TA
A c -- hyb.TA b I q -- _q_b (1 + 6.39)
zo.s (51)
f
where the heat transfer coefficient is given by

(52)

Since Fr = Ab and combining Equations 50 and 51, we get

(53)

If we set the temperature rise to 50 K, and vary the heat flux to 10 and 20 W/em"
we get Figures 9 and 10
The value of the heat flux, length of the heat sink, y, and velocity can be
changed to develop different curves. The figures show the maximum thermally
allowed chip density at a given pin count. The model was generated under the
assumption that the internal thermal resistance is negligible and the convective
resistance is the only one. Therefore, it advances the best scenario since the
current technology is limited and does not allow for this situation to exist
However, the data and methodology are helpful in creating potential bounds on
the problem.
There are a couple of salient points that merit mentioning. Generalization
of problem-specific issues can create false information-alluding to the point
of generating a "number" as yard stick for determining a limit As Hannemann
carefully pointed out, the model developed is for a very specific application
dealing with chips placed on an extended surface. The second point is the de-
pendency of the model on physical parameters and heat transfer coefficient (ve-
locity). Both of these parameters can significantly alter the results if their values
are changed as shown by Equations 51-53, further substantiating the need to
look at each case individually.
Jacobs (1989, 1990) advances the application of Reynolds' analogy in elec-
tronics cooling. He suggests the use of the analogy both for forced and natural
convection cases to determine the limit. Reynolds' analogy is widely used in
the heat exchanger industry and Jacob's method is an interesting approach in
this concept for electronics equipment.
Consider the forced convection case where the equipment is constrained by
the acoustic noise limit and a maximum component-to-air temperature rise-
Limits of Air Cooling-A Methodical Approach 229

2.0

1.0
u
1\~0.64
0.4 q .. 10 W/eml
.:"' (y .. 2 em: v "' 5 mfs) \
~ \
~ \
0.2

-"'
<(
w \
a:
<(
\
\
Q. 0.1
:f
(.)
0.08

0.04

-PERIPHERAL
---AREA ARRAY

FIGURE 9. Thermal limits- q =l 0 W/em' [Hanneman, 1990].

a situation commonly encountered. Figure ll shows a general schematic of a

system that contains fan, filter and the equipment.
The noise generated in the system is proportional to the volumetric flow rate
(R) and generated pressure (P)

R P2 =A (54)

Based on Equation 4, pressure drop as a result of physical constraints is a

function of the flow resistance (D) and the volumetric flow rate. Reynolds anal-
ogy simply states that the drag force created as the result of fluid passing a
heated rigid body is proportional to its heat transfer. This is expressed by the
friction coefficient as a function of other nondimensional numbers:

Cr = 2Nu/(Re Pr 113 ) (55)

If the top area of the shelf shown in Figure 11 is Arop, the volumetric flow rate
is defined by

R = V A,op (56)
230 Air Cooling Technology

r-------------~r-----~~--r------------,
2.0 \

O.l:J \
' \

q•lOW!cm3 \
1.0 (y-2 Clll; "'"'' mimi
IU
\
\
\
0.4
'
~
\
\
\
\
0.2
~ \
'
~
''
-
\
~
~ 0.1 \ 0.64
\
::c 0.08 \
u \
\
''
'
0.04
''
'
(1.112

10

DEVICE PINCOUNT

FIGURE 10. Thennallimits- q = 20W/cm', [Hanneman, 1990],

and air flowing in a circuit pack channel at a velocity V and surface area,
the pressure drop created is

(57)

where n is the number of circuit cards in the rack. The pressure drop can be ex-
pressed in terms of internal and external resistances, which is given by the fol-
lowing equation:
(58)

Defining f3 as P/R2 we get

(59)

The heat transfer from two sides of a board is

(60)

and the heat transfer coefficient is deduced from Equations 55 and 60:
Limits of Air Cooling-A Methodical Approach 231

Circuit
Pack
Carrier

Filter

Fans

FIGURE ll. Schematic of a shelf with filter and fan [Jacobs, 1989].

h = ~J:C t RkPr 113 I 2VA lop (61)

where~= 0.7 is the calibration factor to fit the Reynolds analogy to the circuit
pack problem. Solving for R from Equations 54 and 58 results in

R- [ A J
l/5

(62)
- (P+ Df

By successive substitution of equation 62 into 61 to eliminate Rand equa-

tion 61 into 60, we can obtain an expression for maximum power dissipation:

(63)

In equation 63, <I> corresponds to the percentage of the area of the circuit
board that is heated. Jacobs suggests a value equaling 0.02 for the friction co-
efficient (Cr). This result can also be obtained graphically if one plots system
flow resistance and acoustic noise on the pressure vs. flow rate coordinate
systems. The point of intersection of these two curves correspond to the max-
imum power dissipation.
232 Air Cooling Technology

The derivation for natural convection is similar to the above. The pressure
drop created by air flow across a rack of cards was given by Equation 57 and
the heat transfer from the cards by Equation 60. Note that Equation 60 can be
corrected by a factor of <P introduced earlier. Further manipulation of equation
60 by using pressure drop and friction coefficient yields
t:(Pr)lt3 A 2 ]
Q = PI R [ !J.T <Pk"' top (64)
3a vp

The air temperature rise through a rack of cards is given

l'il'u;r = RgCp : :; ATmax (65)

by substituting Equation 65 into Equation 64, maximum dissipation in watts is

given by
$f!k(PR) 1' 3 ] 112
Q =A top [ PCp AT.max ATVsa':>
(66)

and the associated volumetric flow rate (R in m'/s) is given by Equation 67.

R= Atop[P&T..a<P~k(Pr) 113 ] 112

P VCP!J.Tmax (66 )
Equations 66 and 67 require that the maximum temperature rise be speci-
fied and the pressure drop be given. An interesting point of Equation 67 is its
independence from the number of circuit packs in the rack. At a first glance,
one may think that the equations are independent from the channel height, i.e.,
circuit pack height This parameter, an essential element in natural convection
flows, appears inherently in the expression for pressure.
Again, there are a few points that merit a discussion. Equations 63 and 66
are based on Reynolds analogy whose foundation has constraints. The analogy,
although useful, is for cases when the Pr = 1, flow is laminar and the walls are
smooth. The issue of the Prandtl number can be argued successfully, since for
most air-cooled applications, Pr = 0.72. However, the other constraints may
create some doubt in the utility of this approach or the equations.
Smooth walls and laminar flow are not common occurrences in most elec-
tronic equipment. Then the question of utility of these equations needs to be
answered. The response can be formulated in several envelops. Reynolds' anal-
ogy has been successfully applied to many heat exchanger designs that may
have not agreed with the constraints. Hence, the rigidity of the constraint is
questionable, since many successful applications are observed and the analy-
sis was in satisfactory agreement with the data of Bird ( 1960), Bejan ( 1984 ),
and Cheremisioff (1984). The literature also suggests that a more suitable re-
lation for geometry that is more representative of channels formed by electronic
circuit packs is Colburn analogy:

StPr 213 = C 1 /2 (68)

Limits of Air Cooling-A Methodical Approach 233

The derivation of maximum dissipation based on Equation 68 is identical to

that shown so far.
The Reynolds analogy as shown here has a strong utility in the design process.
The outlined methodology can certainly be used for establishing bounds on the
problem before proceeding with the detailed analysis. It also provides some
guidelines to the direction of the design. Hence, its usefulness, despite its lim-
itation, should not be underestimated by the reader.
Similar to what we saw with Hannemann's approach, although methodol-
ogy suggested by Jacobs appear to be more general in nature, it still requires
system-specific data. This again highlights my earlier argument of avoiding a
yard stick type of approach for gauging air cooling limits. System configura-
tion and operation requirements are as diverse as the designers who make them,
and these are the limits that a designer needs to consider in evaluating air cool-
ing. This discussion may become more clear if we look at natural convection
more closely.

NATURAL CONVECTION OPTIMIZATION

In attempting to maximize heat transfer from the board in a natural convec-
tion cooled system, Chung ( 1987) developed a series of equations based on
available heat transfer coefficient correlation for the following conditions:

1. Symmetric and asymmetric isothermal boards

2. Symmetric and asymmetric isoflux boards

The intent was to obtain the maximum value for the channel spacing (board
to board) beyond which air temperature rise will remain unchanged.
The convective heat transfer is given by the following:

Q = hAtotal AT= [Nu (k/b)] [2LSn]AT (69)

where A,01a1 is the total surface area of the boards and n = W /(b +d). The other
parameters are defined in Figure 12.
Table 3 shows the optimum spacing the boards can have for the given tem-
perature or heat flux condition.
In table 3

(70)
and

(71)
In Equations 70 and 71, L is the height of the circuit board, q' is the heat
flux, and A. = 6.17 x 10-4 + 1.57 x 10_, T 1• The results of the above are shown
in Figures 13 and 14.
234 Air Cooling Technology

Typical PCB arrangements

• Rear cover
Electronic equipment typically
consists of vertically arranged circuit
boards that form an array of parallel
channels. Cooling is most often by
free convection. Optimum spacing
provides maximum heat transfer rate
and minimum package size.

FIGURE 12. Typical PCB arrangement [Chung. 1987]. Reprinted with permission from Machine
Design. March 1987, a Penton Publication.

TABLE3
Board-to-Board Optimum Spacing
for Natural Convection Cooled System

Condition Optimum Spacing

Symmetrically isothermal 2b + 3d --0.005P' 'b' =0
Asymmetrically isothermal 2b + 3d --0.005P' 'b' = 0
Symmetrically isoflux b + 3d --0.3133R""b" =0
Asymmetrically isoflux b + 3d --0.626R""b" = 0

The analysis presented by Chung shows the effect of heating and channel
width on thermal response of the system. The analysis reveals the optimum
board spacing for maximum heat transfer from the board. However, to further
highlight these effects consider the work of Saxena.
Saxena ( 1981) highlighted the impact of other parameters that influence
board or air temperature rise in a natural convection cooled system. He con-
 Limits of Air Cooling-A Methodical Approach 235

Typical design curves for symmetrically

isoflux plates
10

...
~ftw! (W/11\.a)
"0.@1
~~-~­
T. :~~~ :WOC

-i
;;
•
·~~:::
•0.01
<~0.03
~>0.1»11
1..=1111\.
1:111\.
d = 0.3111\.
n=l
i -~
17~ . . .

i
!
~ 30

J:1 20

10

00.1

FIGURE 13. Maximum temperature rise vs. channel width for a isoflux system cooled by nat-
ural convection [Chung, 1987]. Reprinted with permission from Machine Design, March 1987, a
Penton Publication.

sidered channel height, board spacing, and circuit card power dissipation in a
channel. Each circuit pack was 20 em high by 34 em deep. The channel height
was increased in increments of 20 em by stacking the boards above each other.
Figures 15 through 18 show the impact of these parameters.
These figures are self-explanatory and require very little discussion. The
reader should note the variations the data show when parameters such as board
spacing or power dissipation are varied. The impact of channel height shown
in Figures 15-18 also are noteworthy.
In our earlier discussions, I raised the issue of component placement and its
impact in overall cooling or component response. Note the work shown by Saxena
and Chung does not include the impact of component placement. Another para-
meter to consider was work reported by Lee (1994). He considered the natural
convection heat transfer for an array of parallel plates with unheated entry and exit.
One can interpret that as component placement--do you place the hot compo-
nents near the entrance or exit of the channel? Since the study was of a numerical
nature and the data reported are nondimensional numbers the discussion may be-
come lengthy, so I will only reflect on his observations. Lee concluded that:

I. Unheated exit has a higher fluid draw resulting in a higher heat transfer
coefficient.
236 Air Cooling Technology

Typical configuration curves for asymmetry

versus symmetry
~~-
1'.,=~
L = 1'11n.

....
I• SIR.
d•~IR.

1'~ Md !ill 111.1.,

Noi~II~~--­
Sywne~~--

0.5 1. 'I

FIGURE 14. Effect of symmetry and nonsymmetry on thermal response of natural convection
cooled system [Chung, 1987]. Reprinted with permission from Machine Design, March 1987, a
Penton Publication.

10 WATTS. CKT PACK

e· • I 3·· CKT PACKS ON '•·iNCH CENTERS
0 2 .. COMPONENT FEIGHT

0 8 12 16 20 24 28 32
CHANNEL HEIGHT, INCHES

FIGURE 15. Maximum board surface temperature rise vs. channel height for 1.3 em board spac-
ing at different power dissipation [Saxena, 1981 ].
Limits of Air Cooling-A Methodical Approach 237

50 8" • 13''CKTPACKSON liNCH CENTERS

0.2" COMPONENT HEIGHT

IQ

0 a 12 16 20 2-1 28 32 Ju ~o
CHANNEL HEIGHT INCHES

FIGURE 16. Maximum board surface temperature rise vs. channel height for 2.5 em board spac-
ing at different power dissipation [Saxena, 198 IJ.

50 CHANNEL HEIGHT .. 8 IO<CHES

.
u

..."'
.Q
40
<l
w
(/)
ii!
Q. 15 WATTS CKT PACK
:::i 30
w
w
u
....a:< 10
:l
C/l 20
0
a:
<
0
!¥!
....
::.::
(,) 10
><
<
:::i

0 0.5 1 0 15 2.0
CIRCUIT r>A.CK CENTEA-TD-CENTEA SPACING UNCHESI

FIGURE 17. Effect of circuit pack spacing on board surface temperature rise for 20 em chan-
nel height at different power dissipation [Saxena, 1981].
238 Air Cooling Technology

u Hoell • a.-eM CICT ~ACICS ON J...CIII CaNftRI

·-

<I
.
\'
...
30 X

e
•
• 14.&-CM CA-OX Ill CARD t.A-TH AWAY ~ROM ONtN ~RONT)

11 WATTS Nta
I I WATTS Nta
CIACUIT
CIACUIT
~ACIC
~ACIC
• 10 WATTS Nta CIACUIT ~ACK
Ill e 8 WATTS Nta ClaCUIT ~ACK

c
VI

Ill 20
IE
!)

~
IE
It
2
10

c=
a:
....
Ill
zz
c
:z:
u 0 20 10 10 100

CHANNEL HEIGHT, V- (CM)

FIGURE 18. Circuit pack channel air temperature rise as a function of channel heights at dif-
ferent power dissipation levels [Saxena, 1981].

2. For uniform heat flux, unheated entry will result in higher board tem-
perature than the unheated exit.
3. Effects of unheated entry or exit on heat transfer characteristics are sig-
nificant, especially for the case of uniform heat flux and unheated exit.

Although the actual data may not be of immediate application in the design,
the observations are definitely noteworthy. They clearly suggest that compo-
nent placement can play a significant role in the thermal response of the chan-
nel. Although Lee dealt with flat plates, I used the word thermal to highlight
the heat transfer and fluid flow effects that component placement has on heat
transfer.
Let's revisit why we looked at natural convection. My intention was to raise
the flag on the use of the yard stick type of concept when we deal with elec-
tronics cooling problems-especially with concept of limit. The data by
Saxena and Chung, and others not cited here, clearly show how these para-
meters can influence, say, air temperature rise in the channel, board tempera-
ture rise, or heat dissipation. We should also note that the data shown here are
very much system specific, and the effect of system configuration and its in-
teraction with the environment is inherent in the data-adding another uncer-
tainty to the problem. As a side note, the data are very useful for bounding the
solution as we did with other works shown here. The work of Lee further cor-
roborates the inadequacies of using a yard stick type of concept. Hence, once
confronted with this situation, my recommendation is the grounds-up ap-
proach-integral approach.
Limits of Air Cooling-A Methodical Approach 239

METHODICAL PROCEDURE-AN INTEGRAL APPROACH

In the last two sections, I showed the difficulty and perhaps danger of es-
tablishing a limit for measuring the capability of cooling with convection heat
transfer. Data, if available, that is based on a specific system may have been
developed for another industry segment. This data may not have universal ap-
plicability and will not be suitable to telecommunications or consumer elec-
tronics. Furthermore, based on experience, we as designers are typically
constrained by the temperature limit and not by how much heat a board can
dissipate. The thermal engineer typically gets the design, if lucky, when the de-
sign is established and an estimate of its power is available. One never sees the
reverse-that you can only dissipate x-Watts on a board-so let's design the
electronics that meet the power limit (wouldn't that be a luxury). Regardless,
the question "whether I can cool a given system by forced or natural convec-
tion" still lingers. Because of the diversity in design and system application, I
suggest a ground-up approach that I call-integral. This approach simply states
that we should:

Consider each problem individually

Develop a model for the junction (or critical point, e.g., lead) tempera-
ture of the components
.. Determine whether it meets the temperature criterion

The reader can appreciate that for a given class of systems, e.g., PCs or
telecommunication equipment, limits that can be used as a guideline may
evolve. Nevertheless, the need to look at each problem individually, even in
a single industry category, becomes a necessity-as some of us in the field
have experienced.
How does the integral approach work? In the sections on .... I proposed a
model for fluid flow and heat transfer in the circuit pack. The model resulted
in an expression for junction temperature as a function of parameters impact-
ing it. The intent is to look at the junction temperature of the component(s) or
the temperature rise inside the channel to see whether it satisfies the design cri-
teria. This was done by the following equation, as was discussed in the section
on heat transfer from a component

Tj - Ti = (Tb - Ti) + Rerr {hAL (Tb - Ti)- hAL Q3/m2.JCp (70)

+ 2hAb/CP [(Q3/mh.3 + (Q4/m)3.4l}
For example, many power supplies that reside on the boards have an ambi-
ent temperature limit of approximately 75°C (50°C less than the l25°C for the
ICs). Hence, the designer needs to examine what the power supply's ambient
temperature might be for the worst environmental ambient-typically 49°C,
the sections on .... Then a model, such as the one suggested above, is used to
determine the "junction" temperature (typically a location on the side of the
240 Air Cooling Technology

device in the case of the power supply). The junction temperature rise above
ambient, obtained from the analysis, is compared with the imposed tempera-
ture limit, T], Equation 11. The values of heat transfer coefficient set the cool-
ing mode-natural or forced convection. If T] is satisfied then the cooling mode
selected can be implemented. If the forced convection results based on this
analysis do not satisfy T] then higher cooling modes must be considered or the
system needs to be redesigned.
As a side note, the author is not suggesting that the equations shown in
this chapter are the sole model for these studies. The emphasis is on the ap-
proach rather than the equations. Nevertheless, the equations shown here are
based on the laws of physics and is void of heuristic knowledge or system-
specific parameters. The reader can tailor these equations to his/her specific
problem without jeopardizing the accuracy of the results or the integrity of
the approach.
In these analyses, the design temperature limits play a pivotal role. These
are temperature limits that are set by standardization bodies (Bellcore or
UL), market specifications, customer requirement, or any such entity. For
example, in the telecommunication equipment that are used in outside en-
closed cabinets, the air temperature rise in the circuit pack channel cannot
exceed l0°C, or the junction temperature of the ICs cannot exceed l25°C.
In most military applications, this limit is l00°C. For optical components,
the case temperature is the gauge and is set at, typically, 85°C. All these lim-
its are for the worst ambient of 49°C. As a point to note: The highest air tem-
perature recorded on Earth was ll5°F, at the city of Abadan, Iran. Hence,
limit or cooling system qualification by the integral method requires tem-
perature limit specifications.
Combinations of modeling and design temperature limits can provide us
with the answers to the question of whether natural or forced convection by air
can be used for cooling. We should also note that the modeling technique sug-
gested here can also address the use of extended surfaces (heat sinks) typically
used to reduce the junction temperature. The model allows for parameters such
as heat transfer coefficient, areas or Q1 and Q2, section 5.2, to be altered such
that they include either the heat sink or the effect of the heat sink. A typical ex-
ample is to upgrade the value of heat transfer coefficient such that the effect of
heat sink is included. Hence, the suitability of the cooling system can be de-
termined by the use of a set of algebraic equations without the need to adhere
to any unsubstantiated yard sticks.

CONCLUSION

The desire to cool with air continues to persist because of readily available
and its relative ease of use, e.g., natural or forced convection. Hence, engineers
in the electronics industry are continuously challenged with the question of
whether a system can be cooled by the simplest mode of air cooling-natural
Limits of Air Cooling-A Methodical Approach 241

convection. Therefore, the desire for a quick and readily available answer to
this question continues to persist. The answer to this question is often sought
by searching the literature for similar work.
Seeking a solution from the literature may be fruitless. When we take a cur-
sory look at the electronic products in the market it clearly reminds the reader
of the product diversity in the electronic industry. This diversity has resulted in
many excellent thermal works that are industry or product specific, thus, diffi-
cult, if not impossible, to generalize. This difficulty stems from the design and
application variations that electronics products face. The use of these types of
information without understanding its domain of development and application
can result in major flaws with a costly associated fix.
Consequently, in this chapter, a simple methodology based on conservation
laws are advocated, and the use of yardsticks or presolved solutions is strongly
discouraged. The methodology suggests using the integral approach to calcu-
late the junction temperature and then use this temperature as the parameter to
judge design integrity versus "heat flux in a given volume," which is typically
used. The methodology emphasizes the need to obtain and understand the tem-
perature constraints that govern one's problem. These constraints can be im-
posed by standardization agencies or the end user. This approach allows the
engineer to analyze the problem based on its own merits and, hence, is void of
any assumptions that may not be applicable to the specific problem at hand.
This process empowers the engineer to design based on the governing physics
and eliminates the need to use vague industry-specific yardsticks or literature-
based data that may not be applicable to the problem at hand.

APPENDIX A: NOMENCLATURE

A- area
Ab- component thermal foot print, twice component planar area
Ac chip area
A, surface area parallel to the stream
Acr open flow area at the channel entrance
Cf friction factor
Cp specific heat at constant pressure
D drag force
Fhc view factor
g gravity
H height
h heat transfer coefficient
k thermal conductivity
K loss factor
L- half length of the component (due to symmetry)
m mass tlow rate
n number of circuit packs
242 Air Cooling Technology

Nu Nusselt number
P pressure
Pr Prandtl number
q"- component power dissipation
Q heat transfer rate
Qc- conduction heat transfer through the lead
Qh.L- convection heat transfer from the lead
Qh.b- convection heat transfer from the board
R volumetric flow rate
Re Reynolds number
St Stanton number
T temperature
Ti- junction temperature
T,- lead temperature
Tamb- component ambient temperature
Tb- board or base temperature
Tc- component reference surface temperature
Tr fluid temperature
Tm mean temperature
T m,b- mean air temperature on non-component side
T, lead surface temperature
T,,L- lead temperature
V velocity

13 coefficient of thermal expansion

8 boundary layer thickness
£ emissivity
~ viscosity
'tw shear stress
p fluid density
cr Boltzman constant
'P convection constants defined by [Hannemann, 1990]
A fan acoustic noise limit

SUBSCRIPTS
amb ambient
b board
c cold
h hot
m inlet
L lead
out outlet
sa surface to ambient
top top open area of the shelf
Limits of Air Cooling-A Methodical Approach 243

REFERENCES

Azar, K. 1992. Thermal design considerations with applications to multichip modules, in Multi chip
Module Technology and Alternatives-The Basic Approach, Van Nostrand Reinhold, New
York, chap. 12.
Azar. K. 1994. Electronics Cooling-Theory and Application, short course. Class Notes.
Azar, K., Mcleod, R.S., and Caron, R.E. 1992. Narrow channel heat sink for cooling of high pow-
eredelectronic components, IEEE Semiconductor Temp. and Thermal Manage. Symp.,Austin,
TX.
Azar, K., Pan, S.S., Parry, J., and Rosten, H. 1994. Effect of circuit pack parameters on thermal
performance of electronic components in natural convection cooling, in Proc. lOth Annu.
IEEE SEMI-THERM Symp., San Jose, CA.
Azar, K. and Russell, E.T. 1991. Effect of component layout and geometry on the flow distribu-
tion in electronic circuit packs, ASME J. Electron. Packag., 114: 50-57, March.
Bejan, A. 1984. Convection Heat Transfer, John Wiley & Sons, New York.
Bird, R.B., Stewart, W.E., and Lightfoot, E.N. 1960. Transport Phenomenon, John Wiley & Sons,
New York.
Cheremisioff, N.P. 1984. Heat Transfer Handbook, Gulf Publishing.
Chung, J. 1987. Maximizing heat transfer from PCBs, Machine Design, 87-92, March 26.
Hannemann, R. 1990. Physical technology and the air cooling interconnection limits for mini- and
microcomputers, Advances for Thermal Modeling of Electronic Components and Systems,
ASME Press, 1-40.
Hilbert, C., Sommerfelt, S., Gupta, 0., and Derrell, D.J. 1990. High performance micro-channel
air cooling, in Proc. 6th Annu IEEE SEMI- THERM Symp., l 08-113, Scottsdale, AZ.
lncropera, F.P. and Dewitt, D.P. 1990. Introduction to Heat Transfer, Wiley and Sons, New York.
Jacobs, M.E. 1989. The practical limits of forced air-cooling of electronic equipment, Applied
Power Electron. Conf
Jacobs, M.E. 1990. Application of reynolds analogy to the estimation of the limits of cooling by
natural convection, Applied Power Electron. Conf, 431-437.
Kays, W.M. and Crawford, M.E. 1980. Convective heat and mass transfer, 2nd ed., McGraw Hill,
New York, 139.
Klinger, DJ. Nakada, Y., and Menendez, M.A. 1990. AT&T Reliability Handbook, Van Nostrand
Reinhold, New York.
Kraus, A.D. and Bar-Cohen, A. 1983. Thermal Analysis and Control of Electronic Equipment,
McGraw-Hill, New York.
Lee, K.T. 1994. Natural convection in vertical parallel plates with an unheated entry or unheated
exit. J. Numerical Heat Transfer, Part A, 25:477-493.
Manno, V.P. and Azar, K. 1992. The effect of neighboring components on thermal perfortnance of
air-cooled circuit packs, ASME 1. of Electron. Packag., 113:50-57, March.
Murray, J.C. 1990. A Thertnal Model of Parallel Circuit Board In Electronic Enclosures, Master's
thesis, Tufts University, Boston, MA.
Network Equipment Building System, TR-NWT-000063, Sept. 1993.
Okutani, K., et al. 1984. Packaging design of SiC ceramic multi-chip RAM module, in Pmc. of
Inter. Electron. Packag. Soc., 299-304.
Saxena, L.S. 1981. Electronic equipment frames in free convection, in Proc. 2nd Bell Labs Thermal
Design Forum.
Simons, R.E. 1983. Thermal management of electronic packages, Solid State Techno/., Oct.
Sparrow, E.M. Neithammer, J.E., and Chaboki, A. 1982. Heat transfer and pressure drop charac-
teristics of arrays of rectangular modules encountered in electronic equipment, Inter. 1. of
Heat and Mass Transfer, 25 (7):961-973.
Sridar, S., et al. 1990. Heat transfer behavior including thertnal wake effects in forced air cooling
of rectangular blocks, ASME HTD., WAM, (Dallas TX). 153:15-26.
244 Air Cooling Technology

White, F.M. 1984. Heat Transfer. Addison-Wesley, New York.

Wirtz, R.A. and McAuliffe, W. 1989. Experimental modeling of convective downstream from an
electronic package row, ASME J. of Electron. Packag .. II, 207-212.
Wirtz, R.A. and Dykshoom, P. 1984. Heat transfer from arrays of tlat packs in a channel tlow, in
Pmc. Ins/. of Electron. Packag. Symp., 318-326.
 INDEX
A finite difference/control volume method,
125-132
Acoustic noise, forced convection limits, Channelheruflux,218
228-231 Channel height, maximum board surface
Active destabilization techniques, 188 temperature rise vs., 236, 237
Adiabatic board solution, conjugate heat Channel width, maximum temperature
transfer, 131 rise vs., 235
Adiabatic heat transfer coefficient, h•• 75, Chip height
85-92 Reynolds number based on, 176
Air flow rate, heat flow from component, vortex generators of multiples of, 184
224 Circuit pack problem, 231
Air velocity Circuit packs, 213
calculation of, 208-210 fluid flow in, 208-210
in channels, 221 heat transfer in, 211
Array flow, 83, 84 heat transfer in channels, 219
Array temperature prediction, program for, spacing of, 237
75-77 Classical roughness, 182-183
Asymmetry, and thermal response of natural Colburn analogy, 232-233
convection systems, 236 Colburn factor, 183
Average heat transfer coefficient, uniform Compact heat sinks, 193- 194
flow effective diffusivity model, Components, 214
159-160 heat transfer in, 210,211,221-224
Average Nusselt number placement
conjugate heat transfer, 2-D, 131 channel height, 236-238
uniform flow effective diffusivity model, channel width, 237
164-167 circuit pack spacing, 238
thermal coupling, 212-213
B Conduction heat transfer, 205, 208
conductivity ofcommon packaging
Backplane, thermal coupling, 213-214 materials, 114
Blasius solution, 9 1, 119, 120 between shelves and frame, 214
Board, 225 Conduction losses, 65
layout of, 225 Conjugate heat transfer in forced air cooling
thermal coupling, 213-214 background, 104-114
Board materials, 195 classification of problems, 114
Board spacing, 234 2-D situations, ll7- 125
Boundary element method, conjugate heat conjugate models. 125- 133
transfer, 2-D experimental observations, 117-125
laminar boundary layer flow, 142-144 fluid side model, 136-138
laminar channel flow, 133-141 laminar boundary layer flow, 141-145
Boundary layer flow, conjugate heat transfer laminar channel flow, boundary element
2-D, 117- 145 method, 13}-141
3-D. 145-153 laminar channel flow, finite difference/
Bypass flow, 83, 84 control volume method, 125-133
solid-fluid coupling, 138-141
c solid side model, 133-136
3-D situations, 145-166
Channel, heat transfer in, 219-221 boundary layer flow, 145-153
Channel flow, 2-D conjugate models. 117-145 uniform flow, 153-166
boundary element method, 133- 141 Contraction (cc) coefficients, 209

245
246 Air Cooling Technology for Electronic Equipment

Control volume approach Destabilization techniques, 188

conjugate heat transfer, 2-D Dimensionless pumping power, as function
laminar boundary layer flow 141-144 of Nusselt number, 189
laminar channel flow, 125-132 Dimensionless surface temperature, 121, 122
heat transfer from component, 221-224 Dirichlet boundary conditions, 2-D conjugate
Convective heat transfer, 205-206 problem, 134
from component, 222 Displaced promoters and flow modulation,
low-profile package arrays, 83-85 187-189
optimization of, 233-239 D-type 2-D roughness, 182
Convective heat transfer coefficient, 65, 123, Duct aspect ratio, critical Reynolds number
see also Heat transfer coefficient as function of, 176
Convective wake, 107, 112 Duhamel superposition technique, 136
Coolant pressure drop, uniform, in-line
arrays, 94-98 E
Coolants
heat fluxes for, 216 Effective diffusivity, 153-166
heat transfer coefficients, 207, 217 Effective fluid conductivity, 153
Cooling limits Electronic components, see Components
concept of, 215-218 Electronic enclosures, see Enclosures
heat transfer, factors affecting, 218-225 Enclosures
heat transfer from component, 221-224 thermal coupling in, 212-215
heat transfer in channel, 219-221 thermal phenomena in, 204-212
significant parameters, 224-225 Enhanced air cooling
in natural and forced convection cooling, nomenclature, 196--197
226--240 performance evaluation criteria, 195-196
forced convection limit analysis, strategies, 174-176
226--233 techniques, 176--196
integral approach, 239-240 classicial roughness, 182-183
natural convection optimization, 233-238 compact heat sinks, I 93-194
nomenclature, 241 displaced promoters and flow
thermal coupling in electronic enclosures, modulation, I 87-189
212-215 enhanced surfaces, 179
thermal phenomena in electronic enhancement by controlling geometric
enclosures, 204-212 layout, 176--179
fluid flow in circuit packs, 208-210 hybrid techniques, 196
heal transfer in circuit packs and its impinging jets, 189-193
effect of components, 211 improved substrate conduction, 195
heat transfer in electronic components inherent enhancement at electronic
(modules), 210 component surfaces, 176
heat transfer mechanisms, 205-208 large roughness elements (2-D ribs and
Critical Reynolds number, 176 3-D protrusions), 179-181
Cross-sectional area, 207 porous metallic matrices, 194
Cylinder bundle vortex generators and barriers, 183-187
forced convection cooling, 25-27 Enhancement cooling, 207
heat sinks, 38-40 Entrance design correlations
natural convection cooling, 6--8 application of proposed correlations, 74--75
Cylindrical rods, 187 array temperature prediction, program for,
75-77
D data reduction, 65
experimental apparatus and procedure,
Darcy flow, 44 49-52
Darcy's law, 194 experimental uncertainty for pressure data,
Delta-wing vortex generators, 183-187 52-53
Index 247

expermiental setup and procedure, 63-65 Forced convection limit analysis, 226--233
flow visualization, 53-54 Form drag, 209
heat transfer and wake effect, 63-7 5 Frame, see Enclosures
heat transfer results and discussion, 66--74 Free convection, 205
nomenclature, 78-79 Free stream, stack of parallel plates immersed
pressure drop, 49--62 in, 17-25
pressure results, 55--62 Free-stream turbulence, 179
uncertainty analysis for heat transfer Friction factor
data, 65 Colburn factor ratios, 183
Entry region roughness element and, 180
conjugate problem, 2-D, 125-126
low-profile package arrays, 83 G
Environment, thermal coupling, 214
Expansion (ec) coefficients, 209 Galerkin finite element method, uniform flow
Extended surface chip cooling, 227, 228 effective diffusivity model, !55
Gauss-Seidel iteration method, 156
F Geometric optimization
bundle of cylinders cooled by forced
Finite difference method, 2-D conjugate convection, 25-27
behavior enhancement via, 176--179
laminar boundary layer flow, 141-144 heat generating plate cooled inside parallel
laminar channel flow, 125-132 plate channel, 27-32
Finite-element method, uniform flow board with finite thermal conductive in
effective diffusivity model, 155 transversal direction, 30--32
Fins board with large thermal conductance in
pin, see Pin fins transversal direction, 27-30
porous metallic inserts as, 194 heat sinks with pin fms and plate fins, 37-44
Flow destabilization, 188 with pin fin arrays, 38--40
Flow modulation, displaced promoters, 187-189 with plate fin arrays, 40-44
Flow pattern in entrance region, visualization natural convection
of, 53-54 bundle of horiwntal cylinders (pin fins)
Fluid dynamics, Reynolds number and, 181 cooled by, 6--8
Fluid flow in circuit packs, 208-210 stack of vertical plates cooled by, 2-6
Fluid motion, convection heat transfer, plate fins with variable thickness
205-206 and height, 35-37
Fluid side model, conjugate problem, !36--137 stack of parallel plates
Flush-mounted heat sources, parallel plate with flush-mounted and protruding heat
stacks with, 12-15 surfaces, 12-15
Flush-source model, 145 immersed in free stream, 17-25
Forced-air cooling, see also Conjugate heat laminar forced convection, 8-12
transfer in forced-air cooling; turbulent forced convection, 15-17
Entrance design correlations; stacks of plates shielded by porous
Low-profile package arrays screens, 32-35
Forced convection, 205 Geometry, low-profile package arrays, 82-83
bundle of cylinder cooled by, 25--27 Granular (3-D) roughness, 182-183
junction temperature calculations, 218 Green's function, 135
laminar, stack of parallel plates cooled by,
8-12 H
limit to, 228-229
stacks of plates shielded by porous Hagen-Poiseuille flow in parallel plate
screens, 33-34 channel, 56
turbulent, stacks of parallel plates cooled Heat generating plate, cooling inside parallel
by, 15-17 plate channel, 27-32
248 Air Cooling Technology for Electronic Equipment

Heat sinks, 37--44 L

compact, 193-194
nonuniform array, 98-99 Laminar flow
pin fin arrays, 38-40 conjugate heat transfer, 117, 141-144
plate fin array, 40-44 pressure drop, 56
Heat transfer roughness element and, 179
in channel, 219-221 transition to turbulence, see also Transition
in electronic components (modules), 210 to turbulence
in entrance region, 63-75 Laminar forced convection, stack of parallel
factors affecting, 218-225 plates cooled by, 8-12
mechanisms of, 205-208 Large-D asymptote, 4, 5
Heat transfer coefficient, 208 Leads, energy balance on, 223
for coolants, 207, 217 Limit analysis, 214,215
convective, 65, 123 Limits, cooling, see Cooling limits
correlations for, 205-206 Liquid cooling, 195
inline arrays, 177 Liquid Encapsulated Module (LEM), 195
low-profile package arrays, 85-89 Local heat transfer coefficient, 106
Heat transfer in circuit packs, 211 Local Nusselt number, 123, 124, 130
ht, see Heat transfer coefficient Loss factor (K), 209
Horizontal cylinders, see Pin fins Low-profile package arrays
Horizontal orientation, enhancement adiabatic heat transfer coefficient and
techniques, 178 wake effect, 86-89
Horseshoe vortices, 115, 185, 187 array geometry, 82-83
Hot spot, 2 convection processes, 83-85
Hybrid enhancement techniques, 195 determiniation of adiablatic heat transfer
coefficient and wake effect,
I prototype experiment 86, model
experiments 87-89
IBM techniques, 195 nomenclature, 99-l 00
Impinging jets, 189-193 nonuniform arrays, 98-99
Inlet area, 209 package temperature estimation,
Inlet channel temperature, circuit pack 85-86
channels, 220 uniform in-line arrays coolant pressure
Inlet flux, 208 drop, 96-98
Inlet temperature, 225 uniform in-line arrays package heat
Inline arrays, 177 transfer, 89-96
Intersection of asymptotes method, 2-5 adiabatic heat transfer coefficient, 89-92
Isoflux boards, 233, 235 example calculation, 94-96
Isothermal boards, 233 thermal wake function, 92-94

J M

Jets, impinging, 189-193 Mass flow rate, in channels, 221

Junction temperature Material property, 206, 208
calculation considerations, 218 Matrix equation, uniform flow effective
factors affecting, 225-226 diffusivity model, 155-156
heat flow from component, 224 Metallic inserts, porous, 194
Microchannel heat sink, 193-194
K Microfin structure, 193
Mixed boundary condition, 2-D conjugate
Kernel solutions, 153 problem, 134
K-type roughness, ribs, 182-183 Mixed convection, 205
Index 249

Mouromtseff number, 174, I 75 inside channel, 27-32

Multichip modules (MCM), 210 shielded by porous screens, 31-35
vertical, natural convection, 3-8
N Optimization, natural convection, 233-238

Natural convection, 205 p

board-to-board optimum spacing, 234
bundles of pin fins cooled by, 6-8 Package size, nonuniform, 98
junction temperature calculations, 218 Package temperature estimation, low-profile
optimization of, 233-238 package arrays, 85-86
stack of vertical plates cooled by, 2-6 Packaging materials, thermal conductivity,
stacks of plates shielded by porous 1!4
screens, 34-35 Parallel plates
Neumann boundary conditions, 2-D flush-mounted and protruding heat
conjugate problem, 134 surfaces, 12-15
Newton's cooling law, 205 forced convection
Noise, 228-231 laminar, 8-12
forced convection limits, 228, 229 turbulent, 15-17
Nondimensional surface temperature, heat sources, 12-15
definition of, 121, 122 immersed in free stream, 17-25
Nonuniform arrays, low-profile package inside channel, 27-32
arrays, 98-99 shielded by porous screens, 31-35
Numerical solution, 2-D laminar channel vertical, natural convection, 2-8
flow, 128 Peclet number, 13!, see also Uniform flow
Nusselt number, 106, 139, 140 effective diffusivity model
conjugate heat transfer, 123, 124 Performance evaluation criteria, 195-196
conjugate heat transfer, 2-D, 142 Pin fins
boundary layer flow, 142-144 heat sinks, 38-40
channel flow, 129-131 natural convection cooling, 6-8
conjugate heat transfer, 3-D, 146-147 Plate fins
dimensionless pumping power as function heat sinks, 40-44
of, 189 with variable thickness and height, 35-37
in entrance region, 66-67 Plate stacks, porous screen shielding, 32-35
uniform flow effective diffusivity model, Plate-to-plate spacing, optimal, 5-6
159-160 Pohlhausen solution, II
Nusselt-Reynolds relation, 91 Porous medium, flow through, 41-44
Porous metallic matrices, 194
0 Porous screen shields, stacks of plates, 32-35
Prandtl number, 91, 106, 189, 193
Optimal pin fin diameter, 39, 40 Pressure drop
Optimal plate fin thickness, 43 coolant, in uniform, in-line arrays, 96-98
Optimal spacing in entrance region, 49-62
cylinder-to-cylinder Pressure drop number, I 1, 13, 16, 25, 26, 33
forced convection, 25-27 Protruding heat sources, parallel plate stacks
natural convection, 6-8 with, 12-15
pin fins, 7 Protrusions, 3-D, 179-181
plate-to-plate, 5-6
forced convection, laminar, 8-15 R
forced convection, turbulent, 15-17,29,
30 Radiative heat transfer, 206-207
heat sources, 12-15 circuit pack channels, 220
immersed in free stream, 17-25 between shelves and frame, 214
250 Air Cooling Technology for Electronic Equipment

Recirculating bubble, 54 Thermal Conduction Module (TCM), 195

Reference temperature, component, 222 Thermal conductance number, 30, 31
Reynolds' analogy, 228, 231-233 Thermal conductivity, 208
Reynolds number, 42, 189 of common packaging materials, 114
channel height and, 176, !77 Thermal coupling in electronic enclosures,
channel versus chip height and, 176 212-215
conjugate heat transfer, 2-D channel flow, Thermal resistance, component, 224
128 Thermal wake, 107, 112
definition, 51 in entrance region, 63-75
flush-source model, 145 low-profile package arrays, 85-89,
roughness elements and, 180, 181 92-94
Rib (2-D) roughness, 182-183 Reynolds number and, 181
Ribs, 2-D, 179-181 uniform flow effective diffusivity model,
Rods, 187 158-159
Roughness elements uniform in-line low-profile package
classicial, 182-183 arrays, 92-94
large, 179-181 Thin board model, 148
Three-dimensional conjugate heat transfer,
s 145-166
boundary layer flow, 145-153
Scoops, 185, 186 uniform flow, 153-166
Self-heating effect, 85 Three-dimensional protrusions, 179-181
Shear stress, 208 Three-dimensional roughness, granular,
Single-chip modules (SCM), 210 182-183
Smaii-D asymptote, 4 Total source power, definition of,
Solid-fluid coupling, boundary element 121, 122
method, 138-141 Transition to turbulence, 176
Solid side model, 2-D conjugate problem, conjugate problem, 121
133-136 free-stream turbulence and, 179
Stack of parallel plates, see Parallel plates Turbulence, free stream, 179
Staggered array, 98, 177 Turbulent flow
Staggered cylinders, 44 control-volume-based method, 125
Stanton number roughness element and, 179
free-stream turbulence and, 179 transition to, see Transition to turbulence
roughness element and, 180 Turbulent forced convection, stack of parallel
Strip source of heat in boundary layer and plates cooled by, 15-17
channel flow, 117-141 Turbulent regions, 115
Substrate conduction, I 95 Two-dimensional conjugate heat transfer,
Superposition method, 48 117-141
Surface area, 207 conjugate models, 125-133
Surface emissivity, 208 experimental observations, 117-125
Surface enhancements, 207 fluid side model, 136-138
Symmetry, and thermal response of natural laminar boundary layer flow, 141-145
convection systems, 236 laminar channel flow
System-level approach to thermal boundary element method, 133-141
management, 215 finite difference/control volume
method, 125-133
T solid-fluid coupling, 138-141
solid side model, 133-136
Tap number, 55 Two-dimensional ribs, 179-181
Temperature prediction, Fortran program Two-dimensional roughness, ribs,
for, 75-77 182-183
Index 251

u v
Uncertainty analysis for heat transfer data, 65 Vertical plates, natural convection cooling,
Uniform flow, conjugate heat transfer, 3-D, 2-6
153-166 Vortex generators and barriers, 183-187
Uniform flow effective diffusivity (UFED)
model, 153-166 w
Uniform in-line arrays, low-profile
coolant pressure drop, 96-98 Wake effect, see Thermal wake
package heat transfer, 89-96