Physics of

Nonmetallic Thin Films

N ATO ADVANCED STUDY INSTITUTES SERIES
A series of edited volumes comprising multifaceted studies of contemporary
scientific issues by some of the best scientific minds in the world, as-
sembled in cooperation with NATO Scientific Affairs Division.
Series B: Physics
RECENT VOLUMES IN THIS SERIES
Volume 7 - LowDimensional Cooperative Phenomena
edited by H. J. Keller
Volume 8 Optical Properties of Ions in Solids
edited by Baldassare Di Bartolo
Volume 9 - Electronic Structure of Polymers and Molecular Crystals
edited by Jean-Marie Andre and Hnos Ladik
Volume 10 - Progress in Electro-Optics
edited by Ezio Camatini
Volume 11 - Fluctuations, Instabilities, and Phase Transitions
edited by Tormod Riste
Volume 12 - Spectroscopy of the Excited State
edited by Baldassare Di Bartolo
Volume 13 - Weak and Electromagnetic Interactions at High Energies
(parts A and B)
edited by Maurice Levy, Jean-Louis Basdevant,
David Speiser, and Raymond Gastmans
Volume 14 - Physics of Nonmetallic Thin Films
edited by C.H.S. Dupuy and A. Cachard
Volume 15 - Nuclear and Particle Physics at Intermediate Energies
edited by J. B. Warren
Volume 16 - Electronic Structure and Reactivity of Metal Surfaces
edited by E. G. Derouane and A. A. Lucas
Volume 17 - Linear and Nonlinear Electron Transport in Solids
edited by J.T. Devreese and V. van Doren
The series is published by an international board of publishers in con-
junction with NATO Scientific Affairs Division
A Life Sciences Plenum Publishing Corporation
B Physics New York and London
C Mathematical and D. Reidel Publishing Company
Physical Sciences Dordrecht and Boston
D Behavioral and Sijthoff International Publishing Company
Social Sciences Leiden
E Applied Sciences Noordhoff International Publishing
Leiden
Physics of
Nonmetallic Thin Films
Edited by
C. H. S. Dupuy and A. Cachard
Universite Claude Bernard Lyon 1
Villeurbanne,Jlrance
PLENUM PRESS. NEW YORK AND LONDON
Published in cooperation with NATO Scientific Affairs Division
Library of Congress Cataloging in Publication Data
Nato Summer School on Metallic and Nonmetallic Thin Films, 2d, Corsica, 1974.
Physics of nonmetallic thin films.
(NATO advanced study institutes series: Series B. Physics; v. 14)
Includes index.
1. Nonmetallic materials- Addresses, essays. lectures. 2. Thin fIlms- Addresses, es-
says. lectures. I. Dupuy. Claude H. S. II. Cachard, A. III. Title. IV. Series.
QC176.N38 1974 530.4'1 76-8385
ISBN-13: 978-1-4684-0849-2 e-ISBN-13: 978-1-4684-0847-8
DOl: 10.1007/978-1-4684-0847-8
Lectures presented at the Second NATO Summer School on
Metallic and Nonmetallic Thin Films held in Corsica. Serra di Ferro.
September 1-5. 1974
1976 Plenum Press, New York
Softcover reprint of the hardcover 1st edition 1976
A Division of Plenum Publishing Corporation
227 West 17th Street, New York. N. Y. 10011
United Kingdom edition published by Plenum Press, London
A Division of Plenum Publishing Company. Ltd.
Davis House (4th Floor), 8 Scrubs Lane, Harlesden, London, NW10 6SE, England
All rights reserved
No part of this book may be reproduced. stored in a retrieval system, or transmitted,
in any form or by any means, electronic, mechanical, photocopying. microirlming.
recording, or otherwise, without written permission from the Publisher
Preface
For several years now the intense development in the field of
microelectronics, the interest in coating materials, and activity
in integrated optics have produced many advances in the field of
thin solid
The research activity has become so intensive and so broad
that it is necessary to divide the field into metallic and non
metallic thin films. A summer school in the area of non metallic
thin films appeared to be a very fruitful concept and, hence, in
October, 1973, A.S.l.M.S. made a proposal to N.A.T.O to hold this
second summer school in Corsica in September 1974.
The basic idea behind this summer school was essentially to
stress and synthesize physical properties and structure of non
metallic thin films. The main reason for this was the feeling that
many laboratories are very specialized and that few engage in both
physical and structural analysis of these films.
The program included a large section on' physical studies:
electrical (transport, interface effects, switching), mechanical
and optical. There was also a large section
crystal structure, chemical composition (stoichiometry is always a
difficult problem), bonding and electronic structure.
Certainly it is different for every laboratory to pursue all
these avenues of research. However, perhaps a good result of this
summer school will be a deeper realization among scientists of the
connection between physical properties and structure. Collaboration
is probably necessary in many of these areas and in this respect I
am sure that the summer school was profitable.
I want to acknowledge the many people who helped me organize
this summer school. I want to especially acknowledge the members of
the scientific committee and the lecturers. I wish to give special
mention to the scientific codirectors, Dr. A. Cachard, Universite
Claude Bernard (Lyon), and Dr. S. Fonash, Penn State University,
and Ms. pivot and Chassagne, treasurer and secretary, respectively.
v
PREFACt
My special thanks to the staff of the Village de Detente et
de Loisirs in Serra di Ferro.
I want to mention especially Mr. A. Paolini, Director and
Mrs Pelamourgue, Assistant.
To all the people of Corsica I give my thanks. I am sure that
no one will forget the friendly reception of Serra di Ferro and its
mayor Mr. J.R. Tomi.
As was the case for the first A.S.I.M.S. summer school in
Porto Vecchio, Mr. J. de Rocca Serra, Chef de la Mission Regionale,
helped us and I thank him.
During this summer school Dr. J. A. Roger presented and defend-
ed his Ph. D. Thesis. It is very pleasant for me to thank him for
presenting this first Ph. D. in Corsica.
The help from N.A.T.O allowed the school to be organized and
permitted it to take place. The D.G.R.S.T. helped us too, and I
thank them.
C.H.S. DUPUY
Contents
General Introduction
A.K. Jonscher
Part I
BASIC NOTIONS
Preparation Methods for Thin Films
D.S. Campbell
Growth Processes
R. Niedermayer
Electronic States in Semiconductors
D.A. Greenwood
Part II
CHARACTERIZATION
Structure Determination of Thin Films .
H. Raether
1
9
49
93
123
Physico-Chemical Analysis of Thin Films . . . . .. 141
A. Cachard
Thickness Measurements
D.S. Campbell
Part III
PHYSICAL PROPERTIES
Electronic Transport Properties . . .
R.M. Hill
vii
163
189
viii CONTENTS
Ionic Transport in Thin Films . . . . 219
S.J. Fonash
Dielectric Properties of Thin Films: Polarization
and Effective Polarization . . . . . . .
S.J. Fonash
Threshold Switching: A Discussion of Thermal and
Electronic Issues . . . .
H.K. Henisch and C. Popescu
Mechanical Properties of Non-Metallic Thin Films
R. W. Hoffman
Thin Films in Optics
G. Baldini and L. Rigaldi
Radiation Effects in Thin Films
A. Holmes-Siedle
Part IV
APPLICATIONS
Thin Film Applications in Microelectronics
V. Le Goascoz
Application of Thin Non-Metallic Films in Optics
E. Pelletier
Some Applications of Non-Metallic Thin Films
M.H. Francombe
List of Participants
Index
225
253
273
355
383
417
445
459
495
501
GENERAL INTRODUCTION
A.K. Jonscher
Chelsea College, University of London
Pulton Place, London SW6 5PR. U.K.
The purpose of this Introduction is to attempt to place the
subject of Non-Metallic Thin Films into proper perspective against
the more general background of modern physics and technology and
also to discuss some problems relevant to research policiy in this
field.
The fist question that may be asked is : "What is so signifi-
cant in this subject that it should have been chosen as the theme
of a specialist Summer School? ". The answer to this, in my opi-
nion, is that thin non-metallic films are commonly found in natu-
re, that they have many useful applications and that their physi-
cal properties present certain unique feature which are not normal-
ly associated with bulk materials.
From the standpoint of a Summer School, with its definite di-
dactic purpose, the subject of Thin Films is uniquely suited as a
vehicle for conveying to the student a wide range of specialist
topics - it is a multidisciplinary field which is both challenging
and rewarding asa subject of study, provided care is taken to ap-
proach it in the correct manner- I shall have something more to say
about this later.
It is desirable at the outset to define what is meant by the
term "Thin Films". The commonly accepted limitation at the upper
end of thickness is the concept of Ii Thick Films", typically some
tens of micrometers in thickness. Thus it would appear that we would
be justified to describe films up to one micrometer in thickness
as "thin", without placing too much significance on the precise fi-
gure. However, even 1 ~ contains some four thousand atomic layers
and it may have many properties of "bulk" material. Figure 1 shows
schematically some relevant orders of magnitude.
2
Mono-
Ia.yer
1A II I
Thin
ShAric
Acid
0.1
5 I.
1 10
cree rlln g
distanc.e
in meta.1s
..... .1
lunnel'''f
A.K. JONSCHER
Thick
1mm
nm
Fig. 1 Typical order of magnitude relating to thin non-metallic
films.
Whether the behaviour of a thin film is distinct from that of
the corresponding bulk material depends upon the precise context.
Thus the structure may be influenced by interfacial processes, e.g.
epitaxy or strain due to lattice mismatch and this may carryon
for hundreds of atomic layers from the interface. The elec-trical
properties will be determined by the relation between the mean free
path and the film thickness - in very disordered non-metallic films
the mean free path is effectively the tunnelling distance, of the
order of 5 nm and thus most films are bulk-like from this point of
view. However, this is not so in metallic or highly crystalline
films in which the mean free path may be significantly large than,.
say, 10 nm and thus film thickness may become a limitation. The
electrical effect of the interface may extend to several Debye
lengths into the material and this, as we shall see in Dr. Fonash's
lectures, may be a serious consideration. in non-metallic films, but
it is not important in metallic ones where the screening length is
of the order of 0.1 nm, or less than one atomic layer.
Not much will be said in the context of the present Summer
School of very thin films - of tunnelling thickness - nor of the
problem of quantisat.ion of energy levels in the direction normal
to the plane of the film. Some of these questions are however im-
plicit, for example, in the operation of MNOS transistors, as dis-
cussed by Dr. Le Goascoz.
The presence on metallic surfaces of naturally occuring oxide
and other films was known for a long time to influence their opti-
cal and electrical properties. Where the films are mechanically ro-
bust and chemically inert, as in the case of aluminium, tantalum
and silicon, they may act as protective coatings and they have been
GENERAL INTRODUCTION
3
used extensively as capacitor materials. It is hardly necessary to
stress the fact that silicon owes its predominant position as a se-
miconductor device material to the excellent chemical, dielectric
and mechanical properties of its dioxide.
Anodically produced films of oxides on aluminium and tantalum
are the basis of a long-established and important capacitor indus-
try.
We shall hear a good deal during this Summer School about the
preparation of thin films. There are many specialised methods for
doing this but on the whole their common feature is that films pre-
pared in this way are significantly less perfect from the point of
view of structure and purity than the best bulk materials. This is
an important practical limitation which has to be borne in mind
when choosing between samples prepared by film deposition techni-
ques and samples obtained by thinning of bulk-grown material.
In many cases the methods most suitable for the deposition of
thin films are not the same as those suitable for the preparation
of good bulk samples : anodisation, evaporation, sputtering and
chemical vapour deposition are all restricted in the attainable
thickness of the deposits. It may even be difficult to grow films
of nominally the same structure and composition but of different
thickness - a serious limitation when it comes to testing the effect
of the thickness on, say, the electrical properties.
It is understandable, therefore, that the problem of reliable
characterisation of thin films is central to this branch of science
and that it receives due attention in the second part of this Sum-
mer School.
Those methods of deposition which employ the condensation of
the material from the vapour phase are at the same time most sui-
table for the formation of heavily disordered layers, including
completely amorphous ones, even for those materials which do not
occur naturally in the amorphous phase in the bulk, e.g. silicon,
germanium and other non-glass-forming semiconductors. For this rea-
son the subject of non-metallic thin films is closely related to
the study of amorphous materials which has acquired considerable
popularity in recent years, even though the time is ripe to ques-
tion the continuing justification for it.
The advancement of our understanding of the growth processes
of thin films represents one of the most significant developments
in modern physics and may be attributable in a large measure to the
interest in the science and technology of thin films. Professor
Niedermayer's lectures will discuss the various processes gover-
ning the growth of thin films at the interface between the solid
and the gaseous or liquid phase ouside.
4
A.K. JONSCHER
Dr. Geenwood's lectures on the electronic properties of thin
films reflect many of the problems encountered iIL this area of phy-
sics. The disordered nature of the films has immediate consequences
for the bulk electronic energy structure, a subject which has recei-
ved a considerable amountof attention in the last ten years or so.
In addition, the very proximity of the surface has its own conse-
quences and the study of surface and interface states is of great
interest in this context.
Any discussion of the electrical conduction in thin films must
distinguish between transport along and across the film. The former
is of principal interest in metallic films, while in insulating films
we are usually forced to study transverse conduction because of the
excessively high impedance. However. conduction in the plane of the
films is of great significance in some semiconductor films ; one
notable examle is the conduction in the channel of a Metal-Insula-
tor-Semiconductor Field Effect Transistor (Insulated Gate FET).
where transport o ~ u r s in a very narrow potential trough near the
insulator-semiconductor interface. There is no chemical film pre-
sent here but an effective thin film situation arises nonetheless.
Dr. Hill's contribution to this Summer School is concerned en-
tirely with transport across the film and he is effectively refer-
ring to modes of direct current conduction characteristics of he,a-
vily disordered solids, i.e. hopping conduction. The last decade
has seen some spectacular advances in our understanding of this
area oE transport and Dr. Hill will outline the most important as-
pects, stressing the significance of a distribution of localised
states in the forbidden gap of material.
The problem of dielctric characterisation of non-metallic
films receives its proper attention in this Summer School in the
lectures of Dr. Fonash. Dielectric measurements on thin films re-
present a favoured test method, perhaps next in popularity only to
optical measurements and the interpretation of the experimental
data is frequently very complicated, since the dielectric proper-
ties are determined by the interplay of the barrier characteristics
and the bulk region response.
One of the most distinctive applications of non-metallic thin
films is as optical filters and anti-reflection coatings. The lec-
tures of Professor Baldini and of Dr. Pelletier deal with this as-
pect of optical properties and applications and show the degree of
sophistication in design and in the execution of optimal structures
for particular purposes which are made possible by modern methods
of film deposition.
GENERAL INTRODUCTION
Dr. Francombe's and Dr. Le Goascoz's lectures provide further
insight into different applications of thin films in technology
and both make it clear how important it is to back up any develop-
ment programme with adequate technological facilities if success
is to be achieved.
5
After this brief look forward to the present Volume 1 wish to
turn, as it were, to a retrospective appraisal of the Summer School.
A survey of the content of the present Volume will reveal what
I consider to be the most striking common characteristic of this
vho]e subject - it is its complex nature. This complexity derives
from the combination of boundary conditions, from the uncertainty
of the exact structural con.figuration, from the mUltiplicity of
the possible materials, from the complexity of the various deposi-
tion and testing techniques, and so on.
An immediate consequence of this complexity is that the instru-
mentation required for the successful deposition, testing and cha-
racterisation of thin film structures has become very expensive
and the processes of research and development are correspondingly
time- and money - consuming. The study of thin films has cessed to
be thetlcheap" subject which could be pursued with minimal resources,
the proverbial "string and sealing wax" approach so typical of ma-
ny university laboratories.
This brings me to the point of research policy in relation to
thin films. The complexity and the great potential depth of the
subject, with its manifold possibilities with regard to the choice
of materials and geometrical configurations demands a careful choi-
ce of research topics for specific studies. This choice has to be
based on certain criteria regarding the potential value of the ex-
pected results. I suggest that one question that may be asked, pro-
fitably, is whether it is intended ~ make a scientific or a tech-
nological study. Since a scientific study is intended to elucidate
fundamental questions regarding the mechanisms or processes in
question, it is desirable that the system under investigation should
be as simple and well-defined as possible. However, this is preci-
sely what is very difficult to obtain in the majority of thin film
systems and this is where the principal problems arise. In order
to be scientifically meaningful, the study should proceed from sim-
pler systems to more complex ones, while in practice it is often
impossible to devise experiments on simple systems.
By contrast with the scientific approach in which the princi-
pal objective is the understanding of the laws of nature, the tech-
nological approach is aimed at mastering a set of processes leading
to some desired practical end. Technology is mostly concerned with
6 A.K. JONSCHER
complex systems, arrived at empirically and even accidentally and
there is often little hope of elucidating fully the processes in-
volved. However, good technology is, by definition, a technology
relevant to some important objective - a device, a process, a sys-
tem. Once the objective ceases to be important, for example becau-
se of obsolescence, because the market for it has changed or even
disappeared, because it has been overtaken by other developments,
etc., the technology concerned ceases to be good technology.
While good science contributing to our increased understanding
of the material universe and good technology aimed at producing de-
sirable goods are both fully justified in their respective diverse
pursuits, what is never justified is the pursuit of bad science ai-
med at obsolete or irrelevant tecnology.
I must confess to a feeling of unease in this respect when
looking at much of the published material relating to the general
field of thin films. While there is some good science and some good
technology being pursued, there is also a great deal of work done
at a considerable expense of public money which would not be scien-
tifically justified in its own right while its technological justi-
fication has also disappeared long ago.
I hope that this Summer School may help its participants and
also the readers of the present Volume to avoid these pitfalls in
the future and may give them fresh inspiration for the pursuit of
really good science in support of relevant technology. Many exam-
ples of this may be found in the lectures that follow this Intro-
duction.
Basic Notions
PREPARATION METHODS FOR THIN FILMS
D.S. Campbell
Department of Electronic and Electrical Engineering
University of Technology
Loughborough, Leicestershire, U.K.
1 INTRODUCTION
It is possible to divide the methods of making thin layers
(i.e. l ~ e r s which are less than 1 ~ m thick) in several different
ways(l)( ). One classification is in terms of the separate groups,
(1) chemical methods and (2) physical methods. Physical methods
cover deposition techniques which depend on the evaporation or
ejection of material from a source, i.e. evaporation or sputtering,
whereas chemical methods depend on a specific chemical reaction.
This chemical reaction may depend on the electrical separation of
ions as in electro-plating and anodisation or it may depend on ther-
mal effects as in vapour phase deposition and thermal growth. How-
ever, in all these cases a definite chemical reaction is required
to obtain the final film.
When one seeks to classify deposition of films by chemical
methods, one finds that it is possible to further sub-divide the
methods that are available, into two more classes. The first of
these classes is concerned with the chemical formation of the film
from the medium, and typical methods involved are electro-plating,
chemical reduction plating and vapour phase deposition. A second
class, however, is that of formation from the substrate and exam-
ples are anodisation, gaseous anodisation and thermal growth.
It must be emphasised that there is often considerable overlap
between the physical and chemical classification, and also between
the sub-classification of formation from the medium and formation
from the substrate.
9
10
D.S. CAMPBELL
The methods summarized under the classifications given are,
in some cases, capable of producing both films less than and also
greater than I However, there are certain techniques that are
only capable of producing thick films (i.e. greater than I
and these include methods such as scree.n printing, as in thick
film technology, glazing, electro-phoretic deposition, flame spray-
ing and painting. However, these techniques will not be examined
in the context of this chapter. (See Chapman Anderson(3) for
surveys of these techniques)
The aim of this paper therefore is to examine the various
methods available for film deposition, be they physical or chemi-
cal. At the end of this; examination it will be possible to iden-
tify the techniques which are suitable for the preparation of non-
metallic thin films. (4)
2. PHYSICAL DEPOSITION TECHNIQUES
2.1 Evaporation(5)
2.1.1. Introduction
Evaporation techniques are widely used for the
of thin layers. A very large number of materials can be evaporated
and if this evaporation is effected in a vacuum system, then the
evaporation temperature will be very considerably lowered, the for-
mation of oxides will be considerably reduced, the amount of impu-
rities included in the growing layer will be reduced and finally,
straight line propagation will occur from the source to the sub-
strate and this will allow for reproduction of finely defined pat-
terns on the substrate if a mask with the necessary holes in it is
placed between the source and the substrate itself. Substrates can
be of a wide variety of materials and can be held at a temperature
appropriate to the properties of the deposited films that are re-
quired. Most materials can be boiled in a suitable crucible, but
there are several that can be sublimecl-before their melting point
is reached.
The pressures that are required in a vacuum system to obtain
satisfactory deposition, both in terms of the reduction of oxides,
reduction in included impurities in the deposition, and the obtai::;
ning of a sharply defined pattern on tile substrate due to t4e pre-
sence of a mask between the source and the substrate, is less than
10-
4
torr with an ideal pressure for normal evaporation work being
10-5 torr.
Rates of evaporation and condensation can vary over very wide
limits, dependent upon the type and temperature of source and the
material used, and a typical curve for gold as a function of source
PREPARATION METHODS FOR THIN FILMS
temperature is given in Fig. 1. An average figure for growth rate
can be reckoned as 10 A per second, but it is possible to obtain
rates as high as 10+
4
A/second, using special boats. Very low ra-
tes of growth are also possible the work of Walton et al on
Nucleation studies(6.
Control of the deposition rate and film thickness can be ef-
fected by several systems which are discussed elsewhere(S)(7).
2.1.2. Basic Principles
11
The thermodynamics of the evaporation have been
considered in detail by various authors (c.f. GIang . The rate
of evaporation G from a surface at a temperature T is given by the
Langmuir expression
G = P .JM./21TRT (1)
where p is the vapour pressure of the material and M the molecular
weight. R is the gas constant per mole. The relationships between
p and T have been published by various authors for a wide variety
Fig. 1
"
<OJ
0
10
3
&..
.J:.
<OJ

0
&..
C7I
"
>
<OJ
10
3
0
"
0::
800 1000 1200
Source temperature (Oe)
Relative growth rate of gold as a function of source tem-
perature.
12 D.S. CAMPBELL
of materials {Honig (8 and Figure 2 shows a typical curve for
gold. A temperature which is normally considered suitable for eva-
poration is that at which the vapour pressure of the material is
equal to 10-2 torr. As an example. for gold this corresponds to a
source temperature of 1650C and at this temperature the rate of
evaporation from the gold surface is 0.2 x 10-
3
gms/cm
2
/second,
which is equivalent to 6 x 10
17
atoms/cm
2
/second.
The average energy of atoms from the evaporating source is
given by
E = 2. kT
2
(2)
where k is the gas constant per atom. A usual value of E is 0.25 eV
compared with the average energy for gas atoms at room temperature
(the normal temperature of the vacuum in which the evaporation is
occurring) of 0.03 eV.
The velocity of the atoms evaporated from the source will be
distributed in a roughly Gaussian manner, about the most probable
,..........
L..
L..
o 10
3

til
L..
:J
III
III
til
L..
Q..
1000 2000
Temperature(OC)
3000
Fig. 2 Vapour pressure of gold as a function of temperature
PREPARATION METHODS FOR THIN FILMS
value of velocity given by :
ex. = .J2 kT/m (3)
where m is the mass of an evaporant atom. The root mean square
velocity will be given by
c = h kT/m (4)
13
(a value for c for gold at temperature of 1800
0
K will be 10
5
cms.per
second). This value of root mean square velocity will depend on the
source geometry and the figure quoted is for an open filament ty-
pe of evaporation. If, for example, the emission occurs through a
small hole in a crucible type of source then c will be modified to
be equal to
c = /4 kT/m (5)
Such a source with such a characteristic is known as a Knudsen
source.
The mean free path of the evaporating material in the residu-
al gas will be a function of the residual gas. A typical figure
for a pressure of 10-5 torr is 500 cms. and this value will be in-
versely proportional to the residual pressure.
The rate of arrival n of molecules or atoms of molecular
weight M of evaporant at the substrate, is given by
n (6)
where t is the thickness deposited in unit time, p is the density
of the deposition and N is Avogadro's number. ~ i s value of n needs
to be compared with the arrival rate of residual gas atoms in the
vacuum which is given by :
(7)
where Mg is the molecular weight of the residual gas. In normal
systems it is found that a ratio of 100 : 1 exists between the
number leaving a source/unit area and the number arriving, and
this typical figure holds for a source-substrate distance of around
15 cms. Under these circumstances an arrival rate of 1015 atoms
per cm2/second is fairly normal, and this corresponds to a growth
.
rate of 10 A per second. However, the value for the rate of arr1-
val of residual gas atoms at 10-5 torr for nitrogen and/or oxygen,
14 D.S. CAMPBELL
is also lOIS, so that it can be seen that in order to be sure of
obtaining a film with the least number of impurities, either the
source-substrate temperature has to be considerably increased so
as to raise the rate of evaporation, the source substrate distance
must be reduced (with the subsequent unavoidable heating of the
substrate that will result), or the residual gas pressure must be
reduced below 10-
5
torr. It is this final solution which is normal-
ly used if very pure films are required, and the techniques for ob-
taining very low pressures in vacuum systems have been widely dis-
cussed by many authors in the literature(9). It is now possible,
at least in experimental systems, to work at pressure of 10-
10
torr or less, but these sorts of pressures are not widely used in
commercial production.
2.1.3. Types of Evaporation Sources
In order to evaporate materials in a vacuum, a vapor source
is required that will support the evaporant and supply heat of va-
porisation while allowing the charge of evaporant to reach a tem-
perature sufficiently high to produce the desired vapor pressure,
and hence rate of evaporation, without reacting with the source.
In order to avoid the contamination of the evaporant and hence of
the growing film, the support material itself must have a negligi-
ble vapor and dissociation pressure at the operating temperature.
There are, therefore, two types of material that can be used for
this, either refractory metals or certain non-metallic materialu
such as oxides, nitrides etc. The form in which these support ma-
terial are used depends very much on the evaporant and tables are
now available, summarising the best support materials used for the
evaporation of the elements (Glang(IO)), inorganic compounds (GIang
(11)and alloys (Glang(12)) to which the reader is referred.
A. Wire and Metal Foil Structures
Of the two basic types of material used for sources men-
tioned, wire and metal foil structures are very widely used for a
wide variety of evaporants. The simplest vapor sources are resis-
tance heated wires and metal foils o.f various types, examples of
which are shown in Figure 3. These wires and foils are generally
made out of tungsten, molybdenum or tantalum. Wire source are ge-
nerally made from wire of diameter 0.02" to 0.06" and their use is
limited to evaporants which wet the filament upon melting and are
then held on by surface tension. However, it is also possible to
use wires of the material to be deposited, provided that the wires
will sublime. This implies that a vapor pressure of 10-2 torr is
reached before the melting point of the wire itself. Such a tech-
nique has been widely used for the deposition of nickel(13) and al-
so for nickel-chromium,(14) to mention but two, although in the lat-
ter case it is important to note that the composition of the depo-
PREPARATION METHODS FOR THIN FILMS 15
A.
D.
E.
Fig. 3 Wire and metal foil sources. a) Hairpin. b) Wire Helix
c) Wire Basket. d) Dimpled foil. e) Dimpled foil with alu-
mina coat. f) Trough type.
sited film will change with time because of the different sublima-
tion rates of nickel and chromium from the wire.
Metal foil structures can take the form of open boats of va-
rious shapes, or for films that are apt to de-gas on evaporation,
structures which prevent the ejection of solid particles directly
from the source. A particularly well-known example of this type of
source is that developed by Drumheller(15) for the evaporation of
silicon monoxide; (see Figure 4.)
It is possible to use metal foil structures for the deposi-
tion of alloys as well as elements and oxides, provided enough
information is available on the relative vapor pressures of the
alloy constituents. In this context two alloys have been examined
in considerable detail ; nickel-iron (permalloy) and nickel-chro-
mium. The nickel-iorn alloy has been widely used for magnetic mem-
ory elements and nickel-chromium for resistors. Both are often eva-
16
SiO
-
SiO __
LUM PERFORATED TANTA
HEATER TUBE

TUNGSTEN FILAMENT
ANTALUM
1TT""t-__ T
(a)
CRUCIBLE
ANTALUM T.
R ADIATION--
SHIELDS

[
o.s. CAMPBELL
\\\(((
1\11((f-

b



:.f';1;4 1

J i.
,
(b)
Fig. 4. Chimney evaporation sources (after Drumheller (IS))
porated from tungsten boats. However, change of composition with
time during the evaporation process, is a considerable disadvan-
tage(16), particularly in the case of Ni-Cr, where the electrical
properties are dependent on the Ni-Cr ratio of the final deposi-
tion. These problems have been overcome by using evaporation from
two separate sources with the rates of evaporation separately
trolled, or by the technique known as flash evaporation. Various
arrangements for flash evaporation have been derived(17)(18) and
a typical solution is shown in Figure 5. The principle is that
a finely divided powder of the alloy is vibrated on to a very hot
tungsten strip where it evaporates immediately on contact with the
strip. In these circumstances the composition of the deposited film
is the same as the original alloy powder. Care has to be taken,
however, to shield the source in order to prevent any loose powder
from finding its way into the pumping system of the vacuum chamber.
B. Non-Metallic Structures
Crucibles of non-metallic materials are often used as eva-
porating sources(20). These crucibles, which are non-conductive of
PREPARATION METHODS FOR THIN FILMS 17
y---Powder container
L . t in. mild - steel rod
Thin mild-steel strip
Copper wire
Fig. 5 Flash evaporation source (after Campbell and Hendry(16))
electricity, have to be supported in a suitable metal cradle, the
cradle then being heated in the normal way by the direct passage
of an electric current. This cradle can take the form of a wire
coil directly wound round the crucible or of a foil structure. It
should also be noted that another form of heating of such crucibles
is that of an R.F. coil which is placed around the cru-
cible but not actually touching it. Such a source has been used
for depositing aluminium. (19)
Alumina is a widely used crucible material which can be used
up to a temperature of 1900C and which has a fairly good thermal
conductivity (0.014 calories per second per degree per cm
2
) ena-
bling heat to be transferred easily from the heated metal coil or
boat. A better material even than alumina, from the thermal conduc-
tion point of view, is beryllia which has a thermal conduction o-
ver three times that of alumina. However, there are certain toxic
disadvantages in the use of beryllia, which have to be taken care
of in any commercial use of such a source. Other oxide materials
are available and all of these have been summarised by Glang(20).
Other materials that have been used for crucible sources are
boron nitride, carbon either as ordinary graphite, pyrolitic gra-
phite or in the form of vitreous carbon.
C. Electron-Beam Heating(21)
The two main types of source examined so far have been heated
18 D.S. CAMPBELL
either by resistance heating or by induction heating. It is, how-
ever, possible to cause vaporisation of materials by using elec-
tron bombardment. A stream of electrons is accelerated up to 10 kV
and focused on to the evaporant surface. By this means, temperatu-
res exceeding 10,000oe may be obtained, enabling a variety of
otherwise non-evaporatable materials to be used, and also, as on-
ly a very limited portion of the evaporant is heated, reducing
considerably any interaction between the evaporant and the support
materials. This technique is widely used for the preparation of
very pure films.
A wide variety of electron gun structures have been used for
this, and these have been classified by Glang(21), into work-acce-
lerated guns, heavy accelerated guns and bent-beam guns. Two typi-
cal electron beam sources are shown in Figure 6. Figure 6 (a) show-
ing a pendant drop configuration in which the metal to be evapora-
ted is in the form of a rod or wire centred within a cathode loop.
Evaporation takes place from the molten tip. The molten drop is
held on the tip by surface tension, and as result, careful control
of the electrical energy supply is required to prevent the drop
falling off. The second source shown in Figure 6 (b) is that de-
veloped by Unvalaand Booker(22) , which uses a hearth to contain
the evaporant which can be water-cooled. Also, because of the e-
lectron beam configuration, the evaporating unit is self-contained.
ned.
D. Reactive Evaporation(23) (24)
Evaporation from metal wires or foils, from refractory oxide
crucibles, or from electron beam sources, can be used to deposit
HV
PENDANT
DROP
LV =-.. _) _ _ HOT
_ ~ CATHODE
'" I ! \ .....
VAPOR
A. B.
ELECTRON PATHS
EVAPORANT
HOT CATHODE
FOCUSING
ELECTRODE
WATER-COOLED
PEDESTAL
Fig. 6 Electron gun sources. a) Pendant drop. b) Shielded fila-
ment.
PREPARATION METHODS FOR THIN FILMS 19
oxides. However, it should be noted that it is also possible to
evaporate metals in relatively high oxygen pressures. Under these
circumstances the oxide of the metal being evaporated will then be
deposited on the substrate. The oxidation reaction takes place in
the main, at the surface of the depositing film, and such techni-
ques have been used successfully for silicon oxide, tantalum oxide
aluminium oxide and even BaTi0
3

2.1.4 Sunmtary
The various types of evaporation sources have been briefly
described, and the reader is referred to s ~ m m r y tables that are
available (5), with regard to the most suitable source for a par-
ticular evaporant. The applicability of the basic evaporation prin-
ciples outlined, to the practical sources used, depends very much
on the actual geometry of the source being employed. Crucible type
sources with a flat evaporating surface will approximate fairly
well to the basic equations noted for evaporation rate etc. How-
ever, long thin sources of the type associated with evaporation or
sublimation from a wire, will be more complex in behaviour and me-
tal box type structures (Drumheller's silicon monoxide source, Fi-
gure 4) and Unvala's electron beam source, Figure 6 (b), will ap-
proximate to the Knudsen behaviour of emission from a box with a
small exit hole.
2.2. Sputtering(25)(26)(27)
2.2.1 Introduction
If a surface is bombarded with energetic particle, it is pos-
sible to cause ejection of the surface atoms,a process known as
sputtering. These ejected atoms can be condensed on to a substrate
to form a thin film. Such a process nan be realised by forming po-
sitive ions of a heavy neutral gas such as Argon and bombarding
the surface of the target material by making the surface the ca-
thode in an electrical circuit. Such a method of obtaining a film
has various advantages over normal evaporation techniques, in as
much as no container contamination will be obtained. It is possi-
ble to deposit alloys without worrying about any fractionation of
the materials. High melting point materials can be used as easily
as low melting point ones and finally, using an R.F. technique,
both metals and insulators can be deposited.
2.2.2 Basic Principles(28)
When a charged particle strikes a surface, a variety of inter-
actions are possible. The most important reactions are shown dia-
gramatically in Figure 7 and are (a) the ejection of neutral atoms
20 o.s. CAMPBELL
Fig. 7 Ejected species from an ion bombarded surface.
of the surface material, (b) the ejection of a small number of
charged atoms of the surface material (usually only about 1% or
less of the number of un-charged atoms) and finally, (c) the ejec-
tion of free electrons- the number of free electrons usually being
greater than 10 for each arriving i n i ~ e n t ion. The first major
effect of this process is, as has previously been mentioned, that the
neutral ejected atoms (Process a) can be collected on suitably
placed substrates to form a film. The second most important effect
is that the electrons ejected can, if desired, be accelerated away
from the target cathode to a suitable placed anode. On their way
to the anode they can cause further ionisation of neutral gas in the
surrounding space and the positive ions so formed will then be
accelerated towards the cathode target. A self-sustaining system
can therefore be obtained and such a situation is known as a glow
discharge condition. The pressure at which this self-sustaining
system will hold, will depend on the cathode/anode spacing and on
the residual pressure and typical figures are that, for a cathode/
anode spacing of 15 cms., a pressure of 10-
2
torr of residual gas
will be sufficient. Below 10-3 torr the discharge will be elimina-
ted.
PREPARATION METHODS FOR THIN FILMS
To obtain the reactions as shown in Figure 7, the accelera-
ting voltage must be limited. Figure 8 shows a typical graph of
sputtering yield, i.e. the number of neutral atoms ejected for
one incident ion as a function of voltage. The three types of in-
teractions can be identified on this graph, namely :
(a) Hard sphere ejection, below approximately 10 keV.
(b) A region in which the electron clouds of the incident l.on and
surface atoms begin to interact.
(c) Finally, a region in which the nuclei interact.
21
The majority of sputtering situations are concerned with the
region of hard sphere ejection and studies on single crystal mate-
rials have shown that in this region, ejection from a target is a
function of the crystallographic direction. On a polycrystalline
target, however, the effect of crystallographic directions is ave-
raged out.
An important feature of ejection under hard sphere conditions,
from a polycrystalline target, is that the most probable energy ob-
tained for ejected atoms is much higher than that obtained in the
case of evaporation, e.g. 4 eV for Cu bombardment with Hg+ at 1 keV.
Such an average energy in the case of Cu just quoted, corresponds
to a root mean square velocity of 4 x 10
5
cm/second and an equiva-
lent surface temperature of 4 x10
4
OK.
4
"0
bI
>-
3
(7)
c
2
....
bI
....
....
::::I
Q.
tJ)
10 100
Accelerating field (keV)
Fig. 8 Idealised sputtering yield / bombardment energy curve
22 D.S. CAMPBELL
The rate of removal of atoms from the surface under polycrys-
talline conditions, is a function of the number of ions bombarding
the target surface, i.e. a function of current. I mA cm-
2
current
corresponds to an ion impingement rate of 5 x lOIS ions/cm
2
/second.
For sputtering yield figures of around this therefore im-
plies a sputtering rate of 5 x lOIS Comparison
with evaporation situations that has previously been discussed,
implies therefore, that a current of around 100mA/cm-2 would be ne-
cessary to obtain rates of removal from the target, equivalent to
that which can be obtained from a typical evaporation situation
(e.g. gold at 1650 oK).
The rate of arrival of atoms at the substrate surface will
depend on source-substrate distance as with the case of evapora-
tion, and the typical figures quoted for evaporation can also ap-
proximately apply here. The major difference, however, is that if
a glow discharge situation is being used, it will not be possible
to reduce the impingement rate of residual gas atoms to anything
like the level that is possible in evaporation.
2.2.3. Types of Sputtering Systems
A. Glow Discharge
A normal glow discharge sputtering system operates under the
minimum conditions in which sufficient secondary electrons are ge-
nerated to replace those lost to the anode or to the walls of the
discharge chamber. Under these conditions, which are typified by
an 5 cm cathode/anode separation and an accelerating field of
1.5 kV at a residual pressure of 10-
2
torr, material will be sput-
tered from the metal cathode into the space between the cathode
and the anode. Figure 9 shows an experimental arrangement. The po-
sition of the substrate to collect this material would seem, at
first glance, to be best as close to the cathode as possible. How-
ever, such a position would effectively block ions in the dischar-
ge from reaching the target cathode so that as a result, substrates
placed close to the cathode receive very thin coatings of sputtered
materials. This screening effect can be avoided if the substrate
is at some distance from the cathode and a useful rule of thumb
quoted by Maissel(27) is that the cathode substrate distance should
be about twice the length of the Crooke's dark space. This dark
space is the region of low luminosity found adjacent to the catho-
de, and represents the acceleration distance of the electrons from
the cathode, which they require to reach an energy at which ionisa-
tion of the residual gas can occur. It is also, often found conve-
nient to place the substrate directly on the anode.
Films grown in such an environment are often classified as
"dirty" because of the high background pressure of gas, but it
should be noted that such a statement needs qualification, since
PREPARATION METHODS FOR THIN FILMS 23
____ ./Negative lead
Woter
Titanium cathode
Rubber gasket
4,\ cm Sputtering chamber
_ Substrate
1l.lJl--li---- Aluminium anode
Gloss cylinder and support legs

Positive lead
Earthed brass plate
Gasket
Gloss cylinder----+'-
chamber
Gasket
Bose plate
To flOp valve
Fig. 9. Glow discharge sputtering system.
the high pressure in a sputtering system is in the main due to an
inert gas such as Argon and the partial pressure of reactive gases
could be as low as in evaporation systems.
Typical growth figures for substrates placed directly on the
anodes in glow discharge systems can be given. For gold sputtered
by Argon at a cathode substrate distance of 4.8" cms. with a volta-
ge of 1.5 kV between anode and cathode and a total current of 1.7
rna over a cathode area of 48 cm
2
, a growth rate of 0.5 A/second
can be obtained.
B. Reactive Sputtering(29)
If advantage is taken of the residual gas in a glow discharge
system such that a large proportion of the residual gas is a reac-
tive species relative to the film being deposited, then a deposi-
tion can be obtained of a completely reacted material such as an
oxide or nitride. This system has been used for the growth of si-
licon dioxide to prepare capacitors.
Reactive sputtering need not to be confined to a flow discharge
system, but can use any of the sputtering systems which will be
discussed (triode sputtering, R.F. supported sputtering, etc.)
24 D.S. CAMPBELL
The mechanism of reactive sputtering has been considered by
various authors, particularly with regard to the site of the oxi-
dation reaction. Extensive studies have been undertaken on the re-
active sputtering of tantalum(63)(64) in a glow discharge using
resistivity, growth rate and mechanical stress observations. It has
been concluded that at low oxygen pressure, tantalum metal is sput-
tered and reacts at the substrate, whereas at high oxygen pressures
the reaction occurs at the target and the oxide species is sputte-
red.
c. Getter Sputtering (30)
Decreases in the partial pressure of reactive gases beyond
those usually associated with a normal glow discharge can be effec-
ted by utilising the gettering action of the sputtering material
to purify the Argon of the discharge system, before it reaches the
part of the system where coating of the substrate occurs. This is
accomplished by surrounding one or more of the cathodes with an
anode can (see Figure 10). Under these conditions the partial pres-
sure of impurities can be reduced to levels as low as 10-
10
torr.
D. Bias Sputtering(31)
In the systems so far discussed, the substrate has either been
floating electrically or kept at anode potential. It is, however,
possible to give the film, assuming it is conductive, a small ne-
gative bias relative to the anode. The film is then subjected to
ion bombardment throughout its growth, a process which effectively
cleans the film of absorbed gases which would otherwise become trap-
ped in it as impurities. This technique is known as bias sputtering.
Close
fitting
lids
Inert gas
Metal
sputtering
chamber
Heater -r--r-}? ......... =iI:=::;nt--f-- Substrate
strip
I.t.-it--+---=:U--, 11==:111
h.t.- -tt----I.---...LJ u.+----II-I.t.
Fig. 10. Getter sputtering apparatus.
PREPARATION METHODS FOR THIN FILMS
---...
E
u
I
E
s:
0
...
u
"s
'--'
>-
..

..
til
til
til
a:
1200,-----------------------------.
1000
800
600
400
200
x
x
-- Theoretical
x
Expe rimenta I

o -100 -200
Bias(volts)
-300
Fig. 11 Effect of bias sputtering voltage on purity of tantalum
films (Ref. 31)
Figure 11 shows the effectiveness of this process for the case of
25
a tantalum film deposited at various negative biases. The reGisti-
vity is a direct measure of the degree of impurities incorporated
in the film and it can be seen that after a small initial rise.
probably due to stresses introduced into the film by the high ener-
gy ion bombardment, the resistance falls to a very low value for
bias voltages of around -200 volts.
E. Triode Sputtering
Another method of reducing the effect of impurities in a glow
discharge system is to work at lower pressure than 10-
2
torr. Such
a requirement implies that it is necessary to generate electrons
which can be injected into the discharge by other means than the
discharge itself. One such system is known as triode sputteting
and Figure 12 shows a diagram of the apparatus. A filament at a
voltage approximately 50 volts less than the anode is used to in-
ject electrons into the discharge systems and a magnet, external
to the vacuum chamber, is used to the path of the electrons
26
Substrate
support
Field . II
coil'
Filament
assembly
Fig. 12
D.S. CAMPBELL
Target
Triode sputtering apparatus
prior to their collisions with the anode. In this situation, pres-
sures of 10-3 or less can be used in the discharge chamber.
F. Ion Beam Sputtering
If instead of using a discharge system to generate the ions
either at 10-
2
or at lower pressure as with triode sputtering, a
separate ion source is used, then it is possible to grow films
sputtered from targets in a residual pressure as low as 10-
5
torr
or less. A typical apparatus for doing this is shown in Figure 13.
This type of apparatus has been developed commercially by Balzers
and is capable of growing clean sputtered films at high rates of
deposition. It is interesting to note that this technique is very
closely allied to tpat of ion implantation(62), as the ion beam
could be any material that can be ionLzed in the ionisation chamber.
G. R.F. Supported Sputtering
A further method for working at lower pressure than 1 ~ 2 torr
is that of running the discharge in an R.F. field. Such an R.F.
field, typically at a frequency of the order to 5 MHz, will uause
both negative and positive ions to be formed, and the positive ones
will then bombard the cathode in a normal manner. A typical appara-
tus is shown in Figure 14.
H. R.F. Sputtering
The system described" so far have involved the use of a con-
tacting cathode as the material to be sputtered, although it has
been noted that it is possible be reactive sputtering, to grow
PREPARATION METHODS FOR THIN FILMS
Filament
Acceleration grid
(ions)
,. grid
(oloctrons)
Chamber wall --- Gas entry port
I--L..--C-
/

I

I-ve +ve
/1
Target (4 c m Diameter)
Substrate
heater
Substrate
support
Shield ring
Fig. 13 Ion beam sputtering apparatus
Pin
anode
r.f. Coils


Fig. 14 R.F. supported sputtering apparatus
27
28 o.s. CAMPBELL
insulating films on the substrate. It has, however, been found pos-
sible to sputter directly from insulators by applying an R.F. po-
tential between the cathode and anode. This process has become
know as R.F. sputtering and is not be confused with R.F. suppor-
ted sputtering. A simple way of viewing the reactions that occur
is to note that provided the frequency is high enough( greater than
50 KHz), negative charge accumulated on the insulating target will
not be sufficient during the half cycle in which the target is
positive, to prevent positive ions bombarding the target during
the half cycle in which the target is negative. The actual beha-
viour of an R.F. sputtered system is in fact more complex when exa-
mined in detail and the reader is referred to various authors for
further information.(32) The system is important to the preparation
of insulators.
I. Plasma Reactions
Plasma reactions of various types have been used for the depo-
sition of thin films. These reactions are difficult to classify
and they are sometimes considered under sputtering. They are, how-
ever, essentially vapor phase or thermal growth reactions, with
the discharge supplying the energy necessary to effect the chemical
changes.
A plasma reaction that must be noted is that of gaseous anodi-
sation.(33).Figure 15 shows a typical apparatus. Although a glow
discharge is used, this glow discharge is effectively replacing the
liquid electrolyte of a conventional wet process. It is important
in such a system to use a non-reactive anode in the discharge cir-
cuit, otherwise all the voltage in the anodising circuit will be
dropped across the oxide formed on the discharge anode. Various
metals have been successfully oxidised in this way, including alu-
minium and tantalum.
A plasma assisted thermal growth has been used by Ligenza(34) ,
who showed that using an R.F. excited discharge and an oxygen pres-
sure of between 0.1 and 1. torr, silicon could be oxidized to a
thickness of around 3,500 A at around 300C. If the silicon was
made the anode of a 50 volt system then negatively c h r g ~ d oxygen
ions will bombard the surface and thickness above 3,500 A could be
obtained. However, in this latter case, material will be sputtered
from the cathode as well, but the atoms from the c.athode can be
prevented from landing on the anode by placing a suitable bend in
the discharge tube so that the cathode cannot "see" the anode. R.F.
plasma oxidation of this type has recently been used with effect,
for preparing insulating films on superconductors for Josephson
Tunneling Devices(35).
If a normal glow discharge is used and the monomer of an
PREPARATION METHODS FOR THIN FILMS 29
TANTALUM FOIL
r=::5-- COOLING WATER
VALVE
Fig. 15 Gaseous Anodisation apparatus (Ref. 33)
organic or inorganic material is introduced into the discharge cham-
ber, it has been found that the monomer will be polymerised in the
glow discharge so that insulating films can be grown on suitably
placed substrates (36) .Over 40 monomers have been in this
way (37).Glow discharge conditions can also be used to effect va-
por phase reactions such as the deposition of silicon nitride from
a gaseous mixture of silicon hydride, ammonia and hydrogen (38).
R.F. excitation of the discharge has also been used and an appara-
tus has been described by Connel and Gregor(39) that can produce
o
insulating films from styrene at high rates of deposition (20 AI
second).
J. Ion Plating(41)(6l)
The system of ion plating is essentially a combination of eva-
poration from a heated wire and the use of a discharge. The anode
30 D.S. CAMPBELL
of the system is also the evaporation source, so that the e v p o ~
rant is ionised in its passage towards the substrate which is pla-
ced on the cathode and it is therefore accelerated before reaching
the substrate. By this technique, enhanced adhesion of the metal
film to the substrate is obtained.
3. CHEMICAL DEPOSITION TECHNIQUES
3.1 Thermal Growth(42)
Films can be formed on a large variety of metal substrates by
heating them in gases of the required type (oxygen for oxides, ni-
trogen for nitrides, eo for carbides). Films obtained, however,
are limited in thickness because the reaction will become very
slow as the film thickness increases. If thickness is plotted
against time, an exponential relationship is obtained and a typical
curve for the growth of alumina on aluminium is shown in Figure 16.
If a non-coherent film is formed, the film will continue to
grow because sections of the film will continually flake off from
the surface. Such a behaviour, however, is of little use in prepa-
20
r-.....
0
"""-
en
en
til
10 c
~
u
.J:;
I-
10 20
Time (minutes)
Fig. 16 Thermal growth of A1203 on Al at 20
0
e as a function of
time.
PREPARATION METHODS FOR THIN FILMS 31
ring coherent layers of practical importance. Since the mobility
of ions through the oxide is dependent on temperature, the higher
the temperature the greater the thickness that can be obtained.
3.2. Anodisation(43) (44)
3.2.1 Introduction
The thickness of oxide obtained in normal thermal growth is
essentially limited by the ability of the ions to migrate through
the film to the film/metal interface. This migration, however, can
be considerably enhanced if the film are grown on a metal substra-
te in an electrolytic bath. In such a system, the parent metal is
made the anode of an electrolytic cell and a voltage is applied
between the anode and the cathode.
3.2.2 Basic Principles
The reactions that occur at the cathode and anode can be re-
presented by the following equations
M + H20 + MO + 2 H+ + 2 E (anode)
2 E + 2 H20 + H2 t + 2 OH- (cathode)
These equations express the fact that oxide will grow on the metal/
anode surface and hydrogen will be evolved on the cathode. The equa-
tions imply the presence of water and anodisation is usually effec-
ted in aqueous electrolytes, such as a solution of phosphoric acid.
The acidity of the electrolyte is important in obtaining a coherent
film since if the electrolyte is too acid or too alkaline, the film
can dissolve as it grows and porous oxide structures can result.
Thickness as a function of time for coherent films will depend
on the vol tage applied across the electrolytic cell and typical growth
curves are shown in Figure 17. These curves are analogous to the
thermal growth curve previously shown (Figure 16), but in this case
the ultimate thickness is only limited by the applied voltage. The
ultimate thickness is characterised by the "anodisation constant"
that is, the thickness of film that will be obtained after an infi-
nite time, for 1 volt applied. In the case of aluminium this is
13.6 A, tantalum 16.0 A, silicon 3.5 A and niobium 43 A. In prac-
tice, it is usually not possible to keep the metal in the anodisa-
tion bath for sufficient time to obtain thicknesses corresponding
to those expected from the anodisation constant. Therefore, the o-
xide is often somewhat less thick usually between 70-80% of the maxi-
mum that can be obtained. The technique of growth illustrated by
32
r----...
0<{
'---../
III
III
~
c
~
u
.c.
I-
Fig. 17
o.s. CAMPBELL
400
300
To
2
0
S
(to a/sec )
200
100
10 20 30 40
Time (seconds)
Constant voltage growth of Ta205 and Al203 by anodisation.
Thickness/Time for 30 V applied.
Figure 17, assumes that a constant voltage is applied to the cell.
The disadvantage of this approach is that in the initial stages of
growth, very high current densities are required. One way round
this difficulty is to anodise under constant current conditions
rather that constant voltage. Growth will then be as illustrated
in Figure 18. The limiting condition fn these circumstances is the
voltage that needs to be applied as the thickness of oxide increa-
ses and this will eventually lead to breakdown in the film. In the
case of aluminium, this occurs at 1.5 ~ m and with tantalum, 1.1
~ m although the limit is essentially a function of the purity of
substrate of the electrolyte used.
Growth rates obtained under constant current conditions may
be compared with previous figures for evaporation and sputtering.
Aluminium can be anodised at a current density of 2 mA/cm
2
to
PREPARATION METHODS FOR THIN FILMS
500
r-..
0 400
'---"
III
III 300

c:

u 200
.&:.
r-
100
15 30 45
Time (minutes)
33
60
Fig. 18 Constant current growth of Ta205 and Al203 by anodisation
Thickness/Time for a current density of 2 mamp/cm
2

give a growth rate of 11 A/second. This corresponds to the voltage
being increased at the rate of 50 volts/minute. In the case of tan-
talum at 2 ma/cm
2
current density, a growth rate of 10 A/second is
obtained, corresponding to a voltage increase of 40 volts/minute.
3.2.3 Practical Aspects
Anodisation is a widely used technique for obtaining amorphous,
highly insulating films. The mobility of ions in the anodising elec-
trolyte is often such that highly convoluted surfaces can be oxidised
in a very even manner, as is illustrated by the aluminium and tan-
talium structures that are used in electrolytic capacitors.(45)
3.2.4 Gaseous Anodisation
Gaseous anodisation has already been referred to under plasma
reactions in sputtering (see Figure 15).
34 o.s. CAMPBELL
3.3 Vapour Phase Growth(36) (46) (47)
3.3.1 Introduction
The deposition of a film on a surface composed of the same or
a different substrate by means of a chemical reaction from the gas
phase at the surface, is known as vapour phase growth or vapour pha-
se plating. Usually the surface is hotter than the surroundings so
that a heterogenous reaction occurs at the surface. However, other
means of activating the chemical reaction may be used, such as a
glow discharge (see Sputtering-plasma reactions) , electron-beam ex-
citation or ultra violet radiation.
3.3.2 Types of Vapour Phase Reactions
A. Disproportionation
The reaction is typified by the equation
A + AB2 * 2AB
where A and B are the two elements. Tha higher valency state is
more stable at lower temperatures so that if a hot gas of AB is
passed into a colder region, deposition of A can occur.
Figure 19 shows a typical closed tube system that has been
Furnace enclosure
/ /,(,( / ////1/,( / //
Quartz tube
Source
Iodine
vapour
Substr ate
77 / 777777/ 77/r;
e.g. Ge 300C
Si IIOOoC
Fig. 19 Disproportionation reaction vessel
PREPARATION METHODS FOR THIN FILMS 35
employed for the deposition of germanium or silicon. For both the-
se materials two iodides can be formed, and the reaction is used to
transport silicon or germanium from the high temperature source zone
to the low temperature substrate zone.
Alternatively, a continuous flow open system can be used in
which iodine vapour, usually diluted with hydrogen, is continually
passed over the SGurce and then the substrate. Growth rates can
high, for example, germanium can be deposited at rates up to 400 A
per second in the above system.
B. Polymerisation(36)
Both organic and inorganic polymers may be prepared from mo-
nomer vapour by the use of electron beam, ultra violet irradiation
or glow discharge. Insulating films prepared in this manner can
have very desirable properties.
Electron beam irradiation has been applied to a large number
of materials, including styrene, butadiene etc. Recent workers
have described an apparatus for the production of polymer films
from evaporating epoxy resin.
Ultra violet irradiation techniques are widely known in photo-
resist etching(48). In the etching a relatively thick layer of
photo-sensitive material is spread evenly over the surface and then
irradiated through a mask. The irradiated areas polymerise to give
a material that is insoluble in the solvent used for the un-polyme-
rised film. A similar technique has been used to prepare insulating
films. White(49), for exposed metal layers in butadiene
vapour to irradiate and thus built up thin dielectric films. Various
other have also been used (e.g. acrolein) and growth rates
of around 1 A/second are easily obtained.
Glow discharge techniques have already been referred to under
Sputtering-plasma reactions - using either a straightforward glow
discharge or R.F. excited systems. As noted previously, this latter
technique has been used to produce insulating films from a variety
of materials including styrene.
C. Oxidation
This is usually undertaken using a halide of the required me-
tal oxide, because of the high vapour pressure of the halide and the
ease of removal of the by-products. The reaction used is :
H2
0
2 AX
-+
A20 + HX
36
D.S. CAMPBELL
Oxides may be deposited using this technique by the use of steam
mixed with the halide and allowed to flow over a hot substrate.
To ensure thorough mixing and an even temperature in the reactor,
a fluidised bed is often used. Oxides that can easily be prepared
in this way are those of aluminium, titanium, t n t l u ~ and tin,
and high growth rates are possible. (Greater than lOa A/second).
D. Nitriding
If instead of steam, an atmosphere of ammonia is used, then
it is possible to grow layers of nitrides. This reaction has been
used effectively for the growth of silicon nitride.
E. Reduction
Metal films can be prepared if hydrogen is substituted for
steam in the oxidation reaction previously described. This type
of reaction is widely used for the preparation of silicon and ger-
manium. In the silicon case, SiRC13 or SiC14 is used and high growth
rates are obtained (200 A/second at IIOOC). The growth kinetics
of these reactions have been studied in detail by various workers.
F. Decomposition
Decomposition as represented by the equation
AB-+A+B
can be effected both by heat (pyrolysis) and by glow discharge.
Pyrolitic reactions have been widely applied to the preparation of
silicon (from SiR4) , nickel from nickel carbonyl and Si02 from the
decomposition of silicon esters. Rates of growth for nid:.el can be
very high at moderate substrate temperatures (1,000 X/second at
200C)
Glow discharges have also been used to prepare insulating films
by decomposition, either directly or with the discharge excited by
radio-frequency. (50)
3.3.3. Summary
Various possible reactions for vapour plating have been brie-
fly examined. The apparatus is sometimes quite complex with gas
pumping systems or vacuum systems being required, and for pyroli-
tic work, furnaces may be required to run at high temperatures
(2000C for carbon by the decomposition of toluene and benzene).
This high temperature moreover, limits the type of substrate that
can be used.
PREPARATION METHODS FOR THIN FILMS 37
3.4. Electroplating(SI)(S2)(S3)
3.4.1 Introduction
Electruplating has been known for a considerable time, and
many standard textbooks now exist on the subject. The apparatus
involved is basically simple, consisting of an anode and cathode
immersed in a suitable electrolyte. Hetal is deposited on the ca-
thode, and the relationship between the weight of material deposi-
ted and the various parameters, can be expressed by the first and
second laws of electrolysis. These state:
I. The weight of the deposit is proportional to the amount of
electricity passed.
2. The weight of material deposited by the same quaIcti ty of elec-
tricity is proportional to the electrochemical equivalent E.
Expressed as an equation, the weight deposited per unit area G/A
is given by :
G
A
.. JtEa
where J is the current density and t the time. This equation intro-
duces another term, the current efficiency a, which is the ratio
of the experimental to theoretical weight deposited ; it can ge-
nerally be expected to be between unity and O.S.
The above equation can be written in a slightly different form
to give the rate of deposition. If a thickness I is deposited in
time t, then the rate of deposition lit is given by :
t
JEa
p
where p is the film density.
The rate of deposition values can be very high at high current
densities. For example, silver will deposit at 10 A s-I at a cur-
rent density of 1 rnA cm-
2
, and this will rise to 1 ~ s-I at I rnA cm-2
Such a proportionality to current density holds only if a remains
unchanged. This can be expressed in another way by stating that no
secondary reactions must occur.
3.4.2 Basic Principles
Of the 70 metallic elements, it is found possible to plate on-
ly 33 successfully, and of this latter number, only 14 are deposited
commercially. A large variety of baths can be used for the possible
38 D.S. CAMPBELL
elements to improve the adhesion, crystalline structure, current
efficiency etc. However, it is not possible to plate elements out-
side the group of 33, as other reactions (e.g. formation of hydro-
gen) can more readily occur. This can be illustrated by considering
the I-V characteristics of a plating solution. For a simple sys-
tem the curve will be as shown in Figure 20. Such a curve is obtained
with a probe placed near the cathode. The equilibrium potential of
the cathode in the solution is indicated by t ~ intercept value on
the voltage axis. A negative intercept implies that the cathode
will dissolve in the electrolyte at zero voltage (i.e. it will cor-
rode). Equilibrium potentials for the different metals vary from
+ 1.7 to -1.66 V. Saturation is seen in the curve - at a high ca-
thode voltage, ions cannot get to the cathode fast enough.
If two reactions are possible, each will affect the other. Two
I-V curves can now be drawn for the two reactions, and Figure 21.
shows this case. III the case of alloy plating this means that the
composition of the alloy will depend on the voltage used, as indi-
cated by the dashed line in Figure 21. If one of the reactions is the
formation of hydrogen, the curves will now be as in Figure 22.
Saturation
voltage
Current
Voltage
Ei.g. 20 Ideal ilv curve for electroplating bath.
PREPARATION METHODS FOR THIN FILMS 39
Current
B
Voltage
Fig. 21 Ideal i/V curves for two reaction electroplating.
A
Current
H
Voltage
Fig. 22 Ideal i/V curves for one reaction and hydrogen evolution.
40 D.S. CAMPBELL
The voltage drop in the bath must be reduced to as Iowa value as
possible to reduce waste of power in heating, and this is usually
by the addition of conducting salts.
3.4.3. Practical Aspects
In practical deposition systems, care must be taken to control
the current density so as to avoid the inclusion of gas bubbles
etc. in the film. The effect of solution temperature will not ge-
nerally be important unless a changes with temperature, (if it does
a will generally increase with temperature rise). As the deposition
rate can be high, it is possible to use electrochemical deposition
for forming thick layers-a process known as electroforming - and
for refining. An example of electroforming is the preparation of
master disks for gramophone records. In this case, the initial de-
posit to give a suitable cathode is obtained either by using a
colloidal suspension of metal, or by chemical reduction plating.
The many alloys that have been successfully deposited (100 or
so) are discussed in detail in Brenner's two volumes on the elec-
troplating(54). It is not possible to deposit every combination
because of the characteristics of the separate elements, although
it is often possible to slow one of the reactions down by suita-
ble chemical complexing. The effect of complexing is to lower the
equilibrium potential to a more negative value, and this results
in a crowding together of the i-V characteristics for the separate
elements (Figure 21). It is not necessarily the case that the po-
tentials of the separate metals come sufficiently close to permit
co-deposition, and it may also be necessary to vary the individual
concentrations or the concentration of the complexing agent if it
affects both elements. Cyanide is a typical complexing ion for Ag.,
Cd., Zn., and Cu.
Ternary alloys can be deposited by electroplating ; Brenner
lists 15 that can be easily formed.
It is possibie to form oxides of elements successfully on the
anode of the electrode system. The oxides are deposited from the
solution (c.f. anodisation, where it is the oxide of the anode ma-
terial that is formed). Films of Pb and Mn oxides have been grown
in this way, but the method has little importance in thin film tech-
nology.
3.5. Chemical Reduction Plating(55)(56)(57)
3.5.1 Introduction
Films of metal can be deposited directly without any electro-
de potentials being involved, by the chemical reduction of a suita-
ble compound in solution. Such deposition is known as chemical-re-
duction plating or electroless deposition. Four differeht types of
reaction may be distinguished and these are summarised below.
PREPARATION METHODS FOR THIN FILMS
41
3.5.2 Practical Aspects
A. Non-Catalytic Reactions
These take place at any surface submersed in the bath. Silver
mirrors are usually formed in this way, by the use of a mild redu-
cing agent such as formaldehyde in a solution of silver nitrate.
Very thick layers may be built up.
B. Catalytic Reactions
The ability of the metal to deposit on anything can sometimes
be a considerable nuisance, and more controlled reactions are of-
ten more useful. In these, the metal will deposit only on certain
surfaces of other metals and nowhere else. The d,eposition of nickel,
for example, can be achieved by such techniques as the reduction of
NiCl2 by sodium hypophosphite, when the metal will grow on a sur-
face of nickel itself, cobalt,iron, and aluminium - the metal acts
as the catalyst. (Note : the use of sodium hypophosphite as the re-
ducing agent means that between 5 and 10% of phosphorus will become
incorporated in the film). This type of reaction has become so im-
portant that a complete book is now available on chemical reduction
plating of just nickel. (58) Other metals, particularly the Pt group,
can be deposited in t4is manner.
C. Catalytic Reactions Using Activators
The number of metal surfaces that will catalyze deposition is
limited. It is found, however, that it is possible to activate the
surfaces of non-catalytic metals so that deposition will take place
on these surfaces.The role of the activator is to lower activation
energy for the reduction reaction at particular points on the sur-
face so that deposition will occur at these points. Islands of me-
tal will thus grow and spread and eventually, give a continuous
film. The best activators to be used for particular metals are list-
ed in standard texts on the subject(55) ; PdCl2 is often used for
Cu and Ni. Very little of the activator is required - in the case
of PdCl2 a dip in a 0,01% solution, followed by a rinse in water is
all that is required.
D. Catalytic Reactions Using Activators and Sensitisers
For non-metallic surfaces, a sensitisation before activation
is required. For Ni this takes the form of a dip in a 0,1% solu-
tion of SnCl2 followed by a rinse. The activation is then carried
out in the normal way. The advantage of such reactions is that it
is possible to plate onto glass and other non-conducting surfaces.
Also, and this applies in general, it is possible to plate surfa-
ces that are difficult of access, such as the inside of tubes.
42 D.S. CAMPBELL
3.5.3. Sunmary
Chemical reduction plating uses very simple apparatus, provi-
ded a suitable reaction and if necessary, catalysts, are available.
The rates of deposition depend on solution and temperature.
3.6. Solution Deposition(S9)
3.6.1 Oxide Films
It is possible to deposit dielectric films on to non-metallic
substrates, using organic solution techniques. The substrate to be
coated is generally inmersed in a solution of a suitable organic
material so that a thin layer of material is formed on both sides
of the substrate. Uniformity of this liquid film can often be im-
proved by spinning the wetted surface, but after uniformity has
been achieved, the substrate is then baked to a temperature of be-
tween 200-sooDe to convert the liquid layer to a solid, usable
structure. For example, it has been found that a colloidal silicon
dioxide hydrate can be prepared on a substrate by inmersing in a
silicate solution to which acids have been added. This colloidal
hydrate will then yield a film of Si02 on baking.
Film thickness as low as 100 A or less can be obtained by this
technique, and it is possible to easily form multiple layers of dif-
ferent materials. Two materials, Si02 and Ti02 have been extensi-
vely prepared in this manner and the films used for a variety of
optical purposes.
3.6.2. Hydrophilic Films(60)
Multimonolayers of long chain fatty acids can be built up on
a substrate by repeated immersion of the substrate in a liquid, on
which long chain fatty acids are floating on the surface. The layers
obtained at each immersion are around 20 A thick and successful
D
films up to 20,000 A have been prepared. Such films are extremely
stable and electrodes may be applied by normal evaporation techni-
ques.
3.6.3. Liquid Phase Epitaxy(6S)
Under the heading of Solution Deposition it is worth noting
that it is possible to prepare single crystal films from the liquid
phase provided a suitable substrate is available. Normal deposi-
tion from the liquid phase gives randomly nucleated growth and
PREPARATION METHODS FOR THIN FILMS
polycrystalline layers which are often of variable thickness in
the initial stages of growth. However, it has been found possible
to control the system well enough so that material will grow on a
large planar seed in a form that is thin and itself pLanar enough
to be called a layer(65). This technique is now widely used for
43
the growth of semiconductor films (e.g. GaAs from Ga at temperature
down to 700C) and also for other applications such as magnetic gar-
nets for magnetic bubble systems.
One of the major advances of this type of technique is that
it is possible to grow layers at temperatures which are several
hundred degree lower than the melting point of the compound and
thus, to reduce the crystalline defects obtained in the film.
3.6.4. Growth of Polymer Films
Polymer films of materials such as polypropylene, polystyrene
(66) and P.V.C.(67) have been obtained by the simple technique of
direct isothermal immersion of a substrate into a suitable solution
of the polymer, and also by allowing the evaporation of the solvent
from the polymer solution placed on a substrate. Tne former techni-
que appears to be very promising for obtaining durable and useful
films. Detailed studies on P.V.C.(67) have shown that the immersion
technique applied to dilute solutions of P.V.C. in either benzene
plus acetone or in cyclohexanone, yields films of a limited thick-
ness which is not increased by prolonged immersion (e.g. for a so-
lution of 0.6 gms P.V.C. in 100 cm
3
of cyclohexanone held at 40C,
a polrer film will grow on glass to a limiting thickness of
2000 in 15 minutes).
The growth model suggested for P.V.C.(67) allows initially of
the P.V.C. chain segments being adsorbed at suitable, unspecified
sites on the surface. After the initial adsorption, further growth
occurs by adsorption or attachment of new chain segments. However,
the outer layer is also capable of being dissolved by the solvent
so that a final thickness is reached in which the outer layer growth
rate is exactly equal to its solution rate.
4. SUMMARY
Various techniques available for the deposition of thin films
have now been examined. Not all of these techniques are applicable
to the growth of non-metallic thin films. The ones which are, are
summarised in Table 1.
44
Table I
Summary of preparation methods
applicable to dielctric films
o.s. CAMPBELL
TECHNIQUE MAJOR REFERENCES
Evaporation (using all source types)
Sputtering. Reactive.
R.F.
Plasma Reactions
Gaseous
anodisation
assisted
thermal growth
polymerisation
assisted
vapour-phase
Thermal Growth
Anodisation
Vapour Phase - Polymerisation
(E.B. & U.V.)
Oxidation
Nitriding
Solution Deposition - organic
- hydrophilic
- Polymer
(4a) (4b) (5)
(4a) (4b) (26) (27)
(4a) (4b) (26) (27)
(27) (36) (40)
(27) (36)
(27) (36)
(27) (36)
(4a) (4b)
(40)
(4a) (40) (44)
(4b)
(40) (46)
(4b)
(40) (46)
(40)
(59)
(60)
(67)
PREPARATION METHODS FOR THIN FILMS 45
REFERENCES
1. D.S. CampbelL In "The use of thin films in physical examina-
tions " (Ed. J. C. Anderson) Academic Press, London, p. 11-25
1966.
2. K.L. Chopra. "Thin film Phenomena" McGraw Hill, N.Y. p. 10-82
1969.
3. B.N. Chapman and J.C. Anderson. "Science and Technology of
surface coatings".Academic Press, London. 1974.
4(a) P.J. Harrop and D.S. CampbelL In "Handbook of Thin Film Tech-
nology" (Ed. L.r. Maissel and R. Glang) McGraw Hill, N.Y.
p. 16.29-16.33. 1970.
4(b) A. Pliskin, D.R. Kerr, J.A. Perri in "Physics of Thin Films"
(Ed.G. Hass and R.E. Thun)i, p. 257-270. 1967.
5. R. Glang. In "Handbook of Thin Film Technology" (Ed. L.r.
Maissel and R. Glang). McGraw Hill, N.Y. p. 1.3-1.30. 1970.
6. D. Walton, T.N. Rhodin and R. Rollins.
J. Chem. Phys. 38, p. 2695. 1963.
7. D.S. Campbell. Present Volume, 1975.
8. R.E. Honig, R.C.A. Review 23, p. 567. 1962.
9. R. Glang, R.A. Holmwood and J .A. Kurtz. In "Handbook of Thin
Film Technology" (Ed. L. r. Maissel and R. Glane) Mc Graw HilL
N.Y. p. 2.1 - 2.142. 1970.
10. As per 5. p. 1.37
11. Ibid. p. 1.86
12. Ibid. p. 1.87 - 1.88
13. K.H. Behrndt. J. Appl. Phys. 33, p. 193. 1962.
14. P. Huijer, W.T. Langedam and J.H. Laby. Philips Tech. Rev. 24
p. 144-149, 1963.
15. C.E. Drumheller. Trans. 7th Nat. Symp. on Vacuum Technology.
p. 306-312. 1960.
16. D.S. Campbell and B. Hendry. Brit. J. Appl. Phys. p. 1719-
1725. 1965.
46 0.5. CAMPBELL
17. J.L. Richards. In "The use of thin films in Physical Exami-
nations". (Ed. J.e. Anderson) Academic Press, London. p. 71-
86. 1966.
18. As per 5. p. 1.92-1.97
19. I. Ames, L.H. Kaplan and P.A. Roland. Rev. Sci. Inst. 37
p. 1737, 1966.
20. As per 5. p. 1.43-1.50
21. Ibid. p. 1.50-1.54.
22. B.A. Unvala and G.R. Booker, Phil. Mag. 1, p. 691. 1964.
23. R.F. Bunshah. In "Science and Technology of surface coatings"
(Ed. B.N. Chapman and J.C. Anderson) Academic Press, London
p. 361-368. 1974.
24. As per 5. p. 1.80-1.85.
25. M. Kaminsky. "Atomic and Ionic Impact Phenomena on Metal
Surfaces" Springer-Verlag. Berlin. 1965.
26. L.r. Maissel. In "Physics of Thin Films" (Ed. G. Hass and
R.E. Thun) Academic Press, N.Y. l, p. 61-129. 1966.
27. L. r. Maissel. In "Handbook of Thin Film Technology" (Ed. L. r.
Maissel and R. GIang) McGraw Hill, N.Y. p. 4.1-4.44. 1970.
28. G.K. Wehner and G.S. Anderson. In "Handbook of Thin Film
Technology" p. 3.1-3.38. 1970.
29. As per 27. p. 4.26-4.31
30. Ibid. p. 4.21-4.22
31. L. 1. Maissel and P. Schaible, J. App1. Phys. 36, p. 237. 1965.
32. As per 27. p. 4.31-4.39.
33. N.F. Jackson. J. Mat. Sci. ~ p. 12-17. 1967.
34. J.R. Ligenza. J. App1. Phys. 36, p. 2703. 1965.
35. J.H. Greiner. J. Appl. Phys. 45, p. 32. 1974.
36. L.V. Gregor. In "Physics of Thin Films" (Ed. G. Hass and
R.E. Thun). Academic Press, N.Y. l, p. 131-164. 1966.
37. A. Bradley and J.P. Hammes. J. Electrochem. Soc. 110, P .15 1963.
PREPARATION METHODS FOR THIN FILMS
38. H.F. Sterling and R.C.G. Swann. Solid State Elec. ~ p. 653.
1965.
47
39. R.A. Connel and L.V. Gregor. J. Electrochem. Soc. ~ p. 1198
1965.
40. D.S. Campbell. In "Handbook of Thin Film Technology"
(Ed. L.I. Maissel and R. GIang) McGraw Hill, N.Y. p. 5.1-5.25.
1970.
41. G.M. Mattox. Electrochem. Tech. ~ p. 295. 1964.
42. As per 40. p. 5.21
43. Ibid. p. 5.17-5.20
44. C.J. Dell'Oca, D.L. Pulfrey and L. Young. In "Physics of Thin
Films" (Ed. M.H. Francombe and R.W. Hoffman) Academic Press,
N Y. ..' p. 1 _ .. 79 1 971
45. D.S. Campbell. Rad. and Elec. Eng. ~ p. 5-16. 1971.
46. W.M. Feist, S.R. Steele and D.W. Ready. In "Physics of Thin
Films" (Ed. G. Hass and R.E Thun) Academic Press. N.Y. 2"
p.237-322. 1969
47. As per 40. p. 5.12 - 5.16
48. D. 1. Gaffee. In "Thin film microelectronics" (Ed. L. Holland)
Chapman and Hall, London. p. 260-270. 1965.
49. P. White. Elec. Reliab. and Micromin. ~ p. 161. 1963.
50. L.L. Alt, S.W. Ing Jnr. and K.W. Laendle. J. Electrochem.
Soc. ~ p. 465. 1963.
51. F.A. Lowenheim. "Modern Electroplating" John Wiley. N.Y.
1965.
52. K.R. Lawless. In "Physics of Thin Films" (Ed. G. Hass and
R.E. Thun) Academic Press, N.Y. 4. p. 191-225. 1967.
53. As per 40. p. 5.3-5.8
54. A. Brenner. "Electrodeposition of alloys" Volland 2.
Academic Press. N.Y. 1963.
55. W. Goldie. "Metallic Coating of Plastics" Electrochemical
Publications Ltd., U.K. l, p. 55-152. 1968.
48 O.S. CAMPBELL
56. As per 40. p. 5.9-5.11
57. M. Schlesinger. In "Science and Technology of Surface Coatings"
(Ed. B.N. Chapman and J.C. Anderson). Academic Press, London.
p. 176-182. 1974.
58. K.M. Gorbunova and A.A. Nikiforova. "Physical principles of
nickel plating". Israel Prog. for Scientific Translations.
Jerusalem. 1963.
59. H. Scroeder. In "Physics of Thin Films" (Ed. G. Hass and
R.E. Thun) Academic Press, N.Y. i, p. 87-141. 1969.
60. L. Holt. Nature 214 p. 1105. 1967.
61. R. Carpenter. In "Science & Technology of Surface Coatings"
(Ed. B.N. Chapman and J.C. Anderson) Academic Press, London
p. 393-403. 1974.
62. M. Martini. In "Science & Technolgy of Surface Coatings"
(Ed. B.N. Chapman and J.C. Anderson) Academic Press, London.
p. 404-410. 1974.
63. E. Krikorian and R.J. Sneed. In J. Appl. Phys. 37 p. 3674 ..
1966.
64. E. Hollands and D.S. Campbell. In J. Mat. Sci. 2, p. 544-552.
1968.
65. L.R. Dawson. In "Progress in Solid State Chemistry".
(Ed. H. Reiss and J . ~ . McCaldin). Pergamon Press, Oxford,
p. 117-139. 1972.
66. J.A. Koursky, A.G. Walton and E. Baer.' J. Polymer Sci. B.5
p. 177. 1967.
67. A.C. Rastogi and K.L. Chopra. Thin Solid Films ~ p. 187-
200. 1973.
GROWTH PROCESSES
R. Niedermayer
Ruhr Universitat Bochum, Experimentalphysik 4
463 Bochum Postfach 2148, Germany
I. INTRODUCTION
The growth of thin films has attracted considerable attention
during the last years, and several reviews of the field as well as
many original contributions have been published. For a recent revi-
ew, which contains all literature, the article of J.A. Venables and
G.L. Price(l) should be consulted. It is not the aim of this series
of lectures to repeat this work. I shall try to make clear the fun-
damental connection of statistical and kinetic considerations,
which govern the description of growth processes. Though this is not
entirely new, it has been neglected to a certain extent in the lite-
rature. The reason for this is the large weight of the difficulties,
which have been encountered in the solution of the system of kinetic
equations, which have however been overcome now.
The following lectures will consequently be divided into three
parts :
I. the connection between material constants and the rates of
growth and decay of clusters. This is achieved mainly by thermoay-
namic considerations, from the results of which the constants of
the kinetic equations can be derived and which allow a judgement of
the general direction of growth ;
2. methods for the solution of the kinetic equations of film
growth. These methods are new to a certain extent and allow the ea-
sy solution of large systems of almost linear differential equations
by numerical methods ;
3. experimental expectations from these theoretical considerations
and comparison with the evaluation of some experiments
49
50 R. NIEDERMAYER
2 THERMODYNAMIC CONNECTIONS BETWEEN THE PROPERTIES OF SUBSTRATE
AND FILM AND THE RATES OF GROWTH AND DECAY OF CLUSTERS.
2.1. Formulation of the problem.
The formulation of the kintetics of film growth follows essen-
tially the same lines as that of polymerisation of macromolecules
from single atoms. If Bi is the symbol for a cluster of i atoms,
processes of the kind
(1)
are interesting. The general expectance is that the main contribu-
tion to growth comes from the monomers (k=I)(2,3) and so the subse-
quent work will be confined to this case. Some observations and the-
oretical considerations show however, that migration of misaligned
large clusters occurs and is important for the growth of nonepita-
xial thin films(4).
The description of growth will use the following parameters
the impingement rate of monomers on the substrate
Jimp
Jtwin
J.
1-
the rate of twin formation BI + Bl .+ B2
the capture rate of a cluster Bi leading to Bi + Bl + Bi+1
the desorption rate of monomers from the substrate a
the decay rate of Bi leading to Bi + Bi-I + Bl
the size of the last subcritical cluster, a natural number
defined so that Ji/ai 1 Ji +l/ai*+ 1
the number of clusters Bi per unit area, containing i monomers
the total number of supercritical clusters
the total number of condensed atoms
Nc and Na can be defined in terms of ni
00
Nc L
i=i *+ 1
00
Na I
i=i*+
n
1-
(2)
(3)
Definition (3) states that an atoms will be considered conden-
sed, if it has joined a supercritical cluster. The use of these va-
riables allows a description of the in a set
of equations which can be found in Frenkelts which have
been used by many authors in chemical kinetics.
n2
n.
1-
J
twin
- (a
2
+ J2)n2 + a3 n3
Ji-l n
i
- (ai + Ji)ni + ai+lni+l
(4)
(5)
GROWTH PROCESSES 51
for 3 .$. i .$. i*+ m, where m is so defined that ui *+m Jj!r +m These
equations contain the main formulation of the problem. For the com-
parison with experiment the numbers Nc and Na are necessary, which
may exactly be derived from (5) by use of the definitions (2,3) :
(6)
00
(7)
The total number of monomers on the free area Af of the unit area
substrate can be derived from
00
NI = AfJimp- u.N
I
- 2J
tw
in+ u2 n2 - L J
i
- ui)ni) (8)
1.=2
Equations (6, 7, 8) can in principle be obtained by integration, if
the solutions of equs. (4,5) are known. Direct impigement onto nu-
clei and coalescence of islands is not yet incorporated in these
equations and will be considered separately. The infinite sums in
(6, 7, 8) can be approximated by finite expressions containing N
c

The main system contains the rates of capture and decay, ~ v h i h
depend on the size of the clusters and on their thermodynamic pro-
perties. The equations are coupled by the terms ui+l.ni+1 which
cause the main difficulty in the solution of this system, and which
discern a nucleation theory from a simple theory of unidirectional
growth. The solution can be given numerically or by analytical ap-
proximations. The results depend on the transport mechanism con-
tained in the capture probability. Statistical and thermodynamic
arguments for the evaluation of the rates of capture and decay will
be presented in the following paragraphs.
2.2 Some principal considerations on the capture rate on a surface
and on surface diffusion.
The difficulty in the treatment of atomic diffusion on surfa-
ces is the combination of short characteristic times and very low
concentrations. Therefore the question arises what meaning a diffu-
sion mechanism could have in such a system Another difficult ques-
tion is, how the boundary conditions at a cluster edge could be for-
mulated, if the cluster is small and if it emits and absorbs only
single atoms in times which are certainly long compared to the time
necessary for the diffusion from one site to the next one.
This will be illustrated in the following statements :
I. The number of adsorption sites is typically 5.10
14
cm-
2
and
will be denoted lad.
2. The impingement rate in an experiment may be J
im
=I013 cm-
2
s-
l
;
this means that the time for monolayer occupat1.on is very long,
52
3.
4.
salt.
R. NIEDERMAYER
i.e. one minute.
The typical diffusion constant is D = 10-
5
cm
2
s-l ; the corres-
ponding stay time in one site is Tdiff = 10-
1
0 s.
The desorption rate is approximatively a= 10
7
s-l, the corres-
ponding stay time is Tdes = 10-
7
s.
The data for point 3 and 4 have been chosen for gold on rock-
5. The decay rate for gold twins may be a2 = 10
2
s-l and for clus-
ters AuS as = 1 s-l ; T2 = 10-2 s, TS = 1 s. Whenever the bin-
ding energy to a cluster is larger than the binding energy to
the surface similar orders of magnitude will apply at modera-
te temperatures.
6. The capture rate under the same circumstances is J2= 10
3
s-l;
T = 10-3 s.
c .
Obviously all changes at one cluster are very slow compared to
the characteristic times for desorption and diffusion. So the ques-
tion is, whether the few clusters and the few single atoms interact
at all in the sense of a diffusion mechanism with well defined boun-
dary conditions as has been proposed by Halpern(7) and other authors
(8), or if the picture of homogeneous monomer concentration as f.i.
used by Walton(9) is more reasonable.
The following arguments will show, that the diffusion equation
applies to the system for all cluster sizes and for all stages of
growth. Take a small surface with an orthogonal net of adsorption
sites (f.i. NaCI (100)), which may be numbered by their coordinates
(x
n
, Ym). The surface be A, the number of monomers be NI(t), their
concentration nl. This number is increased by the impingement rate
Jimp and by the decay rates ai of clusters; it is decreased by the
desorption rate a and by the capture rates Jtwin, Ji. For the judge-
ment of the changes in such a system the observation of the environ-
ment of a single cluster is useless. We rather choose a Gibbs ensem-
ble of NG identical surfaces, where NG is a very great number of the
order of lad' We divide the time into intervals ~ t Tdiff and we
observe all the systems at each interval. We are then at least in
principle able to count the number of atoms located at a site (xn,Ym)
in the whole ensemble at each time interval, i.e. II(Xn,Ym,t), and
to find the probability p(xn,Ym,t) = II (xn,ym,t)/NG at each moment.
This procedure is not as artifical as it seems to be, because it is
quite analogous to the real experiment, where we observe a macrosco-
pic surface with many clusters and divide the surface into environ-
ments of similar clusters.
If the jump rate from one site in any direction is w = l/Tdiff'
we obtain for a time t + ~ t a slightly different distribution
11(t + ~ t from that at time t, 11(t). From this we can again find
pet + ~ t , which is at a sufficient distance from a cluster edge
GROWTH PROCESSES
p (xn, Ym, t+ilt) P(Xn,Ym,t) - w.ilt.p(xn,Ym,t)
+ w
4
ilt. (p(Xn+I,Ym, t)+P(xn-1 ,Ym, t)
+ p(xn,Ym+I,t)+p(xn,Ym-l,t)
+ J
imp
a
2
.ilt - a.ilt,p(Xn,Ym,t)
53
(9)
The first line represents the probability, that nothing has
changed. The second line is the probability, that an atom from one
of the neighbouring sites jumps into the observed site (n,m) with a
probability {w.ilt)/4 for a jump in one direction. The last line ac-
counts for the impingement and desorption, where a2 = I/lad is the
area occupied by one adsorption site and the prompt sticking coeffi-
cient is assumed unity. Equation (9) is a diffusion equation
ap a2 a
2
at = Jimp.a
2
- a.p + (w.a
2
/4). (ax2 + ay2)p
(10)
The diffusion constant D is related to the jump probability w
by D = w.a
2
/4.
Whatever the single atom will do at the surface, it has no other
possibilities than those enumerated in eq.(9) each of which will be
realized at a certain rate, which must be determined later. An en-
semble of atoms will then behave quite orderly and the probabilities
of occupation will change according to (10). To obtain the boundary
conditions we observe the number of monomers 11 in the immediate en-
vironments of clusters Bi, which may have an edge length s(i), i.e.
the number of adsorption sites occupied by these edge atoms is appro-
ximately s(i)/a. The rate of generation of an atom from one of the
edge sites is ai.a/s(i) if ai is the decay rate of the whole cluster.
We consider now the situation whene (xn,Ym) is !b + ga, i.e. one step
apart from the boundary position !b,g being the unity vector perpen-
dicular to the boundary, and where !b = (XU-I;Ym)' We find in the
same way as in equation (9)
P(Xn,Ym,t +ilt) = p(xn'Ym,t)-w.ilt.p(xn,Ym,t)
+Jimp.a
2
.ilt-a.ilt.p(x
n
,Ym,t)
+ (w.ilt/4),(p(Xn+I,y
m
,t)+P(Xn'Ym+I,t)
+P(Xn,Ym-l,t
+ (aia/s(i.ilt
(II)
If we formally introduce a boundary probability Pb(xn-I'Ym,t) we can
obtain the diffusion equation (9,10) again from (II), provided we
choose
(12)
54 R. NIEDERMAYER
The connection between the decay probability of a cluster Bi
and the boundary probability, which must be introduced into the dif-
fusion equation is thus given by
(13)
The definition of Pb is such, that a probability of occupation
Pb(Xn-I,Ym,t) would produce the same flux as is procured by the de-
cay rate ai. This means also, that a cluster Bi, which is surroun-
ded by a homogeneous concentration Pb/a2 would statistically receive
and loose an equal number of atoms. Under such conditions the pro-
cesses Bi + BI + Bi+l and Bi + Bi-I + BI would be equally probable.
This observation will be important for the calculation of the decay
probabili ties.
In the frame of these considerations the capture rate is given
by the probability of occupation of an adjacent site, s(i)+ 2TIa)/a).
times the transition rate (w/4).(s(i)/(s(i)+2TIa
J.



(14)
Our statistical interpretation of this capture rate is just
sufficient for the validity of its use in eqs. (5 to 8). Equation
(14) can be rearranged by use of the diffusion constant, D = w.a
2
/4,
and of nl(E,t) = P(E,t)/a
2
, the local monomer concentration that
must be interpreted statistically in the sense of this paragraph.
If we further denote the boundary monomer concentration at a cluster
Bi by 0i= Pb/a2, we obtain
J.

(15)
for the capture rate, which must be calculated with the result of
the solution of the diffusion equation (10), and with a determina-
tion of the boundary monomer concentration 0i.
4 .. 3 . The general solution of the diffusion equation and evaluation
of time constants.
The diffusion equation
= J. . a
2
+ D 'p - '" p
Clt .D u..
(10)
with the boundary condition Pb = 0i.a2 must be solved for possibly
time dependent boundary conditions, because 0i varies with i. The
problem has cylindrical symmetry in two dimensions, which results
in
GROWTH PROCESSES 55
an __ 2 a2p 1 an
at Jimpa - a.p + + r .
(16)
Choice of s .. p - a
2
.J
imp
/a, p = s(r,t)+ a
2
.J
imp
/a results in
the homogeneous
s = - a.s + D.(s" + s'/r) (17)
Separation, s = v(t).u(r), gives equations for any w
-iT = Lw.v
(18)
u" + (u' /r)-a+ Lw)/D).u = 0
the solutions of which can be superposed to find a physically ade-
quate expression. Such particular solutions are
iwt
v = vo.e
co
ul (r,w) = f exp(- f(a + LW)!D '.r.coshx )dX (19)
o
u2(r,w)
=
_
I f1T
1T 0
exp(+f(a + iW)!D: r.cos X) dX
ul is an extension of the modified Bessel function Ko and u2 of 1
0
,
Because u2 diverges for large r, it will not be used for the simple
discussion of this paragraph although it would be necessary in the
presence of more than one cluster within a 7.fD/a of the one
just under consideration. Evaluation of results in
a particular solution
,(r. t) - [ exp [ -V,I";D w" a '. r. co,1t X]. (20)
sin ( wt - .r.cosh
X
). dX
which can easily be discussed. Be the boundary conditions such that
they vary with /.1lt or that the main Fourier component of their varia-
tion has the frequency w. The main contributions come from small va-
lues X. s vanishes exponentially for large r and the range is
given by
(2D/(fa2 + a 1/2
(2J)
The influence of the variation of the boundary condition has a
56 R. NIEDERMAYER
frequency dependent range, which is approximately equal to the ran-
ge ro = ID/a of the stationary diffusion equation if w < a, and
which approximates the range of purely diffusion controlled distri-
butions, rd = 12D/w , if w > 2a.
The perturbations proceed in form of circular waves with a
wavelength A and a corresponding velocity of propagation vdiff=
We find for the main contributions
A = [h.D/ (/a
2
+ w
2
(22)
For fast variations of the boundary conditions (w> 2a) this gives
Af = = In this case we have a real wavelike propaga-
tion of the perturbation.
If the variations are slow(w<a/2) we arrive at
AS = D/a= The range, within which the per-
turbation is in phase with the boundary concentration, is approxi-
mately = (a/w).r
o
Within this range the
stationary solution applies.
In all physically important cases the stationary solution is
a good approximation. Wheneverthe monomer concentration at some
distance from the nuclei is mainly determined by the desorption
equilibrium, nl = Jimp/a, we expect the second case and the vali-
dity of the stationary solution, because the coherence length
is larger than the attenuation range ro of the variations. In the
capture regime, Tdes> the first case should be expected. But
in this case the boundary condition is nearly Pb = 0 even for twins
and thus does not change with time. So w is very small and we can
again expect the validity of the stationary solution. An artificial
case, which contains a maximum of short wavelength components, is
the case of homogeneous monomer concentration. A comparison between
the stationary solution of the diffusion equation and the homoge-
neous monomer concentration may be useful.
2.4 The boundary monomer concentration and the decay rate of a
cluster
The discussion of the equations (12,13) has shown, that a
cluster Bi would grow and decay with equal probability, if surroun-
ded by the boundary concentration Pb/a2 = 0i' If the clusters were
a separated phase, the monomers would be in equilibrium with the
condensed phase, the molecular free energies would be equal for mo-
nomers and condensed phase, the total free energy would not change
from a transition of a molecule from one phase to another.
A cluster is however not a separated phase with a constant
chemical potential. This has the consequence that a system, in
GROWTH PROCESSES 57
which a cluster Bi is in equilibrium with the surrounding monomers
in the sense mentioned, is not stable. Quite contrarily such a sys-
tem will start to condense, because the saturation concentration
for small clusters is greater than for large clusters. Only a sys-
tem, which has a very low monomer concentration, i.e. which is sub-
saturated, so that even in the presence of a condensed phase no con-
densation would occur, does not grow. Such a subsaturated system
can form very few clusters by fluctuations, the relative number of
which can be calculated by the mass action law. Though this contains
the decay rate implicitely, their evaluation(5) can lead to diffi-
culties with our present treatment of the transport process and
we shall use another more general procedure for the determination
of the boundary concentrations Oi, which fits into the frame of the
theory of diffusion transport and which leads to the decay probabi-
lities ai via eq. (]3).
(i) Under the conditions of equal growth and decay the occupation
probability must be equal to p(r )t = 0i.a2 and so the
monomer concentration around the cluster be homogeneous.
(ii) Equal probability of growth and decay means that the change in
free energy for both transitions is equal. Thus the condition for
the existence of a saturation environment is, that the Helmholtz
free energy of formation of a cluster has an extremum
= 0 (23)
which is a maximum because of the pos1t1ve surface energy term in
the cluster binaing energies. The Helmholtz free energy is chosen
because cluster formation on a constant surface is considered ra-
ther than under conditions of constant monomer concentration.
The free energy of the cluster Bi is given by the partition
function Zi
F(Bi) = -kT.ln Zi (24)
and the free energy of N] noninteracting monomers on the surface A
can be calculated from the partition function of such a system
(25)
where z] is the single particle partition function for unity area.
We find thus
(26)
The free energy for the formation of a cluster, is the
difference between F(Bi) + F(N],B]) and F(N1+i,B1), which
1
gives for
58
sufficiently large N
1
, i.e. Nl i :
F(Bi) + F(Nl,B
1
) - F(Nl+ i,B)
kT.(lnZi - i.lnzl + i.ln(NI/A
R. NIEDERMAYER
(27)
The last term is the monomer concentration nl' which may be very
low provided the area A is large enough. The condition (23) permits
the determination of the monomer saturation concentration 0i
(28)
0i ~ zl/zi-l/Zi+l
The latter expression uses a In Zi/ ai ~ (InZi+l - In Zi-l)/2 and
is related to a similar expression in (5), which was however deri-
ved by use of the mass action law and detailed balance of growth
and decay in a sur saturated homogeneous monomer concentration. As
we see here it does not depend on such arguments, which are diffi-
cult to confirm in a system with diffusion transport.
The decay rate ai must be connected with the saturation ~ o n e n
tration by (13)
(29)
The actual evaluation of the quite general relations (28,29) depends
on the model for the cluster and for the adsorbed state, which allow
evaluation of the partition functions. Harmonic models have been
used with relatively good success, though this is open to conside-
rable improvement.
2.4.1 The model of adsorbed monomers.
The adsorption state is characterised by the number of equiva-
lent adsorption sites lad, the desorption energy Edes, the mechani-
cal desorption frequency vodes, the diffusion energy Ediff' the me-
chanical diffusion frequency vodiff, and the mass m of the adsorbed
atom. Double occupation of the same site and interaction between
monomers are unimportant for low concentrations nl. For the case of
fairly strong binding to the surface (Ediff ~ 4 kT) one can find(IO)
"the unity area monomer partition function at temperature T
The interpretation is simple : lad harmonic oscillator sites with
frequency Vodes and vodiff perpendicular and parallel to the surfa-
ce plane are bound with Edes and constitute all possible states of
GROWTH PROCESSES
an adsorbed atom. If the binding to a state become weaker (Ediff
< 4 kT), locations between the sites become important and zl will
be more complicated. Very weak binding (Ediff < kT/4) will lead
to the expression for a two dimensional gas.
For the strong binding case, the desorption rate is given by
(31)
with
(32)
The diffusion constant is in this case
(33)
with
(34)
S9
The difference between mechanical and effective frequency constants
is caused by entropy terms.
2.4.2. The cluster model.
The cluster energies which are necessary for the evaluation o ~
the cluster partition functions are composed of kinetic and poten-
tial parts. A discussion of these energies will shed some light on
the relations between the modes of growth and the different contri-
butions to the energy of a cluster. The energy will again be given
in a harmonic approximation and can be represented for a cluster
Bi by
(i-I)
Vi = Vio + (1/2)3I n m u n ~
(35)
n,m
Vio is the potential energy in the equilibrium position of the clus-
ter Bi, i.e. its binding energy. The coordinates un are the devia-
tions from the equilibrium position, the second term is thus the
elastic energy of the cluster. This elastic energy and the corres-
ponding momentum terms will finally result in harmonic oscillator
contributions to the partition function and to preexponential terms
in the decay rate. The most important and interesting part is the
binding energy.
The real binding energies for large clusters cannot be deter-
mined from first principles. Phenomenological or half experimental
approximations can however give the general dependence of the bin -
60 R. NIEDERMAYER
ding energy on size, shape and position. One such approximation is
Vio = - i. A + (36)
where A is the energy of evaporation and is a correction term
for the different binding of atoms at the cluster surface, which
for very large clusters approaches the normal surface energy. It
has been shown(ll) that approximations of the form (36) agree sur-
prisingly well with atomic calculations even for very small clusters.
The term contains both the weaker binding of atoms at the surfa-
ce to the vacuum and the influence of the binding forces of the sub-
strate at the interface. Forces which lead to epitaxial growth like
those calculated by van der Merwe and other authors(12,13,14,IS,S)
have to be incorporated into the surface energy in a general theo-
ry of growth. These questions are quite complicated in detail and
may contain many surprises. For a general survey however some simple
rules are helpful.
So it can be shown from a collection of data(S),
that the surface energy per surface atom, cr.v
2
/
3
, with v the atomic
volume, is given by
cr.v
2
/
3
= A/6 (37)
for most metals. For many other materials a similar relation hods
(38)
where za is the number of neighbours in an adsorption site and Zw
is the number of neighbours in a growth site(16).
The experimental connection between evaporation energy and
surface energy is shown in Fig. I. On the basis of this result it
can be inferred, that the energy of separation of a cluster from
its substrate would be nearly equal to the desorption energy for
each interfacial atom, if these atoms were not pulled out of their
neighbours. This will cause a correction, which depends on the dif-
fusion energy, on the misfit, and on the elastic constants of the
cluster material. It cannot be larger than the diffusion energy.
Calculat{ons for a variety of simple models indicate, that this cor-
rection can be characterised by a parameter n with 0.4 < n < 0.8.
The separation energy per interfacial atom is thus
E = E - n Ed"ff
sep des .
(39)
For very small clusters or for vanishing misfit n tends to zero(14).
This relation should be applicable to both cluster growth and layer
growth(S,IS). An application of (39) can give a criterium for the
mode of growth and a prediction of the cluster shape(17,5). It will
5

[
1
0
4
J
m
O
r
f
]

,
,
"

H
n

L
u

"
p
I
"

/
Y

'
e
G
O

G
d
.

/
.
.
"

e
u
i
/

N
{
,

/
'
"

"

'
I
t
/
u

L
.

.
.
.

t
o
"
,

.
.
"
;
,
g
}
\
;
;
'
8
o

'
-
1
1
-
-
-
-
-
-
-

C
d

1
f
t
'

G
o

G
e

T
e

S
b

K

$

S
e

A
s

I

7
0

8
0

9
0

[
1
0
"
)
m
o
l
.

-
1
]

1
0

2
0

4
0

5
0

6
0

3
D

F
i
g
.

I
.

D
e
p
e
n
d
e
n
c
e

o
f

t
h
e

e
x
p
e
r
i
m
e
n
t
a
l
l
y

o
b
t
a
i
n
e
d

m
o
l
a
r

s
u
r
f
a
c
e

e
n
e
r
g
i
e
s

I

o
n

t
h
e

h
e
a
t
s

o
f

e
v
a
p
o
r
a
t
i
o
n
(
5
)
.

G
')

:
0

:
I
:

"

:
0

o

(
"
)

m

m

C
/
J

62 R.NIEDERMAYER
finally enable us to calculate the surface energy term
The separation of a cluster from the substrate will eliminate
the interface with its interfacial energy v
2
/
3
.Oif per atom and
create new surfaces of the substrate and the cluster, the respecti-
ve energies of which are v
2
/
3
'Osub and v
2
/
3
'O
c
l' This gives
(4Q)
Eqs. (37, 39, 40) can be combined to relations between surface ener-
gies and binding energies
(41 )
for metal clusters
The shape of the cluster will be near the shape of minimum
surface energy. Detailed considerations of surface energies can
lead to very precise.statements on the shape of clusters (9,18). For
a less detailed information as spherical caps, cylindrical or
rectangular prisms have been widely accepted. Optimal shapes
these models can be calculated quite easily, the only shape para-
meter is the relation between the h and the square root of
the interfacial area of the cluster(1 ,5), i.e. for a cylindrical
prism the proportion between the height and the diameter d
(h/d)opt = (ocl + 0if - 0sub)/2o
cl
- - n.Ediff)/Acl
(42)
(43)
The latter equation results from the use of (41). These relations
permit a statement on the mode of growth of thin films from the
knowledge of the surface energies or of the binding energies : if
> 0 we expect three dimensional erowth, if 0 two dimensional
layer growth will occur. Experimental confirmation has been found
for these relations by several authors(15,19,20).
A consequent persecution of these ideas allows the determina-
tion of the surface and edge contributions to the binding energy
(36). In the two dimensional case we obtain
Vio2 = -(2/3).i.A - i.(E
des
-
(44)
and in the three dimensional case
GROWTH PROCESSES
V. 3=-i.A +

'" -i.A+(i.A)2/3. (A-3. (Edes-Ediff))) /3_ iaiff.Ediff(l -n)
(45)
63
i'
f
is the number of interfacial atoms, which can be determined
ffom (43) to (i.IIT/2s)2/3. All interaction with the substrate is
contained in the last term. The binding energy is strong, if n is
small. Good fit, large diffusion energy, small elastic constants
of the cluster material tend to keep the atoms in the adsorption
sites and to diminish n. The good fit depends on orientation and
position of the clusters and also on their size. If we denote the
binding energies (44,45) by Vio = V(i,n) we see from these equa-
tions, that the dependence on orientation and position is given by
and it is in this way, that epitaxial growth is caused by the inter-
action with the substrate(5). The dependence of the potential ener-
gy on orientation and position will not be treated further here,
because the analysis of epitaxial growth is not the aim of these
lectures.
The calculation of the function requires the know-
ledge of the potential energy, eq. (35), and of the kine-
netic energy parts. The kinetic energy leads together with the elas-
tic parts of the potential energy to the phonon states, the parti-
tion functions of which approach kT/hv for high temperatures. The
frequencies V vary in principle over the phonon spectrum of the
cluster, for an approximate treatment however only a mean vibration
frequency vvib' the translational frequencies of the whole cluster
in the surface plane vs, and its rotational frequency are expli-
citely taken into account. This leads to a partition function of
the cluster Bi
Zi =
(47)
The first term M is the number of equivalent and different po-
sitions, which a cluster can assume on a periodic substrate of area
A. The following terms are the Qscillatory partition function and
the binding energy term. Eq. (47) can evidently be specified to
include the dependence on orientation and position(5).
2.4.3. Evaluation of the boundary monomer concentration and
the decay rate for the described model.
From eq. (28) we need alnZi/ai and lnz) for the evaluation of
64 R. NIEDERMAYER




10 100 1000
Fig.2 Dependence of saturation monomer concentration 0i on cluster
size for Au on NaCl for substrate temperature of SOOoK and
714K
GROWTH PROCESSES
the boundary monomer concentration 0i:
aln Zi/ai = 3. In(kT/hvvib)-(I/kT).aV(i,n)/ai (48)
From (44,45) we obtain
aVi02/ai = -2/3}A+ Edes- nEdiff)+ (1/3). A/Ii
(49)
aVi03/ai=-A+ (2/3).A.(sTI/4)1/3. i -I/3
Together with eq. (30) this gives in the three dimensional case
0' ladd'
V
vib
3
e x ~ E d -A+ f:...(STI/4)1/3. i -I/3)/kT)] (50)
1 V d V d'ff3 es 3
o es 0 1
and the decay rate can be calculated from this by
a.'
1
sCi) D 0'
a . 1
(13)
2 .5. Stationary solutions of the diffusion equation and capture
rates
65
The capture rate is statistically defined in the sense of 2 .2
and we have found an expression for Ji
(15)
For the final evaluation of the capture rate a knowledge of the con-
centration gradient at the cluster boundary, r = Pi, must be obtai-
ned from a solution of the diffusion equation. In 2 .2. we have
seen, that no limitation to the statistical application of the dif-
fusion equations (9,11) exists and that these are equivalent to (16)
whithin the limits of a linear approach. We have further seen in
2 .30 that the stationary solution with the boundary condition
nl(Pi) = 0i should be adequate even for very small clusters, because
the change of the cluster size is usually compared to the desorption
rate.
The solutions (19) for w=O are directly applicable to our pro-
blem, resulting in UI (r,O)=Ko(r!lD/a. ), u2(r,0)=I
o
(r!lD/a.). We thus
obtain for the stationary monomer concentration
(51)
with the diffusion range rl
(52)
66 R. NIEDERMAYER
The derivatives of 10 and Ko are II and -KI respectively and we
have for the capture rate
The coefficients A and B must be determined by the boundary condi-
tions.
From eq.(6) we have Nc clusters of all sizes on unity area.
The mean area per cluster is
I/Nc (54)
which defines the radius rNc of the outer boundary. The cluster
radius is Pi. For the boundary conditions the monomer concentrations
nl(Pi) = 0i and nt(rNc)will be chosen. Another choice would be
nl(Pi) and (dnl/dr)rNc = 0 which is certainly reasonable if all
clusters were of the same size or if the monomer concentration va-
nishes at the edge of each cluster (21). We have however a mixture
of different cluster sizes, some of which are subsaturated,Jimp/a<oi,
some are supersaturated, Jimp/a>oi, so the assumption of zero flux
at the outer boundary is not applicable. This is physically evident
from the flux which goes from the subsaturated to the supersatura-
ted clusters. .
These considerations lead to the following procedure : the
outer boundary concentration is first determined by the zero flux
condition for an assembly of average clusters. With this average
outer boundary condition the final solutions, i.e. the coefficients
A and B, are obtained.
The average cluster has a size
(55)
which leads to a mean radius from eq. (42) with atomic volume v
Q (56)
and to a mean monomer saturation concentration ;S =-' 0-:-. These clus-
ters are supercritical and uniform from the definition of N
c
N
a
,
so that Stowells boundary condition of zero flux(21) can be applied.
8 = Jimp/a+ A.lo(p/r
l
)+ B.Ko(p/rl)
b = A.II(rNC/rl)-B.KI(rNC/rl)
This can be resolved to
(57)
(58)
GROWTH PROCESSES
with
A =
B =
Det = Kt(rNc/rt)Io(p/rt)+ Ko(p/rt).It(rNC/rt)
(59)
(60)
The monomer concentration at the outer boundary is thus given by
ntB = nt(r
Nc
) = Jimp/a + (5-Jimp/a).(A.lo(rNC/rt) +
+ B.Ko(rNc/rt (6t)
67
For large rNc > tOrt) the second term vanishes exponentially
so nt (rNc)+ Ji la. This maximum possible monomer concentrations
is reduced by dTfusion for smaller distances For the monomer
concentration around the average cluster this boundary condition
is of course identical with Stowells. For all other clusters, sub-
critical or supercritical, it has however the_advantage that the
size of the diffusion courts depends on the cluster size and that
material transport from small clusters to large ones is taken into
account.
For a critical cluster we have
(62)
and for subcritical or supercritical clusters we have
< 0,
sub
o (63)
All these physical conditions at the cluster edge can be met with a
fixed average monomer concentration nt(r
Nc
) calculated from (6t).
The different possible cases are illustrated in Fig. 3.
To complete the evaluation, we use the mean boundary
tion for determination of A and B in (51) by use of the conditions
(64)
which results in
B = ntB -Jimp/a).Io(Pi/rt)-(oi-Jimp/a).Io(rNc/rt/Det (65)
Det = Io(Pi/rt).KQ(rNc/rt)-Io(rNc/rt).Ko(Pi/rl)
68 R. NIEDERMAYER
o,..... _________ r ~ ~ I .. f;-...:.Q+-i)-.. (r-Qj)
Fig. 3. Qualitative representation of monomer concentration around
clusters of different sizes with a fixed boundary monomer
concentration. The curvature is determined by
~ n l = a/D(nl-Jimp/a)
GROWTH PROCESSES 69
For large rNc this can be simplified considerably and Halperns
formula(7) is obtained
From these formulas as well as from fig. 3 an interesting
elusion can be drawn: around a subcritical cluster the monomer
centration is larger than Jimp/a in the statistical sense stated
above, for a critical cluster it is equal to this value and for
supercritical clusters it is smaller. This means, that the capture
rate for a critical cluster is equal to the value expected from a
homogeneous monomer concentration, as discussed in 2,2 and 2.4,
but that it is larger for a subcritical cluster, though less than
the decay rate, and it is less for a supercritical cluster, though
larger than the decay rate. Thus the diffusion mechanism diminishes
the dependence of the net growth rate on the cluster size, A
quence of this is, that the nucleation rate is higher in a consequent
diffusion theory, as presented here, than in a homogeneous
tration theory, and that it depends on the impingement rate and on
the temperature to a minor degree than predicted by Walton(9). It
follows furthermore- that the growth rate of stable nuclei will be
reduced in a diffusion theory. This effect has been observed
ready by Halpern(7).
2.6. The rate of twin formation
The local rate of twin formation on an area dA is
2
dJ
twin
= D(s(l)/a).nl (r).dA (67)
The free area can be into Nc regions between the inner and
outer boundary radii p, rNc ' within which approximately the same
monomer concentrations are found. Thus we have
J
twin
rNc
D.(s(1)/a).2nN
c
I ni(r).r.dr
p
(68)
This applies to any monomer distribution and is analogous to the
formula of Routledge and Stowell(21)
Because for the calculations of the solutions of the kinetic
equations (4 to 8) computer work is necessary in any case, the re-
latively complicated expression for the rate constants given here
do not add any important complication to the analysis of the system.
70 R. NIEDERMAYER
3. THE SYSTEM OF GROWTH EQUATIONS AND A METHOD
FOR THEIR ~ ~ R I C A L SOLUTION.
The growth equations (4 to 8) are incomplete in two respects :
They do not include the condensation of monomers by direct impin-
gement on top of already grown clusters and they do not account for
the coalescence of nuclei. Their solution is impeded by the infini-
te sums contained in eqs. (7.8). which should be approximated by
finite expressions. The solution of the system is not a trivial
problem. because instabilities may occur. if standard methods as
Euler. Runge-Kutta. or similar are used. The content of this chap-
ter will thus be the final formulation of the growth system and the
description of the method of its numerical solution.
3.1. Condensation by direct impingement
For direct impingement on clusters a sticking probability of
1 is generally assumed. Because of the small area of the first
clusters direct impingement is important only for large clusters.
so that it can be fully accounted for by introduction into eqs.
(7.8). We. find
(69)
and thus the last term in eq.(7) changes to
00
(70)
i
The additional terms LPI nni is the area covered by stable nuclei.
and the free area defined for eq.(8) is given by
00
- L pI nni
i = i*+1
(71 )
Although these corrections are of minor importance for the observa-
tion of nucleation. they are important for the observation of con-
densation on surfaces.
3.2. Coalescence of nuclei
Coalescence of larger nuclei can occur if the distance between
two nuclei is not greater than the sum of their radii. If the clus-
ters migrate on the substrate. they will meet occasionally within
this distance and the coalescence must be treated on similar lines
GROWTH PROCESSES 71
as the capture of monomers(2). A complication is, that the clusters
grow and decay during their migration. If the clusters are mainly
fixed at their nucleation sites, coalescence occurs by growth only.
For the calculation of the latter kind of coalescence some
knowledge of the distribution of clusters on the substrate is de-
sirable. An estimation can be given by the argument, that most clus-
ters form as long as they are relatively far apart each other,
and that under such conditions the monomer concentration around a
cluster is approximatively given by eq. (66). One result of the
presented here is, that the nucleation rate is proportional
to nl , and that w is a number between 2 and 3, which depends very
weakly on the size of the critical nucleus, contrarily to the case
of homogeneous monomer concentration, where W' = i*+l. Thus we ex-
pect in the statistical sense of 2.2, a iistribution Nc(r)
around a selected cluster, which is given by
1/1Tf
Nc(r) = Nc(nl(rw/ J
o
(72)
which is normalised so that the total number of clusters on unity
surface is N
c
' Eq.(72) results in a certain correlation of the clus-
ter sites.
If we consider a radius r = 2p, it is just the radius, within
which all clusters are situated that have already coalesced with
the central one. The coalescence rate around one cluster is thus
given by
(73)
The total coalescence rate is obtained by multiplication with N
c
/2,
so that we finally arrive at
(74)
Conditions where the homogeneous monomer concentration applies
are included in (74). Migration of the clusters will result in a
more uniform cluster distribution than (72), and can be simulated
72 R. NIEDERMAYER
qualitatively by reduction of the exponent w.
3.3. The equations of cluster growth
The number of eqs. (5) is infinite in principle. Those equa-
tions however, for which ai Ji, i.e. by which the growth of very
stable clusters Bi is described, are not coupled to the equation
for Bi+l any more, because ai+lni+l become very small. In this case
the equations for all i ~ i*+ m can be solved by simple recursion as
it has been done successfully by Zinsmeister(21), provided m is
chosen adequately. This means also, that a solution of eqs. (5) is
necessary only until i = i*+m, if one is interested mainly in the
total number of nuclei and in condensation. Eq. (7,70) contains an
infinite sum over the cluster concentration, which must be simpli-
fied if the solution of the infinite system shall be avoided. It
can be shown that
00
I (Ji-ai+pf nJimp).ni ~ N
C
.(J-
1
. a ~ + p2 nJ. ) (75)
i=i*+1 1 1mp
and we find from eqs. (13,15) that
JI - ar = D.2np.(3nl/dr)_
p
The relation (75) is always true, if
(76)
which is the case for i > i*, and in most cases (75) is a rather
good approximation.
With inclusion of direct impingement, coalescence, and the
approximation to the infinite sum we arrive at a system of equations,
which can be solved by numerical methods, and the solutions of which
are able to describe the growth of films from binding energies and
parameters of lattice geometry.
(77)
n. = J. 1-(a.+J.)n.+a.+
1
n'+
1
1 1- 1 1 1 1 1
(78)
(for 3 ~ i ~ i*+m )
(79)
GROWTH PROCESSES
Na = (i*+l)Ji*.n
i
*- i*.a
i
*+l ni*+l+Nc (JI -aI +
p2
nJ
imp
)
.*
2 ~
J. (l-N P n)-aNI-2J . +a n - I (J.-a.)n.-N (J7-a7)
~ p c t w ~ n 2 2 i=2 ~ ~ ~ c ~ ~
73
(80)
(81)
The system of equations (77 to 81) is non linear, because the
capture rates (53) depend on the mean cluster distance (54) via the
constants A and B calculated in (65,61) and on the mean cluster
radius p (56), which is used directly and for the determination of
8 in eq. (61). Eqs. (79 to 81) depend furthermore explicitely on
these parameters. This dependence is however not very strong, so
that the system can be considered linear for short time intervals.
For long times the nonlinear terms lead to (i) decrease in monomer
concentration nl B (61) and consequent ly to a decrease in all cap-
ture rates, (ii) to an increasing importance of the coalescence
term in (79,74). Both effects lead to saturation and finally de-
crease in the number of nuclei and eventually to a linear dependence
ce of condensation on time.
In special cases analytical solutions of the system are possible
and have been given in (5), for a general survey of growth processes
numerical calculations are however necessary and a method of easy
feasability has been worked out for this purpose.
3.4. A method for the numerical solution of the growth equations
As stated above, the nonlinearity of the system of eql.,1ations
is not strong, so that it can be considered linear for appropriate
time intervals.
For an induction period, tin < 1//a
2
+ 8Ds(I)Jimp/a, the system
can be easily solved by any of the normally available methods, as
Euler, Runge-Kutta or any other program for the solution of systems
of differential equations. During this period the monomer concentra-
tion builds up and a fast change of the capture rates occurs. I
have chosen a modified Euler method for this period. The calcula-
tions are made with a preselected accuracy and an automatic accomo-
dation of step width. This step width does not increase essentially
over the induction time tin and so the further progress of the cal-
culations with such a method is usually very slow, the order of ma-
gnitude of t. beeing 10-
7
s. The reason for this limitation of step
width is the1fipschitz condition.
The variation of capture rates is on the other hand appreciable
only in much larger time intervals, i+ the induction period has ela-
psed ; the same is true for the coefficients of the growth equations.
It is therefore possible to solve the equations exactly for a cer-'
74 R. NIEDERMAYER
tain time interval with constant coefficients, calculate new coef-
ficients from the solution and compare the difference between the
old and new coefficients with some prescribed limit and to adjust
the step width accordingly. The coefficients calculated at the end
of the correct time interval are used for the next interval. The
calculation proceeds with a nearly exponential increase in step
width. For the exact solution mentioned a matrix method is used.
The structure of eqs. (77,78) is given by
with
C.
1.
-J . .0. I
tW1.n 1.
ai,i_1 = J
i
else a. k 0
1.,
a ..
1.,1.
(Kronecker -0)
-(J. IT a.. I)
1.+ 1.+
a. . I
1.,1.+
(82)
(83)
(84)
Because the diagonal coefficients a .. are negative, saturation for
each n. I will occur, so that n. I ~ 1 . 0 for large times. This leads
1.T . 1.+ ()
to stat1.onary solut1.ons x. = n. I 00 and to
1.,00 1.+
n. I (t) = x. + X. (t)
1.+ 1.
00
1.
(85)
where the x.(t) are negative and vanish for long times. We thus
1.
have
X.
1.
-c.
1.
i* +m-I
+ L ak(x. 00 +
k=1 1. k
with the asymptotic condition
c
n
(86)
(87)
The solution of (87), which can be performed by a matrix inversion,
leads directly to the result of the stationary nucleation theory,
from which the stationary Becker-Doring nucleation rate(23) can be
directly obtained by eq. (6). We find
~ o o = ~ I (00) = L
-I -I
ankc
n
= -a
kl
J
twin
(88)
n
GROWTH PROCESSES
75
The time dependent part obeys a homogeneous system of differential
equations
(89)
The solution of this system is obtained by a time dependent trans-
formation from any set of initial values
x (t)
n
L TR lxl(t )
I n 0
(90)
where
(exp(A.(t-t ) I
= 0 n
(91 )
For short time intervals, t-t).max(a
l
)< I), the calculations
of TR can be made with small and precision. For larger
times, TR can be obtained from the short time solution by simple
matrix multiplication, using
= L (92)
]..I
The solution of a system of 25 differential equations is easy and
fast for large times and occurs without the appearance of spurious
oscillations. The method is well known in principle, but does not
seem to be used widely in the numerical solution of differential
equations.
The remaining equations (79,80,81) have the general form
N get) - b.N
2
(93)
c c
N f(t) + c.N (94)
a c

1
.
- I
Ln. - N (95)
NI
J. - a..N
1
1mp
i=2
1 a
They can be solved because get) is known from eqs.
(90,88,85,79) and b liN is assumed from equation (74) to
ccoa c
76 R. NIEDERMAYER
be a slowly varying function of time. This allows the calculation of
N
c
. In (94) the function (t) known again from (90,88,85,80) and
Nc(t) from (93), c is TIJimp and varies slowly again. Finally
the number of monomers is stationnary for the times considered and
thus a straightforward solution is possible. The results are appro-
ximated by
N =
c
Na =
Nco/(l+b.Nco.(t-to))+ Jt f(T) dT
to
N
ao
+ ftf(T)dT +
to
(96)
t
dT (97)
to to
The first terms including the coalescence parameter b are exact
solutions, if coalescence prevails, the integrals are exact solutions
if coalescence is unimportant against nucleation. The 10mplete appro-
ximations are valid, whenver (t-to)2INccoall .lg(t)IN
c
' which can
always be met and is thus suited for numerical analysis.
The nucleation is described by the integrals, which require the
calculation of
(98)
This gives the values of the fundamental integrals occuring in eqs.
(96,97) from (79,80)
I
t n.*(T) dT= (t-t ).x.* 1 +
t 0 - ,co
o
(99)
rr
t t
+
o 0
+kLl
,
These integrals are obtained exactly from the matrices calculated
earlier.
If we finally keep in mind, that the capture probabilities are
GROWTH PROCESSES
approximately proportional to the number of monomers, i.e. that
Ji/NI is nearly constant, we find
;,1,2
J. (I -N P rr)
1.mp c
Ci. + N (J-:-/N
I
)
c 1.
(100)
77
The advantage of the method described is the use of solutions,
which are exact in finite intervals and which can be calculated by
algebraic methods. This avoids the difficulties of the usual extra-
polation methods, which result in spurious oscillations or other
instabilities of the solutions. Contrarily the progress of the
computer calculations with our method is very smooth and rapid.
4. RESULTS OF COMPUTER CALCULATIONS AND COMPARISON WITH
EXPERH1ENT
It is possible to obtain separate analytical approximations
for the initial stage, the nucleation stage, and the saturation
stage(S). While these are useful for special purposes, a general
survey is only possible with a complete numerical solution of the
system of equations (77 to 81). This will be presented here for the
case of gold on rocksalt, for which
A = 87.6 kcal/mol, Edes = LD kcal/mol, Ediff= 8'5 kcal/mol
Vodes = 2.4.10
12
s-I , vodiff = 7.4.10
1
1 s-I
are assumed for best fit with the experiments of Robins et al.(3).
Two different approaches have been used in the calculations
(i) homogeneous distribution of monomers, (ii) diffusion governed
distribution of monomers. Both assumptions have been carried
through consequently and the differences are interesting in view of
the fundamental discussions of 2.2 and 2.3.
4.1. Computer results
The growth of gold films is represented by the concentration
of the polymers nl, n2, n3, n4, uS, by the number of stable clus-
ters N ,and by the number of condensed atoms N
a
The mean cluster
p is ~ n o t e r in the figures. At low temperatures or high
impingement rates the monomer concentration is limited by twin for-
mation and capture. This capture regime is characterised by (5)
(10 I)
78 R. NIEDERMAYER
The monomer concentration attains a maximum
n
l
'" IJ .. a/2.D.s(l) < J. /0.
max 1mp 1mp
(102)
The general features of growth are not very different in the two
cases, as can be seen from Fig.4. The twin is already supercritical
in the capture regime. The number of clusters rises steeply during
an induction perioa tin = 1/10.2 + 8Ds(I)Jimp/a, and increases only
slowly at later stages (_t I / 3, _t l / 7). The maximum cluster concen-
tration is higher in the diffusion case, because the cluster growth
is slower and thus the mean monomer concentration is higher here.
The condensation is complete in both cases (Na(t. The number of
polymers develops consecutively, their stationary concentration is
size independent in the diffusion case while it is weakly size de-
pendent in the homogeneous case. The reason for this is, that the
capture rates are size dependent in the homogeneous case and size
independent for high supersaturation in the diffusion case. This
can be seen by an evaluation of (14,IS) for both cases, where (66)
and Ko(x)'" 0.11 - Inx,KI(x) '" I/x can be used in the diffusion ca-
se for small x = p./rl' Ji = is then obtained in the
homogeneous case, Ji = .11+ln(rl/pi) is the
capture rate for the diffusion case.
At high temperature or low impingement rates the monomer con-
centration is determined by the desorption equilibrium. This desor-
ption regime is characterised by
2
a > B.D. s(1).J. /a (103)
1mp
The maximum monomer concentration is
n
l
'" J. /0.
max 1mp
(104)
and is constant for some rime. An example for this case is shown in
Fig. S. In this case of high temperature the first supercritical
nucleus is a cluster of six atoms. Remarkable differences between
the two monomer distribution cases can be observed. The mean mono-
mer concentration is constant in both cases and has almost the same
value. The condensation is incomplete, but it is more effective in
the diffusion case. This is connected to the fact that the nuclea-
tion rate in the diffusion case is 10
4
times larger than in the
homogeneous case under these circumstances (i*= 5). The reason for
the higher nucleation rate is the local concentration around a sub-
critical cluster, which is enhanced from eq. (63) and Fig. 3.
Another consequence of this local enhancement of monomer con-
centration around subcritical clusters is the inversion of the se-
quence of subcritical polymer concentration in the diffusion case,
where nS > n4 > n3 > n2, while in the homogeneous case n2>n3>n4>nS'
1
6

1
5

1
4

1
3

1
2

1
1

1
0

2

I
g
N

H
o
m
o
g
e
n
e
o
u
s

d
i
s
t
r
i
b
u
t
i
o
n

T

=
5
0
0
K

J
i
m
p

=
1
0
1
4

c
m
-
2

s
-
1

-
6

-
5

-
4

-
3

-
2

-
1

0

2
3
4

I
g
N

a

1
6

D
i
f
f
u
s
i
o
n

d
i
s
t
r
i
b
u
t
i
o
n

1
5
t

T
=

5
0
0

K

J
i
m
p
=

1
0
1
4

c
m
-
2

-
1
4

1
3

1
2

1
1

1
0

3

2

-
7

-
6

-
5

-
4

-
3

-
2

-
1

0

_
t
1
/
3
,

-
'

-
I
g
t

2

3

4

5

C
i
)

:
r
J

:
J
:

.
"

:
r
J

o

(
"
)

m

m

e
n

F
i
g
.
4
.

D
e
v
e
l
o
p
m
e
n
t

o
f

c
l
u
s
t
e
r

c
o
n
c
e
n
t
r
a
t
i
o
n

c
o
n
d
e
n
s
a
t
i
o
n

N
a

a
n
d

m
e
a
n

r
a
d
i
u
s

r

w
i
t
h

t
i
m
e

f
o
r

t
h
e

c
a
s
e
s

o
f

h
o
m
o
g
e
n
e
o
u
s

a
n
d

d
1
f
f
u
s
i
o
n

m
o
n
o
m
e
r

d
i
s
t
r
i
b
u
t
i
o
n
.

T
i
m
e

t

i
n

s
e
c
o
n
d
s
,

T

=

5
0
0

K
.

J

i
m
p

=

1
0

1
4

c
m
-
2

s
-
I
,

o
t
h
e
r

d
a
t
a

f
o
r

g
o
l
d

o
n

r
o
c
k
s
a
l
t
.

:
:
c
f

I
g
N

1
6
.
-
H
o
m
o
g
e
n
e
o
u
s

d
i
s
t
r
i
b
u
t
i
o
n

1
5

t

T
=

8
3
4
K

J
i
m
p
=
1
0
1
4

e
m
-
2
s
-
1

1
4

1
3
[

1
2

1
1

1
0

I

/

I

7

-
1

6

-
2

5

-
3

n
1

4

t

t
-
4

3

-
5

n
2

2
t

-
6

-
7

,

~

~

-
6

-
5

-
4

-
3

-
2

-
1

0

2

3

I
g

t

i

i

4

5

.
I
g

N

1
6

t

Q
;
'
I
u
s
;
o
n

d
;
5
"
,
,
"
'
;
o
n

1
5

T
=

8
3
4

K

J

.

1
0
1
4

-
2
-
1

1
4

i
m
p
=

e
m

s

1
3

1
2

1
1

9

8

7

6

5
L

L
.
,

n
1

-
-
2
.
3

.
.

1
0
9

4

_
n
5

~

.
c

~

n
4

3

2
-
6

-
5

-
4

-
3

-
2

-
1

0

n
3

n
2

-
I
g
t

2

3

4

5

F
i
g
.
S

D
e
v
e
l
o
p
m
e
n
t

o
f

c
l
u
s
t
e
r

c
o
n
c
e
n
t
r
a
t
i
o
s

n
l
,

n
2
,
n
3
,
n
4
,
n
S
,
N
c

c
o
n
d
e
n
s
a
t
i
o
n

N
a
,

a
n
d

m
e
a
n

c
l
u
s
t
e
r

r
a
d
i
u
s

r

w
i
t
h

t
i
m
e

f
o
r

h
o
m
o
g
e
n
e
o
u
s

a
n
d

d
i
f
f
u
s
i
o
n

d
i
s
t
r
i
b
u
t
i
o
n
.

T
i
m
e

t

i
n

s
e
c
o
n
d
s
,

T

=

8
3
4
K

J
i
m
p

=

1
0
1
4

c
m
-
2

s
-
l
,

o
t
h
e
r

d
a
t
a

f
o
r

g
o
l
d

o
n

r
o
c
k
s
a
l
t
.

~

:
D

Z

m

c

m

:
D

3
i
:

~

m

:
D

GROWTH PROCESSES 81
Because the number of clusters is stationary here, an appropriate
stationary nucleation theory would be able to predict the nuclea-
tion rate in both cases. The saturation effects and the maximum
cluster concentration could however not be treated by such theo-
ries. The numerical result is, that the maximum cluster concentra-
tion N
csat
would be smaller by a factor of 10
3
in the homogeneous
case, the cluster radius would be slightly larger. The incomplete
condensation would obey law ~ t
5
/
2
, or _t
2
respectively, if the
effect of direct impingement on stable clusters were neglected; the
direct impingement obscures this difference however.
A rather complete impression on the development of the total
cluster concentration Nc for one impingement rate (10
14
cm-
2
s-l)
can be gained from Fig. 6. Generally the maximum cluster concentra-
tions Ncsat are larger in the diffusion case. But as long as the
critical nucleus is 'smaller than a trimer, no other remarkable dif-
ferences exist between the two models. A slight, though perhaps ex-
perimentally remarkable difference is the somewhat stronger depen-
dence of the nucleus concentration on time for the diffusion dis-
tribution at low temperatures. The differences for larger critical
nuclei will be discussed later.
The development of the condensation for several temperatures
is shown in Fig. 7. Both cases look very much alike, so that an ex-
perimental decision on the mechanism from condensation measurements
for gold on rocksa1t is very difficult. The general behaviour is in
agreement with experimental results ; no condensation experiments
are known for the system Au/NaC1.
From Figs. 4 to 7 it is seen, that the growth of a film may
be crudely characterised by the maximum rate of nucleation N ,
which is simply denoted Nc in the following figures , by the ~ h
mum cluster concentration Ncsat and by the mean cluster radius r
sat
at the time of maximum cluster concentration. The maximum nuc1ea -
tion rate is well defined theoretically, its determination might
however be difficult in experiments, which are made in the capture
regime. So the number of nuclei after one second might be experi-
mentally a more useful parameter. In the desorption regime both
numbers agree.
The dependence of Ncsat, N
cmax
' and rsat is plotted against
the impingement rate for both monomer distributions in Fig. 8. It
has been predicted for the homogeneous case(9,5) that
(a1nNc/a1nJimp)T = i*+ 1, which is also obtained in the left part
of Fig. 8 for 834 K. The corresponding curve in the right part does
however not show this dependence. The difference is again caused by
the local enhancement of the monomer concentration around a subcri-
tical cluster. It is also observed that the maximum cluster concen-
tration is generally larger in the diffusion case, especially so at
1
6

1
5

1
4

1
3

1
2

1
1

1
0

9

8

7

6

5

4

3

2

I
g

N
C

I

H
o
m
o
g
e
n
e
o
u
s

d
i
s
t
r
i
b
u
t
i
o
n

J

=
1
0
1
4
c
m
-
2
s
-
1

I
m
p

_
t
1
/
?

_

-
6

-
5

-
4

-
3

-
2

-
1

0

.
.
.
.
.

*
-
-
~

:
)
.

2

3

4

5

I
g
N
c

I

1
6

D
i
f
f
u
s
i
o
n

d
i
s
t
r
i
b
u
t
i
o
n

1
5

1
4

J
.

=
1
0
1
4

c
m
-
2
s
-
1

I
m
p

1
3

1
2

1
1

1
0

9

8

7

6

5

4

3

2

-
l
o
t

-
6

-
5

-
4

-
3

-
2

-
1

0

2

3
4
5

F
i
g
.
6
.

D
e
v
e
l
o
p
m
e
n
t

o
f

t
h
e

t
o
t
a
l

n
u
m
b
e
r

N
c

o
f

s
u
p
e
r
c
r
i
t
i
c
a
l

c
l
u
s
t
e
r
s

a
t

d
i
f
f
e
r
e
n
t

t
e
m
p
e
r
a
t
u
r
e
s
.

T
i
m
e

t

i
n

s
e
c
o
n
d
s
,

J
i
m
p
=
I
0
1
4

c
m
-
2

s
-
I
,

o
t
h
e
r

d
a
t
a

f
o
r

g
o
l
d

o
n

r
o
c
k
s
a
l
t
.

Q
)

.
.
.
,

:
0

2

m

o

m

:
0

s
:

-
<

m

:
0

1
6

1
5

1
4

1
3

1
2

1
1

1
0

7

6

5

4

2

I
g
N
a

H
o
m
o
g
e
n
e
o
u
s

d
i
s
t
r
i
b
u
t
i
o
n

J
.

=

1
0
1
4
c
m
-
2
s
-
1

I
m
p

1
6

1
5

1
4

1
3

1
2

1
1

1
0

7

6

5

4

2

I
g
N
a

D
i
f
f
u
s
i
o
n

d
i
s
t
r
i
b
u
t
i
o
n

J
.

=

1
0
1
4
c
m
-
2
s
-
1

I
m
p

-
I
a
t

-
6

-
5

-
4

-
3

-
2

-
1

0

2

3

4

5

-
6

-
5

-
4

-
3

-
2

-
1

0

2

3

4

5

F
i
g
.

7

D
e
v
e
l
o
p
m
e
n
t

o
f

t
h
e

n
u
m
b
e
r

o
f

c
o
n
d
e
n
s
e
d

a
t
o
m
s

N
a

a
t

d
i
f
f
e
r
e
n
t

t
e
m
p
e
r
a
t
u
r
e
s
.

D
a
t
a

a
s

i
n

F
i
g
.

6
.

C
)

:
:
D

:
:
I
:

"
'
0

:
:
D

o

o

m

m

C
I
)

1
1

1
0

9

8

6

5

4

2

F
i
g
.

8
.

H
o
m
o
g
e
n
e
o
u
s

d
i
s
t
r
i
b
u
t
i
o
n

I
g

r
.

-
4

I

-
5

I

-
6

I

-
7

~

~
J
i
m
p

I

I

.
.

I

I

1
3

1
4

1
5

1
6

1
7

1
8

I
g
N

1
2
1

1

1
1

9

8

7

6

5

4

3

1
4

1
5

D
i
f
f
u
s
i
o
n

d
i
s
t
r
i
b
u
t
i
o
n
I
g
r

-
t

-
5

-
6

1
6

-
-
r
r
+
l
g

J
i
m
p

D
e
p
e
n
d
e
n
c
e

o
f

t
h
e

m
a
x
i
m
u
m

c
l
u
s
t
e
r

c
o
n
c
e
n
t
r
a
t
i
o
n

N
c
s
a
t
'

m
a
x
i
m
u
m

n
u
c
l
e
a
r
i
o
n

r
a
t
e

N
c
'

m
e
a
n

c
l
u
s
t
e
r

r
a
d
i
u
s

r
s
a
t

o
n

t
h
e

i
m
p
i
n
g
e
m
e
n
t

r
a
t
e

f
o
r

t
h
r
e
e

t
e
m
p
e
r
a
t
u
r
e
s
.

e
x
>

"
'
-
:
l
J

Z

m

o

m

:
l
J

s
:

-
<

m

:
l
J

GROWTH PROCESSES 85
elevated temperatures, and that the clusters are smaller in the dif-
fusion case. The nucleation rates are almost equal for small criti-
cal clusters in both cases, which is obviously caused by the fact,
that the nucleation rate is equal to the rate of twin formation at
high supersaturations.
The size of the critical cluster does not influences directly
any of the measurable quantities in a consequent diffusion theory.
If this theory applies, no means is visible, by which the size of
the critical cluster could be determined, the Walton slope being
2 < (3InNc/3InJimp) < 3 for any critical cluster size. The local va-
riationsin the monomer concentration, which have been discussed in
2.2, tend to smooth the effect of the size dependent cluster decay.
This "diffusion barrier" is in many respects equivalent to a constant
small critical size of i* < 2, which has been postulated in previous
analyses.
Corresponding observations are made in the plots of the depen-
dence of the characteristic growth variables on l/T, fig. 9. The
maximum cluster concentrations do not depend appreciably on tempe-
rature except at high temperatures in both cases. The dependence in
the homogeneous case is stronger than in the diffusion case. The
s o p ~ of Nc is for the homogeneous case given by the relation (9)
(3InN
c
/3(1/kT))= -Vi*o+ Edes-Ediff; in the diffusion case no simple
interpretation can be given.
The number of growth defects in a thin film will be connected
to the maximum cluster concentration, so a low value of Ncsat may be
desirable. From the combination of nucleation rate and maximum clus-
ter concentration shown in Fig. 9 it can be concluded, that this is
achieved in shorter times for high temperatures and high impingement
rates, i.e. the preparation of films with low defect concentration
should be tried at high temperatures and appropriate impingement ra-
tes rather than by slow evaporation.
4.2. Comparision with experiments
Comparison of the considerations presented here with experiments
is possible in several respects.
1. It is possible to predict the general mode of growth by applica-
tion of eq.(43) to the system investigated. This has been done in
some cases and relation (43) has been justified, as can be seen from
Fig: 10.
The validity of this criterium is important for the qualitative under-
standing of film growth.
2. In the case of cluster growth a comparison with extended experi-
ments is possible. Both the time dependence of Na,Nc and the rela-
1
2

1
1

1
0

9

8

6

5

4

3

F
i
g
.

9
.

I
g
N

I
g
N

4
~

1

1
3

1
2

H
o
m
o
g
e
n
e
o
u
s

1
1

d
i
s
t
r
i
b
u
t
i
o
n

1
f
/
1

Z

I

r

1
9
r

d
i
s
t
r
i
b
u
t
i
o
n

-
4

8
.
j
.
f

/

/

-
4

-
5

7
~

\

I

/

-
5

-
6

6
~

"
L

~

-
6

r
s
a
t

-
7
t

5
t

I

/

-
7

1
.
5

2
.
0

1
0
0
0
l
T

1

1
.
5

2

-
-
-
+

1
0
0
0

I
T

D
e
p
e
n
d
e
n
c
e

o
f

N
c
s
a
t
'

N
c
'

r
s
a
t

o
n

s
u
b
s
t
r
a
t
e

t
e
m
p
e
r
a
t
u
r
e

T

f
o
r

t
h
e

i
m
p
i
n
g
e
m
e
n
t

r
a
t
e
s

1
0
1
2
,

1
0
1
4
,

1
0
1
6

c
m
-
2

s
-
I
.

e
x
>

0
-
:
0

2

m

o

m

:
0

s
:

-
<

m

:
0

GROWTH PROCESSES
/\leV
5
cluster growth
3
2
1
1 2 3 4
layer growth
!= Na/WI I
!=K/WI !
87
Fig. 10 Modes of growth, experimental: three dim. growth ~ ) ,
two dim. growth ( ), uncertainties in the experimen-
tal binding energies and n are indicated by the width of
the bars, border region extended because of uncertainty
in n.
88 R. NIEDERMAYER
tions of Figs 8,9. can be used. Values of the binding energies can
be chosen for best fit between theory and experiment and the overall
behaviour of the experimental values can be used to check the theo-
ry. An example for this is shown in Fig. 11, where the results of
Robins and coworkers are plotted together with the theoretical cur-
ves.
The experiments selected here have been made for the growth of
gold on rocksalt. They are thus limited to temperature T < 650 K,
because rocksalt begins to evaporate at higher temperatures. From
Figs. 8,9 it can be seen, that the nucleation rate is the same for
the homogeneous case and for the diffusion case in this temperature
range. The maximum cluster concentration is however different for
the two cases with respect to the absolute value as well as with
respect to the dependence on temperature and impingement rate.
Thus the excellent agreement indicates the validity of the diffusion
case.
It must however be remembered, that the agreement between the-
se and similar measurements and the diffusion theory does not real-
ly proof the diffusion treatment of nucleation presented here, i.e.
the local enhancement of monomers around a subcritical cluster.
This question could only be decided, if a larger range of supersatu-
rations could be covered experimentally. For differences between dif-
fusion governed nucleation and homogeaeous nucleation can only be
observed, where from Edes' Ediff' A critical nuclei can be expected,
which are larger than twins. This range can obviously not be covered
for the system Au/NaCI because of the high vapour pressure of the
substrate and the low vapour pressure of the film material. The eva-
poration of Pb or Zn on rocksalt might be interesting, though expe-
rimentally tedious. Similar experiments have been performed(24),
the amount of data is however not large enough for valid conclusions.
5. FINAL REMARKS
It has been shown in these lectures, that the process of film
growth can be described theroretically in some detail, starting
from rather fundamental data and assumptions. The numerical results
agree with the experimental findings. They are not very sensitive
to the details of the nucleation theory.
The present evaluation of the size of the critical nucleus is
only valid for homogeneous nucleation and in view of the interpre-
tation of the experimental data the validity of the diffusion gover-
ned nucleation mechanism and its associated "diffusion barrier"
should be investigated.
The question of epitaxial growth is closely connected to the
I
g
N

1
9
N

1
2
'

T
=
5
7
3

K

/

1
2
~

J
i
m
o
=
1
0
1
3

c
r
r
{
2
s
1

~
~
I
~
I

1
1

I

~
I

1
0

1
9
r

I

1
9
r

9
+

.
/

5
~

9

-
5

:
1

-
6
+

8

-
6

-
7
+

7
t

~

-
7

1
2

1
3

1
4

1
9

J
i
m
p

1
.
5

2

2
.
5

1
0
0
0
/
T

F
i
g
.

1
1
.

C
o
m
p
a
r
i
s
o
n

b
e
t
w
e
e
n

e
x
p
e
r
i
m
e
n
t
a
l

a
n
d

t
h
e
o
r
e
t
i
c
a
l

v
a
l
u
e
s

f
o
r

t
h
e

n
u
c
l
e
a
t
i
o
n

r
a
t
e

a
n
d

t
h
e

m
a
x
i
m
u
m

n
u
c
l
e
u
s

c
o
n
c
e
n
t
r
a
t
i
o
n

f
o
r

g
o
l
d

o
n

r
o
c
k
s
a
l
t

(
3
)
.

G
')

:
D

0

~

J
:

"
'
t
I

:
D

0

0

m

~

m

e
n

0
0

'
0

90 R. NIEDERMAYER
phenomena discussed here. Its solution depends certainly on the de-
tails of the nucleation mechanism and on the relative importance of
cluster rotation.
REFERENCES
I. J.A. Venables, G.L. Price in Epitaxy, chapter 4
J.W. Matthews ed. Academic Press (in print)
2. J.A. Venables, Phil. Mag. 697, (1973).
3. V.N.E. Robinson, J.L. Robins, Thin Sol. Films, 20, 155,(1974)
I am very much obliged to J.L. Robins for unpublished
mental data, which I used in addition to those published in(3)
and in the other papers quoted there.
4. J.J. Metois, M. Gauch, A. Masson, R. Kern, Thin Sol. Films. l.!.,
205, (1972) and several other papers of this group.
5. R. Niedermayerin Advances in Epitaxy and Endotaxy, p. 21.
H.G. Schneider, V. Ruth eds. Leipzig 1971.
6. J.I. Frenkel, Kinetische Theorie der Flussigkeiten, Berlin 1957.
7. V. Halpern, J. Appl. Phys. 40,4627, (1969).
8. B. Lewis, D.S. Campbell, J. Vac. Sci. Tech., 209, (1967).
9. D. Walton, J. Chem. Phys., 37, 2182, (1962)
D. Walton, T.N. Rhodin, J. Chem. Phys., 38, 2698
(1963) .
10. R. Niedermayer, Habilitationsschiift Clausthal (1969).
II. B. Lewis, Thin Sol. Films 1..,85, (1967).
12. F.C. Frank, J.H. vander Merwe, Proc. Roy. Soc. 198, 205, 217
(1949)
13. J.A. Snyman, J.H. van der Merwe, Surface Sci. 190, (1974)
and Surface Sci. 45, 619, (1974).
14. R. Niedermayer, Thin Films 1.., 25, (1968)
15. A. Mlynczak, R. Niedermayer, Thin Sol. Films. July 1975 (in
print)
16. K.C. \V'olf, Physik und Cemie der Grenzlache, Berlin (1957).
GROWTH PROCESSES 91
17. E. Bauer, Z. Kristallogr. 110,372,395, (1958).
18. M. Drechsler, N.F. Nicholas, J. Phys. Chem. Solids, 28, 2609,
(1967).
19. P.W. Steinhage, H. Mayer, Thin Solid Film, July 1975 (in print) .
20. K. Hartig, Thesis Bochum 1975
21. K.J. Routledge, M.J. Stowell. Thin Solid Films, ~ 407, (1970).
22. G. Zinsmeister in Basic Problems in Thin Film Physics.
R. Niedermayer, H. Mayer eds., GOttingen(1966).
23. R. Becker, W. Doring, Ann.d.Phys., 719 (1935).
24. H. Poppa, J. Appl. Phys., ~ 3883, (1967).
ELECTRONIC STATES IN SEMICONDUCTORS
D. A. Greenwood
H. H. Wills Physics Laboratory, University of Bristol
Bristol BS8 ITL, UK
I. INTRODUCTION
In these lectures, we shall be discussing the nature of elec-
tronic states in crystalline and amorphous semiconductors. The
Bloch states of the perfect infinite crystals are well known ; we
shall be more concerned with what happens when periodicity breaks
down, as it does in amorphous semiconductors, and as it does at
the surface of a crystal. Finally, we shall outline some of the
consequencies of this electronic structure. It is intended to
stress the physical principles involved, rather than details of the
mathematical formalism. Indeed, for some of the problems we shall
consider there is, as yet, no very adequate formalism available.
2. THE ONE-ELECTRON HAMILTONIAN
'Many-body' aspects of electronic structure will not be con-
sidered. We shall use the one-electron description for the valen-
ce electrons of a semiconductor, in which the electronic states ~
are solutions of a Schrodinger equation,
(2.1)
and the potential V(!) incorpora.tes exchange and correlation ef-
fects. (We use atomic units : the unit of length is the Bohr ra-
dius ao = li?/mc
2
and the unit of energy the Rydberg, I Ry =
e2/2ao = 13.6 eV. In these units, n = 2m = e
2
/2 = I).
93
94 D.A. GREENWOOD
If the potential V(x) is periodic, we have to hand all the
familiar methods, based on Bloch's theorem, for obtaining the ener-
gy bands. In practice, a good representation of the band structure
of silicon and germanium can be obtained taking V(x) is a sum of
pseudopotentials centred on the ion sites !i (J) :-
I u (x - x)
. ps -

(2.2)
The approximation in which V(x) is of 'muffin-tin' type
will also be useful. In this approximation, V(x) is a sum of sphe-
rically symmetrical, non-overlapping, potentials centred on the ion
sites, with a constant interstitial potential :
(2.3)
A 'first principles' muffin-tin potential can be constructed by the
Mattheis recipe of superposing neutral atomic charge distributions
and adding an exchange potential term(2). This method incorporates
into the potential the local environment of an ion to a considera-
ble extent, but we should not expect it to be completely satisfac-
tory in a semiconductor without some further adju.stment. The dia ....
mond type lattice is a very open structure : even if the muffin-
tins touch, the interstitial, constant potential, region fills
about two thirds of the volume. Nevertheless, the approximation is,
both computationally and conceptually, a useful one.
In writting down a Schrodinger equation, we are adopting a
physicist's approach. The energy band structure appears as a con-
sequence of the translational symmetry of the lattice : band gaps
are seen as Bragg reflections. A chemist will regard the tetrahe-
drally coordinated lattice of a group IV semiconductor as a large
covalently bonded molecule, and its electronic structure
in terms of bonding and anti-bonding orbitals. They are constructed
from the familiar sp3 hybridized atomic orbitals of the theoretical
chemist (Fig. 1). In this approximation, the electron states are
linear combinations of orthogonal directed orbitals lai> at each
site a, with i 1, 2, 3, 4 :
I I/! >
I
a,i
a . I ai >

(2.4)
Then, in the simplest approximattion, a full valence band corres-
ponds essentially to doubly occupied bonding orbitals. An addition-
al electron must go into an anti-bonding orbital, which has a
higher energy. Hence there is an energy gap. From this point of
view, the energy gap is more a consequence of the tetrahedral en-
vironment of each ion, than the long-range order of the crystal
ELECTRONIC STATES IN SEMICONDUCTORS
,-,
I ... _-
'-.....
,
,
\
\
\
,
"-
'-
,-,
I I
"A(
,," \
/" \
/ \
I \
, I
I
/
/
--,
,
\
\
\
\
---,
_-. I
'-'
Fig. I Pairing of directed orbitals.
95
lattice. However, as we shall see later, it is not really possible
to obtain a good representation of the band structure without ex-
tending the range of basis states very considerably, and the con-
ceptual simplicity of the picture is then lost.
Retaining only the basis states of (2.4) the simple two-para-
meter model of Weaire(3) assumes that orbitals on different sites
are also othogonal, and neglects all matrix elements of the Hamil-
tonian except those between orbitals on the same site,
< ai IHI aj > = VI, i # j and < ai IHI ai > = 0,
and those between orbitals making a bond between neighbouring ions
< ai IHI 8j > = V2, for ai, 8j making a bond pair
In Dirac notation, we can write
H I VI lai><aj I
a,i#j
+ I V ai >< 8 j I
bonds
(2.5)
96 D.A. GREENWOOD
In fitting V}, V2 to a real semiconductor, both parameters are
negative ; the first term, for an isolated atom, gives a splitting
4V}, between s and triply-degenerate p orbitals, and the second
term gives a splitting 2V2 between a bonding and an anti-bonding
orbital. The Hamiltonian (2.5) formalizes a simple 'chemical' ap-
proach.
3. AMORPHOUS SEMICONDUCTORS
The band structure of crystalline semiconductors is well
understood. How far can we understand electronic states in amor-
phous semiconductors ?
We need, first, a model for an amorphous structure. Amorphous
films have been extensively investigated experimentally. The inter-
pretation of the experiments is not universally agreed, but the
continuous random network model seems favoured over alternative
micro-crystallitemodels(4). It is possible to generate (either
by hand or by computer) networks in which each ion sits in a locally
ordered, nearly perfectly tetrahedral, environment. Bond lengths and
bond angles are slightly ~ 5%) distorted from their corresponding
perfect crystal values, and the number of atoms in a ring, which in
a diamond or wurtzite structure is always 6, can be either 5 or 6.
Such random networks can be continued indefinitely with no long
range order(5). Real amorphous films probably differ from the ideal,
completely saturated, random network by the presence of voids.
The mathematical description of the random network - other than
by the enumeration of }023 coordinates - raises difficulties. The
pair correlation function is known experimentally, but this is far
from sufficient. Third and fourth order correlation functions are
necessary to describe the local tetrahedral structure, and these
are not known either experimentally or theoretically. Since we are
dealing with a random structure, the only quantities we can hope to
calculate are ensemble averages. Thus we can set out to calculate
the density of electron states,
g(E) = < ~ I 0 (E - En
n
(2.6)
where Q is the volume of the system, and a factor of 2 for electron
spin is included. The brackets denote an average over an ensemble
of representative random networks. But we cannot calculate a parti-
cular wave function Wn. The density of states is the simplest phy-
sically significant quantity we can calculate ; it is related im-
mediately to optical and transport properties. In particular, we
wish to understand the presence of a gap or quasi-gap, in the den-
sity of states, which experiment clearly shows to be there(4).
ELECTRONIC STATES IN SEMICONDUCTORS 97
Making the approximation that the potential at all sites in
the random network is the same, the Hamiltonian (2.1) and either
(2.2) or (2.3) can be used, with Xi now given by the random network
model. The Hamiltonian (2.5) for an amorphous structure,
differing from that for a crystalline structure in the enumeration
of the bonds.
4. MULTIPLE SCATTERING APPROACH TO THE DENSITY OF STATES
The Weaire model Hamiltonian does indeed have a gap in the
density of states for all structures, crystalline or amorphous.
This can be rigorously proved, but the gap has so very much been
built into the model that it is not completely satisfying an ex-
planation. How can a gap arise if we start from (2.1), which is
a more fundamental d"escription ?
One can think of the band-structure problem as a mUltiple
scattering problem, simplified by the periodicity of the crystal.
The KKR method of band-structure calculation exemplifies this, and
it seems natural to adopt this approach to the amorphous case.
The muffin-tin potential (2.3) is most convenient, and we set VMFO
as zero of energy. In a semiconductor the energy gap region in -
which we are interested lies above the muffin-tin zero, so we con-
sider only scattering at energies e: = k2. > O. Thus electrons pro-
pagate freely in the interstitial region, and are scattered by
the muffin-tin potentials.
The scattering of a plane wave by a single muffin-tin poten-
tial is described by the phase shifts Given the muffin-tin
potential u(r), these are determined by integrating the radial
Schrodinger equation out to the muffin-tin radius and matching
to the appropriate free-space solution :
- sin nR,
jR, - sin tiR, ]r =
(4.1 )
Equation (4.1) we can solve for tan n. It is possible to chose
nR, (e) = 0 at = O. Fig. 2 shows phase shifts calculated for crys-
talline germanium. Phase shifts reflect atome structure. In a se-
miconductor both s- and p-phase shifts are large at the Fermi ener-
gy. The d-phase shift is small, and higher phase shifts are negli-
gible.
Conversely, from (4.1) the logarithmic derivative of the ra-
dial wave function is specified by the phase shifts. The eigensta-
tes of the multiple scattering problem can be constructed by mat-
ching the wave function in the interstitial region, which will be
some superposition of the plane waves, to solutions inside the
98 D.A. GREENWOOD
1T/2
r ~ ~ ~
p
d
&
-1T/2
Fig. 2 Phase shifts for germanium (from unpublished calculations
of P.V. Smith).
muffin-tins. Thus in principle the eigenstates of the multiple
scattering problem are determined by the positions 4i, and the
phase shifts n
t
() , of the muffin-tin potentials. A formalism which
makes this explicit has been developed by Lloyd(6). To understand
the physics of its application to amorphous semiconductors we re-
view a little more scattering theory.
A wave packet, scattered by a potential, is delayed in the
neighbourhood of the scatterer bya t i m ~ Tw, given by
= 2 dn
Tw dE:
(4.2)
This is the Wigner delay time(7). If n(E:) is a rapidly varying
function of energy, Tw can become very large, as near a 'scatte-
ring resonance' where an electron enters a metastable state. Ne-
gative values of Tw are possible also, but are restricted by cau-
sality : a wave packet cannot be advanced in space by a range grea-
ter than twice the range of the potential, so that
TW > -
2RM
u
(4.3)
We shall call a large negative value of Tw an 'anti-resonance'.
ELECTRONIC STATES IN SEMICONDUCTORS 99
The Friedel sum rule is easily understood in terms of the
Wigner Delay time. Consider a single scatterer in a volume ~ . An
incoming flux of electrons of angular momentum t and energy E= k
2
is delayed by Tw before being converted to outgoing flux, so that
there is an accumulation (or deficit) of electron density in the
neighbourhood of the scatterer, proportional to dn/dE. At large
distances from the scatterer, the electron density must remain the
same. A straightforward calculation gives for the integrated den-
sity of states N(E)
(4.4)
which is the well known Friedel sum rule. Here No(E) is the free
electron contribution. Thus
geE)
1 2 d
n
t
go(E) + - - \' (2+1) -
~ 7T ~ dE
(4.5)
If there are N scatterers in the volume ~ and we neglect in-
terference between scatterers, a zeroth order approximation to the
density of states is
geE) (4.6)
This is a crude approximation. Neverthless, if a scattering reso-
nance is important (as in a transition metal) it immediately picks
out the resonant contribution to the density of states.
The basic idea of the cluster scattering method of McGill and
Klima(8) can now be very simply explained: a generalized form of
the Friedel sum expression (4.4) is applied, but with clusters of
ions, rather than single ions, as the scattering elements. The clus-
ters are chosen to represent typical configurations found in the
random network model. Thus the short range order inside a cluster
is treated exactly. The size if clusters which can be considered is
limited by the resulting computational ptoblem.Scattering between
clusters is neglected, but since there is no long range order there
should be some partial cancellation of multiple scattering effects.
The actual calculations use Lloyd's result(6), which replaces
the evaluation of generalized phase shifts from a cluster by the e-

100 D.A. GREENWOOD
valuation of a determinant, in a mixed position-angular momentum re-
presentation. If is the volume of a cluster occuring with weight
Po.
(4.7)
In this expression, = - tan nt(E)/k depend on the muffin-tin
phase shifts at the s1te Xj and the motion of electrons between si-
tes and !j is described by the 'propagator'
.t" + ... L"
j
ik L 4 Tr 1. ht" (kR) eLL' , I!I> 0
L"
= ik aLL' , I! I 0
(4.8)
where
(4.9)
are integrals over spherical harmonics.
SlaOQared
Eclipsed
Fig. 3 Clusters of eight atoms in the staggered and eclipsed con-
figuration.
ELECTRONIC STATES IN SEMICONDUCTORS
30
i'
,..
Q:
..
j
..
g
20
..
..

$I
..
li
,..
i
c

10
o
Density of states for silicon
-One center
-- - Two centers
_. - EilJht centers (Staggered
Configuration)
.\
I'
/.
..
/
. \ I .\
\/ \\
I
. \
,. I\.\
" / \ \
/' / ' .....
'/ ',--"';. -;::.:..,,--
01 02 0-3 04
Energy (Ry)
Fig. 4 Density of states for clusters of one, two and eight Si
sites as a function of energy(8).
101
McGill and Klima took clusters of up to eight atoms in 'stag-
gered' and 'eclipsed' configurations' (Fig. 3),and performed calcula-
tions for C, Si and Ge. Results for silicon are shown in Fig. 4.
Note the peak in the single atom cluster result, which corresponds
to the sharp rise of the p-phase shift at this energy. The remar-
kable feature of these results is that, as the size of the cluster
increases, a dip appears, and the density of states falls below
its free-electron value. This pseudo-gap is of about the same width
as the energy gap in a perfect crystal, but appears at a rather
higher energy. This is shown more clearly in Fig. 5. The phase
shifts used in the calculations were adjusted slightly to give
a reasonable band structure for the crystal.
102
.,
>-
0::
..
15
' ...
.c 10
.8
N
g
II

E
en
....
o
;:-
"in
c
5
I
,
,
,
,
,
J
J
,
Ii
1
-I
,I
, I
, 1
, 1
, I
1 1
, 1
, \
\.,.1
,
I,
"

Density of states for clusters
of eight silicon centers
--Staggered Configuration
D.A. GREENWOOD
- - - Eclipsed Configura!ian
I
\
\
\
\
\
\
" -" Free Particle Density of States
-"
-"
05 0-6
Energy (Ryl
Fig. 5 Density of states for clusters of eight Si sites in the
staggered configurations, and the free particle density of states,
as a function of energy(8).
Calculations on larger crystals, of up to thirty carbon atoms,
have been made by Keller(9). Fig. 6 shows the behaviour of the pseu-
do-gap as a function of cluster size. Keller also found that the
gap tended to disappear if clusters in which bond angles were high-
ly strained were used.
These results show conclusively that short range order is suf-
ficient alone to give a pseudo-gap. The valence and conduction bands
can be thought of as resulting from resonances in the scattering by
a cluster. In the gap, an 'anti-resonance' makes it extremely dif-
ficult for an electron to propagate through the crystal. It is,
therefore, possible to comprehend the electronic density of states
in an amorphous semiconductor from the m'.lltiple scattering point of
view. The question of whether a true energy gap in the density of
states should appear in an infinite random network is not one which
ELECTRONIC STATES IN SEMICONDUCTORS 103
01
-
..
'3
i

..
c
e
1:;

FE

005 ..
J!
!
...
0
l:'
iIi
c

0
07
EnerOy(Ry)
Fig. 6 Electronic density of states in the "energy gap" region
for clusters of 10, 18 and 30 carbon atoms. The dashed line is the
free-electron density of states(9).
computation can answer. Some way of incorporating inter-cluster
scattering into the calculation is necessary so that the results
become sufficiently accurate to compare with experiment. Recent
analytic work by Jones(IO) based on (4.7) is promising and shows
how the free electron part of the density of states can be cancel-
led by a part from the logarithm, but the effect of remaining terms
is not completely elucidated.
The Weaire Hamiltonian (2.5) shows that one can construct mo-
dels in which a true gap appears between the valence band and con-
duction band, though a more realistic model in which the parameters
VI and V2 are allowed some degree of variation would lead to a tail-
ing of states into the gap. Phillips has suggested that a carefully
prepared amorphous semiconductor relaxes in such a way as to exclu-
de states appearing in the gap, in order to minimize bond energies
(II),
Theory cannot yet give an unambiguous answer to the question
of the sharpness of the band edges at the gap. The interpretation
of the experimental data is a matter of controversy. Paul et al(4)
104 D.A. GREENWOOD
suggest that the Fermi level is pinned at the centre of the gap,
and relative to this energy

g (0) exp ES
with Es'" 0.1 eVand g(O)'" 10
19
cm
3
eV-
I
. Much sharper band edges
- as shar jl as thos e of a crystal- have been report ed by Donovan and
Spicer (12) and co-workers, but a tailing of states into the gap in
an amorphous semiconductor is undoubtedly easier to accept theore-
tically.
5. GERMANIUM AND SILICON POLYMORPHS
The multiple scattering calculations, or calculations based
on a simple model Hamiltonian, give only broad qualitative features
of the density of states in amorphous semiconductors, whereas band
structure calculations for perfect crystals give more detailed den-
sity of state functions. Comparison of the density of states for dif-
ferent tetrahedrally bonded crystalline polymorphs of germanium and
silicon which differ in the nature and complexity of their unit cells,
can, it is thought, give us eful information on the relation between
electronic structure and short range order.
This is an attractive idea sInce we have already established
th e dominant rol e of short range ord er. A drawback appears to be
the technical difficulty of predicting reliable band structures in
the absence of experimental data to fit the parameters of the empi-
rical pseudopotential. It is necessary to interpolate between values
of the pseudopotential form factor known for the diamond structure.
(It is assumed that the pseudopotential is independent of the struc-
ture). Aymerich and Smith(13), in their study of the hypothetical
germanium wurtzite structure, found that the calculated energy gap
was very sensitive to the method of in'terpolation employed, varying
between 0.11 eV and 0.60 eV for apparently equally reasonable inter-
polations, and presumably there are similar uncertainties in other
work.
Calculations for the GeIII polymorph are most relevant to amor-
phous germanium. This is a metastable structure under normal condi-
tions, but little is known about its electronic structure experimen-
tally. In the Ge III structure there are 5-fold as well as 6-fold
rings, and distorted tetrahedral bonding, so that in some respects
it is similar in its local structure to amorphous germanium. The
detailed structure in the density of states is due to long-range
order. After smoothing, the forms of the valence band and conduc-
tion band are very similar to those determined experimentally(14).
ELECTRONIC STATES IN SEMICONDUCTORS 105
This appears at the moment to be the most proml.Sl.ng technique
for calculating densities of states which can be used in the detai-
led interpretation of experimental data, though of course it says
nothing about the tailing of states into the gap, or the nature of
the wave functions, in the amorphous material.
6. THE NATURE OF THE ELECTRON STATES
To discuss the nature of the wave functions in an amorphous
semiconductor, it is useful to return to the 'tight-binding chemical
approach in which states are constructed out of bonding and anti-
bonding orbitals. This is a reasonable description for the valence
band, but quite inadequate for the conduction band. We, therefore,
concentrate on the valence band, in which the wave function can be
constructed out of bonding orbitals alone to a .good approximation
where
I > = L bail B,ai >
bonds
B,ai >
)
=-
12
(I ai > + I Sj
and ISj > is the state pairing with
(6. )
ai >.
In the Weaire model, at the bottom of the valence band,bonding
orbitals are all combined with the same sign, so that the interac-
tion energy V)/2 0) between bonding orbitals at an atom is mi-
nimized. In a perfect crystal the Bloch functions near the bottom
of the valence band are of the form
= L
p
where Xp is the centre of the pth bond orbital Xp Ziman has sug-
gested- ()5) that in the amorphous crystal the electron states must
be very similar. The wave vector k is no longer a good quantum
number, but it will be 'nearly good' and states constructed in this
way may be a useful basis set. Such states are, clearly, extended
throughout the crystal.
At the top of the valence band there are, in the perfect dia-
mond structure, two bonds at each atom with coefficients of positi-
ve sign in the expansion (6.), and two with negative sign. Thus
maximum advantage is gained from 'anti-linking' matrix elements.
The bands are triply degenerate at k = 0 (corresponding to the three
ways in which four bonds can be divided into pairs). Ziman suggests
that the bonds in an amorphous crystal attempt a 'best fit' to the
106 D.A. GREENWOOD
arrangement on a perfect crystal. The triple degeneracy is probably
preserved, and 'hole states' at the top of the valence band a
'nearly good' quantum number k will exist. Again, such states are
not localized.
In the Weaire model, even if we include anti-bonding orbitals
into the expansion, there is a tailing of states into the gap above
the valence band only if VI and V2 are allowed to vary. Such a va-
riation would, of course, be very reasonable in an amorphous stru-
cture. We might also expect there to be 'dangling bonds' where
there are deviations from the ideal random network model. Electron
states associated with dangling bonds are probably localized near
the structural defect, and will have energies lying in the gap.
7. SURFACESTATES
So far, we have been discussing the electronic states of bulk
material, and have ignored effects which might arise the fact
that crystals, or amorphous films, have surfaces. We turn now to
discuss what happens at the surface of a crystal.
It is perhaps instructive to begin with a one-dimensional mo-
del for a 'semi-infinite' crystal (Fig. 7). For x > 0 the pot,ential
v

Fig. 7
One-dimensional periodic potential terminated by 'surface'
ELECTRONIC STATES IN SEMICONDUCTORS 107
is that of a perfect crystal, with V(x+a)=V(x). For x < the poten-
tial may be terminated by some 'surface layer' before rising to the
external vacuum potential Vo' In one dimension the Schrodinger equa-
tion is a second-order ordinary differential equation. and the dis-
cussion can be based on the well-known mathematical properties of
such equations. At a given energy E the equation has two indepen -
dent solutions. Finding appropriate eigenfunctions which satisfy
the physical boundary conditions is essentially a matching problem.
For an energy E V
o
). integrating inwards from x = -00. our solution
must start like exp(Kx), where K2 - Vo-E. At x = 0, must be
matched to ' the solution for x > 0. Suppose now that E lies in
an allowed energy band of the perfect crystal. Then we can write
where are Bloch states of the perfect latti-
ce satufying
+ika
(x+a) = e- . E(k) = E(-k) = E (7.1)
Since we have two constants a, 8 at our disposal, we can always sa-
tisfy the matching condition, at x = 0. Hence. at any
energy in the band there is always an iigenstate extending uniform-
ly through the crystal and decaying away ouside the crystal, as we
should expect.
Consider now an energy E lying in a bandgap. We can still find
two independent solutions in the perfect lattice, such
that
Here A+, A_ are real. If IA+I> 1, then IA_I< 1. One solution decays
exponentially to the right, and the other to the left. These solu-
tions in the gap are 'evanescent solutions'. In the infinite crys-
tal they cannot be normalised, but in our semi-infinite crystal
solution is physically acceptable for the region x > 0.
It is not, in general, an eigenstate; at x =.0, (E) is de-
termined and will not match for an arbitrary-energy in
the gap. However N_ (E) varies continuously with E (since the
independent solutions of the Schrodinger equation from which is
constructed vary continuously with E), from its value at the bottom
of the energy gap to its value at the top of the gap. For a lattice
with reflection symmetry it is easy to show(16) that the state at
one edge of a bandgap is odd,and at the other is even (e.g.cos(TIx/a),
sin(TIx/a) in the nearly free electron limit). Hence goes
from zero to infinity across an energy gap. Outside the material
is positive and smoothly varying with energy. One might the-
refore expect there to be just one energy in the gap at which the
matching condition is satified (Fig. 8). Such an eigenstate lying
in a bandgap is called a surface state, and an electron in this sta-
te is localized near the surface of the crystal.
108 D.A. GREENWOOD
________ ____________
Fig. 8 Variation of logarithmic derivative with energy
Note, however, that the behaviour of can be modified
by the surface layer potential so that it for example, be-
come negative at x = O. Thus a surface state in the gap does nbt
necessarily exist. It depends on the nature of the surface layer.
On the other hand, a strong surface potential could, clearly ,
bind an electron in a localized state below the bottom of the va-
lence band. This might be a model for a surface state associated
with an adsorbed impurity.
We can regard the evanescent states in the gap as Bloch
states with a complex wave vector k = kr+iki, where kr = 0 or
TI/a, for gaps at the centre of the zone or at the band edges res-
pectively. As the energy increases across the gap,ki increases to
a maximum value and then decreases to ' zero (Fig. 9). This conti-
nuity follows, again, from the continuity of the solutions of the
Schrodinger equation with respect to energy(17).
The algebra of various special one dimensional models is
tempting and has been extensively pursued(18) ; the essential phy-
sics is, I feel, fully expressed in Fig. 8. The real problem lies
in three dimensions, and is immensely more complicated. It is also
an important problem, since, in a semiconductor, particularly, ad-
ditional states in the gap between valence and conduction bands
will have a very considerable effect on electronic properties.
ELECTRONIC STATES IN SEMICONDUCTORS
Fig. 9 Band-structure in one-dimension, showing the complex-k
values for energies in the band-gap.
109
The features we have picked out in the one dimensional model
remain. The Schrodinger equation has evanescent solutions, with
1/J(E + = exp(i where If is complex : = + say.
Standard methods of band structure calculation can be extended to
complex wave vectors, since the wave functions and energies are the
analytic continuations of those with real wave vectors. To take a
simple example, an effective mass approximation valid near k. = 0
would give for complex
tI
2
E(k) = -2 *1kI
2
=
- m-
and is real if = 0
However, the topology of the energy bands in complex-k space
is complicated, and an infinite set of states is needed in-the cons-
truction of a surface state. In the simplest possible model of a
crystal surface, the matching surface x = 0 is taken as a plane
midway between planes of atoms in the crystal. Let us, for simpli-
city, take the (1,0,0) surface of a simple cubic crystal. For x>O
110 D.A. GREENWOOD
the potential is that of a perfect crystal, and for x < 0 the po-
tential is taken to be a constant Vo, to be determined by reference
to the experimental work function for the crystal. Then ky, kz re-
main good quantum numbers. since periodicity is maintened in the y
and z directions, and we can look for surface states with fixed re-'
al ky and kz At x = 0 the exterior and interior wave functions,
and their gradients, have tobe matched over the entire y-z plane.
Outside the crystal, the wave function is a linear combina-
tion of states of energy E and appropriate translational symmetry
1/iL (E) = I amn exp {llmnX + i (k
y
} m7fa ) y + i (k
z
+ 2na1T )}
m,n
where
II
mn
Inside the crystal, similarly, for a given E and given ky, k
z
there is an infinite set of solutions, from which 1/iR is to be cons-
tructed
r
If E lies in an energy band of the constant ky' k
z
cross-sec-
tion of the band structure, the set of states ~ r w111 include Bloch
states ~ k ~ k with real k. The matching will always be possible,
though an infinite set of evanescent solutions must also contribute
to the stateWR for the boundary conditions to be satisfied. Howe-
ver, if E lies in an energy gap the matching conditions lead to an
eigenvalue equation for E. If this has a solution, a localized sur-
face state exists. In this case, as ky, kz vary a band of surface
states will be obtained.
Since energy gaps in different directions in k space do not
all coincide, it is possible for the band of surface states to over-
lap with the valence and conduction bands. In a real crystal, a sur-
face state which overlaps with the conduction or valence band may
be expected to hybridize with bulk states of the same energy to mo-
dify their behaviour at the surface, and will not provide a locali-
zed state.
In a practical calculation, of course, only a finite number
of states can be included, and some sort of compromise has to be
made over the matching conditions. The first calculation along these
lines was made by Jones (19) for the (110) surface of silicon, using
a pseudo-potential scheme for the band structure, and the results
are shown in Fig. 10. There are two bands of surface states, dege-
ELECTRONIC STATES IN SEMICONDUCTORS
111
kz
r--_-'-x;.... -+_"'Y_-1
M
2i' Z
f
(a)
(b)
Fig. 10 Surface states on the (110) surface of silicon (a)
Two-dimensional Brillouin zone (b) E{kx,k
z
) along symmetry lines
in the Brillouin zone. The energies at the top of the valence band
and bottom of the conduction band are shown as dotted lines.
(After Jones(19.
nerate along XM, with a total width of about 3 eV, and overlapping
with the valence and conduction bands. A later paper(20) extends
the calculation to the (100) and (Ill) surfaces.
This sort of rigorous model calculation is important in that
it tells us what to expect. It is, clearly,difficult to modify the
formalism to study the effect of different types of surface condi-
tions. Beeby has set up a multiple scattering formalism for a se-
mi-infinite lattice of muffin-tin potentials(21), which takes pla-
nes of atoms parallel to the surface as scattering elements. In
this approach it might be possible to consider more complex models
112 D.A. GREENWOOD
for the surface potential. But the computational problems in apply-
ing the method are considerable, and no results have been reported.
An ingenious approach by Appelbaum and Hamann(22) moves the match-
ing plane sufficiently into the interior of the crystal that most
of the evanescent solutions of the Schrodinger equation can be
neglected on the interior side of the matching plane. This equation
is integrated numerically through the last layers of atoms of the
crystal and an optimized matching procedure is used. The potential
in these layers can be varied arbitrarily (except that appropriate
periodicity parallel to the surface must be maintained). The cal-
culations can be reduced to computable size if the ionic potential
can be described by a weak pseudopotential. The method was first
applied to find a self-consistent surface potential for sodium,
and in a more recent paper to the (Ill) surface of silicon(23).
We shall refer to these results below. But we need more flexible
- if cruder ~ theoretical approaches to obtain a qualitative under-
standing of, say, the effect of different surface structures on
surface states, or to even begin to think about surface states as-
sociated with amorphous films.
For covalently bonded semiconductors, we can use the picture
of bonding and antibonding orbitals, which we have already discus-
sed in connection with the Weaire Hamiltonian. If no rearrangement
of ions takes place at the surface, then (Fig. 11) we are left with
just one directed orbital which cannot be paired, associated with
each surface atom. The bonding and anti-bonding orbitals construc-
Fig. 11 'Dangling b ~ d s at a surface.
ELECTRONIC STATES IN SEMICONDUCTORS 113
ted from paired a1rected orbitals make up the valence and conduc-
tion bands of the bulk crystal. The orbitals left over at the sur-
face, or 'dangling bonds', interact to form a band of surface sta-
tes. Since the interaction is indirect, we expect the energy of
states in the surface band to lie between the valence and conduc-
tion bands. Counting spin degeneracy, we see that there are two
electron states per surface atom in the surface band. In a perfect
crystal, the available electrons will fill the valence band, and
half-fill the band of surface states. Electrons in surface states
will have a high electron density in the dangling orbitals, and a
lower density in the other three orbitals of the surface atoms.
However, since the surface atoms do not share the wholly tetrahe-
dral environment of bulk atoms, we might expect the dangling bond
orbitals to be somewhat modified.
Calculations of surface states in. the Hall model for diamond
by Koutecky and support this discussion, though, as we
have stressed in an earlier section, we cannot expect numerically
exact results for semiconductors from an LCAO type of calculations.
The calculations of Appelbaum and Hamann(23) for the Si (Ill) sur-
face are in principle more accurate, and were carried out both with
surface atoms in their ideal positions, and with the last atomic
plane relaxed inwards in the normal direction. Chemical bonding ar-
guments suggest there should be such a relaxation. A smooth model
potential is constructed, to be consistent with the experimental
vacuum level. In both cases, a band of surface states is found
which lies in the absolute band gap between the conduction and
valence bands. The electron density for a state in this band is
shown in Fig. 12. Rather remarkably, we see that the electron den-
sity is consistent with the 'dangling bond' picture, though the
starting point of the calculation is entirely different. The re-
laxed structure, though not the ideal structure, is found to be
'highly self-consistent'.
In these calculations, the surface bands are found to be high-
ly to the surface potential. The degree of self-consisten-
cy is dependent on the choice of surface structure. Further theore-
tical progress would seem to depend on detailed experimental know-
ledge of the positions of surface layers of atoms.
On the basis of this hugely simplified picture, we can spe-
culate on the nature of surface states in an amorphous semiconduc-
tor. The number of dangling bonds per unit area of surface may be
fewer, since in the formation of the structure there will be a pre-
ference for saturated bonds and a consequent lower energy. There
may, then, be a tendency for states associated with dangling bonds
to be localized, rather than to be extended over the surface.
114 D.A. GREENWOOD
";c
..,
,.
>-
>-
iii
z
0.15
...
c
z
C
II:
>-
u
0.5
...
..J
...
...
>-
c
>-
0.25
III
...
U
C
~
2 3 0 I 2 3
Z-AXIS 11, Z-AXIS 11,
Fig. 12 Electron density in a 'dangling-bond' state plotted along
a line normal to the surface and passing through a surface atom.
Also shown is the density along a line in a plane parallel to the
surface whose direction is between the two n e r e ~ t neighbour sur-
face atoms. (a) Unrelaxed and (b) relaxed structure. The heavy dots
on the z-axis are the atom sites and the origin is the matching
plane. (After Appelbaum and Hamann (23) '
8. EFFECTS OF SURFACE STATES
In the last section, the principles which lie behind the exis-
tence of surface states were outlined, and the present state of
theoretical calculation was indicated. We consider now some of the
consequences of surface states.
The surfaces of real crystals will usually be more complicated
than in our theoretical discussion. As well as the surface states
associated with a periodic surface and extended over the surface,
there may be localized surface states at sites where periodicity
breaks down. Thus, states may be associated with randomly adsorbed
impurities, monoatomic steps on the surface, or other imperfections.
On clean surfaces, states will vary with surface index, and with
the nature of the surface. Low energy electron diffraction studies
show that a surface may be 'reconstructed' from the ideal lattice,
and new periodicities may appear. Unfortunately, the surface struc-
ELECTRONIC STATES IN SEMICONDUCTORS 115
cture cannot be found uniquely by LEED : the unit mesh is determi-
ned, but not the positions of the atoms in the mesh. One can only
choose a 'reasonable' model for the surface which is consistent
with the experimental data. Clean surfaces of Si and Ge offer the
bes t hope of comparison betl\1een theory and experiment (25) .
Since electrons in surface states are confined to the surface
region of a crystal, the occupation of surface states can lead to
a transfer of charge from the interior of a crystal to the surface.
In a semiconductor, the density of charge carriers is low; a sur-
face charge will be screened out by a space charge layer over a
distance of the order of the Debye screening length L, where
Here K is the dielectric constant, and n is the maximum possible
density of charge carriers. L is about 1 ~ for intrinsic germanium
(K = 16) at room temperature, and ~ 100 A for a doped sample with
n ~ 1017 cm-
3
. Thus the space charge region in a semiconductor ex-
tends over a considerable dista.nce (whereas in a metal the corres-
ponding Fermi screening parameter for the degenerate Fermi gas of
conduction electrons is ~ 1 A, typically).
In analogy with bulk m p u ~ t y states, surface states are clas-
sified as donor-like, if they are neutral when occupied, or accep-
tor-like, if they are negatively charged when occupied. Bearing
in mind the 'dangling-bond' picture, we can expect donor levels
to lie below acceptor levels.
The occup.ation of surface states will be determined by Fermi-
Dirac statistics. Thus the bulk doping of a semiconductor must be
precisely known to calculate the distribution of electrons over
levels. A specific example is illustrated in Fig. 13, which shows
an n-type semiconductor with a surface band of acceptors below the
Fermi-level. These are filled in thermal equilibrium so that the-
re is a surface layer of charge. This repels conduction electrons,
to ieave a positive space-charge layer. In the (semi-classical)
diagram, there is therefore 'band-bending' relative to the Fermi
level. It is usual to assume that the position of the stirface bands
relative to the conduction and valence bands does not change as
the occupancy of the surface states changes. It is obvious that
there are difficult problems of self-consistency in constructing
the atomic potential near the surface, which may make this simple
picture inappropriate.
116
Density of
Positive
space charge
region
D.A. GREENWOOD
Conduction band
-- - --

..-
Valence band
Fig. 13 Formation of a space-charge region and 'band-bending'
due to the occupation of surface states.
The potential is given by the Poisson equation
where
peE) = eJ + - f(E)]dE
Here geE) is the bulk density of states andf(E) the Fermi function.
The Fermi energy is determined by the density of surface states
and the overall neutrality condition. In this equation, V is an
averaged macroscopic potential, and to a good approximation will
be a function only of distance from the surface. The solution is
discussed in standard texts(26), though we should note that in thin
films with a low density of carriers, for which L is comparable
with the thickness of the film, modifications will be necessary.
Similar considerations apply discussing the interface bet-
ween a semiconductor and a metal, or a semiconductor and an insu-
ELECTRONIC STATES IN SEMICONDUCTORS
117
lator. The importance of surface states in the interpretation of
such systems was first recognized by An interesting
discussion of a metal-semiconductor contact in terms of wave func-
tion matching has been given by Heine(28).
The extent of the band-bending, V
s
' can be determined expe-
rimentally, for example by combining measurements of the work func-
tion and photoelectric threshold. However, it is clear that the re-
lation between the theoretical calculations of the previous section,
and measurements of band-bending, is not one which allows immediate
detailed comparison between theory and experiment.
More directly, light can be used as a probe for surface
tes. Electrons from the valence band can be excited into empty sur-
face states, or from full surface states into the conduction band.
However, these effects have to be disentangled from various bulk
effects. A useful technique, which can be combined with various
measurements, is to vary the doping in a set of crystals, so that
the Fermi energy is moved through the surface band. Field effect
techniques. based on the induction of surface charge by an exter-
nal electric field. can be adapted to study surface states. Expe-
rimental techniques. and data,are reviewed by Davison and Levine(18)
It is agreed that surface states in Si and Ge exist, but there is
no clear consensus on the distribution and nature of the states.
Experiment and theory are still far apart.
In discussing amorphous films. we had difficulty enough in
treating bulk states with any rigour. Elementary considerations of
bonding and electron affinity suggest that surface states, perhaps
localized, will exist. It may be difficult, however, to distinguish
experimentally between these, and other states in the quasi-gap
which are associated with dangling bonds in the bulk of the material.
REFERENCES
1. M. L. Cohen, V. Heine, Solid State Phy., 24, 38 (1970)
2. The Mattheis method is described in detail in :
T.L. Louks, "Augmented Plane Wave Method", Benjamin, New York
(1967).
3. D. Weaire, Phys Rev. Lett. 26, 1541 (1971).
4. W. Paul, G. A. N. Conell, R.J. Temkin, Adv. Phys., 22, 531
(1973)
118 D.A. GREENWOOD
5. R. Grigorovici, "Electronic and Structural Properties of Amor-
phous Semiconductors", (ed. P.G. Le Comber, J. Mort), Academic
Press, London and New York (1973).
6. P. Lloyd, Proc. Phys. Soc., iQ, 207, 217 (1967).
7. A. Messiah, "Quantum Mechanics", North-Holland, Amsterdam (I962
8. T. C. McGill, J. Klima, Phys. Rev., B5, 1517 (1972).
9. J. Keller, J. Phys. C ~ 3143 (1971).
10. R. Jones, J. Phys. C ~ 2318 (1973).
II. J. C. Phillips, Comments on Solid St. Phys. ~ 9 (1971).
12. T. M. Donovan, W.E. Spicer, E.J. Ashley, J. M. Bennett,
Phys. Rev. B2, 397 (1970).
13. F. Aymerich, P. V. Smith, J. Phys. ~ L41 (1973).
14. J. D. Joannopoulos, M. L. Cohen, Phys. Rev. B7, 2644 (1973).
IS. J. M. Ziman, J. Phys. C4, 3129 (1971).
16. H. Jones, "Theory of Brillouin Zones and Electronic States in
Crystals", North-Holland, Amsterdam (1960)
17. V. Heine, Proc. Phys. Soc. ~ 300 (1963).
18. s. G. Davison, J. D. Levine, Solid State Phys. 25, I (1970)
(The theoretlcal discussion in this review emphasizes one-dimen-
sional and tight-binding calculations).
19. R. O. Jones, Phys. Rev. Lett.,'20, 992 (I 968).
20. R. O. Jones, J. Phys. C5, 1615 (I 972) .
21. J. L. Beeby, J. Phys. C6 1242 (I 973) .
22. J. A. Appelbaum, D. R. Hamann, Phys. Rev. B6, 2166 (1972).
23. J. A. Appelbaum, D. R. Hamann, Phys. Rev. Lett.
l!..,
106 (1973).
24. K. Koutecky, M. Thomasek, Phys. Rev. 120, 1212 (I 960) .
25. M. Henzler, Surface Science,
~
650 (1971) .
ELECTRONIC STATES IN SEMICONDUCTORS 119
26. A. Many, Y. Goldstein, N.B. Grover, "Semiconductor Surfaces",
North Holland, (1965).
27. J. Bardeen, Phys. Rev., 717 (1947).
28. V. Heine., Phys. Rev., 138, A 1689 (1965).
Other relevant review articles include :
D. A. Greenwood, "The electronic density of states in amorphous
materials" in ref. 5.
V. Heine, Surface 1 (1964).
P. Mark, Surface Science 25, 192 (1971).
Characterization
STRUCTURE DETERMINATION OF THIN FILMS
H. Raether
Universitat Hamburg, Institut fur Angewandte Physik
2 Hamburg 36 Jungiusstrasse 11
1 INTRODUCTION
The concept of the structure of a thin film has different as-
pects. We can mean the geometrical exterior or the crystalline
structure. We shall treat both these aspects in the following.
2. GEOMETRY
The ideal film has uniform thickness with a plane and smooth
surface and a homogeneous density. The real film however has a vary-
ing thickness, a rough surface and it will have, especially in the
case of a rather thin film (some 100 A or less), an inhomogeneous
density.
Very thin films of an average thickness of less than 100 l,
can be produced on substrates, but they are in general built up by
islands more or less separated from each other. This is valid in
general for metal films. If the "island-structure" films are made
thicker and thicker, the islands grow in size till they touch each
otter and form a more or less continuous foil.
In an intermediate state, between the island structure and the
continuous film, the thin film has a structure with voids (see
Fig. 1) which makes these films less dense than the bulk material.
This density deficit, which can reach 10% of the bulk density, di-
sappears with increasing coalescence of the crystal units. Films of
some 10
3
A have in general the bulk density. Alkali metals show
another behaviour : films of some 100 A can be produced by vapo-
risation as homogeneous films only on cooled (liquid nitrogen tem-
123
124 H.RAETHER
su bstrate
Fig. 1 Schematic Structure of a rather thin film.
perature) substrates such as quartz or sapphire which have a good
heat conductivity. At higher temperatures they condense in a pro-
nounced island structure.
In contrast to the metal films, nonmetallic material such as
aluminium oxide, beryllium oxide, c r b ~ n can be produced as very
thin continuous films of even some 10 A. They are used as support
for substances to be investigated in the electron microscope : me-
tallic or other thin films, single molecules etc.
3. CRYSTAL STRUCTURE
Having prepared a thin film we wish to know whether it consists
of a monocrystal, of a polycrystalline material or of a so-called
amorphous structure. In other words we are interested in the orien-
tation of the crystalline units and their size. It is useful for
the following, especially for the description of the fine crystal-
line state, to introduce the density n (r) of atoms as a function of
the distance r from an arbitrarily chosen atom. In the case of a
monocrystal this is not an isotropic function ; in the directions
along a lattice row [h Kt] it has equidistant high maxima at the
sites of the atoms, indicating the high order in the crystal, (see
Fig. 2 ). In a very fine crystal powder with units consisting of
only some elementary cells however n (E) is an isotropic function
of r : in the immediate neighbourhood of an atom "short range order"
is still persisting whereas the long range order has disappeared
(see Fig. 2b). Amorphous and liquid substances show the same den-
sity distribution as Fig. 2b. In a liquid the atoms move around
and are thus in thermal equilibrium which may not be the case in
an amorphous solid. There has been a lot of discusssion on the dif-
ference between the fine crystalline and the amorphous state, but
we will not treat this question here.
STRUCTURE DETERMINATION OF THIN FILMS
n(r)
(a)
[hkl]
(b)
r
Fig. 2 The density of atoms n(r) in a perfect crystal (a) and
in a highly disordered state (b)
Methods of looking at the crystal structure are
a) diffraction
b) image formation
125
which we discuss briefly in the following. In both cases we use ei-
ther electrons or X-rays, which penetrate the volume of the film.
Electrons of a velocity of at least 50 KeV are needed for trans-
mission through film thickness of 1000 A ; in this case the elastic
scattering intensity represents still an important part of the to-
tal intensity (elastic and inelastic). Slow electrons of 10 KeV and
less are not useful in transmission experimente since they lose too
much energy. X-rays penetrate easily several lOP A, but are less
scattered by 1000 A foils. This is due to the weaker interaction of
X-rays with the single atom which consists of the scattering by the
electron of the atom, whereas electrons are scattered by the nucleus,
126
H.RAETHER
an interaction which is nearly 10
6
stronger. This high scattering
probability of electrons makes them an important tool for the in-
vestigation of thin films. An further important advantage of elec-
trons is the existence of electron lenses to construct an image,
whereas we are not yet able to produce X-rays lenses.
3.1 Electron Diffraction
Some typical diffraction photographs of fast electron taken
for a silver foil are reproduced in Fig. 3. They are obtained by
an arrangement schematically drawn in Fig. 4 : a beam of about
50 KeV electrons passes a thin film of about 1000 A. The diffracted
electrons are photographed in a distance of about 20-30 cm from the
object. The position of the intensity maxima is given by the Bragg
equation 2d sin 9 = nA with d the lattice connstant, 8 half of the
scattering angle and A the wave length of the primary electrons,
which has a value of about 0,05 A for 50 KeV electrons. The angle
28 is therefore of the order of a few 10-
2
rad. An other form of
describing the diffraction pattern is given by the Laue equation
X - Xo = 2f. where y and Yo are the directions of the incoming and
the outgoing beam and f of the reciprocal lattice. The
construction of the Ewald sphere with radius Ixl/A = IYol/A =l/A
yields immediately the diffraction pattern since the proJection of
the points on the photographic plate touched by the Ewald sphere
represents the diffraction pattern (see Fig. 5).
One sees immediately that Fig. 3a is produced by a specimen
which is nearly a monocrystal of silver. The different Laue spots
(002), (020) and (200) are the projection of a plane of the reci-
procal lattice on the plate. In principle there should be seen on-
ly one or very few spots corresponding to the exact adjustment of
the crystal to fulfil the diffraction condition. Apparently the
specimen is built up of many crystals which are tilted against each
other very slightly. In Fig. 3 b-d the orientation of the different
crystals becomes more and more random; so that finally powder rings
are observed.
It should be mentioned that continuous rings are also obser-
ved if crystals lie on a substrate and are randomly distributed
around the normal of the substrate surface, (Fig. 6). Under this
condition a certain number of powder rings is lacking which corres-
pnd to lattice planes too much inclined to the electron beam. This
situation is revealed by inclining the specimen against the beam
as Fig. 7b shows : those lattice planes perpendicular to the rota-
tion axis remain in the reflection position whereas the others are
out of it.
If on the other hand the specimen consists really of a good
monocrystal the number of diffraction spots reduces substantially.
In addition Kikuchi lines appear which are produced by the primary
STRUCTURE DETERMINATION OF THIN FILMS 127
.. ",
,. ~ . ~ ."
I , \ \
, I
\' 'I'
" ~ - - ~ "
-- . -
'"
Fig. 3. Electron diffraction diagrams in transmission of thin sil-
ver films, showing the transition of a Laue diagram into
a powder diagram.
diffracted beams
000
f i lm
Fig. 4. Scheme of an electron diffraction apparatus .
128
H.RAETHER
,
,
,
h k l reflex
"
I
/
/
....
....
... -........ -'-
... >
000
Fig. 5. Reciprocal lattice with the Ewald sphere.
t
o Q [] 11111111
., '." .. ,,: ... 'su' 'L:s"t'r""a"'te' .... :-- ,"', ,.' :, .. :
.' .. , ... f" ., .. ~ u . ,.," "'.. !".. .".. "'; ...
" .. ~ . f1I "," ~ ..... " ., < .... t .. '" .... , r""., .. -
electron
beam
Fig. 6. The crystals are turned around the normal of the substra-
te but their (100) planes remain nearly parallel to the
substrate surface.
STRUCTURE DETERMINATION OF THIN FILMS
129
Fig. 7.
( a) ( b)
......
.
.
. .. ....
.
. .
((
) )
d
.
... .. -
.
.
.
. .....
normal oblique
( a)
inci dence
( b)
(a) Diffraction diagram of the crystals of Fig. 6. The
substrate is turned around an axis lying in the substra-
te (see at rigth) which gives a diffraction pattern as
Fig. 7 (b) shows.
electrons scattered at the entrance of the specimen into a cone ;
the aperture of this cone increases with the penetration depth of
the electrons. Now a lattice plane which fulfils the reflection
condition takes out of this cone a whole "line" of electrons and
puts it at another place. One sees immediately that the Kikuchi li-
nes become broader the more the monocrystal breaks up into slightly
desoriented units each producing a Kikuchi line. With increasing
desorientation of the units the lines become very broad and disap-
pear in the background. The number of diffractlon spots becomes mo-
re numerous corresponding to the desorientation as mentioned above.
(1 )(2).
2.2. The size of the crystals
The powder rings give information not only on the randomness
of the crystal orientation but also to a certain extent on the size
of the crystals. In the simple approximation of geometrical diffrac-
tion theory the width of the diffraction spots is given by the num-
ber of atoms : the angular half width can be represented by AlMa
with the lattice constant a and M the number of atoms lying along
the diameter of the crystal. This means that the powder rings be-
come larger with decreasing M till they flow together.
The width of the diffraction spots produced by the crystal
130 H.RAETHER
size is independent of the value of d or of the angle 6, so every
rings is enlarged by the same amount. Thus the Debye-Scherrer rings
approach each other, especially at higher angles where they are
close together and disappear in the background first with decrea-
sing crystal size. Finally there remain in general two broad rings
in the fine cry tal line state. It may be mentioned that there exist
other reasons for a broadening of diffraction spots too. One of
these is a displacement of the atoms 8d varying from zero up to
a maximum value which causes an enlargement of the interference
maxima 86 proportional to the reflection angle 6 : by differentia-
ting the Bragg equation one obtains 86 6 which indicates
the dependance on the angle 6. This has to be taken into account
when interpreting the width of the diffraction maxima.
The above mentioned powder diagram with two or more broad rings
can be used to derive more information than just the rough esti-
mate of the crystal size. If we choose for example a monoatomic
substance whose atomic form factor f is given and assume that the
density of the atoms at distance r is n(r), we can derive rather
simply the diffraction intensity as a function of the diffraction
angle 6 and the wave length of the irradiating beam A as :
I(s) = const. f2. J
with s = sin 6.
sin sr n(r). 4
sr
We can obtain the unknown density n(r) by a Fourier transform
of the measured I(s) :
n(r) = const. f I(s).s.sin sr. ds
The behaviour of n(r) looks like Fig. 2 (b) : it shows (besi-
des the maximum ar r = 0) a pronounced short range order indicating
in a certain distance a shell of nearest neighbours whose number
can be obtained by integrating over tbe first peak. If further ma-
xima are obtained, information on further order can be derived.
This procedure can be applied to liquids as well as to amorphous
substances using electrons, X-rays and neutrons. The general result
of these diffraction experiments is that it is not possible to dis-
tinguish between the liquid and the amorphous state as well as bet-
ween the amorphous and the finest crystalline state. Other methods
are necessary to distinguish between these states.
3.3. Electron Microscopic Methods
The diffraction method gives information on properties avera-
ged over an extended region of the specimen.
More frequently we are interested in localised qualities of
STRUCTURE DETERMINATION OF THIN FILMS 131
the specimen such as tha size and the shape of the individual crys'-
tals which wompose the film. This information can be obtained by
collecting the diffracted beams whith the help of electron lenses
and thus constructing the image from the scattered amplitudes, to-
gether with their phases.
Now looking at the sreen of the usual electron microscope with
a polycrystalline films as object We see the boundaries of the dif-
ferent crystals, but we do not observe the lattice of atoms. This
comes from the fact that the lenses are not good enough to gather
besides the undeflected electrons the diffracted beams of at least
the first order which modulate the background of the undeflected beam
and produce a periodic intensity on the screen. However the quality
of the lenses is sufficient to gather those beans which are diffra-
cted by the boundaries of the crystals, whose diffraction angle is
much smaller than that of the first order, and which are therefore
summed up by the lenses. The diffraction angle of the boundary of
a linear lattice of length Ma amounts to sin a = AlMa for the first
minimum, see Fig. 8 and about 1,5 AlMa for the first maximum which
is essential for image formation. This explains the ability of the
electron microscope to see the crystals, but not their lattices.
Fig. 8.
Ma
Diffraction by a row of atoms. {The first subsidiary
minimum is indicated.
132 H.RAETHER
Fig, 9. Electron Microscope image of a thin gold film. The perpen-
dicularly crossed lines correspond to lattice planes of
2,04 A (by courtesy of Fa. Siemens, Berlin)
The intense work on the developement of better electron lenses
has successed in the last years in obtaining a spatial resolution
of a few A which is well below the spacing of some lattice planes
in some crystals, so that it is possible to observe the periodic
arrangement of these lattice planes as Fig. 9 shows.
We shall point out another development of electron microscopic
observation ; the scanning which has become of great
importance. Here the wave properties of the electrons are neglec-
ted, but the production of secondary electrons and the backscatte-
ring of electrons are of importance. The principle is rather simple:
0
an electron beam focussed by an assembly of lenses to about 100 A
and scanned over the surface (or film) which has to be investigated.
The electrons, backscattered or produced by secondary emission in
the focus are registered on the same side of the object point by
point as Fig, 10 demonstrates; thus the "image" is decomposed in
a series of intensity values following each other in time which are
gathered again on a screen storing these values ; an image of vary-
ing electron intensity is produced by this procedure and informs
us about the "structure" of the surface.
This scanning electron microscope thus gives information on
STRUCTURE DETERMINATION OF THIN FILMS 133
Fig. 10.
electron source
lens
object
Scheme of a scanning electron microscope : the primary
electrons focussed on a small spotcof the object pro-
duce secondary and backscattered electrons which are re-
gistered by an electron detector. By scanning the focus-
sed spot over the whole surface one obtains an "image"
of the object.
changes in the composition of the surface (the backscattering is
dependent on the atomic number Z of the substance) and on the geo-
metrical structure of the irradiated surface ; this comes from the
fact that the number of secondary electrons depends strongly on
the angle the incoming electrons make with the irradiated surface(3)
As already mentioned the electrons are registered on the "re-
flexion" side. It is also possible to look at the transmitted elec-
134 H. RAETHER
trons and to scan over the film observing the varying transmitti-
vity. However the type of scanning microscope commercially avai-
lable in general is that first described.
An interesting developement of the transf'lission scanning mi-
croscope is the following ; if one reduces the diameter of the fo-
o
cussed electron beam to 10 A or less and depos1tes a few molecules
well separated from each other on a very thin homogeneous substra-
te one can register these single molecules or atomic units by scan-
ning the electron beam over the molecules and observing them by
the change in the scattered intensity. This is another way to ob-
serve single molecules(4).
4. INELASTIC SCATTERING
The diffracti;on method uses the elastically scattered electrons.
This means that the electrons (or X-rays or neutrons) interacting
with the crystal have transferred or obtained a large momentum of
h.f (see the Laue equation mentioned above, h is Plank's constant)
or h+ (hi: Miller indices in the case of a cubic crystal).
The energy is negligibly small since the crystal has
an enormous mass compared with the electrons. The inelastic proces-
ses on the contrary are in general characterised by a rather large
energy transfer 10 eV) and a small momentum exchange h.f)
they lead to an excitation of the solid as e.g. a band transition
of crystal electrons, production of exitons or plasmons. To dis-
tinguish these different phenomena one has to analyse carefully
the energy losses of the electrons which have passed through the
film. Thus one obtains diagrams like Fig. 11, in which the inten-
sity of those electrons which have suffered an energy loss between
E and E + are registered. Peaks in this diagram indicate proces-
ses of high probability as just mentioned.
The inelastically scattered electrons thus contains rich in-
formation on the crystals composing the film ; more exactly one has
to say that the information is given by the complex dielectric fun-
ction = 1 + 2, since the intensity of the inelastic electrons
is proportional to + as deduced by the dielectric theory(S).
The field of energy losses has an interesting aspect : the re-
gistration of loss diagrams of organic substances has shown that
certain peaks change their intensity during the irradiation. If
other work has succeeded in relatLlg the different loss peaks to the
structure of the molecule, these changes indicate where the radia-
tion damage has occured.
STRUCTURE DETERMINATION OF THIN FILMS
135
Fig. 11.
J
4.9 9.4 15.0
AE (eV)
Energy Loss Diagram of 50 KeV electrons of a thin A1-
film: the peak at 15 eV represents the volume loss, the
two peaks at 4.9 and 9.4 eV come from surface plasma os-
cillations.
5. THE STRUCTURE OF SURFACES
Regarding the structure of a surface several questions arise
what is its roughness, what is its composition compared with the
bulk (especially in the case of alloys), are the lattice dimensions
and with them the band structure different at the surface, do sur-
face states exist ? etc. These problems are under investigation
with modern means such as photoelectron emission, LEED, etc., but
the results are not yet sufficiently settled to be presented here.
We shall discuss questions connected with the geometric struc-
ture, especially the roughness of the surface.
136
Fig. 12a.
H.RAETHER
( a )




x x
The reflexion of electrons on a rough surface :
polycrystalline (left) and monocrystalline (right).
STRUCTURE DETERMINATION OF THIN FILMS
( b)
Fig. 12b.
x
The reflexion of electrons on a smooth surface of
a monocrystal.
137
138 H.RAETHER
An approximate procedure is to look at the boundary with fast
electrons using the reflection method.
In the case of a rough surface the electrons hitting the sur-
face at grazing incidence give information on the peaks of the crys-
talline substance powder rings are produced, in the case of a mono-
crystalline material (e.g. an etched boundary) a point diagram (see
fig. 3) is expected; see fig. IZ. However, smooth surfaces, produ-
ced by cleaving monocrystals such as mica, graphite, NaCl etc.
give a different pattern since apparently the monocrystalline sur-
face is broken up into a mosaic by the cleaving process, in general
many Bragg spots are observed as in Fig. 12 (above), but the diffra-
ction maxima are elongated into streaks. This can be explained by
the finite penetration depth of the electron waves which limits the
resolution of the third Laue condition (perpendicular to the boun-
dary) and smears out the maxima in both directions perpendicular to
the surface. The penetration depth {D) is connected with the mean
free path (I) of the electrons (the length after which they suffer
an inelastic collision) by D = l/2 sin e. i is of the order of
1000 A for 50 KeV electrons and e 10-
2
so that D 10 A, a depth
which changes the interference maxima nearly into streaks. (In ad-
dition the electron waves are refracted crossing the boundary, but
this effect is neglected here). These streak patterns are characte-
ristic of a smooth surface. This small value of D demonstrates the
sensitivity of this method for surface observations. (2)
The same phenomenon is seen in low energy electron diffraction:
here the angle e is larger (sin e 1/2), but t is strongly reduced
to values of the order of 10 A so that the same value of penetra-
tion depth results and thus similar diffraction patterns. The sensi-
tivity of these methods makes it possible to recognise mono layers
of adsorbed material on the surface if the absorbed atoms are re-
gularly arranged on the monocrystalline substrate and become thus
observable as additional streaks.
More quantitative information is obtained by the method of
light scattering : we know from optics that a light beam falling on
a smooth metal surface at oblique incidence splits into a reflected
and a traasmitted beam. We understand this phenomenon by saying
that the incoming light beam excites a polarisation current at eve-
ry point of the surface by the component of the electric field per-
pendicular to the surface. Every point radiates light like a Hertz
dipole into nearly all directions. Summing up all the amplitudes
radiated from the whole surface there remains only those with phase
differences zero, the others being destroyed by interference ; in
other words the reflected and transmitted beam are the only ones
to be observed. If roughness exists on the smooth boundary this ra-
diates in addition into nearly all directions producing diffuse
light scattering by the irradiated surface.
In recent years the observation of surface plasmons has given
STRUCTURE DETERMINATION OF THIN FILMS 139
a new possibility of looking at surface roughness on metals : the
electrons of many metals can be regarded as a quasifree electron
plasma of high density (no)' It is possible to produce density
fluctuations of this plasma which travel along the as
a plane wave n = no sin(wt - kx) ; w is the frequency and k the
wave vector. The corresponding quanta are called plasmons. These
fluctuations may occur in the volume (volume plasma oscillations)
and in the surface (surface plasma oscillations). Surface waves
have their wave vector parallel to the surface. along which the
charge density fluctuations travel ; the electric fields belonging
to these charges decay exponentially into the metal and into the
vacuum. The frequency and the wave vector of these oscillations are
connected by a dispersion relation w(k) which determines also the
phase and group velocity of these waves.
Fig. 13.
light line
/
W 5 ( 00 ) - f- -- -- -::...:-;,..:;-.--.;;;..-----

t ....
The dependency of the frequency Ws of surface plasma
oscillations on the wave vector Ks (dispersion relation)
in the case of a quasifree electron gas. The arrows in-
dicate how the dispersion relation is displaced to hi-
gher Ks values on a rough surface.
140 H.RAETHER
Now if these waves run along a surface with a certain rough-
ness (or any periodic perturbation), they can be scattered and
transformed into light of the same frequency. By analysing the an-
gular distribution of this light and its polarisation it is possi-
ble to deduce the mean roughness of the surface. For example silver
films vaporised on a w ~ polished quartz substrate show a mean
roughness of about 10 A. This method can be further developed to
give still more information, from non-metallic surfaces too. (6)
As mentioned above the surface plasma oscillations have to
satisfy a dispersion relation ws(k
s
) which is drawn schematically
in Fig. 13. Now it has been found that this behaviour is slightly
but measurably changed, if the surface becomes rougher : the fre-
quency Ws is displaced to higher Ks values with increasing rougp-
ness. This gives a simple method of measuring the roughness of a
surface, a mathod which has to be further developed (7).
6. CONCLUSION.
It was the intention of these lectures to give a short insight
into various aspects of the structure of thin films and of the most
important methods of its determination. The references at the end
may be useful for more detailed information.
REFERENCES
1. L. Reimer, Electronenmikroskopische Untersuchungsmethoden
(Springer Verlag, Berlin) (1967).
2. H. Raether, Elektroneninterferenzen, Handbuch der Physik,
Vol. E, p. 443 (Springer Verlag, Berlin) (1957).
3. L. Reimer, G. Pfefferkorn, Rasterelektronenmikroskopie,
(Springer Verlag, Berlin) (1973).
4. A.V. Crewe, J. Wall, L.M. Welter, J. Appl. Phys. 39, 5861
(1965)
5. H. Raether, Solid State Excitations by Electrons, Springer
Tracts Mod. Phys. 38, 85 (1965).
J, Daniels et al.,:Springer Tracts Mod. Phys. 54, 78 (1970).
6. E. Kretschmann, Optics Communications, Vol. lQ, n04, p.353
(1974).
7. D. Hornauer, H. Kapitza and H. Raether, J. Phys. D. Appl. Phys.
Vol 2, L 100 (1974).
PHYSICO-CHEMICAL ANALYSIS OF THIN FILMS
A.CACHARD
Universite Claude Bernard /Lyon I
Departement de Physique des Materiaux
43, boulevard du 11 Novembre 1918
69621 - Villeurbanne France
1 . INTRODUCTION
In order to fully understand the behavior of thin films and
to ascertain the interpretation of their physical properties, a
number of basic parameters must be kno\vu and monitored.
The single most significant film parameter is its thickness
and details about the measurements are given in the paper of D.S.
Campbell. The problems involved with structure determination are
treated in the paper of H.Raether. In this paper, we will be con-
cerned with the analytical techniques for the determination of the
atomic composition (stoichiometry), the impurities and the bon-
dings of thin films.
A large number of analytical techniques have been developped
or modified for thin film specimens which typically consist of 1
to 10 cm
2
surface area and 10 to 1000 ~ material. A detailed dis-
cussion of all these techniques is beyond the scope of this paper
and we will not attempt to reproduce here that which already
exists in general reviews and texts(1-3). This paper will concen-
trate upon rather new techniques which have been developped and
increasingly used in the last few years; the nuclear microanalysis
(MeV ion beams), the secondary ion mass spectrometry (keV ion
beams) and X-ray or photoelectron spectroscopies.
Special attention is given on lateral or in depth resolu-
tions, detection limits, and quantitative or qualitative possibi-
li ties.
141
142 A. CACHARD
2. COMPOSITION DETERMINATION BY MeV ENERGY RANGE ION BEAM
ANALYSIS
2.1. Introduction
Basically the method consists of placing a target in a mono-
energetic beam of ions of hydrogen, helium or any other lightele-
ments and of detecting the interaction products which provide in-
formation on the surface layer. When incoming ions A bombard
atoms B of the target there are three main interaction possible :
their trajectory can be deviated without energy loss and
this is called elastic scattering ; or there can be nuclear reac-
tion : A+B + C+D + Q, where Q is the energy dissipation, D the
remaining nucleus and C the emitted species (proton, a-particles,
y rays, etc.) ; or only an electronic interaction can occur where
B is excited and then desexcites via X-ray emission (the so-called
ion induced X-ray emission). The activation process case where
the result of the reaction is a radioactive nucleus will not be
considered here since in general it needs higher energy ions.
The instrumentation required is standard in nuclear physics.
A particle accelerator provides a mono-energetic ion beam, the
size of which is of the order of 1 mm
2
and the intensity varies
from nA (inbackscattering measurements) to (nuclear reaction,
case). The interaction takes place in a target chamber in which
the vacuum is of the order of 10-
5
- 10-
6
torr. Most of the time,
the interaction products are detected using a semiconductor de-
tector which has the advantages of simplicity,ability to be used
with large solid angles,insensitivity to the charge state of the de-
tected particles and the capability of non dispersive measurement.
The electronic instruments (preamplifier and linear amplifier)
required for the signal analysis deliver pulses whose heights are
proportionnal to the energy of the incident particles. A multi-
channel pulse height analyser gives the energy spectrum of the
emitted particles.
2.2. Backscattering analysis
The elastic scattering of an incoming particle of mass m by
a target atom of mass M is characterized by the conservation of
the momentum and of the kinetic energy. This gives a simple rela-
tionship between the energy E of the incoming and its
energy E' after scattering in the e direction( -6)
where k
m cos e + IM2-m
2
sin
2
e
m+M
( I)
PHYSICO-CHEMICAL ANALYSIS OF THIN FILMS
For 9 = 180 the formula is very simple
k
M-m
M+m
143
For a given particle and scattering angle, k (and thus E') is
dependent only on the atomic mass of the target atoms. Backscatte-
ring provides mass-sensitive analysis. The selectivity ~ dk
2
/dM)
decreases with increasing M, and increases with increasing 9 and
m. However, the selectivity is rather poor and it allows the dis-
tinction of elements and isotopes only up to about 40. Above this
weight, two elements can be distinguished only if their mass dif-
ference is larger than 10 atomic units. However, in special
cases(4-7), isotopes such as 63Cu and 65Cu have been separated.
The sensitivity of a backscattering analysis is directly re-
lated to the differential scattering cross-section which for
E < 3 MeV is given by the Rutherford formula :
do
d ~
(2)
where z and Z are the atomic number of the incident ion and target
atom, respectively. For a given ion, the sensitivity is a decrea-
sing function of 9 and E and an increasing function of Z.
Typically in backscattering analyis, one uses large angles
(9 > 160
0
),energies in the MeV range, and heavy ions. Most of the
time, 4He+ ion beams are used; different analysing particles such
as lithium(8), carbon(9) and oxygen(7) beams have also been used.
The limit of detection is routinely in the submonolayer range and
can be as low as 10
10
atoms/cm
2
in special cases (gold on silicon
using 12C beams(9)).
Thick film analysis
For film thickness larger than 300 A, the ion beam loses
energy as it penetrates the film, and the signals from greater
depth will have lower energy.
In figure I, the incoming ion beam has an energy Eo. After
scattering by a surface atom the energy in the 9 direction is
E ~ = k2Eo. At a depth x, the incoming particle energy is :
I
x dE
EI = Eo - \ dx dx\
o
144 A.CACHARD
x
target
Fig. 1 Particle trajectories in a thick film
countsr-__ ________ ________ __________ __________
2000
1000
....
. e.- .
. . ..
-
-
.

. ':. :': :". :':';:';'JAI
.,
.
.'
-'
.. -
.............. ., .................... . chan pel
.'
100 200

A
300 400

A
... y"
500
Fig.2
0
Energy spectrum for 1.9 MeV 7Li+ ions backscattered from
1200 A self-supporting anodic A1203
PHYSICO-CHEMICAL ANALYSIS OF THIN FILMS
where dE/dx is the energy loss depending on the target material
and on the ion. The particles scattered at the depth x have an
energy E2 = k2El' and emerge from the target in the e direction
wi th the energy
E'
x
x
dE
dx
dxl
(3)
E' is significantly lower than E' and the receives
145
x ""lb (0" )"
a e eam for a target mass as
does in the case for very thin films. For for
in alumina dE/dx is of the order of 30 eV/A which, for a 1000 A
thick film, gives an energy difference of 70 keV between surface
scattered particles and back-side scattered particles. With a ty-
pical IS keV electronic resolution, such a difference is easily
measured. Figure 2 gives the energy spectrum for 1.9 MeV 7Li+
ions back-scattered from 1200 A self-supporting anodic A1203' The
correspondance between the energy scale and the depth scale is
obtained by the knowledge of dE/dx. The energy loss is given in
semi-empirical tables(IO-II) or can be measured in the case where
more accurate information is needed. dE/dx depends on E, but for
thicknesses lower than 0.5 vm, one can consider dE/dx to be
nearly constant thus, the energy difference = - is
nearly equal to
(4)
where [S] the energy loss parameter is given by
[S] (5)
and El Eo and E2 E2
[S] depends on the mass and energy of the incident particle,on
the composition and on the nucleus of the target under study. In
the case of alumina [S] is different for particles scattered by
oxygen and by aluminium.
Quantitative analysis
The total number of a certain atom (per cm
2
) can be obtained
from the area in counts under the peak characteristic of that
atom. For a total number Q of incident particles and a detector
solid angle n, the area of the peak A is given by :
A = Qn do N
dn
(6)
146 A.CACHARD
where N is the number of atoms per cm
z
. This formula allows the
determination of N as long as do/drl is constant; for thick films
a correction should be made to account for the change of do/drl
with energy.
The composition of a compound can be also obtained from the
spectrum using the same formula (6). In the case shown in figure 2
the ratio of the number of oxygen to the number of aluminium is
given by :
No Ao do I do
-N- = A (dn) Al (drl)O
Al Al .
(7)
The ratio of the cross sections can be calculated from the
Rutherford formula. Actual values of compositions can be obtained
within 1 %.
An in-depth analysis (concentration profile determination)
can be performed using the characteristic shape of the peak under
study. The multichannel analyser gives the energy spectrum in num-
ber of counts per channel (figure 3) where each channel represents
a step oE in energy. At energy E (actually between E and E + oE),
the total number of counts is t.n{E) and the "peak area" is t.nGE)oE.
This peak area represents the number of detected particles scatte-
red by a thin film of thickness ox = oEIIS] located at a depth
x = Eo-E/[S). If n{x) is the atomic concentration (in cm-
3
) , ex-
pression (6) can be written
t.n(E)oE
n(x)
do
Qrl drl n(x)dx
t.n(E)[S]
Qn do
drl
do
Qrl drl n(x)[S]oE
or
(8)
The measurement of t.n{E) (peak shape) and the knowledge of
[S] and do/drl allow the concentration profile determination. Fi-
gure 4 gives the 4He+ backscattering spectrum from 5400 A SiOz
grown on a Si substrate and implanted with 64Zn at 280 keV to a
dose of 1015/ cm
Z
(12). The two plateaus of silicon are indicative
of the concentration of Si in SiOz and of Si in silicon. The ratio
of the heights of these plateaus indicates stoechiometric SiOz, and
the ratio of total area of oxygen and silicon (in SiOz) gives the
same result. This example shows that the determination of a light
element (here oxygen) in a heavier matrix (here SiOz) results in
a superposition of a peak on a continuous background (here due to
bulk silicon) and is not the best case of the application of back-
scattering. The more favorable situation where a heavy impurity is
PHYSICOCHEMICALANALYSIS OF THIN FILMS 147
Counts
lIn(E)

Fig.3 Energy spectrum for ions backscattered from a thick film
Counts (xlO-
3
)
SiZo


12

2.0 MeV
Si
4He+

Rp
I Si
8
LAEo
'.\
. .1

4
1
r.
I
Energy (MeV)
0.4
0.8 16
Fig.4 spectrum of 2 MeV 4He+ ions backscattered from a
sample (5400 A of Si02 on Si) implanted at 280 keV with 64Zn.
The oxygen and silicon components of Si0
2
are the shaded areas,
and 6
4
Zn is the dotted area (taken from ref. 12).
148 A. CACHARD
in a light element matrix, is shown by the peak of Zn located in
the background free region of the spectrum. This peak is shifted
to lower energy by an amount oE
Zn
corresponding to the depth dis-
tribution of Zn below the surface.
o
Typically the depth is of the order of 200 A, and
can be made as low as 100 A in special cases, by proper choice of
analysing ions and of element to be analysed.
In conclusion ion backscattering analysis is a method perfec-
tly adapted to the study of films that are self-supporting or on
light substrate (carbone). It is also well adapted to the study
of heavy impurities in light element matrices. This method has
been used to measure the composition of dielectric layers(l3-16)
to study the formation of anodic oxide films(17) and of silici-
des(18-20) and to determine the implantation profile of dopants
in silicon(21,22).
2.3. Analysis by nuclear reactions
The bombardment of nuclei of mass M by MeV energy range ions
of mass m can induce nuclear reactions, yielding emitted parti-
cles of mass m' and remaining nuclei of mass M'; the defect mass
energy is the Q value of the reaction. Such reactions are usually
written: M(m,m')M'.
The basic advantage of nuclear reaction analysis with respett
to back-scattering is that medium and high Z nuclei practically
do not contribute to nuclear reactions, owing to Coulomb barrier
repulsion ; this allows background-free detection of light ele-
ments on heavier substrate. Due to the high positive Q values of
many such reactions, the emitted particles have energies well
above that of the beam, allowing the elastically scattered parti-
cles to be stopped in a suitable absorber. The second main advan-
tage is that nuclear reactions are spe"cific : two isotopes of the
same element behave in a completely different way, and in many ca-
ses only one of the isotopes has a positive Q value, the other being
ignored less regard of its concentration. This is in contrast to
backscattering in which only the k factor is slightly changed.
The basic example is the first use of nuclear reactions in
1962(23) for the study of oxygen transport in anodic aluminium
oxide, using 160(d,p)017 and 180(p,a)15N'reactions. The nuclei
now routinely determined this way are 2H, 6Li, 7Li, lIB, 12C, 14N,
19F, 27Al, 28Si, 31p. More details may be found in references 24 to
31.
The selectivity of this method is not easily discussed due
to the specificity of the nuclear reactions. The only limitations
PHYSICO-CHEMICAL ANALYSIS OF THIN FI LMS
...
..
"-
.Q
E
I:.
...
..
....
.Q
e
10
b 5
.,.01' .,.
.'
.............................
400 100
60
...
40i-
e
1:. 20
...........
:.
.....
.......
J ....
,.'"
500 100
' .
.'
... ...............
1.5
"
"
"
,
, .
. ,
14
N1d
,d.J 12C
0.5
1000 Ed KeV
" ......... , .
900 Ep KeV

, . .
'-______ -'-_______ .L.-__ Ed KeV
ISOO 1100 1900
149
Fig.5 Differential cross-sections of the reactions used to
analyse 16 0,
18
0 (at 165) and 14N (at 160).
16
0 and
18
0 curves
are taken from ref.24, and 14N curve from ref.32.
150
o


C1l
..c::
(
0 +, -=* ====::::::=:=====:
8
Z1
'/0 p) N+'l
o
o
.0
, \.0
TV 8Z( 1+0 d' p) IVa + -=::::::::::::
"'''''

+<
-::t
8Z[ (1'/0' p) N+'l
NS1(Z+ld'p) N+'l + <.
,

81 (od'p) 8
Z1
+ C')
N
S1
(Ed'p)N+'l +
.
-+
o
...
"

N Sl ( 9 d' P) N +,1 -7- ---====:;::::.-:..._?

,.' 0
-'l"1-----l..------'--------'--------' ...... 0
o
-.....
z
A.CACHARD
PHYSICO-CHEMICAL ANALYSIS OF THIN FILMS 151
are possible interferences between two peaks in the energy spec-
trum arising when several light nuclei are simultaneously present
in the sample. Each case is to be studied, and the optimal nuclear
reaction at the optimal energy must be determined.
The sensitivity, as for backscattering, depends on the diffe-
rential cross-section of the reaction under study, but unlike
backscattering there is no analytical expression. The cross-
section curves have to be determined experimentally. Examples are
shown in figure 5 for 16 0,
18
0 and 1
4
N. In the case
16
0, there is
a plateau between 800 and 900 keV for the 160 (d,Pl)
17
0* reaction
(this reaction is more efficient to use than that leading to the
fundamental level of
17
0). For
18
0, the 180(p,<l) 15N reaction pre-
sents a stationary cross-section near 730 keV, and a strong narrow
resonance at 629 keV. For 1
4
N, there are a variety of deuteron
induced reactions, but those leading to <l-particles emission are
the more interesting. The <lo-particle from 14N (d,<lo) 12C is very
energetic and less liable to interfere with other reactions ; its
cross section presents a good plateau between 1650 and 1730 keVS32)
Figure 6 gives the typical energ spectrum from the deuteron
bombardment (E = 1700 keV) of a 1700 A thick aluminum nitride
film on nickel backing. The peaks of the 27 Al (d,po) 28AI and 14N
(d,<lo) 12C reactions are located in a background free part of the
spectrum and allow a precise determination of the composition of
the film. Simultaneously the oxygen and carbon contaminants are
visible.
Quantitative analysis
The total number N of atoms (per cm
2
) in the surface region
of the sample is obtained from formula (6) as for backscattering
experiments :
N
A
The bombarding energy is chosen near a plateau of the diffe-
rential cross-section curve, so that do/dO appears as a constant.
The absolute quantity N may be obtained by comparison to refe-
rence standards regardless of their physical or chemical states
since the reaction yield depends only on nuclear cross sections.
As an example, by using
16
0 (d,Pl)
17
0'* reaction, less than
a monolayer of oxygen (10
14
atom/cm
2
) can be detected within mi-
nutes with a precision of a few percent.
The composition of a compound can be obtained in the same
152 A.CACHARD
way. From figure 6, the ratio of the number of nitrogen to the
number of aluminium can be calculated
using the 1ltN(d,ao)12C and 27Al(d'Po)28Al deuteron induced reac-
tions. The ratio of the cross sections is determined from a stan-
dard of known nitrogen/aluminium ratio (in this case, aluminium
oxinate(32. The composition can be measured within a few percent
o
prec1s10n even for films less than 100 A thick. Such a method has
been used also for SiO
x
(30), A1203(31) and Si
3
N
lt
(33,15).
The concentration profile may be obtained by analysing the
spectrum of the detected particles. The method is the same as that
developped for backscattering spectra, but for nuclear reactions
the depth resolution is rather poor. This is due to the lower
masses of the bombarding particles (usually protons or deuterons
instead of a-particles or heavier ions common in backscattering)
giving lower energy loss in the film, and to energ straggling
in the absorber. Typical depth resolution is 3000 A.
In the particular case where there exists a strong narrow re-
sonance in the cross-section curve, the concentration profile may
be obtained with better depth resolution(23,24). Figure 7 gives
the principle of concentration profile measurements using narrow
resonance.
The yield of the reaction induced by protons is almost zero
compared with its value at the energy E
R
. If the beam energy is
Eo > ER the protons, upon slowing down, reach the resonance energy
ER at a depth x = Eo-ER/[S] where [S] is the energy loss parameter
for the protons in the matrix under study. For energy Eo the reac-
tion yield comes only from a narrow sJab of the film near the depth
x, and the yield is proportional to the concentration C(x). By
varying Eo from ER towards higher energies, the resonance is dis-
placed from the surface to the interior of the film. Hence the va-
riations of the yield versus Eo give an image of the concentration
profile C(x).
Such a method has been extensively used by Amsel and co-
workers especially for lliO tracing to study oxygen transport during
anodic oxydation of metals(23,24,34). They used the 6&9 keV reso-
nance of the 180(p,a)lSN reaction which allows a 200 A depth reso-
lution in Ta
2
0
S
' Figure 8 shows a typical result obtained with the
27Al(p,y)28Si very narrow resonance at 992 keV. The target is a
0.8 ~ thick evaporated aluminium layer on tantalum backing with
a 475 A thick anodic oxide film(24).
PHYS!COCHEMICAL ANALYSIS OF THIN FILMS
do
dn
film
~
r::
/
/
/
beam
/ ER
Eo
/
/
/
/
/
~
surface
x
a. b
Fig.7 Principle of concentration profile measurements
using narrow resonances.
N/ 1 0 0 0 . - - - . . - - - . . - - . . - - ~
10
130 eV/channel
5
/"
20 40 60
Fig.8 Typical excitation curve ~ f the 27Al (p,y)
28
Si
resonance near 992 keV for a 475 A thick A1203 film on
0.8 pm aluminium (taken from ref. 24).
153
154
A.CACHARD
The sharp r i ~ and the good plateau correspond to a depth
resolution of 60 A in Alz03' The ratio of the plateau heights ~ l
in Alz03 and ~ z in aluminium indicates an actual Alz03 composition
for the oxide film.
The same reaction on aluminium has been used by Dunning(29)
to determine the concentration profile of aluminium implanted si-
licon oxide.
2.4. Ion induced X-ray emission
A target placed in a particle beam (either electrons or ions)
exhibits X-ray emission due to the vacancies created by the ejec-
tion of inner orbital electrons. The subsequent reordering of the
electronic shells induces X-ray emission which is characteristic
of the target atom. The analysis of the energy and intensity of
the emitted X-rays allows the determination of the type and the
concentration of the atoms present respectively.
Ion induced X-ray emission(36,37) has two main advantages
over the electron microprobe(38). Calculations show that the ma-
gnitude of the continuous electromagnetic radiation (bremsstrah-
lung) which accompagnies X-rays produced by electron irradiation
is reduced by a ratio of (Mion/Me)Z, a 10
6
factor under ideal con-
ditions. Evenmore this bremsstrahlung varies as (Z/M)Z which gives
an advantage to heavier ions. In fact, however, a continuous back-
ground exists due to secondary electrons ejected along the ingoing
particle path. The second advantage is that the X-ray production .
cross-section for heavy ions exhibits an energy cut-off above
which the cross-section increases very rapidly as a function of
energy. Thus, by varying the ion energy selective excitation(39)
~ s possible.
The use of solid state Si(Li) detectors with a high efficien-
cy and a good resolution (150 eV) results in good selectivity;
in principle all elements can be separated, except in the case
where elements adjacent on the periodic table exist in the target
in greatly differing concentrations. In such a case the peak of
the more highly concentrated element overshadows that of the other.
For example Musket and Bauer (40) have analysed vacuum annealed
stainless steel by means of proton induced X-rays. They were able
to separate the contributions of iron, nickel and chromium. Such
selectivity cannot be achieved by backscattering analysis.
The sensitivity of this X-ray technique is very high, and
Gray et al(41) claimed to have measured arsenic in silicon implan-
ted at a level of 2.10
13
As/cm
z
.
On the other hand, the depth resolution ~ s very poor, if any,
PHYSICOCHEMICAL ANALYSIS OF THIN FILMS 155
because the result of X-ray absorption is the decrease of intensity
without changes in energy. Thus, a priori we can get no informa-
tions from the peak shape. However, using the cross-section versus
ion energy relationship, Musket(40) was able to obtain a concen-
tration profile but with an in-depth resolutions higher than
0.25 )Jm.
In conclusion, ion induced X-ray emission appears as a pro-
mising tool for surface and thin film analyses. This is a very
new technique and both the fundamental theory behind it and prac-
tical applications are presently the subject of a number of inves-
tigations.
2.5. Conclusion
Backscattering and nuclear reaction techniques have been used
successfully in microanalytic studies of surfaces and thin films.
To a large extent the optimum technique depends on the nature of
the problem. For example, nuclear reactions are more suitable for
low Z elements in heavier substrates and backscattering for heavy
impurities in light matrices. For light element dielectric films
such as Si0
2
, Si3N4' A1203, etc. either of the two methods can be
used by varying the substrate (heavy for nuclear reactions and
carbon for backscattering). These two methods appear to be com-
plementary.
For both methods the results are quantitative, a 1 % preci-
sion being easily reached. The sensitivity may be very high (10
12
atoms/cmf) and the analysis is non destructive. The results are
generally independent of the physical or chemical states of the
nuclei in the matrix, the only exceptions being the channeling
effect in precisely oriented single crystals(35). Depth distribu-
tions may be obtained in the first micron of a film with a reso-
D
lution of the order of 100 A, but the lateral resolut10n is poor
(larger than 0.1 rom).
Ion induced X-ray emission, a more recently developped tech-
nique, is of great interest both as an analytical tool in itself
and as a complement to backscattering analyses.
3. SECONDARY ION MASS SPECTROMETRY
When bombarding a target by a 5-30 keV beam of primary ions,
the surface layers are sputtered off and a variety of secondary
species is produced including neutral atoms or molecules, positive
or negative ions, electrons and photons. The mass spectrometric
analysis of the positive and negative sputtered secondary ions is
a useful method for characterizing the target(42-44). The primary
beam size can be varied from 1 to 300 )Jm with scanning possibili-
156 A.CACHARD
ties and the secondary ions are collected and analysed either by
a classical mass spectrometer or by a direct imaging analyser(42);
both cases ensure good lateral resolution.
The ion production arises by two different processes : "kine-
tic" and "chemical". The former process is the transfer of kinetic
energy between primary ions and target atoms, which can be ejected
from the surface. Unbound electrons in the excitation volume cause
the ejected ions to be neutralized before they escape into the
vacuum where their state is metastable. This metastable atom be-
comes ionized by Auger electron emission and can be ion mass ana-
lysed. This "kinetic" ionization process predominates in inert gas
bombardment of metals and semi-conductors.
The "chemical" ionization process is due to the reduction of
the number of free electrons available for ion neutralization be-
fore escaping. Such reduction increases the production of seconda-
ry ions and occurs when reactive species such as superfical oxide
or bulk chemical compounds (oxides, chlorides, etc.) are present.
It is assumed that the two processes exist together, and this
can lead to ambiguities in the interpretation of secondary ion
intensities because chemical variations in the sample can cause the
intensity to change even if the actual concentration remains cons-
tant In order to avoid such difficulties and to ensure high
emission yield two techniques are currently used. In the first, a
reactive species such as oxygen is introduced into the residual
vacuum. In the second, the primary beam consists of oxygen ions. In
both cases a surface oxide is continously formed and ensures the
predominance of the chemical ionization process.
The quantitative analysis makes use of the erosion capabili-
ties of the sputter ion source to obtain three dimensional analy-
sis. Lateral analyses (x,y microanalysis) are possible with less
than 1 lateral resolution the electron microprobe).
In depth analysis is a result of the sputter removal of material
by the primary ion beam. If we can assure a constant erosion rate,
the ion current versus time relationship is easily converted into
a concentration versus depth curve, but the correlation of time
and depth requires careful attention due to the great variety of
factors influencing the eros!on rate. In the best cases the depth
resolution is less than 100 A.
Quantitative measurements are'achieved using the comparative
standard technique which requires a precharacterized sample of the
element to be analysed in the matrix tobe employed. Such a compli-
cated process explains why up to now such a technique is only
routinely employed in a few cases like silicon technology. Most of
the time in other applications the analysis is only semi-quantita-
tive.
PHYSICOCHEMICAL ANALYSIS OF THIN FILMS 157
However, secondary ion mass spectrometry is very useful for
thin film study(42,44) mainly because of its superior lateral and
depth resolutions. All elements can be analyzed with good isotopic
separation and great sensitivity (10-
15
to 10-
19
grange).
4. X-RAY AND PHOTOELECTRON SPECTROSCOPIES
The methods described above in sections 2 and 3 give informa-
tion about the atomic composition of a film but not about the che-
mical surroundings of the atoms. For nuclear methods this uncapa-
bility is due to the physics involved in the phenomena. However,
this is not the case for ion induced X-ray analyses where the che-
mical shift of characteristic X-rays would be evaluated by use of
a crystal detector.
Evaluation of the chemical bonding in various glass films has
been performed by Pliskin(45,46) using the infrared absorption
technique which deal with direct absorption of electromagnetic ra-
diations by electric dipoles oscillating in a molecule.
Information on changes in the electronic structure of atoms
due to differences in the chemical state of these atoms can be
obtained using either X-ray or photoelectron spectroscopy.
4.1. X-ray spectroscopy
When an atom is irradiated by an X-ray beam transitions bet-
ween electronic levels (either emission or absorption) occur which
cause peaks in the subsequent energy spectrum. In the case of
changes in chemical bonding, the electronic distribution around
the atom changes and the observed peaks shift in energy.
For a solid only the inner shell maintains its atomic charac-
ter while the outer shell exhibits a band structure. The X-ray
emission results from the transition between a conduction band
state (metal) or a valence band state (semi-conductor or insulator)
and an inner shell energy level with rather atomic character. A
broad emission band is thus observed which give a picture of the
density of states of the ~ u p i e d levels.
The X-ray absorption is the result of the transition between
an inner orbital level and an empty state of the conduction band.
Thus, an X-ray absorption spectrum gives an image of the empty le-
vel density of states.
In the case of an insulator or a semi-conductor, the energy
difference between the X-ray emission and absorption band edges
(with the same inner shell level) sometimes yields data about the
energy gap. However, in many compounds such a difference has no
real physical meaning because of the existence of intense absorp-
158 A. CACHARD
tion lines superimposed on the absorption discontinuity. Such
peaks are indicative of transitions towards localized orbitals
with atomic character.
By varying the chemical surroundings of an atom, a shift is
induced either in the emission or in the absorption band edge and
can give information, for instance, on the degree of oxidation of
a metal in different compounds.
An exhaustive review of X-ray spectroscopy has recently been
compiled by C.Bonnelle(47).
4.2. Photoelectron spectroscopy
Photoelectron spectroscopy is often referred to as ESCA
(Electron Spectroscopy for Chemical Analysis). In this technique,
the surface of a solid is bombarded by a monoenergetic X-ray beam
and as a result photoelectrons are emitted with a kinetic energy
which is roughly the difference between the primary energy of the
excitation beam and the binding energy of the photoelectrons. The
use of a high energy resolution electron spectrometer allows the
measurement of the energy levels in the solid(48,49).
The advantages of this method from X-ray absorption are that
the core levels can be seen alone giving sharp line spectra and
that the absolute binding energy can be obtained. On the other
hand, only occupied states are accessible, and in that sense,
the two methods are complementary.
The two main applications in thin film field are the bonding
determination in a compound and the band structure measurement
in solids(50). Quantitative concentration measurements can be ob-
tained from the intensities of the emission lines. For such an
elemental analysis both heavy and light elements can be detected
with good sensitivity (less than 10-
8
g). ESCA, however, is res-
tricted to surface analysis due to the low average depth (10 A)
from where photoelectrons can be observed.
5. CONCLUSION
There is a large variety of techniques for physico-chemical
evaluation of thin films, and those described in this paper have
proved to be extremely useful. Of course, the choice of analytical
methods always depends on the purpose of the particular investiga-
tion (atomic composition or bondings), the nature of the film and
the facilities and financial support available.
It must be stressed that a single technique is seldom suf-
ficient to satisfactorily characterize a film ; in the vast majo-
rity of cases, it is necessary to make use of several complementa-
ry methods.
PHYSICO-CHEMICAL ANALYSIS OF THIN FILMS
REFERENCES
1. The Use of Thin Films in Physical Investigations, Edited by
J.C.Anderson, Academic Press (1966)
2. Thin Film Phenomena, K.L. Chopra, Mc Graw-Hill Book Company
( 1969)
3. Handbook of Thin Film Technology, Edited by L.I.Maissel and
R.Glang, Mc Graw-Hill Book Company (1970)
4. M.A.Nicolet, J.W.Mayer and I.V.Mitchell, Science, 177, 841
(1972)
5. M.Peisach and D.O.Poole, Ana1.Chem. 38, 1345 (1966)
159
6. W.K.Chu, J.W.Mayer, M.A.Nicolet, T.M.Buck, G.Amsel and F.Eisen,
Thin Solid Films 12, 1 (1973)
7. S.Petterson, P.A.Tove, O.Meyer, B.Sundquist and A.Johansson,
Thin Solid F i l m s ~ 157 (1973)
8. JP.Thomas, A.Cachard, M.Fallavier, J.Tardy, S.Marsaud and
J.Pivot, Rev.Phys.Appl., to be published
9. F.Abel, G.Amsel, M.Bruneaux, C.Cohen, B.Maurel, S.Rigo and
J.Roussel, J.Radioanal.Chem. ~ 587 (1973)
10. C.F.Williamson, J.P.Boujot and J.Picard, CEA Report R.3042
(1966)
11. L.C.Northcliffe and R.F.Schilling. Nucl.Data Tables, ~ 233
(1970)
12. W.K.Chu, B.L.Crowder, J.W.Mayer and J.F.Ziegler, Appl.Phys.
Letters, 22, 490 (1973)
13. I.V.Mitchell, M.Kamoshida and J.W.Mayer, J.Appl.Phys.42,
4378 (1971)
14. O.Meyer, J.Gyulai and J.W.Mayer, Surface Sci. 22, 263 (1970)
15. M.Croset, S.Rigo and G.Amsel, Appl.Phys.Letters, ~ 22 (1971)
16. D.V.Morgan and R.P.Gittins, Phys.Status Solidi (a), 11,517
( 1972)
17. M.Kamoshida and J.W.Mayer, J.Electrochem.Soc., ~ 1084
(1972)
160 A.CACHARD
18. A.Hiraki, M.A.Nicolet and J.W.Mayer, Appl.Phys.Letters, ~
178 (1971)
19. R.W.Bower and J.W.Mayer, Appl.Phys.Letters, 20, 359 (1972)
20. C.J.Kircher, J.W.Mayer, K.N.Tu and J.F.Ziegler, Appl.Phys.
Letters, 22, 81 (1973)
21. J.Haskell, E.Rimini and J.W.Mayer, J.Appl.Phys., 43, 3425
(1972)
22. J.W.Mayer, L.Eriksson and J.A.Davies, Ion Implantation in
Semi-conductors, Academic Press, New-York (1970)
23. G.Amsel and D.Samuel, J.Phys.Chem.Solids, 23, 1707 (1962)
24. G.Amsel, J.P.Nadai, E.D'Artemare, D.David, E.Girard and
J.Moulin, Nucl.lnst.and Meth., 92, 481 (1971)
25. G.Amsel and D.David, Rev.Phys.Appl., ~ 383 (1969)
26. E.Ligeon, A.Bontemps, J.Fontenille and G.Guernet, Proc.
Microelectronics Conf., Paris, 1970, Chiron, Paris, p.50 (1970)
27. B.Maurel, D.Dieumegard and G.Amsel, J.Electrochem.Soc., !.!2.,
1715 (1972)
28. M.Croset and D.Dieumegard, J.Electrochem.Soc., 120,526 (1973)
29. K.L.Dunning, G.K.Hubler, J.Comas, W.H.Lucke and H.L.Hughes,
Thin Solid Films, ~ 145 (1973)
30. A.Cachard, J.Pivot, A.Roger, M.Talvat and J.P.Thomas, Rev.
Phys.Appl., ~ 279 (1971)
31. C.Diaine, J.A.Roger, J.Pivot, A.Cachard and C.Dupuy, Congres
AVISEM, Paris (1971)
32. J.Tardy, A.Cachard, J.P.Thomas and J.Engerran, Le Vide, 173,
359 (1974)
33. A.Cachard and J.Tardy, unpublished.
PHYSICOCHEMICAL ANALYSIS OF THIN FI LMS
34. C.Cherki, Thesis, Paris (1969)
35. D. S. Gennnel, Rev. Modern Phys., 46, 129 (1974)
36. J.A.Cairns, A.D. Marwick and I.V.Mitchell, Thin Solid Films,
~ 91 (1973)
37. J.D.Garcia, Phys.Rev. A4, 955 (1971)
3S. E.Reuter, Surface Sci., 25, SO (1970)
39. J.A.Cairns, Surface Sci., 34, 638 (1973)
40. R.G.Musket and W.Bauer, Thin Solid Films, ~ 69 (1973)
161
41. T.J.Gray, R.Lear, R.J.Dexter, F.N.Schwettmann and K.C.Wiemer,
Thin Solid F i l m s ~ 103 (1973)
42. G.Slodzian, Rev.Phys.Appl., 1, 360 (1968)
43. C.Evans Jr, Thin Solid Films, ~ 11 (1973)
44. A.Socha, Surface Sci., 25, 147 (1971)
45. W.A.Pliskin, J.A.Perri and D.R.Kerr, Physics of Thin Films,
(Ed. G.Hass and R.E.Thun), Academic Press, New-York, ~ 257
( 1967)
46. W.A.Pliskin, Progress in Analytical Chemistry, Vol.2 : Physi-
cal measurementc and analysis of thin films, Plenum Press,
New-York, p.168-192 (1969)
47. C.Bonnelle, in Physical Methods in Advanced Inorganic Chemis-
try, Ed.H.A.Hill and P.Day, Interscience Publishers, p.45-73
( 1968)
48. K.Siegbahm,C.Nordling, A.Fahlman et aI, Nova Acta Regiae Soc.
Sci. Ups aliens is Ser.IV, 20 (1967)
49. K.Siegbahm, Trans.Roy.Soc.London, Ser.A, 268, 33 (1970)
50. T.Di Stephano and D.Eastman, Phys.Rev.Letters, 27, 1560 (1971)
THICKNESS MEASUREMENTS
D.S. CAMPBELL
Department of Electronic and Electrical Engineering
University of Technology
Loughborough, Leicestershire, U.K.
I. INTRODUCTION
The measurement of the thickness of thin films is usually of
great importance for examining the properties of the films, as it
is the very thinness which often gives rise to the properties
which cause the film to be different from those of bulk materials.
As a result various summaries have been given previously of the
techniques availables(I)(2).
Methods for measuring thickness can best be divided into two,
and these are :
A. Direct measurement. In this technique a direct measure of the
thickness of the film is obtained after deposition. Not all direct
methods of measurement are absolute as often some previous cali-
bration against another technique has to be employed. Nevertheless,
such systems are widely used.
B. Indirect methods. Indirect methods depend on the control of
the rate of deposition of the film so that provided a previous ca-
libration has been made, it is possible to prepare a layer with
an exact thickness. Rate control devices by their very nature,
will depend on a calibration against direct thickness measurements
using either absolute or derived systems.
This paper examines the various techniques for films prepared
by the various methods that are available.
163
164 D.S. CAMPBELL
2. DIRECT THICKNESS MEASUREMENTS
2.1 Interference Methods
The fundamental method of measuring films most commonly used
is that using interference technique(3)(4)(5) and this method has
provided the reference technique against which most other film
thickness measurements have been calibrated. The general require-
ment for a satisfactory interference measurement is that the film
is smooth and is supported on a smooth substrate (an optical flat).
Also, a well-defined edge is required to the film. If the substra-
te and film are sufficiently reflective, the substrate/film combi-
nation can then be made part of an interference system as shown in
Fig. 1. With the top plate of the interference system semi-silvered
on the surface nearest to the film/substrate combination, inter-
ference fringes will occur, corresponding to positions of constant
displacement t between the top and bottom plate. If the top plate
is then tilted very slightly relative to the bottom substrate/film
combination, equal path length fringes will be formed running nor-
mal to the direction of slope of the top plate. If this normal di-
rection is arranged to be a right angles to the direction of the
edge of the film on the substrate, then the interference fringes
will show a distinct displacement at the step and observationof
this displacement in a suitable optical microscope arrangement,
will allow a direct measurement of the film thickness.
Light source
(monochromatic)
- -------
-------
Optical
microscope
I

I I
I I
I
I I
--1--
I I
I
t
i Semi-silvered
i ; / top plate

I ,1 (Film
Fig. 1
flat
substrate
Simple interferometer arrangement.
THICKNESS MEASUREMENTS
165
For satisfatory fringes to be obtained the surface of the sub-
strate and film must be reflecting. Therefore it is usually neces-
sary to evaporate a coat of silver or other reflecting metal on to
the film/substrate combination, completely covering the step at
the edge of the film.
The general equation relating I the relative intensity of the
reflected light from a position in the system is given by :
I = Imax/(l + F Sin
2
'IT { ~ t Cos q,} ) ( 1 )
where ~ is the refractive index of the medium between the plates,
t is the thickness between the plates, A is the wavelength of the
incident light and q, is the angle of incidence of the incident
light (see Fig. 2) F is given by
F =
4R
where R is the reflectivity of the film on the top plate.
It can be seen from equation (J) that fringes can be formed
in a variety of ways and these are summarised in Table 1.
Fig. 2
Incident beam
wavelength \
A
\
\
\qS
\
\
Reflectivity of
top plate
/
,
/
i
/
/
/
"/
/
t
Parameters of interference system.
166
D.S. CAMPBELL
Table 1
Summary of interference systems
Constant terms in
Type of fringes Name
equation (1)
A, t Equal inclination Fabry-Perot
A, cP Equal thickness Fizeau
cP Equal t/A Fringes of
equal chroma-
tic order
(FECO)
t Equal t COS CP/A White light
Fabry-Perot
If monochromatic light is used, then it is possible to ob'tain
very satisfactory interference effects using either Fizeau fringes
or fringes of equal chromatic order. The system initially descri-
bed in Fig. 1 was for the production of Fizeau fringes and these
are the ones most normally used. The best sharpness in such frin-
ges is obtained with R as high as possible, while allowing the
fringe to remain visible, and the R value of 0.94 is found most satis-
factory. The light source used for such a measurement is usually
a mercury lamp but this source is not truly monochromatic. There
is a blue line associated with the mercury green which if not fil-
tered out will cause every fourth friqge appear to have a brown
edge.
Differences between successive fringes will correspond to to-
tal thickness differences of A/2 and thus a direct measurement of
the side of a step can be obtained by comparing the displacement
of the fringe produced by a step with the distance between succes-
sive fringes. The accuracy of this type of measurement is limited
essentially by thermal effects in the interferometer and a limi-
ting value of 10 A is usually quoted.
Fringes of equal chromatic order can also be used effectively
but in this case it is necessary to employ a spectrometer to sepa-
rate out the fringes. Fig. 3 shows a diagram of a typical set-up
used, and fringes are obtained at the plane HK where they can be
observed either with an eyepiece or photographed. A similar accu-
THICKNESS MEASUREMENTS
Spectrometer
slit
Mirror .,t-J
/ft-- I
I \
I I
Focussing ...J--l.......
lens n
0<-- -- :
White I
source I
T ,
+
I I Interferometer
I I
167
Fig. 3 Arrangement for observation of FECO interference patterns.
racy is obtained to that produced by Fizeau fringe observation.
Methods of improving the accuracy of measurement have been
studied. One example is the photoelectric method of detecting interfer-
ence minima that has been developed by Dyson (6,7,5). Fig. 4 shows
one arrangement of his apparatus. If the beams A and B fallon
identical surfaces, then the compensator can be adjusted to allow
zero light through. However, if A and B travel different distances
then light will be observed. The null position is best detected
by applying an A/c field to the ADP and detecting this via the com-
pensator with a photo-multiplier using a phase conscious rectifier
which feeds a centre zero meter. The setting then consists of ope-
rating the compensator until the meter reads zero.
By this means it is found possible to make settings with a
standard deviation corresponding to an uncertainty of film thick-
ness of 0.1 X. Such a figure is somewhat academic however becau-
se of the variation in flatness of substrates. Settings are possi-
ble to I X with certainty and this method is the most sensitive
available for determining film thickness.
Before leaving interference techniques it is worth noting
that in certain cases the interference colours produced by oxide
films on metal can themselves be used as a thickness guide. This
168 D.S. CAMPBELL
} \
Photo detector
I \
~
I I
~ Compensator
r-r (V4 ).. plate)
: IV
'"'-- Ammonium
I Dihydrogen
+ Phosphate
O
[ti
l (A.D.p.)plate
I /[-... I
0---..... I I
Source I "....... ~
WolI<;Jston *
prism T T
Fig. 4
AI 'B
+ + Film
I' 4
\
Substrate
Dyson's interferometer.
is particularly true for oxides such as Si02, Ta205 and Nb205 on
the parent metal. A sensitivity of around 40 A is possible around
a thickness of 400 A butothe technique cannot be used for low
150 A) or high (>5000 A) thicknesses.
2.2 Photometric Methods(1)(29)
It is possible to determine the thickness and refractive in-
dex of a film by measurement of the transmittance and reflectance
of a film. The simplest form this type of observation can take is
that of a direct measurement of the intensity of a transmitted
beam through the film. If the absorption of the film is known as
THICKNESS MEASUREMENTS 169
a function of thickness then the technique can be used for thick-
ness determination. However, such a simple approach will only ap-
ply to a continuous film with small grain size and more generally,
measurement of reflectance is required as well.
If transmittance and reflectance are known, then the mathema-
tical equations to be solved are not complex. However, various pu-
rely numerical methods have been derived which can ease the pro-
blem. (30) (31)
2.3 Polarimetric Methods
Polarimetric methods depend for their operation on the fact
that a metal surface can be directly detected by measuring the
ellipticity of the light after reflection from the surface. Such
measurements, however, involve the knowledge of the real and ima-
ginary parts of the refractive index of the film as well as that
of the thickness under examination. A rather difficult mathematical
procedure is usually necessary and although graphical and arithme-
tical procedures for fitting refractive index and thickness to the
experimental data are available, the process is somewhat cumber-
Some. Details of the theory and techniques can be found in other
texts. (8) (9)
The method has the basic attraction that no step is required
in the film, but the film itself must be transparent to light. The
accuracy that can be finally obtained is considerable, figures of
0.5 A have been claimed if refractive index measurements are
known sufficiently accurately but 10 A is more normal.
2.4 Mechanical Methods
Mechanical methods of measuring film thickness can involve
either the scanning of a film surface with a probe, the weighing
of a film after deposition or the rotation of a cylinder or disc.
All these methods will however depend ultimately on an optical de-
termination of thickness as a calibration.
2.4.1 Probe Systems
The most widely used probe system available for thin film mea-
surements is that developed by the Rank-Taylor-Hobson division of
the ~ n k Organisation, Leicester, U.K. The instrument was initial-
ly designed for the measurement of metal finishes and consists of
a diamond tipped probe that is mechanically driven across the work-
piece(10). Movement of the probe relative to an average datum pro-
vided by a steel skid that also traverses the surface or an op-
tical flat ,is detected by means of a 30 kHz inductance bridge, one
arm of which contains a cored coil, the core being mechanically
170
O.S. CAMPBELL
linked to the probe. The amplified out of balance signal from the
bridge is usually displayed on a recorder. Fig. 5 is a diagram of
the apparatus.
Such an instrument can be applied to the measurement of film
thickness provided the probe pressure can be reduced sufficiently
to prevent damage to the film when the probe traverses the film
and also that, as in optical measurements, a well defined edge is
available(11).
Such an apparatus has now been fully developed for the speci-
fic measurement of very thin layers and is marketed under the na-
me 'Talystep'. The pressure on the stylus has been made very light,
although the diamond tip used is of very small radius. However,
the load can be reduced to less than 3 milligrams. The final ampli-
fication of the system can be up to 1 million times so that itois
possible to measure films down to thickness accuracies of 1 A.
Other transducer systems have been used to monitor the probe
movement and these have included devices and semi-
conductor displacement transduces 12). However none of these are
as yet as sensitive as the 'Talystep' inductance bridge arrange-
ment.
2.4.2 Micro-Balances
Micro-balances have been used for measuring film thicknesses,
and such systems have the advantage that they can be used for in
situ measurements even in ultra-high vacuum systems.
Reference
datum
Optical flat
Bridge I
Diamond
probe
Substrate
Direction of movement
of probe an d datum.
Fig. 5 Diagram of 'Talysurf'probe.
THICKNESS MEASUREMENTS 171
Balances can be roughly classified(13) as of the torsion type
in which the deflections of the balance beam is balanced by the
torsion of the wire supporting the beam or the type in which some
external force, usually either magnetic or electrostatic is used
to restore the balance to a zero position.
Fig. 6 illustrates the basic idea of a magnetically restored
micro-balance(J4). Such a system could have a sensitivity of .10-8
gms per rom deflection at 3.5 metres. The apparatus can be baked
to 400C. With this sensitivity it is possible to determine the
mass of a mono layer equivalent to 3 A thick on a 1 em square.
However, the final calculation of thickness depends on a knowledge
of the density of the film and the state of aggregation.
2.4.3 Rotating Systems
It is possible to use a pivotted eddy current damped rotor
without restoring couple for measuring the thickness of deposition
in an evaporation system(15). If the vapour stream is only allowed
to land on one side of the rotor this will cause the rotor to ro-
tate. As the rotation generates an opposing couple proportional to
the angular velocity, a measurement of the thickness deposition is
given by the number of turns of the rotor. Furthermore the angular
velocity is proportional to the rate of arrival of the atoms.

CALIBRATION PAN
SPRING
VANE
Fig. 6 Quartz microbalance (after Meyer et al. (14
172 D.S. CAMPBELL
Two were evolved, a cylinder type for evaporation in
a horizontal direction (Fig.7) and a disc type for evaporation in
a near vertical direction (Fig. 8). Fig. 9 shows a typical result
using a cylinder type. A sensitivity of 10 A/revolution was obtai-
ned with the disc type.
Fig. 7
H
55 em
r
12cm
i
I Beoring
Needl'
--
(
, "
: '
Thin aluminium
alloy cylinder
mild stul cylinder
---Riqid support
Diagram of rotating cylinder thickness monitor
(after Beavitt. Ref. 15)
Serrations for
pulse counting
-----5cm----

Direction of / Needle
evoporot ion
disk
Fig. 8 Diagram of rotating disc thickness monitor
(after Beavitt. Ref. 15)
THICKNESS MEASUREMENTS 173
8 iT-y-----.-------,r---,-/-....,
6iT
"
III
<:
C
"
C
'-
'---"
<: 4iT
0
....
c
....
AI
0
a::
OAu wilh
2iT
correcled
rolation

o 005 010 015
Evaporant moss (q)
Fig. 9 Evaporant mass plotted against rotation for cylinder
thickness monitor (AI evaporant ; cylinder length 2.5 em;
cylinder radius 1.6 cm ; source distance 10 cm).
2.5 Electrical Methods
2.5.1 Quartz Crystal Systems(16)
The frequency of oscillation of a quartz crystal depends on
the mass of the crystal. Therefore, if a film is allowed to grow
on one side of a quartz crystal plate, the frequency change,
will be proportional to the mass M of the film and will be given
by
-
Np A
(2)
where f is the resonant frequency of the quartz plate, p is the
density of the quartz, A is the plate area, and N is a constant.
This equation was verified by Sauerbrey in 1959(17) and since
then a large number of papers have been published on quartz crys-
tal thickness measuring systems.
As an example of the use of equation (2), using a 6 MHz crys-
tal it is found that an area of deposition 0.3 cm
2
in area and 1 A
thick, will give a change of 1 Hz (for film density = 5). Fig. 10
-174
Fig. 10
D.S. CAMPBELL
Meter
Circuit diagram for use of Quartz crystal thickness
measurement
shows a block circuit diagram of a typical electronic arrangement
that can be used.
As with the micro-balance, a knowledge of the density of the
film is -required or the system needs to be calibrated against: so-
me other method, before it can be used for thickness determination
and furthermore the substrate must be the quartz crystal.
2.5.2 Capacitance Measurements
It is sometimes possible to determine the thickness of a di-
electric film which has initially been deposited on a conducting
substrate, by depositing a second electrode on the top surface.
A capacitance measurement of the resultant parallel plate structu-
re can allow of the thickness of the film to be determined, provi-
ded the permittivity of the dielectric is known. A typical capaci-
tance value that can be obtained is 0003 ~ per sq. em. for a film
of permittivity 6 and thickness 3000 A.
Another arrangement of electrodes that has been used(18) is
one in which a comb pattern(Fig. 11)is initially deposited in alu-
minium on the substrate which is usually quarz. If a dielectric
film is deposited on the comb the change in capacitance can be re-
lated to the thickness of dielectric.
Fig. 12 shows a typical result for SiD deposited on a comb of
line width and space 0.0075". The initial capacitance using a quartz
substrate was 65 pF.
THICKNESS MEASUREMENTS
175
Fig. 11 Comb structure of electrodes for capacitor method of mea-
suring thickness.
,--....
e

Q 20
V)
....
16
0
CII
12
CII
"
c
8

u
.-
4 .s::.
I-
10 20
0/0
/e Change in Capacitance
Fig. 12 % capacitance change for film thickness on comb electro-
des. (SiO ; electrode width and spacing 0.0075", Quartz
substrate, Initial Capacitance 65 pF);
Studies have also been made using a proximity meter approach
in which the second electrode is not deposited on the film but is
a plate brought into as intimate contact as possible with the top
surface of the film. However, this type of system suffers from the
difficulty of having, invariably, an air gap between the film and
the plate.
176 D.S. CAMPBELL
2.5.3 Resistance measurements
The resistance of a metal conducting film is a function of
film thickness and it is therefore possible to measure film thick-
ness using a direct measurement of resistance(I9). However, this
technique suffers from the disadvantage that it is not suitable
for very thin non-continuous films and also that the method needs
a knowledge of the resistivity of the deposited layer. Fig. 13
shows a typical resistance/thickness curve for silver on glass and
it can be seen how rapidly resistance increases at a thickness of
around 500 A as the film becomes discontinuous(20).
However the system therefore is of interest in determining
thickness reproducibility in metal films and has found
in controlling the length of deposition of a metal film( I). Con-
trol situations are possible for sheet resistance values around
100-500 Q/sq.
10
3
t.I
U
c:
0
-
1/1
10
2
1/1
t.I
c::
t.I
>
-
10
0
t.I
a:
500 1000
Film thickness (.a)
Fig. 13 Resistance against thickness for silver on glass.
THICKNESS MEASUREMENTS
177
2.6 Magnetic Methods
Studies have been made of the use of magnetic techniques for
measuring film thickness. If the film is deposited on a magnetic
material then the observation of the magnetic characteristics of
a probe placed on the top surface of the film, can give data as
to the film thickness. However, such techniques are generally not
sensitive enough for thickness measurements on thin films.
2.7 Radioactive Techniques
It is sometimes radioactive methods for mea-
suring film thickness t )(2){ 4)t 2).From the various techniques
available for making this type of observation, the simplest sys-
tem is that of measuring the radioactivity of the depositing ma-
terial. However, only a very limited number of materials have the
right high saturation activity and a half life that is not too short
or too long (3 hours is considered the lowest limit in order to
allow time to take the irradiated film from the atomic pile, and
six months is too long as the decay will cause contamination)
Gold is found to be very suitable material, as it has an ac-
tivity of 8 000 millicuries per gm. and a half life of 2 1/2 days.
Using detector systems that are readily availahle, in principle it
is possible to detect 1/IOOOth of a monolayer by this technique,
using gold.
A similar technique is to deposit the non-radioactive mate-
rial on a radioactive substrate and to measure the absorption of
the radioactivity due to the deposited film.
More sophisticated radioactive techniques which can be adop-
ted for measuring thickness have been summarised by Cachard.(32)
These include the use of back scattering from a 1-3 MeV ion beam,
the detection of nuclear reaction products on bombardment, the use
of electron micro probe, ion induced X-ray emission and secondary
ion beam mass spectroscopy. All these methods, apart from the last,
are non-destructive. Detection sensitivity is often to less than
a monolayer and depth resolution can be to 100 K in certain cases.
2.8 Chemical Techniques
In principle, it is possible to chemically dissolve films
from their substrate and measure the changing weight of the subs-
trate. Microchemical balance methods using such a on dis-
solvable films have been found capable of measuring 100 A at I cm
2
to about 0.5%. The method is, however, destructive and depend on
a knowledge of the film density.
178 O.S. CAMPBELL
3. RATE CONTROL SYSTEMS
3.1 Introduction
As mentioned initially, it is often possible to determine the
thickness of a film by calibrating the rate of deposition. The rate
control available depends on the deposition technique being used
and Table 2 summarises the rate control parameters against various
deposition methods.
Table 2
Rate control parameters for various deposition systems
Method
Electroplating
Chemical reduction
Thermal growth
Anodization
Vapour phase
Evaporation
Sputtering
Rate control
Current density
Temperature, pH
Temperature
Current density
Temperature of substrate
Temperature of source
Current density, Target poten-
tial
In the case of evaporation, sputtering and vapour phase depo-
sition, direct control of rate of deposition is not as easy as with
other systems and therefore other methods of measurements of rate
control have been evolved. Such rate control systems will be only
calibrated for a particular geometry inside the vacuum chamber and
any alteration of this geometry, whether with regard to the source-
substrate distance or the position of the rate measuring device in
the system, must be allowed for.
3.2 Mechanical Methods
Mechanical systems have been derived for use in evaporation
deposition that depend for their operation on the momentum of the
arriving evaporated atoms or molecules and Fig. 14 shows a typical
example. (11)(25) The vane rate meter can be restored into an equi-
librium position by passing a current through the meter coil, the
amount of the current being dependent on the rate of arrival of the
evaporated species. Such a system can be used for a feedback control
to alter the boat temperature of the evaporation source, and by
THICKNESS MEASUREMENTS 179
this technique 5% control of rate of deposition is possible. The
rate control with such a system can be as long as 1 A per second
though normally control is down to 10 A/sec.
Rotating cylinders oro_discs can also be used for evaporation
systems either with the suspension giving the restoring torque(26)
or by monitoring the speed of rotation of the disc.(15) For a sus-
pended type, Neugebauer obtained a sensitiv!ty of 1.5 A sec/radian,
and with a disc type Beavitt obtained 0.04 A sec/radian.
3.3 Electrical Methods
3.3.1 Quartz Crystal Systems
It is relatively straightforward to introduce a differentia -
ting network into the electrical circuit associated with the ouput
of a quartz crystal thickness measuring system Wig. 10) If this is
done a direct measurement of aate of deposition can be obtained.
Rates can be measured to 0.2 A/sec for Ai using commercial equip-
ment.
3.3.2 Ionisation Gauge Systems
A so-called nude ionisation gauge can be used for determining
the rate of arrival of depositing atoms in a vacuum system. (27).
Mirror scalll End ItopS Counterbalance
Fig. 14
F'!'trv"""'---;---- Moving coil
-.c-___ Polc pill CIS
.upport (20 gougll AI Wirll)
Vanll------____
(2cm X 2cm X 0.005- AI Shut)
Diagram of vane ratemeter (After Campbell and Blackburn
Ref. 11)
180
Grid Centre filament
Anode
c:J------
Rotating
Chopper
tEvaporant
D.S. CAMPBELL
Fig. 15 Diagram of ionization gauge ratemeter.
Fig. 15 shows a diagram of the basic principle. A chopper is intro-
duced bet"Jeen the source of material and the gauge with the chopper
usually running between 10-20 Hz. The pulsed current obtained from
the gauge is then observed and via suitable electronic circuitry
can be used to provide a direct measurement of rate. Levels down
to 0.2 A/sec can be obtained.
A refinement to the nude gauge system is to use two gauges,
one of which sees the vapour source and the other of which, does
not. The difference between the two out-puts will then give the
rate measurement directly.
It should be noted that if an electron beam source is used
for the evaporation system, then sufficient of the evaporant will
become ionised, and no separate filament will be required in the
gauge itself.
3.3.3 Capacitance Measurements
It is possible to use capacitance measurements for rate con-
trol based on the system discussed in section 2.4.2.
It can be seen from Fig. 12 that the relationship between
T
A
B
L
E

3

S
u
m
m
a
r
y

o
f

t
h
i
c
k
n
e
s
s

m
e
a
s
u
r
i
n
g

t
e
c
h
n
i
q
u
e
s

d
i
s
c
u
s
s
e
d

(
p

=

f
i
l
m

d
e
n
s
i
t
y

;

=

r
e
f
r
a
c
t
i
v
e

i
n
d
e
x

o
f

f
i
l
m
)

T
h
i
c
k
n
e
s
s

R
a
t
e

N
a
t
u
r
e

I
n
t
e
r
f
e
r
e
n
c
e

(
F
i
z
e
a
u
)

I

F

I
n
t
e
r
f
e
r
e
n
c
e

(
F
E
C
O
)

I

F

I
n
t
e
r
f
e
r
e
n
c
e

(
P
h
o
t
o
e
l
e
c
t
r
i
c
)

/

F

I
n
t
e
r
f
e
r
e
n
c
e

(
C
o
l
o
u
r
)

I

F

P
o
l
a
r
i
m
e
t
r
i
c

I

F

M
e
c
h
a
n
i
c
a
l

(
P
r
o
b
e

I

D

M
e
c
h
a
n
i
c
a
l

(
B
a
l
a
n
c
e
)

I

D

M
e
c
h
a
n
i
c
a
l

(
C
y
l
i
n
d
e
r

o
r

D
i
s
c

I

I

D

Q
u
a
r
t
z

C
r
y
s
t
a
l

I

I

D

C
a
p
a
c
i

t
i
v
e

I

I

D

R
e
s
i
s
t
a
n
c
e

I

D

R
a
d
i
o
a
c
t
i
v
e

I

F

C
h
e
m
i
c
a
l

I

D

I
o
n
i
z
a
t
i
o
n

g
a
u
g
e

I

D

B
e
s
t

A
c
c
u
r
a
c
y

1
0

A

2

A

0

0
.
]

A

4
0

A

0

1
0

A

0

]

A

3

A
0

0
.
5

A

0

5
0
0

*

1
0
0

A

0
.
1

A

0

0
.
5

A

R
e
m
a
r
k
s

S
t
e
p

a
n
d

r
e
f
l
e
c
t
i
n
g

c
o
a
t
i
n
g

n
e
e
d
e
d

L
i
m
i
t
e
d

r
a
n
g
e

a
n
d

a
p
p
l
i
c
a
b
i
l
i
t
y

m
u
s
t

b
e

k
n
o
w
n
.

C
o
m
p
l
e
x

S
t
e
p

n
e
e
d
e
d
.

P
r
o
b
e

m
u
s
t

n
o
t

d
a
m
a
g
e

s
u
r
f
a
c
e

p

r
e
q
u
i
r
e
d

C
a
l
i
b
r
a
t
e
d

f
o
r

p
a
r
t
i
c
u
l
a
r

m
a
t
e
r
i
a
l

a
n
d

s
o
u
r
c
e

t
e
m
p
e
r
a
t
u
r
e

p

r
e
q
u
i
r
e
d

L
i
m
i
t
e
d

t
o

d
i
e
l
e
c
t
r
i
c
s

L
i
m
i
t
e
d

t
o

c
o
n
t
i
n
u
o
u
s

m
e
r
a
l

f
i
l
m
s

H
a
z
a
r
d
o
u
s

D
e
s
t
r
u
c
t
i
v
e
.

p

r
e
q
u
i
r
e
d
.

-
l

:
:
I
:

n

"

2
:

m

3
:

m

e
n

C

:
:
c

m

3
:

m

2
:

c
;
l

0
0

182 D.S. CAMPBELL
thickness and capacitance is essentially non-linear and this means
that such a device is in practice, slightly complex to use. How-
ever, using an initial capacitance of 65 pF it has been found pos-
sible to measure rates of deposition down to 4% over the total
thickness range 1 ]Jm to 30]Jm. Rates in this thickness range could
be measured up to 3 ]Jm par minute.
4. SUMMARY
The various direct and indirect methods of measuring thick-
ness and rate of deposition have been examined. The method actual-
ly used will depend very much on the system and requirements.
Table 3 summarizeG the methods available (see also the table
given by Pliskin and Zanin(28. The table shows whether the tech-
nique can be used for thickness or rate measurements or both, and
also whether the measurement is fundamental (F) or must be eventual-
ly calibrated against a fundamental system (D). The best accuracy
figures given for thickness determinations will only be applicable
for limited conditions of thickness.
In e r t ~ i n case quoted, it should be noted that accuracies of
less than 1 A are possible. It can be commented that such an accu-
racy is often meaningless with regard to the physical structure of
the surface and these figures just indicate the reliability of re-
sults that are possible.
THICKNESS MEASUREMENTS
REFERENCES
I. K. H. Behrndt. "Physics of thin films" (Ed. G. Hass and
R. E. Thun). Academic Press. N. Y. l p. 1-40. 1966.
183
2. W. A. Pliskin and S. J. Zanin. "Handbook of thin film tech-
nology" (Ed: L. 1. Maissel and R. GIang) McGraw Hill. N.Y.
P . I l.l -I L34. I 970 .
3. H. E. Bennett and 1. M. Bennett. "Physics of thin films"
(Ed. G. Hass and R. E. Thun). Academic Press N. Y. ~ p. 21-
41. 1967.
4. S. Tolansky "Multiple-Beam Interferometry of Surfaces and
Films" O. V.P. London 1948
5. o. S. Heavens "Physics of thin films". (Ed. G. Hass and
R. E. Thun). Academic Press. N. Y. ~ p. 229-235 1964.
6. F. Dyson. Nature, March 23rd'1963. p. 1193.
7. F. Dyson. Physica 24, p. 532. 1958
8. o. S. Heavens. "Physics of thin films" (Ed. G. Hass and
R. E. Thun). 2 p. 218-227 ; 1964. W. A. Pliskin & S.J. Zanin.
"Handbook of thin film technology" (Ed. L. 1. Maissel and
R. GIang). McGraw Hill. N.Y. p. 11.21-11.24. 1970.
9. H. E. Bennett and 1. M. Bennett. "Physics of thin films"
(Ed. G. Hass and R.E. Thun). Academic Press N.Y. i, p.79-84
1967.
10. R. E. Reason. Symposium on the properties of metallic surfa-
ces. Institute of Metals Monogram N 13: p. 327. 1953.
II. D. S. Campbell and H. Blackburn. Trans. 7th National Sympo-
sium on Vacuum Technology. p. 313-318. 1960.
12. G. V. Planer and L. S. Phillips. "Thick film circuits"
Butterworth. London p. 24-25. 1972.
13. K. H. Behrndt. "Physics of thin films". (Ed. G. Hass and
R. E. Thun). Academic Press. N.Y. l p. 27-40. 1966
14. H. Mayer, R. Niedermayer, W. SChroen, D. Stunkel and H. Gohre.
"Vacuum Microbalance Techniques" Plenum Press. N.Y. l p. 76
1963.
184 D.S. CAMPBELL
15. A. R. Beavitt. J. Sci. Inst. 43, p. 182-185. 1966.
16. K. H. Behrndt. "Physics of thin films". (Ed. G. Hass and
R. E. Thun). Academic Press. N.Y. l, p. 21-27. 196G.
17. G. Sauerbrey. Z. Physik 155. p. 206. 1959.
18. F. Z. Keister and R. Y. Scapple. Trans. 9th Nat. Symp. on
Vacuum Technology. p. 116. 1962.
19. K. H. Behrndt. "Physics of thin films". (Ed. G. Hass and
R. E. Thun). Academic Press. N.Y. l. p. 18-19. 1966.
20. L. Maisselo "Handbook of thin film technology". (Ed. L.1. Ma1s-
sel and R.Glang) McGraw Hill. N.Y. p.18.25. 1970.
21. F. W. Bishop. Rev. Sci. Inst. 20. p. 527. 1949.
22. K. H. Behrndt. "Physics of thin films". (Ed. G. Hass and
R. E. Thun). Academic Press. N.Y. l. p. 38-40. 1966.
23. W. A. Pliskin and S. J. "Handbook of thin film tech-
nology". (Ed. L. 1. Maissel and R. GIang). McGraw HilloN.Y.
p.lI.31-11.32. 1970.
24. T. P. Flanagan and J. A. Bennett. Int. J. App. Rad. & Iso-
topes. p. 19, Jan. 1962.
25. R. E. Hayes and A.R.V. Roberts. J. Sci. Inst. 39. p. 428.
1962.
26. c. A. Neugebauer. J. Appl. Phys. "35. p. 3599. 1964.
27. K. H. Behrndt. "Physics of thin films" (Ed. Hass and
R. E. Thun). Academic Press. N.Y. l. p. 3-5. 1966.
28. W. A. Pliskin and S. J. Zanin. ftHandbook of thin film tech-
nology" (Ed. L. 1. Maissel and R. GIang). McGraw Hill. N.Y.
p. 11-33. 1970.
29. F. Abeles. In "Physics of thin films". (Ed. M. Francombe
& R. W. Hoffman). Academic Press. New York. ~ . p. 151-204.
1971.
30. J. M. Bennett and M. J. Bostry. Appl. Opt. ~ p. 41. 1966.
31. F. Abeles and M. L. Theye. Surface Sci. ~ p. 325. 1966.
THICKNESS MEASUREMENTS 185
32. A. Cachard. In "Physics of Non-Metallic Thin Films" Procee-
ding of N.A.T.O. Advanced Study Institute, Corsica 19th Au-
gust to 12th Septembrer, 1974. Plenum Press. London (this pre-
sent volume). 1975.
Physical Properties
ELECTRONIC TRANSPORT PROPERTIES
Robert M. Hill
Chelsea College, University of London
Pulton Place, London SW6 5PR
I. INTRODUCTION
The broad classification of materials, by their ability to
transport electric current, into metals, semiconductors and insu-
lators, can be generally applied to the same materials when depo-
sited in the form of thin films. It is not then the similarities
of films to bulk that give rise to the study of transport proper-
ties but the differences between the two, and in particular, the
differences that are inherent in the preparation of thin non-metal-
lic films. As an example, if we wish to investigate high electric
field phenomena, a bulk specimen 10-2m thick would require a power
supply capable of giving a potential of 10
6
volts to give a field
of 10
8
V.m-
I
, whereas a film 1,000 A thick only requires 10 V for
the same field. A second example is that highly disordered struc-
tures can be prepared in the form of thin films by deposition onto
cold substrates. Below the deposition temperature these structures
are at least metastable and their electrical properties can be in-
vestigated. Indeed, without extreme care it is difficult to pre-
pare films which have good crystalline properties and the investi-
gation of transport of charge in non-metallic films is essentially
an investigation of the interaction of quasi-mobile charge with the
non-crystalline media. Here the first example, of high field, is
of interest, for if a non-linear regime of current and voltage can
be induced the interaction between electric field and structure
can be used to elucidate information about the electronic structu-
re of the material. It is the extraction of this information that
is of interest to us here.
We can make some general comments about conductivity in poorly
crystalline materials, as in section 2, but because of the lack of
189
190 R.M. HILL
a detailed theory it is necessary to set up models of behaviour
and compare the predictions from the models with experimental re-
sults. This is done in section 3. Firstly, however, we shall take
a general look at the properties of poorly crystalline materials
in order to obtain some information to use as a basis for model
structures.
1.1. Summary of Experimental Results
Four examples of the behaviour of thin non-metallic films and
bulk disordered materials are sho\yn in Fig. 1-4. These can be ta-
ken as typical of the experimental results that have been reported
in the literature over the last few years. Fig. I shows the opti-
cal properties of the elemental semiconductors germanium, silicon
and selenium, and the criterion for the choice of these particular
results has only been that they, individually, show a typical fea-
ture.
The first diagram(I), (Fig. la) compares the reflectivity of
crystalline, amorphous and re-crystallised germanium over the band
of energies up to twelve electron volts. The crystalline material
exhibits sharp fine structure which arises from the exact shape of
the conduction and valence bands, the reflectivity being an inte-
grated function of the product of the band states at an energy se-
paration given by the incident radiation. In contrast, the amor-
phous material has a much smoother reflectivity curve, although
there is still a broad peak corresponding to the maximum crystal-
line reflectivity, and the general form still follows the same sha-
pe. The re-crystallised material however is much more similar to
the first curve with strong evidence for the presence of fine
structure.
The transition from amorphous to re-crystallised is shown in
more detail in the second diagram(2), '(Fig. Ib). The material an-
nealed at the low temperature is truly amorphous and shows an ill-
defined absorption edge which is less than that of the crystalline
material by about 0.3 eV. As the annealing temperature is raied
the absorption curve moves to higher energies, for the same value
of absorption coefficient, until at a temperature of about 500C the
crystalline optical gap is attained. Even here, however, at ener-
gies just above the optical gap, the absorption coefficient is
still larger than that for the perfect crystalline material. Fig.
Ic shows similar measurements for selenium(3), but here the range
of temperatures over which the measurements have been made inclu-
des the melting point at 130C, and shows no indication of this
physical transition. However when selenium melts local order wi-
thin chains is maintained although the long range order between
chains is lost.
ELECTRONIC TRANSPORT PROPERTIES
>-
I-
>
-
I-
u
I.J.J
...J
I.J...
I.J.J
a::
.6
.3
2
024681012
flw (eV)
,
E
u
1-10
4
z:
I.J.J
U
I.J...
I.J...

u
z:
o
I-
0..
1.0
en
a:l

o
'El0
5
u
I-
z:
I.J.J

I.J...
I.J...
I.J.J
o
u
z:10
3
o
I-
(b)
0..

.8 1.6 2.4
-11w (eV)
(c)
191
Fig. 1 (a) Reflectivity of germanium (after ref. I). 1 - single
crystal ; 2 - amorphous thin film ; " - recrystallised thin film.
(b) Absorption edge in an annealed silicon film (after
ref. 2). Annealing temperatures. I - 20C; 2 - 223C; 3 - 500C;
4 - 949C. Curve 5 is for crystalline silicon.
(c) Temperature dependence of the absorption edge in se-
lenium (after ref. 3). Measuring temperatures I - 400C; 2 -
3 - 200C; 4 - 100C; 5 - 25C; 6 - -100C; 7 - -200C. The
melting point of selenium is 130C and the material was non-crys-
talline in the solid state.
One method of examining the physical state of an amorphous ma-
terial is to use diffraction measurements to construct a radial
distribution function(4). Such a curve is shown in Fig. 2 for amor-
phous germanium(S), the vertical lines along the radius axis indi-
cating the positions, and relative densities, of atoms in the crys-
talline material. It can be seen that the first and second peaks
192
Fig. 2
ter ref.
rial are
3.0
2.0
N (T)
1.0
4
'.
' .
...
A
R.M. HILL
6 8 10
T_
Radial distribution function for amorphous germanium (af-
5). The positions of the atoms in the crystalline mate-
shown by the vertical bars.
in the RDF curve correspond reasonably well with the first and se-
cond crystalline neighbours but the resolution for succeeding
neighbours is poorer and by the eighth neighbour the material looks
uniform in its properties. This result is similar to that obtained
in liquids and indicates a degree of short-range order but a com-
plete lack of the normal crystalline long-range order.
For the third type of informatioll Fig. 3 exhibits an Arrhe-
nius plot for the alloy system Cd TlxAs2 as x is varied from 0.1
to 0.8,(6). This diagram contains two pieces of information which
are of interest. Firstly, there is a surprisingly small dependen-
ce of the magnitude of the conductivity on the value of x at any
single temperature and, secondly, none of the plots is linear, i.e.
there is no well-defined activ.ation energy in this system. Again,
both these observations are common in the field of disordered ma-
terials.
A different type of information is shown in Fig. 4. Here, the
current voltage characteristics of a thin insulating film have
been measured over an extensive temperature range. A casual exami-
nation of the characteristics shows that a common pattern exists
ELECTRONIC TRANSPORT PROPERTIES
5
x = 0.8
>-
10- 8
.....
>
.....
U
::J
"'C
10-
10 c
0
<...J
2 4
8
Fig. 3 Temperature variation of conductivity in vitreous
CdTlxAsZ (after reference 6).
193
throughout the range of variables examined. At low enough voltages
ohmic behaviour is observed and then there is a transition into an
exponential or higher power dependence as the voltage is raised.
At very low temperatures the current appears to be insensitive to
the temperature but highly dependent on the voltage. This particu-
lar specinen is of silicon monoxide but very similar results have
been reported on a wide range of materials prepared in thin film
form.
194
R.M. HILL
10-
6
11'1
0..
E
III
I-
z
10-
8 LLl
0:::
0:::
=>
(J
10-
10
10- 12L.-__ "'--__ ...... __ ...... __ --I ___ _
10-
2
10-
1
1.0 10
APPLIED VOLTAGE (Volts)
Fig. 4 Current/voltage characteristics of silieon monoxide as a
function of temperature (after reference 7). 1 - 413K ; 2 - 333K
3 - 256K ; 4 - 16SoK ; 5 - 77K ; 6 - 4.2K.
We can summarise the evidence contained in Figures 1-4 and use
these experimental observations as a basis for examining transport
of charge. We know that the broad classification of solids into me-
tals, semiconductors and insulators, carries over from the bulk ma-
terials into the thin film form, hence there must be some retention
of energy band structure, as we have seen from the optical evidence.
However, there is no longer the sharp well-defined bands characte-
ristics of a clear E-k relationship but certainly some degree of
diffuseness in the region of the band edges. The results on sili-
con in particular show that this diffuseness can be related in so-
me way to the degree of disorder in the material. Locally, order ap-
pears to be maintained implying that the bonding between atoms is
of the same form in crystalline or disordered structures, but the
ELECTRONIC TRANSPORT PROPERTIES 195
loss of long-range order implies that individual groups of atoms
and their nearest neighbours are rotationally distorted with res-
pect to each other. This can only occur with physical distortion
of the structure and lead to a fraction of the bonds being strai-
ned and possibly a smaller fraction actually broken, i.e. a large
density of "intrinsic" defects will be present. Supporting evidence
for this conclusion is given by the alloy results where the lack of
sensitivity to doping, in the semiconductor sense, implies the pre-
sence of an initially high density of "impurities".
Figure 4 shows that one of the tools at our disposal is that
of non-linear field effects. A simple measurement of conductivity
can only yield information about the density/mobility product. Non-
linear effects, if interpreted correctly, will yield information
about the detail of the interaction of the carrier with the elec-
tronic structure of the material.
1. 2 Examination of the concept of Localisation and Disorder
We will assume that the basic transport processes is crystal-
line materials are well known(8) and concentrate here on the effect
of disorder. The seminal work in this area was carried out by Ander-
son(9) who examined a Kronig-Penney-1ike array of potential wells
spaced a distance 'a' apart to which was added a random potential
U of root mean square value U
o
In the absence of U the wells gave
a band of allowed, extended, states of width J. For the perturbed
system three cases were delineated,
i) U /J < 0.5. Within an electronic mean free path of length
L ~ a J ~ U o 2 ) phase coherence exists and the electron states are
extended.
ii) 0.5 < Uo/J < 5. A diffusive region for transport where no
wave solution exists and the mean free path isof the order of 'a'.
iii) Uo/J > 5. At zero temperature all charges are localised in
the potential wells and no transport occurs. However, at finite
temperatures transport is possible by quantum mechanical tunneling
between the wells. For this to occur, however, the carrier must cap-
ture or emit a phonon of energy equal to the difference in energy
between the initial and final states. This mode of transport is ter-
med hopping.
Intuitively we would expect a small potential perturbation in
a basically crystalline meaium to smear out the band edges. Within
the original band regions case (i) will apply, the bands will sup-
port extended wave functions, and the carriers will still possess
something like the crystalline mobility. At the edges of the bands,
however, there will be relatively more perturbation yielding a dif-
196 R.M. HILL
fusive region,(ii), where the physical extent of the states is fi-
nite but there is continuity in energy. For states displaced into
the normally forbidden gap the local density of states will be low
and the only process of transport will be hopping, either within
the localised states or into the extended states within the bands.
In the extended state region the scattering of electrons is
weak, and the wave number k is well defined. The uncertainty in
k, is approximately proportional to the inverse of the mean
free path and I. However, as we move down from region (i)
to region (ii) the meap free path decreases, I, and wide
variations from the Elk density of states for the crystalline ma-
terial can be expected. In the localised region k is no longer
well defined, the uncertainty is of the order of k itself, and e-
ven the concept of momentum is no longer useful.
At a finite temperature Einstein's relationship can be used
to define a mobility
J.l

kT
evL:2
z--
kT
where z is the coordination number and V the frequency. In the
extended state region this will be an electronic frequency
:2
s-I), assuming a mean free path L much greater than
atomic spacing. In the localised hopping region diffusion
requires phonon assistance and the relevant frequency will be that
of a phonon ("'10
12
s-l) reduced by the probability of phonon captu-
re or emission. Considering the frequency terms only the ratio ex-
tended state, to localised mobility will be given by
> > 3.10
3
Hence a drastic change in the magnitude of the mobility is
expected as we pass from extended to localised states. This situ-
ation will occur both for the bonding and the anti-bonding bands
and has led to the concept of a mobility gap, equivalent to the
crystalline band gap, in disordered systems, the critical energies
at which the mobility changes rapidly being termed Ec and Ev for
the conduction and valence band respectively.
2. GENERAL THEORY
In this section the Kubo-Grennwood formula, which is suited
fora discussion of transport processes in imperfect crystalline
ELECTRONIC TRANSPORT PROPERTIES 197
materials, will be derived and used to determine the physical mea-
ning of some simple experimental measurements.
2.1 Kubo-Greenwood Formalism
The general relationship for the conductivity in a medium at
a frequency w is (for example see Matt and Davies(lO
a(w) =
J
* d1/iE
where D(E) = 1/I
E
dx
J
D ~ NED
E
+ nw.NE+hw
trw
dE
( 1 )
d
3
x, Q is the volume of the material per
electron, NE the density of states at energy E and 1/1 the electronic
wave function. If we set the product DE. DE+hw to be given by the
square of an average value of D, IDI!v' where the averaging is car-
ried out over all states E to E+bw and let the frequency go to zero,
we can obtain the limiting, D.C. conductivity. Defining
we can write
a = - J
df
a (E)aE" dE
(2)
which is the Kubo-Greenwood formula. The function f is the Fermi-
Dirac function and af/aE (=sinh-
2
E/2kT) has a bell-like shape with
a maximum at the Fermi level and decays exponentially for energies
greater than about 4 kT from the maximum. The physical significan-
ce of a(E) is that it is the conductivity of carriers at energy E
and can only be taken as 'a really well defined at present for ex-
tended states.
A useful method of manipulating the Kubo-Greenwood integral
is available if it can be shown that the product under the inte-
gral sign possesses a well-developed maximum at a particular va-
lue of energy E
Q
In this case integration by the method of stee-
pest descents(IL) gives
a
1 -2 Eo J (E - Eo)2 2 Eo
4kT a(Eo)cosh 2kT exp[- 4(kT)2 {l+tanh 2kT
2(kT)2 a
2
a(E)}]dE
a(Eo) aE02
(3)
E
t

E
o

E
F

F
i
g
.

5
.

I
n

C
1

(
E
)

-
G
R
A
D
I
E
N
T

-
(
k
T
)
-
l

(
a
)

E

E
l

E
F

I
n

C
1

(
E
)

(
b
)

G
r
a
p
h
i
c
a
l

s
o
l
u
t
i
o
n

o
f

e
q
u
a
t
i
o
n

(
5
)
.

T
h
e

e
n
e
r
g
y

a
t

w
h
i
c
h

t
h
e

d
o
m
i
n
a
n
t

p
a
r
t

o
f

t
h
e

c
u
r
r
e
n
t

i
s

c
a
r
r
i
e
d
,

E
o
,

i
s

g
i
v
e
n

b
y

t
h
e

e
n
e
r
g
y

a
t

w
h
i
c
h

t
h
e

I
n

a

(
E
)

v
s

E

c
u
r
v
e

h
a
s

g
r
a
d
i
e
n
t

k
T

f
o
r

E
o

-
E
F

>

k
T
.

I
n

(
a
)

a

m
o
n
o
t
o
n
i
c

r
i
s
e

o
f

E
o

w
i
t
h

t
e
m
p
e
r
a
t
u
r
e

i
s

e
x
p
e
c
t
e
d

b
u
t

i
n

(
b
)

E
o

w
i
l
l

r
e
m
a
i
n

i
n

t
h
e

r
e
g
i
o
n

o
f

E
I

f
o
r

T
I

<

T

<

T
2
'

:
:
0

1
5
:

:
:
c

r
-
r
-
ELECTRONIC TRANSPORT PROPERTIES
which for Eo - EF > kT is equal to
(4)
The energy at which the maximum contribution to the current
is obtained, Eo, is given by

aoE Eo
I
kT {tanh
(5)
199
In equation (3) the principal temperature dependence is contai-
ned in the pre-integral terms and if O(E) is constant over an ener-
gy range which does not include the Fermi level and zero elsewhere,
a well-defined activation energy corresponding to the minimum ener-
gy of the band will be obtained as in the crystalline case. In the
more general case of aCE) being a continuous function of energy,
equation (5) can be expressed graphically to show how the energy Eo
varies with temperature. For Eo > kT the right hand side of equa-
tion (5) has the value (kT)-l and hence Eo will be given by the
energy at which the gradient of a plot of In aCE) against E has this
value, as in Figure 5(a). If aCE) monotonically increases with ener-
gy, Eo also increase monotonically from the Fermi level as the
temperature is raised from zero. However, if, as in Figure 5(b),
aCE) exhibits negative curvature, or even peaks, Eo will be pinned
over a finite temperature range and linear regions will be observed
in the Arrhenius plot.
In principle, it is possible to construct aCE) vs E diagrams
from detailed examination of Arrhenius plots but in practice it is
simpler to assume simple functions for the energy dependence of
aCE) and to look for the equivalent characteristics in the experi-
mental activation energy, which, if the approximation yielding equa-
tion (4) applies, is identical to Eo.
2.2 Activation energy
It is cornmon practice to regard an activation energy as only
having physical significance when linear Arrhenius plots are obtai-
ned. This simple case is characteristic of crystalline solid state
physics where energies are well-defined. Here, we shall show that
the activation energy of transport has a more general meaning which
is useful when considering materials which are not perfectly crys-
talline. We by defining the activation energy to be the gra-
dient of the Arrhenius plot, i.e.
200
/::"E = -
dIna
d(1 /kT)
1 d a
- -
a d(I /kT)
which in the notation of equation (2) can be written as
- 1 { f df daCE) +f () d (af) }
- a aE d ( 1 /kT) dE a E d (I /kT) aE dE
R.M. HILL
(6)
(7)
We can now consider two specific cases to simplify the physi-
cal meaning of (7). If the major part of the integral oc-
curs for IE - EFI > kT, then taking daE/a(l/kT) = 0
(8)
where < >a denotes a conductivity averaged energy.
Hence the activation energy measured experimentally as the gradient
of an Arrhenius plot, when greater than kT, represents the average
energy of the carriers with respect to the Fermi energy, even when
/::,.E is itself a function of T. Obviously, if the carriers can only
move at the bottom or top of well-defined bands, the activation
energy will be constant, as observed for crystalline semiconductors
and insulators. Conversely, a well-defined activation energy can be
taken as evidence for band-like properties in the material under
study.
The second case, IE - EF I < kT, can be conveniently treated by
using a Taylor's series about the Fermi energy, in which case
k f af
/::"E = - - a(E)-- T dE = kT
a aE
(9)
Le., as one might expect, the average, energy of the carriers is
simply the thermal energy and the Fermi function is constant.
2 .3 Thermopower
An alternative experimental measurement is that of thermopo-
wer. Letting dj be the current corresponding to the local conduc-
tivity aCE) under the action of an applied field F, we have(14)
dj =
af
a(E)af F dE
ELECTRONIC TRANSPORT PROPERTIES 201
The energy carried by this incremental current is -(E - EF).e-l.dj
so that the total heat transport is given by
-I J cH
e aE a(E)(E - EF). F. dE = j ~
where ~ is the Peltier coefficient. The thermopower S is defined
as the ratio of the Peltier coefficient to the temperature, hence
in our notation
k J E - EF af
S = eo a(E)( kT ) W dE
k
E - EF
< >
itT a
(10)
:0::_
e
Comparison of equations (9) and (10) shows that the activa-
tion energy and thermopower contain the same basic information but
that the latter is sensitive to the sign of the carriers whereas
the former is not. For either pure band-like transport or conduc-
tion at the Fermi energy it can be shown that
lsi eT = A k T + ~ E (II)
where A is an integer of order unity, the exact value depending on
the form of the density of states function(IO), i.e" if
N(E) ~ Ex/2,A = I + x (x = 0, I, 2).
It would appear that measurement of both the activation ener-
gy and the thermopower should be useful in determining the detailed
structure of transport as a function of energy.
2.4 Mobility
The standard method of discriminating between carrier densi-
ty and mobility in crystalline materials is the use of the Hall mo-
bility. Experimentally, measurements of the Hall constant in disor-
dered materials, that is materials that show a small, and tempe-
rature sensitve, activation energy, have yielded little real infor-
mation. The mobility values appear to be small, as Clarke observed(15)
in germanium ~ I cm
2
v-l S-I). This itself is an indication of
localisation, for, if the mobility is less than about 10 cm
2
V-
1
s-1
the mean free path is of the order, or less than, a lattice spa-
cing, hence the carrier is not moving in classical extended sta-
tes. Friedman(16), starting from the Kubo-Greenwwod formula, exa-
mined mobility in a localized system by using a random phase model
for transport. The effect of the Hall magnetic field was introdu-
ced by a suitable modification of the phase of the transfer inte-
gral between the localised sites,
202 R.M. HILL
/H)
n' ,n
= /0) exp (i a., )
n',n n ,n
(12)
where n' and n are the site numbers and
a.
n' ,n
hec H x n'].
In this way Friedman was able to show that the ratio of the
drift to Hall mobilities could be expressed as
2
J (_Z_)
6kT 1)2
(13)
- ---rr;- 1
where 1) cos 8 ="3 and Z is the average number of sites which
form a closed path.
In general J > kT hence > Friedman also showed that
both are small, the Hall was essentially independent of
temperature, and the sign of the Hall coefficient was such that it
appeared that the carriers were always n-type, as observed expe-
rimentally.
The alternative method of measuring mobility is by transit
time techniques(17). If a sheet of charge is injected into a block
of material and moves under a field of ma@1itude F the transit
time is given by
.F)....J. (14)
where the length of the specimen is L. The magnitude of the field
F has to be large enough to overcome the self field of the charge
sheet and the system can be further perturbed by space charge ef-
fects. The charge is injected into the "conduction band and as it
drifts through the specimen interehange occurs with localised le-
vels and the high mobility band. Once equilibrium has been esta-
blished it can be shown that the effective transit time is given
by
Ec - E
t
exp + (kT )
(15)
where Nt is the density of a single trapping level at an energy
Ec - Et below the conduction band. For traps distributed in ener-
gy similar relationship apply, generally with the energy term ap-
pearing in the pre-exponential(IO). Obviously, in this case, the
leading edge of the deformed sheet of charge will be formed from
ELECTRONIC TRANSPORT PROPERTIES 203
carriers which have been released from the shallowest traps and
there will be a long tail to the charge distribution as carriers
are released slowly from deeper traps. The simplest case of a
single trapping level can be analysed with confidence but it is not
such an easy matter to deconvolute the smeared out signals obtained
from more complex systems.
3. SPECIFIC TRANSPORT PROCESSES
Although conceptually useful the general analysis of conduc-
tivity contained in Section II is not complete at the present time.
In order to obtain more detail of the mechanisms of conduction it
is necessary to postulate models of behaviour for the interaction
between charges and structure, analyse these for field and tempe-
rature behaviour patterns, and critically examine experimental mea-
surements for evidence that supports the original hypotheses. As a
starting point we consider a material with diffusive band edges
extending into the normally forbidden gap region. At a critical
density these states will become localised(lO). If the widths of the
localised tails are large the states precipitated from the conduc-
tion band may overlap equivalent states from the valence band, in
which case charge transfer will occur, the conduction band like
states being empty of charge and the valence band states filled.
The Fermi level will then lie, and be pinned, in the region of
overlap(18) ,(Figure 6(a, the degree of stability of the Fermi level
being directly proportional to the density of localised states in
that region.
The principal characteristic of crystalline non-metallic solids
is the ability to position the Fermi level by doping. Although this
property does not appear to be strong in disordered solids it is
still in principle, possible to dope these materials. Only if the
'intrinsic' doping is large will the effects of 'extrinsic' doping
be small. Hence we are forced to conclude that the disorder gives
rise to an 'intrinsic' doping from, for example, a large density
of broken bonds. For the crystalline materials the ionisation ener-
gy of the dopants is given, to a good approximation, by the hydro-
genic model, i.e.
(J 6)
Even in an imperfectly crystalline material we would expect equa-
tion (16) to apply as the dielectric constant is a macroscopic quan-
tity. The effects of disorder, however, will be to broaden out the
single energy level associated with non-isoelectronic impurities in
the crystalline material to form a band of equivalent donor or ac-
ceptor states in the imperfect material. This arises through both
204
Fig. 6
solid.
R.M. HILL
N(E)
N(E)
(a) (b) (c)
Density of states diagrams for an imperfect non-metallic
(a) Localised conduction and valence band states overlapping
to pin the Fermi level.
(b) Donor and acceptor-like defects forming bands in the re-
gion of their ionisation energy.
(c) For a large density of donor-like defects the Fermi level
will be pinned within the donor band.
interactions between the defects and because of local, spatial,
variations in the energy at which extended states occur(19), these
states being the datum from which the ionisation energy is measured.
In the energy band diagram we can introduce these 'intrinsic'
doping states as shown in Figure 6(b). As a direct consequence of
the presence of these 'impurity' states, particularly if their den-
sities are not equal, the Fermi level will again be pinned but now
within one of the band regions (Figure 6(c)).
Figures 6(a) and 6(b) represent two different types of model
structures which can be examined in detail for characteristic be-
haviour patterns. The former is that first examined by Mott(20),
ELECTRONIC TRANSPORT PROPERTIES
whereas the latter is typical of Poole-Frenkel behavioqr. These
two general cases are examined in detail below.
3.2 Mott Hopping
205
In the absence of detailed information about the distribution
of trap-like states we assume a constant density of localised sta-
tes of Nl cm-
3
distributed randomly in space and over an energy
range 2E
m
. Without loss of generality the Fermi level can be set in
the middle of this energy band and we further assume that the band
is wide with respect to the thermal energy. The density of states
par unit energy, Nl/(2Em), we shall call Nt. Two conditions can be
set for transport in this system, firstly, that of excitation of a
'free' carrier, and secondly, that such a carrier should be able
to pass through the system, the continuity condition. A full exa-
mination of the random system has been made(21), but here we shall
only examine the low temperature asymptotic behaviour.
a) Exci tation
If one particular site is considered as datum the physically
distributed sites surrounding it can be ordered in terms of their
radial distance. The expected value of the radial distance of the
's' site is given by (22)
3 1/3
e(Rs )=( 47f
N
l) r (s+I/3) (17)
and for large values of the ratio of the gamma functions
can be approximated by e
V3
, Within the radially ordered s-set the
expected value for the energy difference between the datum site and
the site of lowest energy is
and again for s large the approximation 2E
m
s-
1
can be used.
In the low temperature regime the probability of a hop occu-
rring can be written as(23)
Ph ex: exp - (2 ex R + E/kT) (19)
where Rand E are the radial distance and energy associated with
a hop, and ex is the localisation parameter of the sites. The maxi-
mum probability of hopping occurs within the s-set where dPh/ds=O,
i.e. when
3E 4N J ..
S = (---.E!) 3/4 (4:-4 (20)
ex kT 3
206
R.M. HILL
and at this value of S the probability is given by
3a J4
exp - 3.88( 4 Nt
kT
)
(21 )
which differs from Mott' s La,., of variable range hopping only by a
numerical factor. The approach used here differs from Mott's but
allows some physical insight through the predicted range of hopping,
equation (20). For large values of Sh, the radial parameter within
which variable range hopping takes place, the temperature should
be low, validating our choice of the low temperature form of the
Miller and Abrahams probability function, and the density of centres
in real space, Nl , has to be large. However, the density per unit
energy (2Em)-1 should be small, i.e. the states have to be of great-
er separation in energy than kT. If we take the conductivity as
being proportional to the probability of hopping, which we would
expect to be the case, the activation energy, as defined in equa-
tion (6) can be determined as
( akT) 3/4
0.971 (
41TN
tpA
(22)
and our choice of the low regime requires that this
should be greater than kT.
b) Continuity
The alternative approach, initiated by Ambegaokar et al(24),
is to consider that the path of a carrier will be determined by
the easy hops within the disordered structure. At low temperatures
particularly, only a limited range of hopping probabilities will
be possible. Defining the critical probability in the same terms
as equation (19), as exp(-hc),we require that from a general site
in the unformly random structure the number of sites available to
which the probablity of hopping is greater than this critical va-
lue should be greater than, or at least equal to, a critical number
nc in order that a continuous path should be the cri-
tical number heing determined from percolation theory(25).
From the definition of hc we have that
hc = 2aR + E/kT
and as the energy of a hop goes to zero a maximum radius over which
hopping can occur, R
o
, is determined as h
c
/2a. The number of sites
within a sphere of this radius which satisfy the probabiliy crite-
rion is then
ELECTRONIC TRANSPORT PROPERTIES 207
f
Ro Ihc-ZexR
n = I f(Rs) feEs) dE. dR
s 0 0
(Z3)
where f(Rs) and feEs) are the probability distribution functions
in real and energy space of the random array of sites. Taking the
summation over s into the first integral sign and integrating yields
7T
n = 3" Nt kT (Z3ex3 )
(Z4)
Taking the critical value of n as 1.5(Z5) gives the magnitude
of hc as
(Z5)
regaining the T-
1
h characteristic derived by Ambegaokar et al(Z4),
Jones and Schaich(Z6) and Maschke et al(Z7) for this form of ana-
lysis. As we have used the same framework for the two criteria we
can make a direct comparison in which case we see that the excita-
tion condition has slightly less probability and will dominate trans-
port by variable range hopping. Indeed, for the continuity argu-
ment to dominate the critical number nc would have to be which
is extremely large(Z5).
The method used here of determining the energy at which the
dominant part of the current flows is equivalent to determining the
pre-integral term of equation (3) and is satisfactory as a means of
investigating the first order temperature dependence.
The effect of high electric fields is to reduce the expected
energy of the sth state by a amount E(Rs).eF for a state lying
in the field direction. Considering only current flow in the field
direction it has been shown(Z8) that the field introduces a correc-
tion term to the current which becomes
J exp - 3.88[(
IF/
ex ) 1/'+ (I _ _e_ ) 1/,+ ]
47TN tKT' ZexkT
(Z6)
The usefulness of this relationship is that the localisation para-
meter ex appears in the second term along with measurable
only so that it is possible, by investigating high field effects, to de-
termine both Nt and ex unambiguously.
208 R.M. HILL
3.2 Poole-Frenkel and Poole behaviour
The essential difference between the trap states considered
in the last section and the donor/acceptor states to be conside-
red here is in the binding force field of the centre from which
emission takes place. It was implicitly assumed in the previous
section that the sites through which hopping takes place have no
long range potential distortion. In the limit this assumption is
invalid but it is a convenient approximation which appears to be
substantiated by experimental observation. When ionisation of a
donor is being considered, however, it is impossible to make use
of this approximation, and indeed, in equation (16) use has already
been made of the coulombic force field to determine the ionisation
energy. Again, two specific cases will be considered, that of a
low density of ionised centres where only a single donor need be
considered and the case of a large ionised density where the cou-
lombic fields interact.
a) Low density
Field assisted ionisation of a centre which is electrically
neutral in the unionised state can be described in terms of the
force field by
F(x) = - eF + e
2
(4nrx2)-1
The maximum in the potential occurs at
x = etA (4nr>-tA,
and the magnitude of the potential on this is Ei - 8ppF
U2
where 8pF is the Poole-Frenkel constant, e
The current can be determined by use of the Kubo-Greenwood
formula,
(27)
where is the free band mobility and Nd the effective density
of unionised donors. The energy range in which the conductivity
is not zero is taken from the peak in the p'qtential. For reason-
able values of ionisation energy Ei - 8PF FlP:kT, in which case
Ei - 8 F1P
J ex: !:! N _( PF)
kT d exp kT
(28)
ELECTRONIC TRANSPORT PROPERTIES
209
It has been pointed out(29,30)that this Poole-Frenkel equa-
tion only applies for emission in the direction of the field. How--
if consideration of the total forward hemisphere is made the
correction is relatively minor. A more serious problem with the
Poole-Frenkel equation is that it must be in error for such low
fields that the probability of emission in the forward direction
differs little from that in the reverse direction. In this case
it has been suggested(31) thflt a hypothetical reverse peak can be
used. of magnitude Ei + and the nett current will be the dif-
ference between that flowing forward and reverse, i.e.
J
nett
eF Ei
ex: kT]Jc Nd exp - IT sinh
(29)
in the low field expansion gives a current proportional to
F Alternatively, it has been suggested(32) that equation (27)
represents the density of carriers available in which case the re-
verse term should be summed to the forward term and a cosh function
is obtained in place of the sinh term, with an ohmic low field ex-
pansion. In practice the high field difference between these appro-
ximations is slight, particularly when the Schubweg or mean free
path of the emitted carrier. and its field dependence. are unknown
and the simple sinh with a square root field dependence
in the pre-exponential appears to give as good an agree-
ment with experimental results as the more detailed formulae.
The use of the Kubo-Greenwood approach emphasises that only
carriers which are excited over the potential barrier contribute
to current flow. At low temperatures the density of such carriers
will be low and the only measurable current will be that which tun-
nels trough the distorted potential barrier. Between these two re-
gions (Figure 7) will be a region of thermally assisted tunneling
where carriers will be thermally excited the coulombic
and tunnel through the barrier at an energJ less than E. - B Fl I
Examination of the current in this regime(31) gives a Jt.: exp-B/TlJ"3
characteristic whereas the pure tunneling current at low tempera-
tures is temperature independent and essentially given by a Fowler-
Nordheim(33) field dependence.
b) High density
When the density of ionised donors is large,ionisation takes
place not into the extended high mobility region but into the cou-
lombic potential associated with one of the ionised centres. Figure
8 shows the energy diagram for this situation. Because of the over-
lap of coulombic potentials the peak in the barrier remains in the
plane midway between the centres over an appreciable field region.
The distortion of the potential is then linear with the field and
obeys Poole's Law(34). It should be noted that here the zero field
210
R.M. HILL
____ . . , ~ ~ c )
Fig. 7 The three emission processes from a donor centre.
(a) Poole-Frenkel thermal emission. (b) Thermally assisted tunnel-
ing. (c) Direct tunneling at zero temperature.
~ ~ - - - - - - - - 5 --------.
Fig. 8 Energy diagram for an array of overlapping coulombic po-
tential wells
ELECTRONIC TRANSPORT PROPERTIES 211
potential barrier, E ~ is less than the true ionisation energy of
a single centre by an amount depending on the distance between the
centres(31). In this case the barrier in the reverse direction is
well defined and considering emission only along and against the
field direction gives a current
J cc exp
E
a
kT
(30)
At high fieldA the barrier distorts from the symmetrical shape
and a return to Poole-Frenkel behaviour is to be expected.
3.3 Injection Limited Conduction
The two specific processes of transport considered above are
bulk limiting in the sense that we have implicitly assumed that the
electrodes supply and extract sufficient charge not to impede cur-
rent flow. For the disordered non-metallic solid this is a reason-
able assumption as there will be localized states in the region of
the Fermi levels of the contacts into which charge can be extracted.
For more crystalline-like materials, or for materials with wide
band gaps, this state of affairs is not obviously true and the cur-
rent may well be contact limited. To simplify consideration of such
processes we shall only look at the case of a single carrier, the
electron, and injection limited currents, i.e. make use of a die-
lectric approach.
When the work function of the solid being investigated(35) is
greater than the work function of the electrode material an enhan-
cement layer of charge is injected into the semi-insulator; the
converse situation leads to charge extraction and the formation of
a depletion layer. The flat band 'ohmic' case is the limit of both
situations and in general, is not to be expected. Conventionally,
enhancement leads to space-charge limited flow(36) and depletion to
Schottky effects(37). However, the essential difference between the-
se two processes is not in the form of the boundary layer but in the
nature of the interaction between the injected.charges.For space-
charge control to be established the excess charge must be suffi-
ciently large to interact with itself and to be definable in terms
of a true local density over any small but finite elemental volume.
For the Schottky emission to be the limiting process the opposite
is the case, the injected charge has to be sufficiently small so
that it can be treated as a set of single charges interacting only
with their individual mirror image charges. The normal Schottky a-
nalysis neglects band bending due to depletion and considers only
the flat band condition. This neglect is justifiable for the case
of a highly insulating, very thin, film as the depletion layers
212 R.M. HILL
will be wide and overlap extensively to give a quasi-flat band con-
dition throughout the system.
a) Space charge Control
As electric fields are additive so are their differentials,
which, from Gauss's theorem, define space charges. Hence, band ben-
ding effects can be introduced into the normal transport equations
as a fictive charge which has to be added to the injected space-
charge. As a boundary condition it is necessary, however, either
to know how the surface charge density due to band bending, Ps'
varies with the electrical stress or to assume that it remains cons-
tant under any experimental conditions. For convenience we assume
the latter to apply. Figure 9 shows the predicted current Ivoltage
characteristics for a range of the ratios of surface charge density
to the thermally free 'intrinsic' carrier density n. For large va-
lues of this ratio the characteristic is initially ohmic, when the
~ P ; : C
100
10
3
JaV
2
! ~ 0
10
2
-0.5
/
I-
-0.4
:z
UJ
10
1
0:::
0:::
::::l
U
1.0
10- 1
VOLTAGE
Fig. 9 Current-voltage characteristics for space-charge limited
conduction. ~ s is the density of injected charge at the injecting
surface and n the thermally free carrier density. The current and
voltage are in units of Jr(n
2
e
2
l..I1)-:1 and Vr(ne1
2
)-1 respec-
tively.
ELECTRONIC TRANSPORT PROPERTIES
213
excess charge is localised at the injecting contact, but as the
space-charge cloud sweeps across the specimen and reaches the anode
the Mott-Gurney square law(39) is followed. In the limit of infi-
nite excess charge at the injecting contact the square law is fol-
lowed throughout the high field region, but for limited surface
charge a second transition to a square root voltage dependence
should be observed. This arises from saturation of the enhancement
charge pool and is a consequence of the boundary condition. How-
ever, even for moderate values of the ratio ps/n the transition
to this third region occurs at sufficiently high voltages that it
may not be observed experimentally, the transition point from the
ohmic to the Mott-Gurney region occuring at the voltage neL
2
(SE
r
)-1
which is generally within two or three orders of the breakdown volt-
age of most materials. In thin film specimens, however, L is small
and ultra-high field effects may be observable.
The figure also shows the situation for injection into a de-
pletion region (ps negative). In place of the ohmic region at low
potentials the square root region is observed and higher voltages
yield an ohmic region as the depletion charges is annulled.
In the absence of charge trapping a simple activation energy
is to be expected for the temperature dependence of the current.
However, if there are localised trap-like states, trap filling ef-
fects will occur(40) and anomalous power law current/voltage rela-
tionships can be expected, together with more complex temperature
dependences.
b) Single Carrier Control
The standard single carrier Schottky analysis is identical to
that already described for process in Section 3.2
with one important difference. In the Poole-Frenkel process the
carrier is freed from the coulombic potential generated by the im-
mobile ionised centre. Under Schottky conditions the interaction
is with a mirror image charge (Figure IOa),obtained by summation
of the real surface charge on the injecting contact. Hence, as the
carrier moves away from the injecting surface so does the image,
by an equivalent distance, and the magnitude of the coulombic field
is halved when compared with the immobile centre case. This reduces
the Poole-Frenkel constant by a factor of two to give the Schottky
constant Ss(= i SpF)' If we consider a crystalline material the in-'
jected charge will be free to move in the conduction band, and, for
a thick specimen, the current will be given by ne exp(SsF
1
/
2
/kT)
where is the conduction band mobility. For a thin film, however,
there is more than a single image charge to be considered (Figure
lOb). In this case the high field characteristics are given by(38).
214 R.M. HILL
(a)
+
Fig. 10 Diagrammatic representation of the image charges and po-
tential in (a) a thick and (b) a very thin semi-insulating layer
(3 I)
i.e., at single temperature there is a power relationship between
the current and the voltage, the power being temperature dependent.
4. CONCLUSION
Critical reviews of experimental work(10,41,42) on thin films
have shown that the specific transport processes discussed in Section
3. have been observed and it has also been shown that it is possi-
ble to extract useful information about the detailed band structu-
re of thin films of non-metallic materials from transport studies.
ELECTRONIC TRANSPORT PROPERTIES 215
.'
E
___ ---:w----:...------ .. '
0.4
'"
."
ENERGY
(eV)
0.8
:
"""
4 ____
t d 570
0
K 520
0
K 400
0
K
310
0
K
1.2
10 17
10
18 10 19
10
20
N(E)
cm-3 eV-l
Ev
1.6
Fig. 11 Density of states for glow discharge deposited silicon
films. Td is the deposition temperature and the arrows indicate
the positions of the Fermi level. The continuous lines were obtain-
ed directly from field effect experiments. (after reference 43.)
As a single example Figure 11 shows diagrammatically the form
of the density of states curve obtained by Spear(43) from drift
mobility measurements on glow discharge deposited silicon as a func-
tion of the temperature of deposition, Td. In this material the po-
sition of the Fermi level was sensitive to Td and allowed almost
the full range of the normally forbiden gap to be explored under
bias conditions.
The physics of crystalline materials, wheth er in the from of
thin films or bulk specimens is well understood. The problem areas
of poorly crystalline or disordered materials have been delineated
but in order for a satisfactory theory of transport or any other
physical property to be established, it is necessary for experimen-
tal work to be carefully examined and used to indicate the prime
features of disorder. Thin films are ideal as specimen for this type
of examination as the nature of deposition allows control over the
structure, and the specimen geometry allows a wide range of experi-
mental techniques to be applied.
216
R.M.HILL
REFERENCES
I. J. Tauc, L. Pajasova, R. Grigorovici and A. Vancu "Proc.
Conf. Non-Cryst. Solids, Delft" North-Holland Amsterdam
p. 606 (1965)
2. M. Brodsky, K. Weiser and G.D. Pettit, Phys. Rev. BI,
2632 (1970).
3. K. J. Siemsen and E.W. Fenton, Phys. Rev. !2!, 632 (1967).
4. V. D. Frechette (Ed) ''Non-Crystalline Solids" John Wiley,
N.Y. (1960).
5. H. Richter and M. Furst, Z. 38 (1951).
6. L. Cervinka, R. Hosemann and W. Vogel, J. Non-Cryst. Solids
l. 294 (1970).
7. A. Servini and A. K. Jonscher Thin Solid Films l, 341 (1969).
8. e.g. J.P. McKelvey "Solid State and Semiconductor Physics"
Harper & Row, N.Y. (1967).
9. P. W. Anderson, Phys. Rev. 109, 1492 (1958).
10. N. F. Mott and E. A. Davis "Electronic Processes in Non-Crys-
talline Materials" Clarendon Press, Oxford (1971).
11. D. A. Proc. Phys. Soc. Z!, 585 (1958).
[2. H. Jeffreys and B. S. Jeffreys "Methods of Mathematical Phy-
sics" University Press, Cambridge p. 503 2nd Ed. (1950).
13. R. M. Hill J. Phys. C 1, L267 (1972).
14. M. Cutler, Phil. Mag. 25, 113 (1972).
15. A. H. Clark, Phys. Rev. 154, 750 (1967).
16. L. Friedman J. Non-Cryst. 329 (1971).
17. W. E. Spear,Proc. Phys. Soc. B76, 826 (1960).
18 N. F. Mott "Electronic and Structural Properties of Amorphous
Semiconductors" Le Comber & Mort, Eds. Academic Press, London
(1973)
ELECTRONIC TRANSPORT PROPERTIES
217
19. H. Fritzsche J. Non-Cryst. 49 (1971).
20. N. F. Mott Festkorperprobleme 22 (1969).
21. R. M. Hill, to be published
22. P. Whittle "Probability" Penguin Books, Harmondsworth (1970).
23. A. Miller and E. Abrahams Phys. Rev. 120, 7A5 (1960).
24. V. Ambegaokar, B. I. Halperin and J. S. Langer Phys. Rev. B
2612 (1971).
25. S. Kirkpatrick, Rev. Mod. Phys. 45, 574 (1973).
26. R. Jones and W. Schaich J. Phys. 43 (1972).
27. K. Maschke, H. Overhof and P. Thomas, Phys. Stat. Sol. (b) 62
113 (1974).
28. R. M. Hill,Phi1. Mag. 24, 1307 (1971).
29. A. K. Jonscher, Thin Solid Films!, 213 (1967).
30. J. L. Hartke J. Appl. Phys. 39, 487 (1968).
31 R. M. Hill, Phil. Mag. 23, 59 (1971).
32. G. A. N. Connel, D. L. Camphausen and W. Paul, Phil. Mag. 26,
541 (I 972).
33. R. H. Fowler and L. W. Nordheim, Proc. Roy. Soc. A119, 515
(I 928).
34 H. H.Poole,Phil. Mag. 32,112 (1916) ; Phil. Mag. 34,195
(1917).
35. D.K. Davies, J. Phys. D,!, 1549 (1969).
36. R. H . Trego1d "Space charge Conduction in Solids" Elsevier,
Amsterdam (1966).
31. W. Schottky, Physik Z. 12, 872 (1914).
38. R. M. Hill, Thin Solid Films 12. 369 (1973).
39. N. F. Mott and R.W. Gurney "Electronic Processes in Ionic
Crystals" Clarendon Press, Oxford (1940).
218 R.M. HILL
40. M. A. Lampert, Reports on Progress in Physics 12, 328 (1964).
41. D. Adler, C.R.C. Crit. Rev. in Solid State Science ~ 317
(I 971).
42. A. K. Jonscher and R. M. Hill, Physics of Thin Films
to be published (1975).
43. W. E. Spear, Proc. 5
th
Int. Conf. on Amorphous and Liquid
Semiconductors, Garmisch (1973).
IONIC TRANSPORT IN THIN FILMS
Stephen J. Fonash
Department of Engineering Science and Mechanics
The Pennsylvania State University
University Park, Pennsylvania 16802
A charged defect - impurity, vacancy. or interstitial - in a
thin film can drift under the influence of an electric field. As
the defect moves through the material, in addition to the imposed
electric field, it is subject to some sort of an oscillatory
tential as a result of its interaction with the constituents of
the material. Assuming that this potential is periodic allows it to
be sketched as shown in Fig. 1. Here the externally applied field
E has been superimposed. For the masses, or effective masses, and
for the energies under consideration the wavelengths of the mobile
defects are such that they are much smaller than the characteritic
distance over which the potential sketched in Fig. 1 varies. Thus
the motion of these defects may be treated classically.
The movement of the defect of charge q way be understood from
Fig. 1 : the probability per attempt that the defect succeeds in
advancing to the right one unit along this potential is
; the corresponding quantity for going against
the field is If there are V attempts per second
then the net probability per second of advancing is (1)

Here is the barrier height and a the distance between minima.
Consequently this results in an electric current density J given by

J = 2vaqn e sinh(qaE/2kT) (1)
for a defect density n.
219
220 S.J. FONASH
From Eq. (1) it is seen that the mobility of the defect is
2va -q, /kT .
= e (2)
This expression, for qaE 2kT, simplifies to
(3)
Thus at room temperature for fields less than 10
5
V/cm, the ionic
current conduction is ohmic. For higher fields Eq. (1) applies.
However, for higher fields one would suspect that at some point it
may become necessary to modify Eq. (1) to allow for the possibility
that the field can change the number density n. Of course, for high
fields Eq. (I) may be written as
J = Vaqn exp[-(Jl _ qa
E
)]
kT 2kT
since jumps against the field are negligible.
(4)
Before examining this point of a field enhanced n some of the
other problems and assumptions inherent in Eq. (I) should be dis-
cussed. First of all, since many thin film materials have high
dielectric constants, it is not clear that the macroscopic electric
field E is the correct field to use in Eq. (1). That is, it is
obvious that a great deal of polarization is possible. Also since
Fig. 1 Representation of total potential energy field in which
the charged defect migrates.
IONIC TRANSPORT IN THIN FILMS 221
bonding in many materials of interest is only partially ionic, it
is not clear if an effective ionic charge should be used for q(2)
However, the results of examining these effects indicate that, even
at very high field wich the net activation energy
(see Eq. (4 goes from (<1>-';-) to a more general power series in
the electric field, the E and q to be used are the macroscopic
field and the actual charge carried by the defect. This has been
definitely established for the case that the potential barrier is
symmetrical (I , 2).
Another point of difficulty is that although a single <p appears
in Eq. (I), for amorphous films there is a range of <p values. There
also is a range of activation distances a, attempt frequencies v,
and local dielectric constants. Thus, to properly derive an expres-
sion for J, all possible combinations of sequential steps, parallel
paths, and branches must be considered(3). If Eq. (I) is used then
for amorphous thin films, it is necessary to consider <p and a to be
functions of the field E and the temperature T. The precise func-
tional forms would depend on the details of the transport network
aSRumed(3).
Further, for crystalline materials one may speak of the defect
density n appearing in Eq. (I) as arising from impurities, Frenkel
defects, and Schottky defects. In this case the computation of n is
rather straight forward(4). Impurities certainly can exist in
amorphous thin film materials; their migration according to Eq. (I)
is easily understood. For full ionization, n ,.,ould be a constant.
However, if the impurity ionization were governed, for example, by
electrons being liberated by the Poole-Frenkel effect then n would
depend on T and E. Frenkel-like defects presumably are present in
amorphous thin films also(3). However, the computation of n is cer-
tainly not as simple as in the crystalline case due to the spread in
parameter values. The same can be said for Schottky-like defects
which could be attributed to regions of local llonstoichiometry(3).
In addition, for glass-like networks there may be intrinsic defects
present which may be imagined as having been created by a rupture
of bonds followed by an exchange of partners. The separated dang-
ling bonds, having acquired or lost an electron, would contribute
to ionic conduction(3). Thus in thin films there are rich and
varied sources of defects possible but the computation of n can be
rather complex.
For the situation where the drift current is carried by an ion
liberated from a fixed ion vacancy, n can also depend on the elec-
tric field, as has been demonstrated(5), by adding to the potential
of Fig. 1 a binding potential between the ion and the vacancy. After
a certain distance AI+A2 from the fixed ion vacancy, this attractive
potential could be modeled as Coulombic. Thus in the direction of
the field the envelope of the oscillatoDY potential would first in-
crease going away from the fixed ion vacancy. Then as the external
222 S.J. FONASH
field takes over, it would decrease. The first several barrier
heights (separation of adjacent relative maxima and minima) would
be modified from the value ; succeeding barrier heights would be
essentially
If it is assumed that either the first (at dl=Al) or second
(at dz= Al+ Az + determines the activation energy for
the production or liberated ions, then, with Ql and Q2 being the
absolute heigths of these respective barriers, it is easily seen
that Eq. (4) is valid with n given by (5)
n N + q;E )/2kT] (5)
Here N is the number of positions from which ions may be liberated
per volume. Whether the first or second total harrier is higher
(that is, whether i = 1 or 2) depends on the field strength. Thus
an Arrhenius plot should give a discontinous change in slope, with
field, at the field strength sufficient to make the second barrier
lower than the first in cases where this model is applicable(5).
It has also been suggested (6) that n may be modified by con-
sidering that the activation energy necessary to liberate the migra-
ting ion is essentially provided to overcome a Coulombic binding
potential energy. Thus, with the presence of a sufficiently strong
electric field the arguement proposes a Schottky barrier lowering
term 2(q3E/
ES
)V2 This analogue to the Poole-Frenkel effect for
electrons would, of course, increase the supply of migrating ions
with electric field.
Although couched in classical terms, it is seen that the pro-
blems encountered in examining ionic transport in thin film are
much the same as those found for the electronic transport mecha-
nisms of hopping and percolation in thin films : difficulties in
determining the number densities of carriers and their mobility.
Also, just as for electronic the rate limiting process
may not be characteristic of the bulk but may interfacial in nature
(7,8).
Considering the experimental study of impurity ion transport
in thin films, such investigations are most easily accomplished
using M-I-S configurations since shifts in the C-V characteristics
result with ion motion(8). With the of bias temperature
stressing(8-ll), a typical example of whieh is shown in Fig. 2,
information can be obtained on impurity mobilities, activation
energies and number densities. The effects of mechanical stress
may be examined also(ll). Due to the intense technological inter-
est in Si02, ion transport in thin film structures using this ma-
terial is particularly well studied with these approaches.
Techniques have been developed for introducing a given ion species
IONIC TRANSPORT IN THIN FILMS
INITIAL
CHARACTERIZATION
-50 -40 -30 -20 -10
BIAS (VOLTS)
0
223
60
40
-
a.
W
I,J
30
z
<:(
t-
13

20

u
10
Fig. 2 High frequency C-V data for Al-Si02-Si structures in which
the thin film Si02 was thermally grown. Results of bias tempera-
ture stressing are shown. The shift in C-V characterization is due
to the drift of impurity ions in a dc field.
into the thin film, drifting it in a field, and collecting it again
(12). Some of this work has indicated that certain species of ions
are more mobile if injected into the Si02 than if present in the
oxide as impurities incorporated during growth on Si. It has even
been shown(J3) that Na, deposited uniformly on the Si02' \lhen drif-
ted across the thin film, does not accumulate uniformy at the Si02-
Si interface. This has been correlated with dielectric breakdown
in the Si02-Si system.
REFERNCES
1. J. O'Dwyer, "The Theory of Electrical Conduction and Breakdown
in Solid Dielectrics", Clarendon Press, Oxford, (1973).
2. M. Dignam, J. Phys. Chem. 249 (1968).
3. M. Dignam "Oxide and Oxide Films" (J. Diggle, Ed.), Dekker
Inc. (1972).
4. N. Mott and R. Gurney, "Electronic Processes in Ionic Crystals"
224 S.J. FONASH
Dover, Inc. N. Y. ( I 964)
5. C. Bean, J. Fischer, and D. Vermilyea, Phys. Rev. lQl, 551
(J 956)
6. J. Hanscomb et al., Proc. Phys. Soc. London 88, 425 (1966).
7. S. Hofstein, IEEE Trans. Electron. Devices ED-13, 222 (1966).
8. E. Snow et al., J. Appl. Phys. ~ 1664 (1965).
9. R. Castagne, C.R. Acad. Sci. (Paris) 267, 866 (1968).
10. M. Kuhn, Solid-St. Electronics ii, 873 (1970).
II. S. Fonash, J. Appl. Phys. 44, 4607 (1973).
12. M. Woods and R. Williams, J. Appl. Phys. 44, 5506 (1973).
13. T. Dunn et al., Phys. Chern. G l a s s e s ~ 16 (I 965)
14. T. Di Stefano, J. Appl. Phys. 44, 527 (1973).
DIELECTRIC PROPERTIES OF THIN FILMS POLARIZATION AND EFFECTIVE
POLARIZATION
Stephen J. Fonash
Department of Engineering Science and Mechanics
The Pennsylvania State University
University Park, Pennsylvania 16802
1. INTRODUCTION
In thin films, with their amorphous or polycrystalline
ture, a large number of dielectric polarization mechanisms are pos-
sible. These can be classified into two rather general groupings :
those which occur due to the presence of interfaces and those which
are characteristic of the thin film material itself. That is, this
latter category does not owe its existence in any way to interfaces
whereas the former exists because of interfaces. The mechanism
basic to the first category is the presence of concentration modifi-
cations at electrodes and at interfaces, which may occur within the
film. Although such interface effects are usually dismissed in deal-
ing with bulk dielectric properties, they may dominate in thin film
structures. Consequently their probable cannot be taken
lightly. It should be pointed out that even in thick "bulk" samples
of material it is well known that the interface effects can dominate
as is the case for electrets.
In the second category are found all those mechanisms which may
be considered "bulk" in origin. These include dipole and dipole-like
polarization effects which result when dipolar molecules and ions
are capable of occupying several closely spaced positions in a gi-
ven region. Another possible polarization mechanism is that arising
from imperfections ; the effective dipole resulting from a cation-
vacancy pair is an example. Electronic mechanisms which can produce
an effective polarization are also in this category. Of course, this
includes the normal polarization arising from the distortion of the
spacial configuration of tightly bound electrons; but much more
importantly for many thin film materials, it includes effective pola-
225
226 S.J. FONASH
larization which may arise from mechanisms involving itinerant elec-
tronic carriers (for example, electron transport by hopping or per-
colation).
In general, at given point in a thin film the flux density D,
the macroscopic electric field E, and the polarization
by the response characteristics of the material, are related by 1-3)
E E = D - P
o
(I)
where E is the permittivity of free space. Of course, in this equa-
tion thg polarization P is due only to those mechanisms in category
II ; interface effects are discussed in terms of macroscopic charge
density p through Poisson's equation:
p (2)
In a static field, the static polarization P
s
should be proportional
to the macroscopic field E ; ie,
(3)
where Es = KsEo is the static permittivity of the material and Ks
is the static dielectric constant(4). For time varying macroscopic
fields E, one would expect that P would try to follow the electric
field. If first order kinetics are applicable (as then apparently are
for at last some of the phenomena in category II) this may be ex-
pressed, for the l' th contributon to the polarization, as (1)
dPj(,
1'1 dt Psl - P
l
(4)
Here Tl is a phenomenological relaxation time for this l'th mecha-
nism and P = LQ.PfI- ; ie, P is the sum of all mechanisms present from
category II. Taus = PfI- (t) obeys some relationship such as Eq. (4)
in a time varying and consequently P = P(t).
For a static electric field, the electric flux configuration
in a thin film structure is static and only a particle current can
be detected flowing through the thin film. Thus in this situation
no information can be obtained on polarization effects from electri-
cal measurements. To actually probe electrically the dielectric pro-
perties of a thin film, it is necessary to study currents passing
through the film under dynamic conditions. Common configurations
employed are metal-thin film-metal and metal-thin film-semiconductor
structures. Considering the metal-thin film-metal configuration of
Fig. 1 for definitiveness, it can be seen that by varying the dif-
ference between the electrochemical potentials (Fermi levels) of
the two metal electrodes by applying a bias V = V(t), changes in
interface polarization can take place. Of course that polarization
DIELECTRIC PROPERTIES OF THIN FILMS 227
1---lC_

Fig. 1 Thin film in an M-I-M configuration. Regions 1, B, and 2
are shown as described in the text. A bias V is applied and positi-
ve current flows right to left.
from mechanisms being referred to as category II will also
be varied in this manner. An additional component of current caused
by these polarization changes will be detected at the electrodes in
this case and information on dielectric properties can be obtained.
To put this in quantitative terms, Eq.(I) together which Eq.(2)
may be used in a straightforward manner to write that the total cur-
rent crossing electrode 1 of Fig. 1, for example, is given by (5)
W
I
dV A J ap
CG---- -dx
dt W 0 at
+ A
x ap
- -)- dx
W at
(5)
where 12 is the particle current crossing the plane x = 0, W is the
width of the thin film and A is the area of the film. The geometri-
cal capacitance C
G
is given by AO/W. In a static situation only the
particle current flows. Under dynamic conditions a particle current
(which may considerably different than that present in the static
situation since some of this particle current may be flowing due to
changes in p or due to changes in stored carriers) crosses x = 0
together with displacement currents arising from the last three
228 S.J. FONASH
therms in Eq. (5). The first of these three is the displacement
current which vould flow if there were no material present. The
second accounts for polarization phenomena characteristic of the
material itself (category II). The third term arises from the pre-
sence of interfaces (category I).
Thus by varying the field (by application of an ac signal(6),
by application of a pulse(72 etc.) or even by thermally stimulating
frozen-in polarization(8-
10
5, the polarization can be made to vary
and information obtained on dielectric properties as may be seen
from Eq. (5). The most prevalent of these techniques is that of
applying an alternating current signal oV about some dc bias V (ex-
perimentally, very often V = 0). Consequently the resulting ac
quantities are related by(5)
d V A rWClop JW x Clop
01 = 01 + C
G
-- - - J -- dx + A (I - - ) -- dx
P dt W 0 dt 0 W Clt
( 6)
Further, in using ac signals to explore dielectric properties,
it is freequently assumed that oV<kT permitting linearization(5) of
Eq. (6). This linearization is tacitly assumed when the current
01 = iejwt is attributed to two channels of flow given by(I,4,II,12)
v . C
... = -- + JW pV
Rp
(7)
Here j = (_I)lk and oV = vejwt. The quantity Rp is a differential
parallel equivalent resistance and Cp is a differential parallel
equivalent capaciatance evaluated at the bias V. Obviously the phy-
sical origins of these small signal equivalent circuit elements lie
in terms on the right hand side of Eq. (6). On consideration of Eq.
(6) it is not surprising to find experimentally that Rp and Cp are
functions of the frequency w of the ac signal.
As an alternative to Eq. (7) an effective complex conductivity
er* = a' + jer" may be defined such that the small signal current i
in Eq. (7) is given by(I,4)
*
i
Aer v
=-W-
(8)
Thus it follows that
Rp
W
Aer
l
( 9)
and
C
p
Aer"
Ww
(10)
DIELECTRIC PROPERTIES OF THIN FILMS
229
It is equally possible to define an effective complex permit-
tivity E:: = E' - jE" (or an effective complex dielectric constant
E:: = K::E
O
) (I ,4) such that
i
. AE::
JW -w v
In this case comparison with Eq. (7) shows that
C =
p
AE'
W
and from Eq. (9), (10), (12) and (13) it follows that
(11 )
(I2)
(I3)
(I4)
(15)
These quantities Cp, Rp, E::, and 0:: are effective ; they
arise from and are combinations of the processes represented in
Eq. (6). The task in interpreting experimental data is that of un-
raveling such effective, measured quantities to determine their
physical origins. It is a task made quite difficult by the fact
that many of the processes in category I and in category II, to
the external measuring circuit, appear rather similar in behavior.
However, studying these polarization phenomena-which in no way
should be construed as limited to dielectrics- can provide very
useful, additional information on the structure and electronic pro-
perties of non-metallic thin films. Thus it can be a very worth-
while undertaking.
2. INTERFACE EFFECTS
Interface effects, which contribute through Eq. (5)-(15) to
the dielectric properties of thin film structures, arise due to
modifications, near the interfaces, in the concentrations of the
various charged species present in thin films. Analysis of this
phenomenon is extremely complicated due to the fact that, even in
the absence of any dc current (ie, V = 0), there can be space char-
ge regions and built-in Hotentials VD(diffusion potentials) at the
interfaces of thin filmst5, 13-15). These are modified with the ap-
plication of a dc bias and further modified if it is varied.
Considering an ac variation, it is seen from Eq. (6) that, if
230
S.J. FONASH
the term involving the bulk polarization is neglected (its effects
for an ac distrubance can be represented by replacing Eo with E::
as is discussed in the section on bulk effects), the quantities
oIp = OIp(x,t) and op = op(x,t) need only to be determined to eva-
luate interface effects(15). Of course the charge density p is com-
posed of contributions from mobile carriers (positive ions or holes
with density Pi ; negative ions ot electrons with density nj) and
fixed charges (a given species having the density Nk)' The particle
current, on the other hand, is composed of contributions from the
various mobile carriers.
The basic equations governing small ac concentration modifi-
cations of the mobile carriers, in linearized form, are(15)
(I6)
daOtn
j
=+l1
J
On. oE + 1 1 n ~ + 11 E oonj + 11. oE ~ n j
J ax J J oX j ax J \1X
a
2
0n; + oG' - oR'
ox2 J J
(17)
Here drift has been modeled by Ohmic conduction ; a more general
model for drift could be employed. The basic equations governing
the concentration modifications of the various fixed species are
of the form
(18)
Added to this system of equations is Poisson's equation
(19)
with op = ~ Zi oPl.' - ~ Z. on.+ EZkoN
k
In these equations op, on,
~ J J J
oN and oE are variations of the number densities and electric field
about the values p, n, Nand E. The quantities G and R represent
generation and recombination; the 11 and D are mobilities and dif-
fusivities, respectively. These equations are linearized forms of
the continuity equation (one for each positive mobile species of
valence zi, one for each negative mobile species of valence Zj and
one for each fixed species) and of Poisson's equation. Of course
the quantities needed to compute the observed polarization current
DIELECTRIC PROPERTIES OF THIN FILMS 231
caused by these small ac variations are op given above and the par-
ticle current
01 =
[
Jli(EOPi + Pi OE )- Di
aOPi]
+

ax
Jli(EOnj + njQE) + Dj
aonj]
(20)
J
ax
Solving the system, Eq (16)-(19), allows these to be computed.
To solve for the op, on, oN and oE from the system of equations
(16)-(19) initial and boundary conditions are needed. Initial con-
ditions are simplified in part by assuming an ac variation. However,
information on the initial (ie, dc) values of Pi' nj' N
k
, and the
electric field is necessary as may seen from the system of equa-
tions. Boundary conditions at the thin film-metal electrode inter-
face are also necessary.
Thus, in general, the dc values of pi, nj, N
k
, and E existing
for some bias V must be obtained as functions of position from ap-
propriately formulated continuity equations, Poisson's equation,
and boundary conditions. This has been discussed in detail in re-
ference(15). These dc values are then substituted in the system
Eqns. (16)-(19). Consequently, with appropriate boundary conditions,
the small ac variations may be calculated.
This calculation scheme - the core of which is equations of the
form (the precise form depending on the model used for drift) of
Eqns. (16)-(19) - constitutes a general approach to determining the
effective polarization currents induced by a small ac signal applied
across a thin film. actual evaluation of such a system can
be a horrendous task. This is true since the boundary conditions may
be difficult to model (definitive information .about interfaces trans-
port is necessitated) and since the dc quantities must be calculated.
The equations for the de values are non-linear and are complicated
by the fact that band bending may be present even for the case V=O
(15,20) .
At least two approaches to this complex analytical problem of
determining the possible effective ac polarization currents arising
from interface effects may be found in the litterature. These repre-
sent two extreme models for thin films. Thus these constitute two
approaches to the system Eqns. (16)-(19). One approach(13,15-18)
neglects the existence of regions of large band bending (>kT) al-
though some progress has been made to correct this deficiency of
the analysis(19,20). In this approach there is never any appearance
of quantities such as 1 and 2 which are the widths, at bias V, of
the regions in Fig. I, for example, where large band bending is occu-
232 S.J. FONASH
ring. Actually in such analyses it is usual to treat the bands as
being flat and thus to ignore completely any band bending at the
interfaces. In this approach it is alsc usual to treat the V = 0
case only ; discussion of its extension to V F 0 may be found in
reference (15) and (20). This approach may be characterized by
saying that the effective polarization arises from the motion of
mobile charges moving back and forth under the action of the ac
field, in essentially a zero diffusion potential situation.
The alternative approach neglects either completely(5,21,22)
or partially(23) any modifications in concentrations which take pla-
ce in region of band bending <kT. In fact this class of approximate
solutions to the system Eq. (J6)-(19) usually considers only modi-
fications to charge in regions where the band bending is >kT(5,21,
22) and allows for charge growth only at the edges of this space
charge region. This alternative approach is essentially a depletion
approximation but the charge being depleted need not necesf1arpy
be mobile carriers. It could, for example, reside in
Using the first approach it is possible to obtain a solution
to the system (16)-(19). With the initial and boundary conditions
that are stipulated in this analysis, one makes the assumptions
which characterize it : the possible existence of regions with lar-
ge band bending is ignored (although the implications of this are
fully realized(15)). That is, to make a general solution tractable,
flat bands are assumed for the unperturbed situation. Further, the
case where V = 0 only is explored (i.e., E = 0)(15). In addition
the so-called Chang-Jaffe conditions Cl6) are imposed as the bounda-
ry conditions on the current contribution, at x = (O,w), for the
i'th species; i.e
(2ri eDiA/W)oni I
(21)
x=O,w x=O,w
In this phenomenological expression ri is a dimensionless discharge
parameter.
From Eqs. (16)-(19), with these assumed conditions, general
solutions for the effective polarization currents to be expected
experimentally have been obtained for the case of one positive and
one negative species present (holes, electrons, ions) with arbitra-
ry valencies and mobilities. An immobile homogeneous charge density
was also assumed present. Recombination, generation effects were
neglected(15). These solutions are such that when Eq.(6) with oP=O
is evaluated and interpreted in terms of Eq.(7), C
p
and Rp are found
to be extremely complicated functions. However, they simplify in a
number of situations of interest for the case of w = O. Polarization
is accounted for here by E.
DIELECTRIC PROPERTIES OF THIN FILMS 233
When the discharge parameters of the two mobile species are
taken as equal (to sar r), it is found for example, that Cpo=Cp(w=O)
is given by Eq. (22)( 5) ; here C6 = ~ o C G
C .. +
G
cot
[I + r/2]2 [I +
C ..
G
(22)
It is interesting to note the manner in which \II appears in this
expression. Obviously if the interfaces at the electrodes did
not modify the concentrations adjacent to them (ie, if r + 00),
there would be no interface effect.
Since LD is the Debye length defined by
(23)
the contribution to the effective dielectric properties of a thin
film coming from the term in Eq. (22) involving LD could be sub-
stantial - depending on the values of nand p. If the mobile charge
carriers are electrons or holes (although it is difficult to con-
ceive of a real situation where they would have the same discharge
parameter except for the case of blocking contacts ; ie, r=O) the
electron and hole densities would have to be something of the order
of I014cm-3 or larger for this analysis to be meaningful for a thin
film (\II < micron). Thus in this case for wide gap thin film mate-
rials, with deep donor and acceptor levels, this origin of a con-
tribution to the effective dielectric constant should not be impor-
tant. This is especially so if there are depletion regions present.
Then these have widths 1 and 2 of the order -of
(24)
where NC is the effective doping density(5) and VD is the diffusion
potential. Since NC can be large even in wide gap materials
capacitance arising from these regions would be quite important.
If the mobile carriers in this situation (Eq. 22) are ions, the
large number densities required for a contribution to th& effective
dielectric properties of a thin film should be detectable in the ma-
terial transport which would take place in a dc experiment. Some
experimental effects seen in thin films have been attributed to this
case with the mobile carriers being ions(25).
234 S.J. FONASH
For the situation in which the contacts are completely bloc-
king to both carriers it is apparent from Eq. (22) that the above
comments are still valid. For the situation in which the
are blocking to one species of mobile carrier but not to another,
it is again found that relatively large number densities of free
charge are required for this phenomenon to contribute significantly
to the effective dielectric properties of thin films. For example,
if positive ions are blocked but the electron density at the inter-
face is totally undisturbed by the variation about V = 0 (ie, rp=O
rn + 00), then, with Zn = Zp = I, Cpo given by(15).
Cpo
C ... +
E:
(
nE
)
G 2LD (n + p)2
+
E: W
[ (n p)
r
c ... r
---p-
r
(25)
2LD
coth 2LD
Gl
(n + p)
Again LD is given by Eq. (23) and again it is intersting to note
the dependence of this on W. Obviously this resulting dependen-
ce on W predicted for this case by this approach to
the system Eqs (16)-(19) could be examined experimentally.
Since the quantities in parentheses are always < unity, the
important difference(that is, the possible origin of larger effects)
between Eq. (25) and Eq. (22) is the appearance in Eq. (25) of the
factor (WiLD)' Thus, much larger effects can be found for given
densities because of this factor ; however, comments made regarding
Eq. (22) should also be valid in this situation.
On the basis of this discussion it would seem that experimen-
tal data, which have been obtained at room temperature for several
wide gap dielectic thin film materials yet which show an apparent
increase in low frequency dielectric cqnstants which is greater than
several orders of magnitude over the high frequency values, must be
interpreted in terms of an effect attributable to mobile ions and
defects (Fig. 2) having their concentrations perturbed at the elec-
trode interfaces(l5,20.25,26). Similar effects have also been seen
in thin films of ionic materials. The interpretation has been the
same(27). It is obvious, however, that there is range of values of
Cpo which could be determined experimentally which are extremely
to interpret without some independently obtained infor-
mation such as that which may be extracted from dc measurements
(28,29), SEM observations(30), etc. This difficulty will become clea-
rer as the analytical predictions resulting from the second approach
(that is, second model for the origins of the effective polarization)
are examined.
DIELECTRIC PROPERTIES OF THIN FILMS 235
10
7
280C
--=-----
- --- ---
10
6
- ....
......
--....
--..... '......
'a
- ~ '"
..
10
5
" , ,
UJ
' ,
i
" , ,
i .
,
10
3
10
2
10-
4 10-
1
10
FREQlENCY, Hz
Fig. Z Dependence of capacitance on frequency for RF sputtered
SiOZ films. Film thickness is 1400 A. Theoretical curves and ex-
perimental data are shown. (After Meaudre and Meaudre, Ref.(Z5)).
For many thin film materiais it is quite evident that the pos-
sibility of large band bending (and, therefore, of space charge
regions) at interfaces is very good. This would occur even at zero
bias(5,14,15,ZO,30-3Z). Thus regions of band bending >kT should
probably exist even for V = 0 essentially because the amorphous
nature of thin films indicates that there should be states in the
energy gap of these materials. These states will be able to empty
or fill near the interfaces to equate Fermi levels.
An analysis of the polarization effects possible because of
the existence these regions has been recently undertaken(5). The
existence of concent,ration modifications in regions of band bending
<kT is ignored. Space charge is assumed to grow only at its edge.
236 S.J. FONASH
For the sake of definitiveness the electronic carrier of principal
interest is assumed to be electrons, the source of which is a donor
level ED below the conduction band edge. The Fermi level is assumed
to lie somewhere above the donor states. With this model a wide gap
dielectric thin film can be represented which has very few mobile
electrons but which, through the mechanism of emptying or filling
of the donor states, can develop a substantial space charge region
at the interfaces. The diagram of Fig. I represents this situation
uneer the general biasing conditions assumed in the analysis.
Equation (6), of course, valid as are equations of the' form
of Eqs (16)-(19). However. from the assumptions made it follows
that oQl' the change in the charge in region I, is related to E
B
,
the electric field in the interior, by
(26)
and
(27)
where here 11 is the particle current at X * 0 and 12 is the cor-
responding quantity at X = W. Also since the charge in regions
and 2 is assumed to grow only at the edge, it is easily shown
that(S)
and
The total bias variation eV is given byeS)
where
OV
1
= - EB e ~ l ~ l eEl
eV2= - EB O ~ 2 ~ 2 eE2
and
(28)
(29)
(30)
(31)
(32)
(33)
DIELECTRIC PROPERTIES OF THIN FILMS 237
In these expressions El and E2 are the electric fields in re-
gions 1 and 2, respectively. These equations are, of course equiva-
lent to the system Eqs (16)-(19).
With = and = the system Eqs (26) -
(33) can be solved in terms of oV by noting that the particle cur-
rent crossing X = 0 can be written in a general way as(5,33)
(34)
A similar expression follows for 012. Using the results from sol-
ving this system in Eq. (6) and assuming the variations are of the
form ejWt allows interpretation in terms of Eq. (7). The expres-
sions for C
p
= Cp(V.w) and Rp = Rp(V.W) which are determined are
also extremely complex(5). However,they may be represented in
terms of the equivalent circuit of Fig. 3.
and
For Fig. 3 the following definitions have been employed(21)
IE = - C
B
E aOv\+
B B aVl at
1
ar EB
at2 1
-
-+
aV
2
B R2 aEB

E
all B at2 1
OV
2
-
aE
B
aV2 B
E
(35)
(38)
(39)
a II 1 0
[ C
B
:
B

+
aE R, B VB
+
aV2
B B
(40)
I =1312 B .!.. OV _ dI2 1 OV
R -(lEB (lV
1
B 1 (lE
B
iBB B
+ fl CBE
B
C E2 1 C E (41 )
l
-B- + B2 n
B
B
(lQ,1 - !! oV
o aVl dEB dt 1 aE
B
dt B
238
S.J. FONASH
In addition, the small-signal circuit elements representing
the interior region B, the region across which aVB is developed,
are given by
Further, in all these definitions
E
Cl :: 11
C2
E
-
I;
C
B
E
-
B
and
B -
have been used, as have the definitions
Rl :: [ ~ ~ ~ J-I
R2 : : [ ~ ~ ~ )-1
(42)
(43)
(44)
(45)
(46)
(47)
(48)
(49)
(50
This involved equivalent circuit representation of Cp(V,w)
and Rp(V,w) reduces to that of Fig. 4 if the field in the inte-
rior, region B, is much less than that in the interface regions
and if changes in the field of the interior are much smaller than
changes in the fields at the interfaces(21). The extremely simple
model of Fig. 4 is nothing more than a Maxwell-Wagner type of mo-
del for three layered regions!4)However, unlike true Maxwell-
DIELECTRIC PROPERTIES OF THIN FILMS 239
-i>
REGION I REGION B REGION 2
CI- C", CR
..
r=-c
Ii:'
1-.;
Iii'
R
"
R
Fig. 3 AC equivalent circuit resulting from the full analysis of
Ref. (5)
REGION I REGION B REGION 2
..
"
..
.-'L
C,
c.
C
z
- -
1'1, R.
It
z
Fig. 4 "Maxwell-Wagner-like model ; circuit elements are defined in
the text Fig. 3 reduces to this under two sets of conditions
described in text.
Wagner models for regions of differing conductivities, these regions
represented by Fig. 4 can grow and shrink with
If Eqs. (35)-(50) are numerically evaluated for parameter va-
lues that insure EB El and E2 (and oE
B
OEl and OE2). results
such as those shown in Figs. 5 and 6 are obtained(5). Since the
circuit of Fig. 4 woald apply for such a choice of parameters. the
240 S.J. FONASH
appearance of only one relaxation time for zero bias is just as ex-
pected for a Maxwell-Wagner-like model. This is so because for n
layers, (n-l) relaxation times result(34). With the application of
bias, two relaxation times appear since there now are three dis-
similar layers in the thin film.
For the conditions EB El and E2(with eEB eEl and eE2)
the low frequency capacitance dictated by the interfaces as
may be seen from Fig. 5 and Fig. 7. Such behavior has been seen
experimentally for certain systems(35). This conclusion applies
equally well to the situation of zero bias (i.e., V = 0). Equation
(22) gives, however, for V = 0 a 6C=C o-C 00 which is independent
of thickness. Equation (25) can give or = 0 a 6C which increases
with film thickness. Equations (22) and (25) can explain increases
in the dielective constant at low frequencies over the high fre-
quency value which exceed several orders of magnitude. The behavior
depicted in Figs. 5 and 6 is more limited; essentially the 6C
values possible are restricted by tl. For very small values of tl
the interfaces would become invisible due to tunneling. Also t1can
be made to grow(5) if Fig. 4 is valid providing another check on
this interpretation of 6C.
,.:-
24
"'e
u
201=------.--J-.-5-X-ld-=---.
"
= 7.,. 10
8
2
"' ------------- ,
-, "
.-. 2-.-107.-.;------. "','
FREQuENCY (sec-I)
Fig. 5 Parallel capacitance (xl07F/cm2) versus fre-
quency at 300
0
K for a 2000 A thick insulator. Numerical evaluation
of the circuit shown in Fig. 3. is done for parameter values such
that Fig. 4. is valid.
DIELECTRIC PROPERTIES OF THIN FILMS
It is interesting to note that a Maxwell-Wagner-like model
may be extracted from Eqs (6), (26)-(33) without stringent
conditions on the fields. This may be accomplished by writting
the approximations
(51)
and
(52)
in place of expressions of the form of Eq. (34) and by defining
and
and
oVl - oVl + OEI
oVz - OV2 + oE2
(53)
(54)
(55)
With these statements Eqs. (26)-(33) may be rewritten as
(56)
(57)
(58)
(59)
oVl + oVi + oVz oV (60)
241
242 S.J. FONASH
These equations are consistent with Eq. (6) and may be easily
seen as resulting in an equivalent circuit like that shown in Fig.
4. However, in this case
and
define
are as
do not
10
<D
'Q
K
10
1
N
E
<J
,
E
.c:
10-2 0
"-
II:
10
3
10
4
10
5
Rl
-1 d II 1
-
- dEl 1"1
(61)
Rz
-1 Hz
-
- dEz I;
(62)
these elements in the equivalent circuit. The other elements
defined previously. Here the three RC parallel combinations
physically represent the three regions of the film since
FREOUENCY
Fig. 6.Parallel equivalent resistance (x 10-
6
cm
2
) versus frequency
at 300
0
K for a 2000 A thick insulatQr. Numerical evaluation of the
circuit shown in Fig. 3. is donee for parameter values such that
Fig. 4. is valid.
DIELECTRIC PROPERTIES OF THIN FILMS 243
the transformation Eqs (53)-(55) has been employed. The circuit
as a whole does represent the thin film (with its assumed interfa-
ces) within the validity of Eqs (51) and (52). Here it is seen
that Cpo is not dominated by the interface; i. e.,
(63)
Experimental data. in which a clear domination of Cpo by the inter-
faces is absent but for which the existence of space char.ge re-
gions has been demonstrated in a rather interesting manner(30) is
shown in Fig. 8.
In general then for this approach to a system of the form
of Eq (16)-(19) the results are such that Fig. 3 is valid. However,
if the interior electric field (as well as changes in the interior
electric fiels) is small, then Fig. 3 reduces to Fig. 4 -- a
I- J. 2
z
=> 1.0
:::J 0.8
a..
L1J
U
0.6
0.4
I-
u
;;:0.0 4 3
5 10- 10- 10- 2 10- 1 100
B I AS (vo 1 ts)
Fig. 7 The quantltles Cl, C2, and C
G
(xI0
7
F/cm
2
) which were used
in the computation of Fig. 5 and Fig. 6
244 S.J. FONASH
Maxwell-Wagner-like model (21). Here each RC combination corres-
ponds to a physical region in the thin film. In this situation
= of course, but C 0 is dominated by the interfaces. With
only the assumption of Eq. t51) and (52) the analysis again reduces
to a circuit model of the type shown in Fig. 4 ; however, the RC
combinations no longer necessarily represent physical regions and
Cpo is given by Eq. (63). The assumtion of Eqs. (51) and (52) is
rather good in general for a reversed biased junction. For a for-
ward biased junction it is quite good if the field in the interior
is small ; however, under bias the expression for the forward bia-
sed contact should not be critical, in general, since this inter-
face's resistance should be much smaller than other resistances in
the thin film structure.
Thus the Maxwell-Wagner model can be used to explain thin film
dielectric behavior in several situations and the model appears re-
peatedly in the literature. It is used for situations where the above
OY
3XI0'
5Y
-
lOY
..
E
2 X 10
4
i::
a.
OV u
5Y
IOV
ISY
10"


I V

RZ 10000 1
RIIZOOOO 1
______ ________ ________ ______ ______ ____
0.1 10 100 IK 10K lOOK
FREOUENCY I HZ I
Fig. 8. Experimental data showing Cp versus frequency for a thin
film with the global stoichiometry of SiO. (After ref.(30)}.
DIELECTRIC PROPERTIES OF THIN FILMS 245
analysis applies; ie., a Maxwell-Wagner-like model. For example,
Fig. 9 shows experimental data on Mo03 in which the interfaces do-
minate(28,35). It is also applied in situations where fixed regions,
with differing conductivities, are believed to exist in films(36,
37). This latter situation, of course, is that for which the model
was first developed. (4)
In the discussion of the second approach presented, the fini-
te emission time required to liberate carriers from states in the
gap has been ignored. This can be a rather important omission for
cases where the space charge comes from emisssion and sweeping-out
of such carriers as it probably does in a wide gap dielectric thin
film(5). As may be seen from Eq. (6), if T is a measure of this
emission time, then for w T-1the capacitance reduces to C
G
(again oP is being ignored here as it has throughout this section
by the standard procedure of incorporating its "bulk" effects into
e:).
Sampll Ihickness 3500 1
---E_perimlnlal
- - - - - - Theorltical
10:1.
...
c:
10

1 Fr8qUlncy (HZ)
Fig. 9. data for the series equivalent capacitance
versus frequency for a Mo03 thin film sample with W = 3500 A.
Comparison is made with theoretical curves computed from Fig. .
with Rl and Rz + 00. (After G. Nadkarni and J. Simmons,
permission).
246 S.J. FONASH
Some aspects of this type of complication have been explored in the
literature(23.30).
3. POLARIZATION CHARACTERISTIC OF THE MATERIAL
Contributions to the polarization arising from the third term
on the right in Eq (5) are characteristic of the thin film material
as opposed to owing their origins to interfacial phenomena. In the
case that these category II contributions are from permanent dipo-
les or from dipole-like phenomena for which the simple bistable con-
figuration is applicable (can be applicable to polarization in the
thin films arising from point defects-impurities. vacancies- and
even to two center electron hopping(I.26).Eq (4) is valid(I). If
an ac variation is assumed. it is seen from Eq (4) that

sn
-
00
In this standard form. P
and
(64)
Poo + P
n
and the definitions
(65)
(66)
have been used; these are, of course. consistent with Eq. (3).
The quantity Poo is the high frequency polarization which is able
to follow the field(I). If Eq. (4) is not valid (as been argued re-
cently for many materials(38). oP still is proportional to OE.
with the proportionality factor depending on the ac frequency. if
the governing for the temporal behavior. replacing Eq. (4)
is linear.
Thus. Eq. (6) becomes (note that the sign conventions of Fig.1
require that - oE dx = OV)
01 == OIp + A;. oV + A J: (I .. dx
(67)
This expression for the ac variations is of general validity ; the
only restriction is that the equation governing the temporal res-
ponse of oP be linear.
DIELECTRIC PROPERTIES OF THIN FILMS 247
o ~ the case where Eq (4) is valid the specific mathematical model
for the complex permittivity E* appearing in Eq (67) can be written
down using Eq (64) ; ie.
E - Eoo
"""1-!-n
W
-
2
,(-zr- (1 - jWTn)
n
(68)
This expression for E*, resulting from the validity of Eq (4), re-
presents the sum of n Debye-like processes.
As was mentioned in the preceding section, the discussion of
category I ac behavior proceeds using Eq. (67) with E* replacing
EO' If no particle current crosses X = 0 and if op is zero, the
thin film structure still appears, in spite of no interfaces effects,
to have a resistance (from Eq. (12 as well as a capacitance (from
Eq. (13. Thus, if both category I and category II mechanisms are
operative in a thin film, one is faced with their joint effects,
as indicated by Eq. (67), in an ac experiment. That is, both bulk
and effective polarization are present in general.
Cole and Cole(39) expressed the superposition of relaxation
times indicated by Eq. (68) as
(69)
which, although it suffers from so.ne possible inconsistencies(40),
is frequently ased to interpret experimental data in terms of bulk
Debye processes. In terms of Cole-Cole diagrams(39) with E" plotted
versus E', a semicircle is obtained for a= 0 (one relaxation time)
with its center on the E' axis. With a > 0, the center moves below
the axis ; a being a measure of the angle between the E axis and a
radius to the closest point of intersection (to the origin) or the
curve and the E' axis(39). For a values" 0, E" becomes fairly flat;
therefore, fairly independent of frequency. In fact, neglecting in-
terfacial contributions to Eq. (67) shows, using Eq. (69), that
cos(I - a) - (70)
and
(71)
248 S.J. FONASH
for the case of a distribution of relaxation times and WT I.
Thus, depending on a, the superposition of relaxation times could
lead to Cp = constant and Rp - w-a. Such behavior is commonly seen
in thin films(41). Figures 8 and 10 show similar behavior; however,
here a = 2. Consequently the explanation of Eq. (71) is not possible
due to restriction of a(39). For a single relaxation time, WT I
leads to both Cp and Rp being constant with frequency for the Debye
model.
It is interesting to note the changes in capacitance
possible from these dipole-like mechanisms : from Eq (68) it follows
that
N M 2
n n
C/Cpoo '" Ln --
kTe
o
(72)
Here Mu is the dipole moment for species n with the number density
N
n
Thus if the N values are of the order of 10
20
cm-
3
, is
ofothe order of 20 per species using as the effective moment arm
5 A. Consequently it is difficult to explain data such as of Fig. 2
in terms of a dipolar-like mechanism without invoking extremely
large dipole densities. Data such as that of Fig. 8 could be inter-
preted in terms of a dipolar-like except that the activa-
tion energy (assuming T = TO (I for Cp is the same as
that for the dc resistance(30). From Eqs (67) and (68) it is seen
that the dc resistance for the channel of current flow due to e" is
infinite. Also scanning electron microscope observations have indi-
cated the presence of interfacial space charge for the data of Fig.
8(30). There are, however, many case in the literature where dipole
or dipolar-like mechanisms are used to explain observed dielectric
properties in thin films(26,42,43).
In amorphous films, electronic conduction by hopping and
percalation(41,44,45)must. be considered as quite possible. The pre-
sence of these transport mechanisms would give rise to another con-
tribution to the dielectric properties from category II type pheno-
mena. In general electronic hopping transport, it has been argued,
should lead to crt varying as wn and e' varying as wn-I with n > 1/2
(41). As has been mentioned, this class of behavior is common in
thin film materials for higher frequencies having been seen in SiO
x
(26,30), At
2
03(26), and chalcogenide glasses(46) to cite a few cases.
For the simple case of two center hopping(26,47), this increa-
se of crt with W is easily demonstrated to arise : here
crt
(73)
DIELECTRIC PROPERTIES OF THIN FILMS 249
and also
s'
1+ W2,2
(74)
These are of the same form as Eq. (68) would yield. However, here
it is possible to postulate that, is very small; i.e., for fre-
quencies of interest I W
2
,2. Consequently, a' varies, in this
case, as w
2
and s' is constant.
Unfortunately, interpretation of this a = a(w) behavior is far
from unambiguous(46). Ionic defect hopping can also make a contri-
bution(48). Conductivity increasing with w can come, as discussed,
from dipolar-like mechanisms. As may be seen from the numerical
evaluation presented in Fig. 6, Rp _ w-
n
can come from polarization
in depletion regions. It can develop from effective polarization
with origins in mobile carrier concentration modification even for
flat bands(I3). In fact, as pointed out in the literature, any in-
homogeneous conduction mechanism - on the microscopic or macrosco-
pic scale - can produce conductivity increasing with frequency(46).
4. CONCLUSIONS
Dielectric properties of thin films are the result of inter-
face effects and of effects arising from the material "bulk" pro-
perties. These mechanisms add according to Eq. (5). If ac varia-
tions are used to modify the Eq. (67) is valid if the
polarization obeys a linear temporal equation. The quantity s* ap-
pearing in Eq. (67), in ,cases where the Debye model is valid (Eq(4,
is given by Eq. (68) in the general situation of many Debye-like
mechanisms.
Thus if, as an example, the dielectric properties of a thin
film owed their existence to space charge coming from deep donors
and to a "bulk" dipolar-like mechanisms described by Eq. (69), it
can be seen from Eq. (67) that for low w the resistance would be
mainly due to particle current crossing the electrodes. The capa-
citance would be > C
G
since it would arise from the last two terms
in Eq. (67). If the deep traps supplied the carriers, the activation
energy for the dc current should be the same as that for Cpo For
w > ,-I, where, is being used here as a measure of the emission
time from the deep states, the contributions from the third term
on the right in Eq. (67) could become small. Thus, Cp would tend
now to C6 and Rp - w-a . The activation energy of the high frequen-
cy process would be totally unrelated to that of the low frequency
process in this example.
250 S.J. FONASH
Unfortunately many phenomena - or combinations of phenomena-
can give rise to the frequency behavior described in the above ex-
ample. As discussed, analysis of interface polarization arising
from depletion regions indicates that it alone can produce similar
behavior as can Maxwell-Wagner and Maxwell-Wagner-like models. Di-
pole, dipole-like, or hopping mechanisms with a constant (in w)
background conductivity can also produce this frequency behavior.
However, these combinations will not produce correlations, descri-
bed above. for the activation energies. It is obvious, however,
that other combinations could be imagined which would.
Thus in examining thin film polarization or effective pola-
rization response, it becomes clear that care must be exercised in
searching for the physical origins of the observed behavior. The
conductance as a function of frequency, bias, and thickness must be
studied. A similar study must be made for Cpo Activation energies
should be measured. Ac properties should be correlated with dc be-
havior and independent information sllould be obtained, if possible,
from thermally stimulated currents, observations of space charge
regions, etc. Thin films necessitate this kind of a continued, in-
tensive experimental effort - together with much more theoretical
work - to unravel fully the causes of their bulk and effective po-
larization.
10
IOV
ISV
Oy----------____ __
5V--------------__ _
__ :: IOV-

10
IOy-_________________ _
o
R3 20.000 A
Ra 10,000 l
RI 5,000 1
0.1 10 100 IK 10K lOOK
FREQUENCY (IN HZ)
Fig. 10 = for a film with stoichiometry of SiO. Note
that the high frequency behavior scales with W but is independent
of bias.
DIELECTRIC PROPERTIES OF THIN FILMS
REFERENCES
1. V. V. Daniel, "Dielectric Relaxation" Academic Press, N. Y.
(1967).
2. H. Frohlich, "Theory of Dielectrics", Oxford Press, Oxford
(1958) .
251
3. C. Kittel, "Intro. to Solid State Physics" Wiley, Inc., N.Y.
(1968).
4. A. Von Hipple in "Handbook of Physics" (E. Condon and H. Odis-
haw, Eds.) McGraw-Hill, N.Y. (1967).
5. s. J. Fonash. J. A. Roger, J. Pivot, and A. Cachard, J. Appl.
Phys. 45, 1223 (1974).
6. M. Perlman and S. Unger, J. Appl. Phys. 45, 2389 (1974).
7. M. Baird, Rev. of Mod. Phys. 40, 219 (1968).
8. J. G. Simmons and G. Nadkarni, Phys. 4815 (1972).
9. Ai Bui, et al., Thin Solid Films 313 (1974).
10. c. Bucci, R. Fieschi, and G. Guidi, Phys. Rev. 148, 816 (1966).
11. D. M. Smyth in "Oxides and Oxide Films" (J. Diggle, Ed), Dekker,
Inc., N. Y. (1973).
12. P. J. Harrop and D.S. Campbell in "Handbook of Thin Films"
(L. Maissel & R. GIang, Eds.) McGraw-Hill, N.Y. (1970). Con-
tains an excellent list of references up to the late 1960's.
13. R. J. Friauf, J. Chem. Phys. 1329 (1954).
14. J. C. Simmons and G. W. Taylor, Phys. Rev. i, 4793 (1972)
15. J. R. Macdonald, J. Chem; 4982. (1973).
16. H. Chang and G. Jaffe, J. Chem. Phys. 1071 (1952).
17. J. R. Macdonald, J. Chem. Phys. 40, 3735 (1964).
18. F. A. Kroger, "The Cehmistry of Imperfect Crystals" Wiley, Inc.
N.Y. (1964). (Contains a review of the literature on this sub-
ject up to 1963 ; see sec. 22.11).
19. E. Fatuzzo and S. Cuppo, J. Appl. Phys. 43, 1457 (1972).
20. J. R. Macdonald, J. Appl. Phys. 73 (1974).
21. s. J. Fonash, J. A. Roger, and C. H . S. Dupuy, J. Appl. Phys.
45, 2907 (1974).
22. J. G. Simmons, G. S. Nadkarni, and M. C. Lancaster, J. Appl.
Phys. 538 (1970).
23. J. Maserjian, J. Vacuum Sci. Technol. i, 843 (1969).
252 S.J. FONASH
24. G. Jaffe, Phys. Rev. 85, 354 (1952).
25. M. Meaudre and R. Meaudre, Solid-St. Electron. 1205 (1973).
26. F. Argall and A. K. Jonscher, Thin Solid Films, 185 (1968).
27. J. C. Macfarlane and C. Weaver, Phil. Pag. ll, 671 (1966).
28. G. Nadkarni and J. Simmons, J. App 1. Phys. 43. 3741 (1972).
29. J. Siuntlons, G. Nadkarni, and M. Lancaster, J. App1.Phys. !,.,
545 (1970).
30. J. A. Roger, C. H. S. Dupuy, and S. J. Fonash. J. Appl. Phys.
46 (1975) ; J. Solid State Chern. g, 238, (1975).
31. J. G. Simmons, J. Phys. Chern. Solids 32, 1987 (1971).
32. J. G. Simmons, J. Phys. Chern. Solids 2581 (1971).
33. S. J. Fonash. Solid-State Electron. 12., 783 (1972).
34. J. Volger in "Progress in Semiconductors" Vol 4 (A.F. Gihson
Ed.) Heywood and Co., London (1960).
35. G. Nadkarni and J. Simmons, J. Appl. Phys. !,.. 545 (1970).
36. I. S. Goldstein, J. Appl. Phys. 45, 2447 (1974).
37. M. C. Lancaster, Thin Solid Films 213 (1971).
38. A. K. Jonscher, Private communication.
39. K. Cole and R. Cole, J. Chern. Phys. 341 (1941).
40. J. Macdonald, J. Chem. Phys. 36, 345 (1962).
41. A. K. Jonscher, J. Non-Crystalline Solids 8-10. 293 (1972).
42. G. Pfister and M. Abkowitz, J. Appl. Phys. 45, 1001 (1974).
43. A. Willoughby and F. Yuen, Thin Solid Films!l, 199 (1972).
44. J. Ziman, J.Phys. C (Proc. Phys: Soc.)..!.., 1532 (1968).
45. D. Adler, L. Flora, and S. Senturia, Solid. St. Communications
g, 9 (1973).
46. A. Owen and J. Robertson, J. Non-Cryst. 40 (1970).
47. M. Pollak and T. Geballe, Phys. Rev. 122, 1742 (1961).
48. E. Snow and P. Gibbs, J. Appl. Phys. 2368 (1964).
THRESHOLD SWITCHING:
A DISCUSSION OF THERMAL AND ELECTRONIC ISSUES
H. K. Henisch and C. Popescu
Materials Research Laboratory
The Pennsylvania State University, University Park
Pennsylvania 16S02, U.S.A.
1 INTRODUCTION
Papers of joint authorship ordinarly result from shared re-
search, but the present case is an exception. Shared principles are
its basis, of course, but it represents above all a search for
common ground, approached from opposite sides of the interpreta-
tional spectrum. The opposite sides are represented by thermal and
non-thermal (electronic) models of threshold switching, an issue
which has been widely (and, at times, hotly) debated by protagonists
at international conferences and in the literature. In the course
of such debates, misunderstandings abound, and some of the sweetest
victories are scored against "opposing" models not actually advanced
by the other party. In the present didactic context, an attempt at
clarification therefore seems particularly appropriate. Once found,
the common ground must serve at the starting point for more dis-
criminating experimentation. Inasmuch as our agreement falls short
of total unanimity, we each take our cue from the clergyman who
ended a bitter dispute with a colleague, saying : "It is clear that
we cannot settle this issue, and we must therefore agree to disagree
you will have to continue to serve the Lord in your way, and I shall
cDntinue to serve Him in His."
Threshold switching, in the form here discussed, is a phenome-
non specifically associated with thin films. It can take place in
an astonishing variety of materials, but few of these appear to be
suitable for commercial applications ; indeed, only the multicom-
ponent glasses (typical composition Te4oAs3SGe7Si1S) have come close
to marketable devices, following their invention by S.R. Ovshinsky(l)
(All the detailed results here discussed refer to them).
253
254 H.K. HENISCH AND C. POPESCU
Nevertheless, the hope and prospect of commercially viable devices
has in large part helped to stimulate the wide-spread interest in
high resistivity materials in general and chalcogenide alloy glas-
ses in particular. The view is often expressed that this interest
can and will be sustained only if it can be shown beyond doubt that
the phenomena involved are primarly electronic, and, conversely,
that it would cease if it were shown that they are primarly ther-
mal. This dichotomy has its simple attraction, but we cannot sup-
port it, because it seeks to draw excessively far-reaching referen-
ces from the right"-answer to the wrong question. Although there
are such connections in limiting cases, it is wrong to use the terms
"electronic" and "thermal" a priori as if they were synonymous with
"stable" and "unstable" respectively. Device design is always a
matter of compromise in which the conflicting demands of performance
and durability must be reconciled, and the fact that thermal aspects
are important in electronic devices of many different kinds has
not automatically precludes such a compromise. Moreover, whereas
there is no prob lem about envisaging a "purely thermal" process,
a "purely electronic" process is clearly impossible. Once and elec-
tronic process is demonstrated, we are immediately concerned with
"thermal overtones" in varying degree, from the functionally trivial
to the functionally essential. In the case of threshold, which
involves power dissipation within-a volume of a few cubic microns,
the functionally trivial limit is unlikely, and that is of course,
one reason why thermal models of various kinds have been so popular.
It is, of course, possible to envisage models in which switch-
ing arises from the simultaneous action of thermal and electronic
effects, neither mechanism being by itself sufficient to secure
the existence of a discontinuity. Such a model would have to be
called "electrothermal" rather than thermal. Accordingly, the choice
is between a thermal model (albeit modified by electronic processes,
e.g. by the field dependent conductivity), and electronic model
(albeit modified by self-heating), and an electrothermal model
(implying a cooperative process). Though the most discussed in qua-
litative terms.
It was certainly hoped during the initial phase of this re-
search that the relative importance of thermal and non-thermal as-
pects was a matter which could be quickly and unambigously resolved
by experiment, but this hope has not yet bean fulfilled to every-
one's satisfaction. To be sure "proofs" of one sor or another fill
the literature, but they carry conviction mostly so kindred spirits,
rarely to outright opponents. Why this should be so is not hard to
see. Because some heating is inevitable, the demonstration of
heating effects as such proves nothing. One must also realize that
a switch is a two-thermal network and, in the last analysis, the
only electrical measurement we can make on it is a measurement of
resistance. Resistance as a function of voltage, current, time,
THRESHOLD SWITCHING 255
temperature, rate-of-address, pulse duration repesent some of the
possible variations, but the fact is that the consequences of elec-
tronic disequilibrium are expected to be rather similar as far as
their effect on overall resistance is concerned. Experimental
results therefore leave a good deal of room for interpretation,
and, in the past, features which have seemed essential, character-
istic and conclusive to one party have often seemed trivial to
another. However, the field of threshold switching cannot continue
to grow in this way, and attempts to define decisive issues are
urgently needed.
2. PHENOMENOLOGICAL CHARACTERISTICS
For practical device purposes, thershold switches must be made
entirely by thin film techniques, and unless alternative stable
electrodes can be found, this involves the preparation and manipu-
lation of graphite films. That this can be done has been demons-
trated by Ovshinsky and co-workers(2). However, most of the results
in the literature (indeed, an overwhelming proportion of them) have
been obtained on other types of structures, usually as shown by the
insert on Fig. 1. Assembled and encapsulated units of this kind
have been known to be highly stable, necessarily, within an appro-
priate range of operating conditions, and on them is based the
(very reasonable) hope that stable all-thin-film devices can also
be produced.
The V-I characteristics of threshold switches have often been
described and only a brief summary is needed here, e.g. see Shanks(3).
A schematic representation of a switch is shown on Fig. 1. It
corresponds to the picture which can be traced out point by point
by applying pulses of varying amplitude and using series resistances
over a wide range of values. Under AC conditions, the oscilloscope
display gives a similar picture, but some controversy is attached
to the exact manner in which the ON-state terminates near the mini-
mum holding current IMN' Under pulse conditions, Henisch and Pryor(4)
found it to be as shown by the full line in Fig. 1. However, under
alternating signals, the slope is often reported as negative (broken
line), just before the return of the OFF-state. There is a general
belief that the exact conditions of measurements (including circuit
capacitance) and the rate of external voltage withdrawal influence
the behavior, but precise relationships have never been ascertained,
as far as we know.
The V-I relationship is also known as the "primary character-
istic". In addi tion, switches have a variety of secondary charac-
teristics, some of which are illustrated by Fig.2. Thus
t
switching
at minimum voltage (the threshold point) is statistical 5) as far
as the switching delay is concerned (Fig. 2a). Voltages higher than
threshold can be applied, with corresponding reductions of the
256
(a)
H.K. HENISCH AND C. POPESCU
( b)
(c)
P = pyrolytic grophite
V vitreous graphite
- = active moterial
Fig. 1 (a-c). Threshold Switching; Pricary Characteristics.
Switch structures with solid electrodes.
THRESHOLD SWITCHING
voltage
,-
c
CI)
...
...
::J
U
ON
+ IMH --
.'
,
,\ "
" "
I ,
I ,
.,.
-lMH
ON
-- - ION
Fig. I Cd) Threshold Switching; Primary Characteristics.
257
(d)
Schematic voltage-current characteristic of threshold
switches made with multi component chalcogenide alloys.
258
(0)
H.K. HENISCH AND C. POPESCU
(b)
~ , ~
I --'"
J
,
I
,
I
I
switching delay
Ume
IV 1500C
temperature
Fig. 2. (a), (b), (c) Threshold Switching; Secondary Characteristics
(multi component. chalcogenide alloys).
THRESHOLD SWITCHING
-
p
c:
CP
I
D'
...
(d)
... 0
:I
(,)
0
-VH
>
D'
c:
+ V
H
.s::
(e)
(,)
-c:
CP
...
...
:I
(,)
-
-
o

en
recovery time
I
threshold voltage----j
opplied voltage
259
Fig. 2 (d), (e), (). Threshold Switching; Secondary Characteristics
(multi component chalcogenide alloys).
260 H.K. HENISCH AND C. POPESCU
switching delay. With increasing overvoltage, the switching delay
ceases to be statistical and soon becomes well defined. During the
switching delay within the statistical regime, the current is con-
stant (2b), which means that there is no general increase of tempera-
ture. A highly localized increase, responible for only a small
fraction of the total current, is not ruled out by this observation.
The threshold voltage V
TH
(measured with AC) diminishes with in-
creasing ambient temperature in a manner which is virtually linear
over a 100C range from -50C upward, a non-linearity then sets in,
as shown on Fig. 2c. Extrapolation of the linear region sometimes
yields a temperature close to the glass transition temperature, a
fact which has given rise to a good deal of speculation(6) (7).
However, it is not known whether this coincidence is fortuitous or
significant. Such measurements are most conveniently made on encapsul-
ated units, but this procedure introduces additional uncertainties
arising from the thermal expansion of the capsule. Moreover, the
relationship between threshold voltage and ambient temperature is
more complicated than is generally realized. On un-encapsulated .
units, one of us (CP) and also Stubb and co-workers(8) have reported
finding a certain low-temperature range over which the threshold
voltage remains constant while the threshold power strongly decreases.
When the ON-state is displaced by rapid transient, a V-I
relationship is obtained as shown on Fig. 2d. Any point of this
relationship is accessible from any other point within 10 nano-
seconds or less. The high resistance portion has been called the
"blocked ON-state"(4) (9). Starting with any operating point, (say),
PI, the blocked ON-state can be reached by means of a suitable volt-
age pulse and allowed to prevail for periods up to 0.3 to 0.4 ~ s e s
after which point PI can be restored with a minimum of delay (e.g.
10 nanoseconds or less), and without switching. Switching is neces-
sary only after longer periods of current interruption, as shown
in Fig. 2e. This curve is independent of the energy dissipated
during the preceding switching pulse, and the position of the knee
is independent of temperature.
There are other secondary characteristics, but those listed
are probably the most important. In addition, there are some basic
(room temperature) observations upon which all interpretations
ultimately depend :
(a) The OFF-current, at any rate before the first (ever) switch-
ing event (see below), is proportional to the electrode
area ;
(b) The OFF-state resistance is roughly proportional to the
film thickness, as is the threshold voltage V
7H
, before
as well as after the first Bwtiching event (forming) ;
(c) The V-I characteristic in the ON-state is insensitive to
changes of film thickness ;
(d) The ON-state characteristic in the ON-state is independent
of electrode area.
THRESHOLD SWITCHING 261
Though experiments as a function of film thickness are highly in-
formative in principle, they are not easily performed, because
there is no simple way of ensuring that thick films have precisely
the same bulk and surface structure as thin films. Published results
must be viewed with this caution in mind. Point (c) implies that
most of the film thickness is free of field, and the last observa-
tion implies filament formation. On these point all models are
agreed. _The term "filament", though widely used (including here)
is actually misleading, inasmuch as the conducting volume often
has a length of I micron and a cross-section of 1 micron
2
Observa-
tion (a) means that the current distribution is uniform. Observation
(b) means that what matters for the threshold point is a critical
field at a given temperature, but whether this is the only condition
is uncertain and remains to be ascertained. Item (b) has some further
implications, namely that there is no appreciable contact resistance
in the OFF-state and that the_field distribution is uniform. Accord-
ingly, any byhavior in the OFF-state can be regarded as
bulk behavior ), 12 This concerns particularly the field depend-
ence of the conductivity which, at high fields, can be approximated
as
(I)
Beyond threshold, on the unstable part of the characteristic which
may be explored with short overvoltage pulses, the field dependence
is even steeper. It is believed(13) to arise from a mobility effect
(rather than from an increase of carrier concentration), but its
precise is not yet understood. This constitutes one of the
most challenging problems in solid state physics.
3. THERMAL THEORY FOR UNIFORM STRUCTURES
Thermal theory begins with the heat equation
(2)
where C = specific heat, j current density, and K = thermal con-
ductivity. In the steady state, it says no more than that whatever
energy goes in must come out. Its integration leads in principle
to a three dimensional temperature distribution contour for any
current, and ultimately to a complete voltage current relationship.
In that sense, it is always desirable to examine how far agreement
with experiment can be obtained on this basis alone. However, the
integration needs boundary conditions, and when these are formulat-
ed in the most general case, eqn. (I) cannot be explicitely solved.
The procedures are therefore open : (i) the introduction of approxi-
mations which will permit explicit solutions, and (ii) numerical
262 H.K. HENISCH AND C. POPESCU
solutions by computer, of which the latter is by far the more satis-
factory. At least two sets of computed solutions are in facL avail-
able, one by Kroll and Cohen(14) , and one by Popescu(15). They differ
somewhat in the assumptions made concerning the nature of the bulk
conductivity and the degree of thermal coupling. Thus Kroll and
Cohen assume a bulk conductivity in the form
(3)
is taken from experimental results and could be inter-
preted as arising from an activation energy which varies with the
temperature. Popescu's conductivity has a constant activation energy,
but the ratio of his thermal coupling constants (lateral to longitud-
inal) can be varied at will. Lateral coupling depends, of course, on
the thermal conductivity of the material and on the film thickness.
A priori, its importance must tend to zero as the film thickness
tends to zero. Longitudinal coupling depends on the thermal con-
nection between the electrode surfaces and the surrounding ambient.
Perfect lateral coupling would imply a uniform temperature and
current density, as shown by curve (b) on Fig. 3. Reduced lateral
coupling leads to filament formation and the appearance of a distinct
branching point, which may be as low as the turnover point. There
are, of course, longitudinal thermal gradients, but it can be shown
that these gradients are relatively unimportant in determining the
V-I relationships.
In accordance with these considerations, thick specimens are
expected to exhibit voltage turnover, followed by a (stabilizable)
negative resistance region. Thin specimens are expected to exhibit
a threshold voltage, followed by an experimentally inaccessible
region and, in due course by a stable ON-state. The branching point
marks the onset of filament formation. On a temporary basis filament
formation can be avoided. Experiments with high overvoltage pulses(17)
of very short duration should therefore reveal the characteristic
which corresponds to the case of a system completely isothermal with
the ambient (see Fig. 3). Practically all the current is carried by
the filament. Between the threshold point and point Vmin, the fila-
ment temperature increases, the diameter decreases. After that, the
current can increase only because the filament widens, its tempera-
ture coefficient being by then close to zero.
The curves on Fig. 3 which correspond to a small lateral-to-
longitudinal coupling ratio are evidently similar to the primary
switching characteristics on Fig. I. (They predict an ON-state which
terminates in accordance with the broken, rather than the full line.)
A field dependent conductivity modifies the shape of the characteris-
tics but does not otherwise contribute critically to the switch-
ing process. In the ON-state the average field is low but 0(F) may
still have a modifying role, inasmuch as a relatively high field
THRESHOLD SWITCHING
-c
cu
...
...
::J
U
263
opp I ied vol tage
Fig. 3. Schematic Voltage-current Characteristic Calculated on the
Basis of Thermal Theory (after C. Popescu). (a) Limiting
of system without heating; (b) System with uniform tempera-
ture and current distribution; (c-d) Filamentary systems
for diminishing ratio of lateral to longitudinal thermal
coupling
high current
(0)
low current
(b)
Fig. 4. Schematic Representation of Current Distribution for High
(a) and Low (b) ON-currents.
264
H.K. HENISCH AND C. POPESCU
may exist across the cooler portions of the glass near the electrodes
as suggested by Kroll.
As far as agreement with the primary characteristic is con-
cerned, the thermal model must be regarded as highly successful.
However, granted that thermal switching is possible in principle,
the question remains whether such a model can account convincingly
for the known secondary characteristics. This will be further dis-
cussed below.
4. ELECTRONIC THEORY FOR LIFETllm SEMICONDUCTORS OF UNIFORM
STRUCTURES
Thermal models aims at a single relationship which describe
the entire V-I characteristic in terms of bulk properties. Of course,
if it were considered necessary at any time, contact and interface
effects could be introduced as additional refinements. In contrast,
electronic models concern themselves separately with three issues :
(i) the shape of the OFF-state characteristic,
(ii) the nature of the instability at the threshold point,
and
(iii) the nature of the ON-state.
Of these issues, (i) need not be considered in detail here, hqwever
interesting it may be (as it certainly is) in principle. It can be
set aside because it is known that threshold switching occur in
systems, i.e. those consisting of polymer films
tl8
), which show only
linear pre-threshold characteristics. Pre-threshold non-linearity
is therefore not a pre-condition for switching.
Most progress has been made with (iii), the most detailed model
that of Mott(19), following suggestions by Henisch(20,21) and
Lee( 2). It itself with the manner in which the ON-state
could be maintained, once established by some mechanism (which
remains to be discussed). In accordance with clause (c), Section 2,
there is almost no electric field in the inclose to the electrodes,
possibly across space charge regions. At one time, these were seen
as regions of trapped charge, but such a system would be expected
to exhibit a strong temperature dependence. In fact, the results on
Fig. 2(e) do not, which suggests that trapping centers are not in-
volved. The alternative is a region of free charge carriers in
excess of the concentration demanded by neutrality. Mott(195 has
shown on a semi-quantitative basis that such layers can be stably
maintained in the presence of double injection and current flow,
assuming a near-intrinsic material. The appearance of space charges
is, of course, immediately likely whenever current passes through
a boundary between media of different conductive properties. Ex-
actly how dense and how spatially extensive these charges are can
be established only through computer-aided calculations, which have
not yet been done with complete generality.
THRESHOLD SWITCHING 265
The notion is that carriers from the electrodes tunnel through
these space charges regions and thereby maintain an electron-hole
plasma of density in the interior of the film. As has been
pointed out( 1), this implies that most electron traps are full of
electrons and most hole traps full of holes, the oucome being a
population inversion. The electron-hole plasma will tend to lose
carriers through recombination (some radiative, as discussed below),
and will be compensated by injection. When the two rates balance,
the system is stable. However, unless a certain minimum injection
rate is maintained, the carrier loss through recombination will
"win", which means that the system will revert to the OFF-state.
This is the electronic expla.nation of the minimum holding current.
The critical quantity is, of course, the ratio of transit time to
carrier lifetime. The minimum holding current should correspond to
equality to the two times, the transit time being, of course, field-
(and therefore current-) dependent. It will diminish sharply as the
diminishes, partly because of the lower average field
bu also for geometrical reasons (Fig. 4), as suggested by Mott.
The V-I relationship in the ON-state would be governed by the
tunneling characteristic of the barriers and by the field distri-
bution as between barriers and bulk. If the voltage drop across the
bulk is neglected (for this purpose only) one can calculate plausible
V-I relationships which can be fitted to experimental results with-
out difficulty. Characteristics which take account of barriers and
bulk have not yet been computed and, in the circumstances, compari-
sons between calculated and observed results are not as realistic
as they should be.
The absence of field in the interior implies an absence of
space charges and that is a priori an astonishing claim consider-
ing that these high resistivity materials co.ntain traps. However,
according to the Cohen, Fritzsche and Ovshinsky(23) model, at any
rate in its simplest form, the concentrations .of electron traps and
hole traps in a multicomponent glass are automatically equal (This
is believed to make it possible for both types of traps to be full
or almost full without generating the major space charges which one
would otherwise expect). The model would thus help to explain the
special position which the multicomponent chalcogenide glasses appear
to occupy in threshold switching.
It remains necessary to discuss electronic "options" for (ii)
above, the nature of the threshold point. The statistical character
of the switching delay makes impact ionization an attractive
sibility in principle. The suggestion is that the threshold field
is the critical field required for impact ionization. Once the
critical field is available, there will be a random waiting time
(switching delay) for an event. At higher fields (overvoltages) that
time s40uld be shorter, because carriers in less-than-ideal location
should be able to initiate the avalanche. If this were correct as
266 H.K. HENISCH AND C. POPESCU
H stands, then illumination should diminish the delay, in fact it
does not, but it has been shown(13) that it leaves the threshold
field unchanged, as-one would expect on the present basis. To
account for the independence of switching delay on illumination,
one would have to argue that the initiating carriers arise from
localities which are always occupied (e.g. from deep traps) or else
from a locality in which the carrier lifetime is very low (e.g. the
electrode interfaces). The matter has not yet been satisfactorily
resolved. However, despite the remaining gaps, there is no doubt
that an electronic model is feasible, based in part on conventional
semiconductor ideas and in part on the available information on the
band structure of mUlticomponent glasses. The same processes should
occur in single crystal materials, provided they are accurately
intrinsic, and it has been shown that this is indeed so(24). A
prediction is therefore fulfilled, but lest this be interpreted as
a decisive triumph of electronic theory, it must be pointed out
that the intrinsic material is also a material of high resistivity
and high temperature coefficient, factors which, in the absence of
more detailed evidence, permit other explanations.
Of course, thermal considerations would eventually have to be
superimposed upon any electronic model, but this has not yet been
attempted.
5. ELECTRONIC THEORY FOR RELAXATION SEMICONDUCTORS OF
STRUCTURE
The concept of relaxation semiconductors was introduced by Van
Roosbroeck and co-workers(25-28). It concerns materials in which
the dielectric relaxation time Td exceeds the carrier lifetime T.
Such materials are expected to behave very differently from normal
("lifetime") semiconductors, particularly in their response to
inj ected minority carriers. Van Roosbroeck has, for instance,
predicted a majority carrier depletion .region as the outcome of
minority carrier injection. On the basis of such considerations, he
has also proposed a theory of threshold switching or, at any rate,
a theory of the ON-state(29). The new concepts are believed to be
of greatest interest for our understanding of highly resistive
semiconductors, particularly when measured and explored in
film form. However, the analyses so far available do not give a
realistic picture of the consequences, inasmuch as they are based
on drastic simplifications, some of which are now known to be im-
permissible. In particular, the role of diffusion processes must be
considered, as must the contribution of minority carriers to the
current. Moreover, the presence of a minority carrier injecting
electrode on one side of the film implies minority carrier extraction
on the other side, a process which yields less easily to analysis.
Once again, the complete equations cannot be explicitly solved ;
a computer analysis is in progress(30) and until it is complete,
THRESHOLD SWITCHING 267
the significance of relaxation concepts for switching cannot be
properly assessed. The matter will therefore not be further pursued
for the moment, beyond pointing out
~ ) that there are electronic mechanisms potentially active
in switching other than those described above, and
~ ) that an analysis of thin film systems consisting of semi-
insulating (relaxation)materials, may not ultimately be
possible without consideration of the relaxation effects.
The expectations ar1s1ng may turn out to be just a com-
pelling as those arising from Joule heating.
6. CRITIQUE OF EXISTING MODELS
All the existing models have shortcomings, and though this
fact is not denied by anyone, it has proved difficult to distinguish
between shortcomings which are intrinsically associated with a
model (and thus cannot be remedied by any conceivable elaboration)
and those arising only from the available approximations and simpli-
fications. No dilemma if this kind is more vexing than that arising
from the observations of radiative emission.Kolomiets and co-workers
(31), and more recently Vezzoli and co-workers(32) have reported
observing radiative emission from the ON-state. In the former case
the wavelength peak corresponded to half the mobility gap, in the
latter to the entire gap. For the Kolomiets radiation to arise from
thermal causes, the source would have had to be at 1500
o
K, i.e. far
above the glass transition temperature. Moreover, in both cases, the
output was proportional to current, not to the power dissipated.
Similar observations have been made by Preudenziati and co-workers(33)
in the course of experiments on switching in boron. The emitting
volume is, of course, very small (e.g. 10-
12
cm3) in all these cases
and the optical signal correspondingly weak , but some such emission
is clearly expected on the basis of electronic models. In particular,
it is expected from the population inversion in the ON-state discuss-
ed above. On the face of it, this evidence is conclusive, and ad-
vocates of electronic models are entitled to find it in the ultimate
vindication of their view. However, we do not in fact know whether
it is intrinsically linked with switching process. More evidence is
needed, under a greater variety of conditions. If it were found
that radiation is always or generally associated with the ON-state,
one would have to conclude that an electronic desequilibrium is in-
volved. Further research of this kind constitues the best chance of
resolving the basic issue.
Thermal theory is at this stage more mathematically polished
and permits more detailed predictions. Among these are the prevail-
ing filament temperatures, corresponding to various ON-state currents.
They are high, e.g. over 500C, and even for quite reasonable operat-
ing currents far above the accepted value of the glass transition
temperature ~ 150C). This remains true, even though the tempera-
268 H.K. HENISCH AND C. POPESCU
ture estimates can be lowered by making additional assumptions, e.g.
by assuming that the device structure is non-uniform. Such a non-
uniformity can also explain the independence of VTH upon illumina-
tion and its exsistence is a priori likely, either because it is
built-in during initial preparation or else because it is the result
of a "first-switch" forming process. Thermal theory can thus deal
with a part of the problem, but the high filament temperature is in
conflict with the high degree of stability which threshold switches
of careful manufacture are known to exhibit. Additional assumptions
concerning possible mechanisms which inhibit phase transformation
and crystallization processes have to be introduced to maintain the
thermal position.
The detailed nature of the changes introduced during "forming"
and, indeed, the extent to which forming processes are active during
the first {ever) operation of different switching systems remain
uncertain. The most drastic manifestations(34) involve a decisive
"first switch" event, after which the system is permanently modified.
Gunthersdorfer(35) has shown that this modification involves (or,at
any rate, can involve) a certain amount of melting. In contrast,
other workers(36) have reported no drastic "first switch" effects
of this kind, only a gradual modification of the system over the
first few hundred switching operations, and sometimes no forming at
all(37). The matter deserves further investigation.
Figure 2e shows the recovery of VTH after a previous switching
event. For switches of I micron thickness, the recovery time (knee
of the curve) is about a microsecond. In terms of thermal theory,
the origin of the recovery curve is clear : it is a consequence of
cooling. It should, then, depend on the filament temperature attain-
ed during the previous switching event,but does not(38). Whatever
the after-effects of that event may be, they are not linked with the
current passed or the energy dissipated during that event. Thermal
theory, in its present form, leaves the matter open.
At low temperatures, slight polarity effects have been noted,
depending on the direction of the preceding switching event, but at
high temperatures (room-temperature and above) the after-effects
appear to be non-polar. Electronic theory ascribes them to residual
space charge, left over from the ON-state pattern of the previous
event, decaying slowly at low temperatures (and hence observable),
fast at high temperatures (and hence unobservable). Thermal theory
has at this stage no competing explanation. However, electronic
theory is not free from corresponding embarrassments. Thus, it is
hard to see how a field configuration corresponding to the ON-state
could establish itself in a time smaller than the carrier transit
time across the film, which is believed to be about 10-
7
seconds.
In fact, switching itself takes place within times which are orders
of magnitude shorter, e.g. 10-9 seconds or less. Neither theory
offers a conclusive explanation of switching with 1-10 nanosecond
THRESHOLD SWITCHING 269
pulses. Under such conditions, the switching voltage tends to
become constant, as indicated on Fig. 2a. This is within the expect-
ations of electronic theory, inasmuch as it implies a critical field
under conditions in which heating was believed to have been avoided.
However, as far as they go, the available results can also be in-
terpreted thermally. The constant voltage is then ascribed to the
field dependent conductivity. Switching voltage is therefore not a
unique and sufficient definition of the switching condition ; what
is needed is a measurement of total energy dissipated during the
short pulse.
The two theories have, of course, quite different interpreta-
tions of the minimum holding current. This point of the
istic has been investigated much less than the threshold point but
is of great significance. On the basis of electronic theory, the
minimum holding current should certainly increase with increasing
film thickness, because thicker films are associated with greater
transit times. Going in the opposite direction, the minimum holding
current should tend to zero as the film thickn.ess tends to zero,
and this is actually observed. Thus very thin films exhibits no
switching process, but a static V-I characteristic, very similar to
the transient charcteristic shown on Fig. 2d, and highly suggestive
of tunnelling. Thermal theory must call upon additional ad-hoc
assumptions in order to explain the transient ON-state (Fig. 2d),
not because it is a priori at fault but because it does not deal
with anything but bulk properties, governed by thermal inertia. Any
experiment which does not relate to such properties is necessarily
outside the compass of thermal theory. On the basis of electronic
theory the transient ON-characteristic is the direct consequence of
the contact barriers. A similar dichotomy arises in connection with
phenomena observed at the interfaces between amorphous and crystal-
line semiconductors(40,41), phenomena which strongly suggest some
form of electronic interaction.
It is clear enough that the resolution of these problems calls
for further experimentation, but the design of experiments is no
longer a simple matter. The obvious measurements have already been
done, and since they proved indecisive, n'ore sophisticated proce-
dures will have to follow. Among the most urgent needs are:
(a) further work on the light emission associated with
- switching;detailed spectral characteristics for various
contacts and switching materials; temperature depend-
ence, etc.;
further work on the nature of the ON-state close to
the minimum holding current, including dependence on
circuit conditions and on rate of voltage withdrawal;
further work on contact effects, in darkness and under
illumination;
further work on the nature of the high-field conduct-
ivity, over a wide range of pulsed fields and tempera-
270
(!)
(>
ACKNOWLEDGMENT
H.K. HENISCH AND C. POPESCU
tures, to determine its association (if any) with
the switching process;
further work on switching _short pulses of high
overvoltage to determine the dependence of the
switching response on energy input;
further work on switching characteristics as a
function of the material constants;
further work on the thickness dependence of all the
parameters involved (co-planar and sandwich structures)
We would like to express our grateful thanks to each other,
for many useful discussions and for forbearance in the
face of intolerable provocation.
REFERENCES
1. S.R. Ovshinsky,Phys. Rev. Letters 1450 (1968).
2. S.R. Ovshinsky, personal communication.
3. R.R. Shanks, J. Non-cryst. Solids 504 (1970).
4. R.W. Pryor and H.K. HeniscQ; J. Non-cryst. Solids 2-. 181
(1972).
5. S.H. Lee and H.K. Henisch, J. Non-cryst. Solids!l, 192 (1972).
6. C.B. Thomas, A.F. Fray and J. Bbsnell, Phil. Mag. 617
(1972) .
7. P.J. Walsh, et al., J. Non-cryst. Solids 107 (1970).
8. T. Stubb, T. Suntola and O.J.A. Tiainen, Solid State Elec-
tronics ..!2., 611 (I 972) .
9. H.K. Henisch, R.W. Pryor and G.J. Vendura, J. Non-cryst.
Solids 8-10, 415 (1972).
10. J. Marshall and A.E. Owen, phys. stat. solidi (a) 181
(1972).
11. P.J. Walsh, R. Vogel and E.J. Evans, Phys. Rev. 178, 1274
(I 969).
THRESHOLD SWITCHING 271
12. M. Telnic, L. Vescan, N. Croitoru and C. Popescu, phys. stat.
solidi (b) ~ 699 (1973).
13. H.K. Henisch, W. Smith and M. Wihl, Fifth Int. Conf. on Amor-
phous and Liquid Semiconductors, Garmisch (1973). Proceedings
in print.
14. D.M. Kroll and M.H. Cohen, J. Non-cryst. Solids 8-10, 544
(1972) .
15. C. Popescu, personal communication.
16. T. Kaplan and D. Adler, J. Non-cryst. Solids 8-10, 538 (1972).
17. D. Buckley and S.H. Holmberg, personal communication.
18. W.R. Smith and H.K. Henisch, phys. stat. solidi (a) 12, K81
(1973).
19. N.F. Mott, Phil. M a g . ~ 911 (1971).
20. H.K. Henisch, E.A. Fagen and S.R. Ovshinsky, J. Non-cryst.
Solids ~ 583 (1970).
21. H.K. Henisch, Scientific American 221,30 (1969).
22. S.H. Lee, Appl. Phys. Letters ll. 544 (1972).
23. M.H. Cohen, H. Fritzsche and S.R Ovshinsky, Phys. Rev. Letters
22, 1065 (1969).
24. R.W. Knepper and A.G. Jordan, Solid State Electronics ~ 45
(1972) .
25. W. Van Roosbroeck and H.C.Casey JL, Phys. Rev. B ~ 2154
(1972) .
26. W. Van Roosbroeck and H.C. Casey JL, Proc. Tenth Int. Conf.
on Semiconductors, Cambridge, Mass. (U.S. Atomic Energy
Commission), 832 (1970).
27. W. Van Roosbroeck, Phys. Rev. Letters ~ 1120 (1972).
28. H.J. Queisser, H.C. Casey JL, and W.Van Roosbroeck, Phys. Rev.
Letters 26, 551 (1971).
29. W. Van Roosbroeck , J. Non-cryst. Solids ~ 232 (1973).
30. C. Popescu and H.K. Henisch, Phys. Rev. !ll, 1563 (1975).
272 H.K. HENISCH AND C. POPESCU
31. B.T. K,olomiets, E.A. Lebedev, N.A. Rogachev and V. Kh. Shpunt,
Soviet 167 (1972).
32. G.C. Vezzoli, P.J. Walsh, P.J. Kisatsky and L.W. Doremus, in
preparation.
33. M. Prudenziati, G. Majni and Alberigi Quaranta, Solid State
Communications 11, 1927 (1973).
34. L.A. Coward, J. Non-cryst. Solids 107 (1971).
35. M. Gunthersdorfer, J. Appl. Phys. 42,2566 (1971).
36. J.R. Bosnell and C.B. Thomas, Solid State Electronics
1261 (I 972) .
37. R. Pryor, Ph.D. Thesis (Physics), The Pennsylvania State
University (December 1972).
38. H.K. Henisch and R.W. Pryor, Solid State Electronics li, 765
(1971).
39. S.R. Ovshinsky, personal communication.
40. G.J. Vendura and H.K. Henisch, J. Non-cryst. Solids l!, 105
(I 972) .
41. K.E. Petersen, D. Adler and M.P. Shaw, in preparation.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
R. W. Hoffman
Professor Department of Physics
Case Western Reserve University - Cleveland, Ohio 44106
I.. INTRODUCTION
The importance of mechanical properties of thin films has
been established over the past several decades. In the first place,
many properties are substantially modified in a condensed f1m. The
literature contains many examples (I) illustrating shifts in band-
gap in semiconductors, transition temperature for super-conducting
films, or expected magnetic anisotropy. Any property which itself
is strain sensitive, may well be modified in a deposited film.
Even a property which is originally isotropic may have lower symr
metry when the effects of the strain are considered. The control
of the stresses by changing deposition conditions is one of the
goals now partially reached in the search to develop new properties
More recently, mechanical constraints introduced by the sub-
strates have been used to produce or stabilize new film structures
that are unknown in the bu1k.(2). In this way properties or devices
may be constructed that fit special needs and are unvai1ab1e by
conventional techniques. This design , through the mechanical cons-
traints in the system, is one of the more excitipg new areas.
One of the earliest evidences of the importance of the mecha-
nical properties in the various modes of failure. These failures
may be the more obvious ones of actual fracture of the film or
buckling as a result of the loss of adhesion to the substrate.
Dislocations or cracks may be introduced in the substrate at high
stress levels. Somewhat more subtile effects such as strain-enhan-
ced diffusion or e1ectromigration may also occur.(3)
273
274 R.W. HOFFMAN
A complete review would define the stress and strain system,
consider the techniques for measurement of the internal stress, as
well as present the data and consider the various models for the
origin of the stress. The tensile properties should also be trea-
ted, including the elastic modulus, the tensile strength, plastic
deformation, and the fracture mode. Indeed, the earlier general
reviews by Hoffman(4,5), Caffibell(6,7), Buckel(8), Kinosita(9),
Scheuerman(IO), and books(l I), have proceeded along these lines.
We will not attempt to reproduce that which already exists in these
review papers. We shall update them, however, and call attention
to the more specialized reviews, especially in those areas where
the information is pertinent to non-metallic films.
We also consider our task to discuss the stress and strain
distributions found in a film so that a given property may, in turn,
be calculated. New experimental techniques and data will, of course,
be included.
Although at this time we begin to understand, at least in prin-
ciple, the mechanical properties of metallic films, our state of
knowledge for non-metallic films is much more rudimentary. In the
first place, the properties seem to be not very reproduceable from
laboratory to laboratory and, indeed, seem to depend sensitively
on the conditions of deposition. In addition, instabilities or pro-
nounced changes in the internal stress may be seen when compound
films are exposed to the atmosphere or other surroundings after
removal from the system. All of these make for a much more diffi-
cult p r o l ~ m in understanding the origin of the internal stress be-
cause the structure of non-metallic films is not so well characte-
rized. We shall see that stoichiometry and rapid diffusion of am-
bient gasses are partially responible. In spite of these difficul-
ties, a qualitative understanding is emmerging.
2. THE ELASTIC PROBLEM
Formulation of the Elastic problem
Since, as we will see, the mechanical properties of a film on
a substrate are primarly a result of constraints in introduced by
the substrate, we wish to understand both qualitatively and quanti-
tatively the nature in which the forces are distributed in the film
and substrates and then transmitted across their interface. Even
though many problems of practical importance can be solved in terms
of isotropic elasticity, we wish to set up the problem generally
enough that we can consider cases of anisotropic strains. We must
also take account of single crystal films and substrates in order
to include epitaxial cases.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 275
Only strains may be experimentally measured. Hence the stres-
ses we talk about are calculated on the basis of measuring defor-
mation, knowing the boundary conditions and assuming that the sam-
ple is in equilibrium. Following Nye(12) and Smith(f3) we use a
mutually orthogonal coordinate system with the xl and x ~ axes in
the plane of the film and the x ~ axis normal to the film plane.
The coordinate system is defined in Fig. I. The first sub-
script refers to the direction of the force on a face perpendicu-
lar to the second index. Thus, oIl, O ~ are the normal stresses
in the p-lane and a ~ 3 is the normal stress perpendicular to the
film plane. The shear stresses o12, a ~ 3 and together with rota-
tional equilibrium, represent force systems attempting to decrease
the angle between the axes indicated. The corresponding normal
strains are sII S ~ and S ~ 3 and shear stress S12, S23, S31'
We point out that the tensor shear strains (s's) are half the en-
gineering shear strains (y's) since the y's represent the decrease
in angle between two original orthogonal directions. The primes
are used since the orthogonal primed axes are sample axes and are
not the crystal (unprimed) axes.
-X2.
Fig. 1 Definition of stress system
276 R.W. HOFFMAN
and
Since elasticity is a fourth-rank tensor property,
E: ! .
1J
, a'
Sijke ke
(I)
where the s's are the elastic compliances and the c's the stiff-
nesses for the particular sample orientation. Fourth rank tensor
properties can be quite complex in the case of low symmetry crys-
tals. The 81 possible components may be reduced by well-known equi-
librium and symmetry arguments.
Even in the cubic system three constants are independent and
two are needed for an isotropic specimen. The elastic equations are
often reduced to the two index notation(12,13)
E: !
Skj
a!
1
J
(2)
a! c! . E:'
1
1J
J
where the subscripts I, 2 and 3 stand for normal components and
4, 5 and 6 refer to the shear components for both stresses and
strains. The reader is reminded that the arrays s .. and c .. are not
second rank tensors in this notation. 1J 1J
For the (001) orientation in the cubic system, we may drop
the primes and find the independent constants Sll, S12 and S44 or
Cll, C12 and C44 listed in the handbooks. For a general orientation
the prfmed constant must be calculated by transforming from the
crystallographic axes where the elastic constants are tabulated to
the primed sample axes. Nye, among others, gives this prescription.
Young's modulus defined as the ratio of the longitudinal stress
to the longitudinal strain, i.e., I/stl' is often desired. For
the cubic system(13).
(3)
where 123 are the direction cosines of the arbitrary direction
t. referred to the crystallographic directions. As the quantity
1
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 277
(S11-S12. -1/2s .... ) is positive for all cubic metals except:mobybde-
num, Young's modulus has a maximum in <Ill> and a minimum in <100>
directions. Yotmg's modulus is independent of direction in the plane
for (111) orientation. Nye lists expression for Young's modulus for
crystals of lower symmetry.
Vook and Witt(14) give expressions for Young's modulus in se-
veral simple directions. Turley and Sines(15), following Thomas(16)
separate the given stiffness in a constant and orientation-depen-
dent part and then suggest several rotation methods to calculate
the transformation. Polar plots are given for several shear cons-
tants for Cu. Mo, and Si.
For convenience in our discussion and following equations, we
will drop the primes from the notation. In order to obtain a better
feeling for the problem, let us consider the simplifications of an
isotropic elastic substrate. Firlmy attached to this substrate is
a film which is also elastically isotropic. We shall regard the di-
mensions in the plane as being semi-infinite. If we now consider a
homogeneous stress in the film, and invoke the usual boundary con-
ditions of no forces at the free surfaces or edges, we have the si-
tuation of Fig. 2, where we depict the forces acting on an interior
section of the film. As we shall see later this corresponds in re-
ality to the situation found in most systems except near the edges
of the film itself, or perhaps at the boundaries of crystallites
within the films. In this figure we see that the film is in a
state of tensile stress indicated by the total force per unit width
of the film F.
We note that this tension is a biaxial one. As one traverses
the interface in a direction normal tc the plane of the film, there
is a discontinuity in the stress but the strain is continuous.
If the film is in tension we find that the substrate is in CORr
pression just below the interface. In fact, if, the film is quite
thin compared to the substrate, the neutral plane is a third of
the substrate thickness from its free surface.
Thus our oversimplified picture of a film under the uniform
tensile strain would give us the situation indicated in Fig. 3.
An interior volume element would see the normal stresses, ~ 1 and
022 but no shear stresses. If we imaginate cutting through the film
in the center we see a uniform tensile stress, assuming the subs-
trate were rigid. If, on the other hand, the origin of stress were
concentrated at the interface, we would see decreasing tensile for-
ces as indicated schematically in the figure when we moved away
from the interface. The free edges of the film can support no for-
ces. Hence, as seen in the right hand side of the figure, we would
expect shears to exist only near the edges of the film.
278
R.W. HOFFMAN
FILM
SUBSTRATE
Fig. 2 Forces acting on an interior section of the film-substra-
te composi te
Fig. 3
'---
,
.--
'-


,
,-
t-
I-
'-
1-
Idealized stress distribution within a constrained film
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
SUBSTRATE
T
,))), I
------;---t-
fiLM Ea ,ll.
t
1
STAt'SS ,0-
Fig. 4
Stress distribution for longitudinal applied force.
----

- \
\
R. RADlliS OF
CURVATIJR

I I R
STRESS,""
Fig. 5
Stress distribution resulting from bending.
279
280 R.W. HOFFMAN
Stress Distribution
Let us consider some of these points in more detail. There are
two problems that need to be attacked. One is the stress distribu-
tion within the interior of the substrate and film, and the second,
the effects near the edges. We will consider the first of these. As
pointed out by Brenner and Senderoff,(17) and treated in the review
by Hoffman, using the case where the elastic constants of film end
substrates are the same, and later extended by Doljack,(18) the pro-
blem is one of longitudinal forces in the film superposed on ben-
ding. It is instructive to look at each of these separately.
If we apply an extenlal force to the composite plate longitu-
dinally, so that all of the fibers strain equally, then the stresses
across the cross section will be those indicated in Fig. 4 for the
case of a biaxial stress system. A discontinuity in the stress exists
at the interface but the stress in each of the substrate and the
film is uniform.
In the case of pure bending, when only an external moment is
applied to each of the elemental pieces and no external force is
applied, the elastic response is only a bending strain. This pro-
duces the stress distribution as shown in Fig. 5. It is seen that
the bending gives rise to non-uniform stress in the direction per-
pendicular to the film, arising from the fact that the stretching
in the longitudinal fibers depends on the distance from the plane
of zero bending strain. If the plate were not a composite one this
plane of zero bending stress would just be in the middle.
We now relate these longitudinal and bending relaxations to
the actual case of depositing a film on the substrate. Consider
first a substrate clamped against bending and contraction during
the deposition of the film. In this case the stress distribution
in the film after deposition is the intrinsic stress. This dis-
tribution is indicated schematically ih Fig. 6, and of course, the
distribution of the stress need not be constant throughout the thi-
ckness of the film. The total force exerted by the film is be i n t ~
gral of the stress over the thickness of the film ; and equal to
the product of the average stress and the film thickness.
F St
I
t
cr(t) dt (4)
o
where (t) represents the stress distribution as a function of the
thickness t.
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS
Fig. 6
SUBSTRATE
________________________ +-__
cr(t.)
FILM

..l
________ :'\
Intrinsic stress distribution for deposition with
substrate clamped against bending and contraction.
281
If we now release the clamping preventing contraction, but
still do not allow anay bending to take place, the stress distri-
bution throughout the plate is modified as in Fig. 7. The uniform
stress distributions 01 and 02 are produced by the net contraction
of the tensile stress distribution, o(t) in the film. The relaxed
stress distribution in the film o'{t) is just the original distri-
bution reduced by constant stress 02. Equilibrium is achieved when
the total force on the cross section depicted in Fig 7, is zero.
Fig. 7
SU8STMTE
FILM
qo'(t) = Ci(-t) - 0"'.
Relaxed stress distribution upon releasing longitudinal
clamps.
282 R.W. HOFFMAN
If the remaining clamps are now removed and the plate is al-
lowed to bend, it reaches an equilibrium radius of curvature. This
bending of the film substrate plate gives rise to a relaxation of
the stress as was indicated in our discussion of pure bending. In
essence, the additional contraction which results from the neces-
sity to achieve bending equilibrium decreases the actual stress
in the film ~ v n more. And, furthermore, makes it non-uniform as
a function of thickness. The radius of curvature may be determined
by equating the moment of the film to the moment produced by the
stresses resulting from the bending. The plane of zero elastic
strain, or neutral plane, is now shifted, and may be determined
from the condition that there is no net external force exerted a-
cross any cross-seetion and thus the sum of the bending forces must
be zero.
Equations for the radius of curvature and the relieved stress
distribution are given by Brenner and Senderoff(17) for the case of
identical eltstic constants of the film and substrate and Doljack(18)
and Klokholm 19) for the general case. The 3tress distribution will
be modified by a bending distribution as depicted in Fig. 4. If aCt)
is constant, Klokholm calculates the relaxation correction M by
which the simple Stoney(20) average stress S should be multiplied
to obtain the true average stress.
M (5)
where S is the substrate to film thickness ratio hit and 11 is the
ratio of elastic constants Ef/l-Vf and Es/l-vs. E is Young's modu-
lus and V Poisson's ratio. The subscripts f and s refer to film and
substrate. For $> 1000 and any 11 or S> 10 and 11<0.2 N is suffi
ciently close 1 to be experimentally negligible.
In practice it is difficult to clamp a substrate sufficiently
well to meet both the conditions of no bending and no expansion
during deposition. A much more common case is that with the substrate
completely free during deposition. During deposition an initially
flat plate begins to bend slightly relieving the stress in the film
already deposited. The continuous bending proceeds, giving rise to
a stress relief in the film in addition to the contraction and ben-
ding just discussed. Under these conditions the relief in each layer
is different when the deposition is completed. If the mechanism gi-
ving rise to the stress in the first place were originally a process
uniform with thickness then the result of such a stress relief would
clearly give rise to larger stresses near the outside grO'i!ing edge
of the film as indicated in Fig. 8.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 283
SUBSTRATE
FILM
Fig. 8 Relaxed stress distribution for free cantilevered
substrate during deposition. After Brenner and Senderhoff(17).
Brenner and Senderhoff, in fact, suggested that this would be
an explanation for the fact that many films curl when detached
from the substrate in a way to suggest that the outer layer has
the higher stress. Fortunately, these stress relief terms are not
large in practical cases since the film is generally much thinner
that the substrate. On the other hand, we must keep in mind that
they will give rise to stress gradients within the film. The stress
distributions have been calculated by the authors previously refe-
renced, and the errors are less than the previous case for the
same geometry.
Chaudhari(21) gives an approximate treatment of
following the thermal stress analysis of Timoshenko.
distribution in the film may be wtritten as :
the relaxation
The stress
(6)
For this treatment Ell is the assumed uniform strain r ~ s ~ n g from
thermal contraction or other mechanism, E and V are the usual elas-
tic constants assumed the same for film and substrate, C = t/2,
the film half-thickness, and D = h+t/2, the substrate plus film
half thickness. X3, the coordinate normal to the film, is 0 on the
free surface of the film.
The first term in the equation represents the stress as one
would calculate it considering a rigid substrate. The second term
284 R.W. HOFFMAN
represents the contraction or compressional relaxation while the
third term represents the bending relaxation. The stress distribu-
tion in the substrate is given by
(7)
The results are presented in Fig. 9 for several values of Gin.
As the stress in the substrate is reduced by the factor Gin, it is
obvious that in most cases of deposition the film is much, much
thinner than the substrate so that the stress in the substrate is
only a very small fraction of the stress in the film. If we consi-
der a film thickness of I ~ and a substrate thickness of 1 rom,
than the maximum value of the stress in the substrate is less than
1/2 % of the stress in the film.
'.0
0 ..
0.&
w=
0 .
~
t./h = 0.1
.,.
!.
0.2
l?
0
-0.2
-0.+
0 0.2
Fig. 9 Approximate stress distribution for several values of film-
substrate thickness ratio. After Ghaudhari(21).
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
285
As Chaudhari has pointed out the critical feature of whether
plastic flow will take place in the substrate is the value of cri-
tical resolved shear stress. If the stress in the film is 1% of the
shear modulus and we consider the case where the ratio of the film
thickness to the substrate thickne$s is 10-4, the maximum value of
stress in the substrate would be

For comparison,most bulk

single crystals will begin to undergo a plastic deformation when
the shear stress is of the order of 10-5 or 10-
4
times the shear
Thus, when the film thickness is very thin, compared to
the substrate thickness, we can neglect plastic flow by a disloca-
tion mechanism in the substrate. When the ratio of the film thick-
ness to the substrate thickness is larger than 10-
4
we have to con-
sider the particular substrate in question to determine whether or
not plastic flow may take place.
There are two cases of practical importance where one has to
be concerned about plastic deformation in the substrate. One of
these, of course, is the obvious one where the material in the sub-
strate is such that it easily undergoes plastic deformation, alkali
halides for instance. Stress concentrations at propagating crack
tips may also introduce dislocations in single crystal substrates.
Dislocations and other local defects may also be introduced in the
substrate during growth. The second case takes place when the sub-
strate is comparable to the film in thickness. As one can see by
examination of Fig. 9, the relaxation in the film and substrate is
sufficient to cause a non-uniform stress distribution within the
film as well.
Similar distributions have been calculated by Oel and Frechett
(22) for both large and small area planar interfaaes for the case
of comparable film and substrate thickness. Model experiments with
thermal strains induced in birefiringent glasses gave good agree-
ment with the calculations except when diffusion took place during
sealing at high temperatures. In this case, from merely smooth-
ing the stress gradient, diffusion-caused stresses had large effects
in magnitude and even sign of the stresses.
No detailed treatment concerning the sharpness of the inter-
face seems to exist. The large stress gradient in this region may
play a role in interfacial dislocation generation or mass flow and
other effects, and represents an area of needed future study.
Edge Effects
We now consider the se"cond part of the question, namely, the
spatial details of the stress distribution. Of course, if the me-
chanism giving rise to the stress in the first place is not uni-
form, then the stress distribution will be non-uniform as well.
Such distributions will be discussed later under origins of the
286
0'".
I
Fig. 10.
R.W. HOFFMAN

I

I
OJ, = 0'" i
033
=0
0'".
I
0'"31
= 0
+
I I
Boundary traction method applied to a thin film under
uniform intrinsic stress constrained to a rigid sub-
strate. After Aleck
23
modified by Doljack
l8
.
intrinsic stress. However, we shall consider here the case where we
initially have a large stress 011 = 022 that lies in the plane
of the film is isotropic, and nearly uniform throughout the thick-
ness, and is uniform over most of the area of the film. If we look
at the cross section of the film-substrate composite, we see that
the film appears to be a rectangular plate attached along its long
edge to what we shall assume to be a semi-infinite rigid substrate.
As a result, a slab of thin film attached to the substrate is under
a state of stress nearly ideNtical to that of a rectangular plate
clamped along an edge and having undergone thermal expansion.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 287
This latter problem has been treated approximately by Aleck(?3)
We shall consider the question of thermal expansion later, but at
the moment, we shall use the results of this treatment to ask the
question as to the stress distribution near the edges of the film.
According to Aleck,the problem is first converted to one of boun-
dary tractions. The film is detached from the substrate and allowed
to relax. To bring it back to a uniform state of stress, one
imagines applying a uniform stress of magnitude Oi normal to the
ends of the film as shown in the top of Fig. 10. Next, the film is
attached to the substrate and a uniform compressive stress or is
applied to its ends. or In this case is equal to 0i because at the
edge if the film there are no net applied normal boundaty tractions.
The solution to the problem of the specified surface traction thus
replaces the original stress problem. Doljack has solved this
10
8
6
b-
/035 X
4
(/)
2 (cr.,
(/)
w
a:::
l-
(/)
o 1.0 t
DISTANCE ALONG INTERFACE
Fig. II Interfacial stresses near the free edge of a semi-infi-
nite film. After Doljack.(I8)
288 R.W. HOFFMAN
problem based on the approximate solution of Aleck for the case of
thickness of the film being much thinner than its length. Fig. 11
shows the results of these calculations. The pertinent pictures are
these: all is just equal to the intrinsic stress until one gets
within a distance of the order of 10 thicknesses to the edge. A large
shear exists very close to the edge. In addition, a concentrated
stress normal to the interface lies at the attached corner. We point
out that there exist no non-zero stresses or stresses normal to the
interface once one gets to a distance several times the thickness
away from the edge. The presence of an extremely large normal force
near the edge points out its relationship to failure at scratches
or film edges. We will return to this phenomenon later.
Haruta and Spencer(24) have observed the spatial stress distri-
bution by X-ray topography which is sensitive to the strain gra-
dients, and qualitatively confirmed the rapid change near the film
edge. A substantial decrease in strain was found after annealing.
To summarize then, we conclude by detailed calculations that
as long as the substrate remains essentially rigid, and as long as
we are not concerned with the stress distribution very close to the
edges of the film, our simplified point of view which says only the
components all and 022 are presents, is indeed the correct one.
3. TECHNIQUES FOR STRESS MEASUREMENTS
Introduction
Although the literature primarly quotes values for the stres-
ses in films, it is actually strains that are measured. Two direct
ways exist. First of all, a measure of the deformation of the sub-
strate upon which the film has been deposited, or secondly, by dif-
fraction techniques, a measurement of the elastic strain within the
film, or in some cases, substrate. These two ways need not yield pre-
cisely the same values because of the distribution of forces across
grain boundaries and whithin the grains of the fine grain deposit
affect each technique differently.
A summary of the various methods used prior 1970 for measuring
the deformations has been well documented in the reviews by
Scheuerman(IO) Cambell(7), and earlier by Hoffman(4,S,2S,2o). We
see no point for reproducing that information here. Many different
techniques and their sensitivities were compared. In the last seve-
ral years progress in measurement technique has been in several
areas. First, the production of automated systems for measuring the
stress on the r.outine basis on production samples(26). The search
for a technique of measurement of the stress on a localized scale
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 289
on real structures has resulted in increased use of x-ray techni-
ques. Holographic methods are not used commonly, perhaps because of
some of the stringent vibrational requirements associated with that
technique.
There has been recent progress in the refinement of the cal-
culations used to relate the observed deflections of the stresses
in the films. As the substrate temperature during deposition is
perhaps the single most important parameter affecting the stress,
its control or the correction for the various temperature factors
in the measurement has also been of extreme importance, and progress
has been made along these lines. We consider each of these areas in
more detail.
Cantilever Plate Method
Perhaps the most common method for measuring the stress is the
cantilever beam technique. Many have contributed to this type of
measurement ; it exists in a wide variety of detection schemes.
Because good data is needed for further progress ih understanding
the origin of the internal stresses, and also because there is still
a lack of approciation for some of the subtleties in the experiment,
we describe the cantilever method in detail ; the information may
be extrapolated to other geometries.
In the cantilever method the deflection of the freee end of
a thin plate is measured, or alternatively a force applied to the
free end to restore the beam position to some fixed point. The
other end of the beam is imagined to be clamped rigidly. This tech-
nique is often used because the deflection can be monitored contin-
uously during deposition of the film and thus obtain information
about the stress distribtuion. If the stress in the film is tension,
as is commonly the case, then the film applies both a compression
to the substrate and a bending moment. The stress is called tensile
if the surface traction compresses the plate. This definition comes
about because the plate applies a tensile stress to the thin film
to prevent it from elastically relaxing. The resultant bending will
be such that the film in tension finds itself on the concave side.
The commonly used equation relating the deflection at the end of
this cantilever beam to the force/unit width of the film is :
(8)
where t is the length of the cantilever leam, Es Young's modulus
and Vs Poisson's ratio for the substrate, and 0 the deflection of
the free end. It should be pointed out, although this is the pre-
sently accepted cantilever beam relationship, occasionaly the I-v
s
290 R.W. HOFFMAN
term, arising from the biaxial film stress, seems to be still ne-
glected.
The concern, then, is how well this relationship actually de-
scribes the free end of the beam. The problem is one of properly
describing the curvature in the transverse direction of the beam
when it undergoes a large longitudinal nurvature. This problem has
been in the literature for a long time having been discussed as
early as 1632 by Gallileo. St Venant in 1864 presented the first
formulation. The more recent treatments by Ashwell and
Greenwood
t27
) and later by Bellow, et al.,(28) have given experi-
mental justification for the treatment put forth by Lamb in 1891.
The results may be summarized as follows : consider a plate
of width b, length i, and thickness h, bent to a longitudinal cur-
vature I/R by uniform moments applied to the ends. If the deformed
shape is to be truly cylindrical, that is, the transverse strips
are to remain straight, then moments equal to the moment in the
longi tudinal direction times Poisson's ratio are required along the
long edges. Since such moment are not present the plate tends to
assume a curvature viR in the transverse direction. This case is
known as "anticlastic bending", and is illustrated in Fig. 12. The
transverse distorsion results in longitudinal membrane forces. As
the curvature becomes larger the effect to these forces must con-
sidered. It results in a stretching in the neutral plane and the
Poisson effect eventually becomes cancelled leading to the case of
." ..1--.""- __ _
;' --
;'
;'
Fig. 12 Anticlastic bending of a flat plate From Bellow. (28)
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 291
cylindrical bending. The pertinent parameter in this problem is the
ratio b
2
/Rh. For values of this ratio from 0 to 1.6, the anticlas-
tic case occurs. If the ratio is larger than 1000, then the surfa-
ce can be considered cylindrical. (27) Under cylindrical conditions,
the deflection of the end of the plate becomes smaller by the fact
tor of 1/(I-V
2
). As the transition from anticlastic to synclastic
bending is a gradual one, the appropriate relationship must be
known for the beam of our particular experiment. At present numeri-
cal calculations have compared favorably with experimental results
for ratios up to 50.(28) By connecting the radius of curvature to
the dimensions of the beam approximately through the rigidity rela-
tionship, one can write the ratio b
2
/Rh in the form 2(b/t)2o/h
It is seen that for a beam whose length to width ratio is a reason-
able 10, one may apply the anticlastic bending formula to the situ-
ation where the deflection is about 100 times the thickness of the
beam itself. It appears that for practical purposes, unless one is
dealing with a beam of rather square geometry, that the anticlas-
tic formula will meet most practical pusposes, and equation (I) or
its equivalent for other geometry will suffice.
For the geometry of the cantilever beam the boundary conditions
require that three edges be free while the fourth is clamped. Thus
the deflection and slope of the plate as is emerges from the clamp
are rigdly fixed. In practice, this seeems to be a much more diffi-
cult condition to meet experimentally than is usually realized. The
problem is often encountered when one uses a concentrated load at
the end in order to determine Yong's modulus for the substrate.
Differences of as high as 20% are encountered between the apparent
modulus values for the same beam measured by a cantilevered loading
a a-free beam loading. Rottmayer and Hoffman(29) pointed out that
it is extremely important that the beam be gripped in the clamps
such that no sliding would take place. Indeed, a rigid glueing is
sometimes necessary. Resonance techniques are sometimes used to ob-
tain values for E.(7,9).
Springer(30) has calculated the cantilever experimental arran-
gement on the basis of the linear bending theory of plates. The
basic assumptions are that E33 = 0, that is, there is no strain in
the direction perpendicular to the plane. Secondly, plane sections
remain plane upon bending which implies that no serious non-uniform
stretching occurs and, thirdly, normal sections remain normal. This
means that the transverse shear deformations are small. With these
assumptions one proceeds as follows. The function for the shape of
the plate is assumed, the stress distribution is given, and the to-
tal work done in forming the plate to a given shape is calculated
and set equal to the strain energy stored in the plate for this
configuration. Numerical calculations were necessary, and the re-
sult shows that the deflection calculated by the usual anticlastic
relationship was 7% larger than the numerical calculations. This
292 R.W. HOFFMAN
difference is probably due to the fact that the boundary conditions
at the clamp is such that the beam may not have any cross curvature
and the transition from this behaviour to the anticlastic situation
probably does not become apparent until about one beam width away
from the clamp. The net result is to reduce the force necessary to
bend the beam at the clamp and thus cause a greater deflection to
appear at the end.
In practice it is probably not worthwhile to go through the
effort of the numerical calculations and one should use the anti-
clastic relationship. It should be pointed out, however, that in
careful experiments the error of about 7% in the bending equation
may be comparable to other experimental errors. In order to compare
numerical values for stress obtained amongst various laboratories
one should make certain that the elastic equations used are describ-
ed in the publication.
Detailed calculations of the deflection for another geometry
have also been carried out. A free circular plate was first consi-
dered by Finegan and Hoffman(31) for the case of determining whe-
ther the stresses were anisotropic. As will become more obvious in
the next section, temperature control demands that any plate be at-
tached to a thermal sink. Doljack and Hoffman(32) have solved the
deflection equation for a circular .plate clamped along an inner
circumference. One solves the problem by superposition of the de-
flection of the free plate and the deflection corresponding to the
forces applied by the central clamp, under the condition that the
intrinsic stress may be anisotropic. The plate itself wants to bend
to the familiar elliptic paraboloid. The clamp at the inner radius
applies those moments and shears to the plate that are necessary to
change the deflection and give zero slope at the inner radius. The
solutions can get rather complex because one must satisfy a fourth
order differential equation, and enforcing the boundary conditions
results in solving two sets of four simultaneous equations for four
unknowns in each set. The resulting solutions are shown in Fig. 13,
wher it is seen that the centrally clamped circular substrate de-
forms in a manner very simi.lar to the free circular substrate, with
the zero deflection point moved out to the clamp.
It is also worthwhile noting the solution to the problem where
the plate is clamped around its outer circumference. It is easy to
show that under these conditions the clamp applies opposite bending
moments that precisely cancel those applied by the surface traction.
Thus, no deflection takes place. If a concentric hole is cut in the
center of the plate, the plate will deflect and the solution will
be the same as the problem previously solved with the roles of in-
ner and outer radius interchanged. As one would expect, the over-
all bending produced by the same surface traction is less than ob-
tained with the central clamp configuration.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 293
......
~
i
~
2.0
>-
ex:
<
~
iii
JIll:
'<
.....
:2!
0
i=
~
1.0
b..
!.oJ
0
o
RADIUS I ..
Fig. 13 Anisotropic bending of a f r ~ e ~ n centrally clamped cir-
cular substrate. Doljack and Hoffman. \32)
The last configuration that is worth mentioning is the case of
a free circular plate with a thin film deposited over the central
circular area of the plate. The solution to this problem shows that
as the outer diameter of the plate becomes very large, the bending
of the plate in the region of the film does not decrease signifi-
cantly. although in practical problems the films deposited are sel-
dom circular, the extrapolation of this statement is important for
the case of many discreet devices on a single substrate.
The second area where a marked improvement has come about in
the last few years is in the temperature stabilization of the sub-
strate during deposition. Since the early work of Campbell and his
co-workers in measuring the stresses in the early stages of depo-
sition, it has been realized that both thermal and momentum effects
294 R.W. HOFFMAN
are important. More recently, Kinosita(9) and his coworkers have
shown how to make corrections to the deflections obtained during
deposition. In their case for depositions of low stress metals on
extremely thin mica substrates, the corrections were substantially
larger than the observed deflections and in some cases of the oppo-
site sign.
To illustrate the thermal contributions, let us take the can-
tilever beam of length ~ width b, and thickness h, and condense
the film of density p or mass per atom m, at a rate R.
The momentum effect can be calculated by knowing the tempera-
ture Tm of the source, because that determines the velocity of the
Maxwellian distribution of atoms as they leave. Assuming the arri-
ving metal atoms do not rebound from the substrate, on can integrate
the effect along the length of the beam and find the resultant
moment. The result for the implied force and deflection from momen-
tum consideration is equal to:(34)
~ 4 p [kTm )1/2
2.38 E
s
h 3 m J R
F
From the point of view of the experiment in which the deflection is
measured as a function of the thickness of the film, this contribu-
tion would be a constant compression from the time the film starts
to deposit until deposition is completed, assuming the rate of depo-
sition is held constant, and the source temperature does not change.
Before we can say anything about the thermal effects, we have
to establish the temperature history of the substrate. Maki and
Kinosita(35), Yoda(36,37) and Namba(38) have measured the tempera-
ture rise of 12 ~ m thick mica substrates during the deposition of
silver films, by measuring the temperature with an iron-nickel ther-
mocouple on the back side of the substrate. As indicated in Fig. 14,
the shape of the curve is sensitively dependent upon the rate of
deposition. Indeed, in films such as silver, the increase in the
reflectivity of the thermal radiation may result in a decrease in
the film temperature as the film becomes thicker. Note that the
temperature rise may be quite large, perhaps 50 and saturate at
thicknesses corresponding to the order of 1000 X. Similar experi-
ments in our laboratory with extremely fine wire thermocouples bon-
ded to the back side, show an increase in the temperature of the
free end which is almost linear in the deposition thickness for de-
position rates of the order of 25 X per sec.
Normally, the cantilever beam is clamped at one end with a
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
295
90
2.2 A/SK
o
500 \000
,500
FILM THICKNESS, t (A)
Fig. 14 Temperature history during deposition of Ag on very thin,
poorly conducting substrate. After Maki and Kinosita.(35)
large thermal mass acting as an heat sink. At room temperature we
assume the incident power must be conducted away from the substra-
te which produces the temperature gradient along the length. Assu-
ming radiation negligible one may integrate the heat flow equations
resulting in the temperature difference along the length
(10)
where P is the power per unit area incident on the substrate and K
the thermal conductivity of the substrate. P includes the condensa-
tion energy as well as the energy radiated from the source and sur-
roundings. For 'hot' sources the condensation energy is usually ne-
gligible, but the kinetic energy may be considerable in the case of
high bias sputtering. For the evaporation geometry used by Sprin-
ger,(30) P is 20 milliwatts per cm
2
resulting in a calculated tem-
perature rise of 33C. This estimate is perhaps 30% high compared
to experimental evidence presumably resulting from the fact that
steady state conditions are not reached and radiation losses are
neglected.
296 R.W. HOFFMAN
Elastically, there can be several effects resulting from ther-
mal expansion. First, the effect of differential thermal expansion
between the film and substrate if the temperature of the specimen
is uniform, but changed from T
so
, the temperature of the substrate
at the begining of film deposition. Secondly, in a cantilever expe-
riment, the temperature gradient along the length of the beam given
rise to a non-uniform bending, and thirdly, there is a temperature
gradient normal to the film plane. All of these conctribute, but
the first two are interrelated.
Rottmayer,(39) Kinosita and Doljack have treated the first
effect, under the model that the intrinsic stress is temperature
independent and no stress relief takes place as the temperature is
changed upon further deposition. A given layer deposits thermal
stress-free, so there is a thermal stress gradient in the film from
maximum at the interface to zero at the free film surface. The ther-
mal s t r ~ n may be calculated from the expansion coefficients af and
as of the film and substrate. The result is, for a rigid substrate,
equivalent stress or end deflection
Ef
SI = -1-- (af-a
s
) (Ts-<Tg
-Vf
3 2Ef (1 -Vs)
Es (1 Vf)
(11)
where Ts is the final substrate temperature, assumed equal to the
film temperature and <Ts> is the thickness averaged substrate tem-
perature.
If the substrate temperature is linearly related to the thick-
ness, then this net differential thermal contribution vanishes when
cooling to the average substrate temperature CTso+Ts)/2 after the
deposition is completed.
We consider now the case of a thermal gradient along the length
of the plate. Elastically, this problem is not the same as the dif-
ferential thermal expansion contribution that arises by changing
the temperature of the substrate and the film by an amount liT 1 The
reason for this is, of course, that the temperature gradient is not
constant, giving rise to a non-uniform bending even though the ave-
rage strain is still the same. The solution to this bending problem
has been approximated by Alexander(34) giving rise to the smaller
stress
(12)
if the temperature difference i'lT2 between the free end of the can-
tilever beam and the clamped end is established before the film is
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 297
TABLE I
Comparison of Extraneous Effects in Cantilever Technique*
Effect Temperature Deflection Force/ Percent
Change c cm Unitwidth of 1000 A
dynes/cm Film
1000 A Film 2.5xlO-
2
10
5
Substrate gravi-
tational 4 x 10-2
I. 6
x
I 0
5
160
Film gravitational
4 x
10-5
I . 6x I 0
2
0.2
Momentum R=20 A/sec 5 x 10-5
2 xI02
0.2
Uniform temperature
3.7XIO-
4 1.5X ,J03
increase, Case I 10 1.5
Gradient along
length - Case 2 30 1.8XIO-
4
7.5x10
2
0.7
Gradient through
10-
6
thickness, Case 3 5 10-4 5 x ' 20 0.02
*
Calculated for a representative metallic substrate, 0.01 cm thick
by 5 cm long and af-a.
s
= 5XIO-
6
C-I
Stress in 1000 A reference film is assumed to be 10
10
dynes/cm
2
298 R.W. HOFFMAN
deposited. We see that this effect, too, may be eliminated entire-
ly if the film and substrate have identical expansion coefficients.
These effects may have either an apparent tensile or compres-
sive contribution but for the case where the af is larger than the
as' the situation realized most often in practice, one finds that
the contribution is an apparent compressive one. When related to
the initial temperature Tso the third thermal effect arises from
the temperature gradient through the thickness of the substrate.
Although this contribution is extremely difficult to measure a
worse case estimate of this gradient can be computed assuming that
the incident power must be conducted from one side to the other.
This gives rise to the equation P h/K. Assuming the thermal
expansion strains very linearly through the substrate the following
expression is obtained for the deflection and equivalent force .
I Eg 1 9,2
F3= 6" 03 = "2 11 (13)
For our experimental conditions the temperature gradient is 0.5
millidegree C. Table I summarizes these effects and compares them
to the force/unit width fora high stress material, such as nickel
at a thickness of 1000 A. Unless the substrates are exceptionally
flexible to obtain high sensitivity, we see that the momentum ef-
fect can be neglected for reasonable rates of deposition. Since
the intrinsic stress is itself a function of the temperature of sub-
strate during deposition, it is then important to maintain the tem-
perature as constant as possible. In terms of good substrate design
one would then pick a substrate of high thermal conductivity and
make the cantilever beam as short and thick as possible. The limi-
tation to such an approach is usually the sensitivity of the deflec-
tion measurement.
Experimental Verification
Data for a cantilevered beam of the approximate dimensions
used in our previous illustrations have been obtained by Springer
and Hoffman. (40) For the case of a nickel substrate initially main-
tened at 32 C, the free end rose by 23 C. This temperature rise
decreased as the temperature increased owing to the contribution
of radiation losses. At approximatively 150 C the temperature in-
crease vanished and at temperature above that, additional power
had to be supplied in order that the substrate temperature did not
fall during the deposition. A reproduction of the actual data for
the case of a nickel film on a nickel substrate was shown. (40) A
small compressional effect upon opening the shutter did not have a
corresponding feature after the close of the shutter. Thus we con-
clude from the lack of a final transient that the compressional ef-
fect at the beginning of the film formation is a real phenomenon.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 299
For this experiment af- as = 0 and only a temperature gradient nor-
mal to the plane would produce an effect.
Even tighter temperature stabilization was achieved in the ex-
periments of Doljack and Hoffman(32) with 2.4 cm dia., 0.015 cm
thick (Ill) Si substrates. In part this was achieved by a shorter
distance to the heat sink plus water cooling of the heat sink. With
this geometry the temperature difference between the inner and
outer radius was no more than over the temperature range from
1C to 200C. The worst case increase in the average temperature
was 10C over this same temperature range. The average temperature
increased linearly with film thickness up to about 100 A and then
remained constant for a deposition rate of 20 A/sec. In these ex-
periments, af > as and differential thermal expansion gives a bend-
ing moment that decreases intially to some compressive value before
coming constant when the substrate temperature reached its equilibrium
value. Upon terminating the deposition, the force curve would rise
toward a tensile value as the substrate temperature cooled to its
original temperature. For the circumstance where the temperature
increase took place up to roughly half the final thickness of the
film the apparent tensile transcient at the end of the deposition
would be about 4 times the amount of the apparent compressive value.
Maki and Kinosita(35) analyzed the more difficult experimental case
of silver films on extremely thin mica substrates. Large transcients
with the thermal time constant took place after the close of the
deposition. As was pointed out by the authors, the negative values
of deflection are inexplicable. Furthermore, the decreasing def--
lection during the course of the deposition indicated that relax-
ation in the film was taking place. The silver films used in these
experiments were probably responible for the aging effects at ex-
tremely low temperatures.
In addition to the techniques treated in some detail in this
section, we now list some of the techniques that have been used re-
cently for deflection measurement. Many of them have been contruc-
ted for special measurements and it is suggested that the referen'-'
ces be consulted for details. For in situ during de-
position, the force restoration and optical in-
terferometers, in some cases using laser sources, (32,42,43,44) are
the most widely used.
Holographic techniques, as proposed by Haines and Hildebrand
(45) for macroscopic objects, were developed by Magill and Young.
(46) To avoid the requirements of flatness and high reflectivity,
GIang et al.(47) have surface profiled Si waters in orthogonal di-
rections using a light-section microscope. The profile has also
been determined by inversion of local radius of curvature data
found an optical lever method proposed by Axelrod and Levin-
stein. ( 8). Moire techniques have been useful in determining the
principal stress direction in anisotropic films. (49)
300 R.W. HOFFMAN
Stress determinations using the bulge method were developed
by Beams(50) in his studies of mechanieal properties of metal films.
Jaccodine and Schlegel(51) used this technique on an unsupported
Si02 film by etching away the Si substrate, or by removing the oxi-
de from one side and observing the resulting deflection. These con-
cepts have been extended to a more local basis first, by Lane(52)
using a moat geometry and later by Lin and Pugacz-Muraszkiewicz(53)
who have developed a technique to make deflection measurements on a
microscale. Their technique uses standard photoresist methods to
etch out a small bar-shaped sample and then free it from the sub-
strate by etching with a second selective solution. If the film were
initially under compression, upon being freed from the substrate
is expands and deflects giving rise to considerable stress relief.
The authors have analyzed the elastic problem in detail and have
developed an expression relating the maximum displacement of the
free oxide film, observed optically as indicated in Fig. 15, to the
strain in terms of elliptic integrals of second kind. They have
found that approximately 10% of the total stress remains in the de-
flected sample. With much less sensitivity this technique can also
be used for films under initial tension by measuring the change in
length of an undercut beam, after it is released from the substrate.
The technique has been applied to films under compression down to
thicknesses of 250 A for the case of Si02 on silicon and the results
will be reported later.
A similar free-film technique is also reported by Wilmsen et
al. (54) who considered the stability of the thermal Si02 film in
relation to the thickness and width.
We summarize this section on the most commonly used technique
for the quantitative measurement of stress in thin films by remar-
king that with sufficient care, the thermal difficulties may be
overcome, the boundary conditions used in the calculations can be
realized experimentally, and that interpretative data may be taken
continuously during the deposition of a film in order to give in-
formation as to the stress distribution. The bending plate techni-
ques have thus been very useful in studies dealing with the origin
of the internal stresses and have even been used in automated sys-
tems on production samples through the use of fiber
Micro-techniques have led to localized stress distributions.
Thus we conclude with proper substrate design and choice of
high thermal conductivity substrates, temperature effects may be
reduced to the point where they are negligible for high stress
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 301
Fig. 15 Moat technique for local measurement. Reference(53).
302
R.W. HOFFMAN
deposits, even if thermal expansion coefficients are not matched.
It is still true, however, that for measuring the stress in the
early stages of growth or for materials which have a very low in-
trinsic stress, these problems have not been completely eliminated
X-ray Methods.
In order to follow the change in stresses through various pro-
cessing steps, or in order to determine local changes in a stress
as a function of position in the sample, x-ray techniques have been
used. The more familiar one is to measure the lattice constant,
usually with a diffractometer arrangement which determines the spa-
cing between planes lying parallel to the plane surface. The
strain in the film is most commonly measured but in a few cases
the strain in the substrate has actually been used to infer the
stress in the film. As usual, the stresses are calculated from
the strain in the appropriate elastic constant, in this case the
stress in the film plane in the simplest formulation is given by
(14)
The elastic strain normal to the film is determined from the
renee between the measured latticed constant of the film and the
bulk lattice parameter.
Sufficient care must be taken in these measurements that chan-
ges in line shapes and shifts arising from stacking faults or hetero-
geneous strains are not confused with the homogeneous component.
Recent progress in diffractometer methods follows. Zosi(56)
measures the strain as the function of the angle from the normal
to the film. Using only the position of the line centers, he finds
a linear relation between the strain and sin The stress in
the direction is then determined from
(J 5)
for a biaxial stress. In this expression is the polar angle and
the azimuthal angle.
Bush and Read(57) simplify the expressions derived by Taylor
(58) and also relate the film stress to normal and inclined diffrac-
tion angles upon specimen rotation. These treatments assumed no
stress gradient normal to the film.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 303
In order to directly measure the elastic strain, and thus
eliminate the difficulty of using tabulated values for ao the film
may be detached from the substrate thus relieving the elastic
strain and the lattice constant measured a second time. Kamins and
Meieran(59) have successively used this technique for the stress
in silicon films on sapphire substrates, through various proces-
sing treatments. The Nelson-Riley extrapolation function has been
used to determine the lattice constant. Microbeam techniques cou-
pled with scanning motion should allow the determination of the
local elastic strain with a reasnnable resolution. No such experi-
ments seem to have been done.
Cullity(60) formulates the more problem to consider
the importance of line shapes. Borie.(61,62) considers the effect
of a linear variation of strain through the thickness. McDowell
and Pilkington(63) derive a five-parameter nonlinear distribution
from the line shape which should be able to consider differences
in structure at the free and substrate surface. In their results
with gold films, immense elastic strains of 2.5% tension exponen-
tially decaying within 100 A of the glass substrate interface,
slowly shift to compression in the body of the film and a large
2.5% compression near the free surface. The antisymmetric strain
distributions in the thickness are a result of the symmetric dif-
fraction peaks. but the authors offer no explanation for large
strains and the lack of agreement with deflection experiments.
Within recent years an understanding of the topological x-ray
contrast of imperfections in crystals has emerged through a sudy
of both kinematical and dynamical effects. (64) As applied to
stresses, contrast may be observed when a film deposited on a sin-
gle crystal substrate provides a sufficient gradient to locally dis-
tort the crystal and give rise to intensity fluctuations. As pre-
viously decribed, Haruta and Spencer(24) used the technique to exa-
mine the stress distribution at the edge of a deposited film.
Meieran and Blech(65,66) used Borrmann contrast to explain the dif-
fraction contrast along interfaces. Schuttke and Howard(67) regard
the model as oversimplified and analyze the transmission geometry
using the scanning oscillator technique following the treatment of
Penning and Polder. (68) Dynamical and the lattice curvature
resulting near the edges of an etched diffusion window give
rise to the observed contrast. The strain gradient is a maximum at
the window edge. Reference to Fig. 16 indicates that when the ra-
dius of curvature and the diffraction vector g are pointing in the
same direction enhanced blackening is found on the film. If the
substrate is found to be in tension when the film must be in com-
pression, as illustrated in the figure. This technique provides good
spatial resolution for the sign of the stress but quantitative in-
formation as to the magnitude is lacking. The authors also give
illustrations of kinematical contrast at the boundaries where the
304
R.W. HOFFMAN
9
TOP08ltAPHICAL CONTItAST
BLACK WHITE
Fig. 16 X-ray topological contrast according to Schuttke. and
Howard. (67)
adhesion is lost. DYLamical images obtained by either adjusting
the substrate thickness or choosing the appropriate radiation give
rise to unambiguous interpretation of the regions that have failed
at the interface. It is worthwhile pointing out that localized
adhesion losses were determined by this technique when the optical
appearance of the film was mirror like and without any observable
flaws. The authors also pointed out that these techniques make it
possible to map the stress buildup in planar semiconductor devices.
It is then possible to obtain information about the relative stress
magnitude in the stress direction after each processing step, and
the stresses associated with the diffusion processes may be deter-
mined.
A x-ray double crystal arrangement in which the resultant to-
pographs show the non-uniform variations of lattice orientation
near near the surface has been developed by Zeyfang(69) for thin
silicon crystals bonded to glass. The origin of the strain may be
inferred from the distinction between a uniformly distorted surface
or one which has non-uniform distortions.
A modified scanning x-ray topographic arrangement has been
developed by Rozgonyi and Ciesielka
l70
) which uses a feedback sys-
tem to maintain a wafer orientation while the beam traverses the
specimen. The radius of curvature may be determined with a spatial
resolution of mm with simultaneous x-ray topographs. Stresses in
both very thin films ( ~ 1 A) and substrate are calculated as
well as effects of diffusion and ion implantation.
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 305
Because of the availability of x-ray topographical techniques
for semiconductor defect studies, their increased application to
stress systems caused by deposited films, diffusion, or implanta-
tion is expected. Both lattice parameter and topological methods
are non-destructive, have good spatial resolution and sensitivity,
but are not convenient nor rapid enough to be used during deposi-
tion.
Other Techniques
A new technique by EerNisse(71,72) has been developed for si-
multaneous measurement of stress and mass change. This double quartz
crystal resonator technique uses AT cut and BT cut crystals such
that the sum of the frequency shift is proportional to the mass
change and the difference of the frequency shift is proportional to
0.2
;;' 0411
e
u
-
'"
OIl
c-
ot
G'g 0
'oJ
:;
~
l-
en -0.1
o . ~
-200 150 -100 -50 0 so 100 UiO 200
DISTANCE XJ C;I(m)
80
,.....
60
.n
~
0=:
~ o
<II
...
0
'-J
20
UI
~
w
0=:
0
W
II.
u-
is
-zo W
<II
<
:s:
-40
~
-60
-80
2.50
Fig. 17 Stress distribution from optical birefringence measure-
ments of Reinhart and Logan(73) in GaAs.
306 R.W. HOFFMAN
the change in the thin film stress. The sensitivity is quoted at
125 dyne/cm for a 10 Hz frequency shift. The technique is suited
best for stress studies when the mass change is quite small, and
has been used in connection with implantation studies on silicon
samples.
Any physical property may be used to determine the stress,
and examples are beginning to appear. Note that this is the inver-
se of the usual care in which the 'unusual' properties of a film
are being explained in terms of the stress, and represents are
more sophisticated point of view. Oel and Frechette(22) have done
this for macroscopic glass discs using birefringence to measure
the strain. Reinhart and Logan(73) calculated the phase difference
and elastic constants from the crystal piezo-optical tensor coef-
ficients, and compare with data to determine the interface stress
of thick epitaxial layer structures. Fig. 17 gives the stress dis-
tribution near the interface for AlO.IGao.9As on (Ill) Ga As.
An increased use of such 'indirect' measures of the strain is
anticipated.
4. THERMAL STRESSES IN THIN FILMS
Introduction
The techniques for measurements that we have discussed in the
preceeding sections have allowed the determination of a strain and
the calculation of an associated stress. We now turn to the ques-
tions of the origin and the various mechanisms that can contribute
to these strains. As a matter of definition the total stress in
the film is regarded as being made up of two possible contributions.
These are defined as the differential thermal expansion stress whose
origin is understood, and whose value s generally amienable to
calculation, and, the intrinsic stress which is the manifestation
of all other contributions. We may write aT anTE + ai, where
the intrinsic stress, ai, may be composed of several terms. Clear-
ly, aT is measured and a property responds to the total stress.
The contributions from either term may be tension or compression,
with the resultant total stress in the film being determined by the
relative magnitudes.
In F.ig. 18, we consider in an idealized way the case of a film
deposited on a substrate at temperature Ts and cooled to the final
temperature Tm. If the thermal expansion coefficient af of the film
is larger than the expansion coefficient of the substrate as then
the contribution to the stress from differential thermal expansion
increases. On the other hand, the intrinsic stress is often a de-
creasing function of the temperature, so that the total stress in
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 307
If)
If)
W
0::
.-
If)
Fig. 18
GROWTH
DEPOSITION
TOTAL
"
"
TEMPERATURE
Thermal and intrinsic stress contributions.
the film may, in fact, have a minimum at some intermediate tempe-
rature.
Even today the literature far too commonly considers diffe-
rential expansion as the only contribution to the stress. For the
higher melting point metals, for example, intrinsic stresses are
quite large up to substrate deposition temperatures of several nun-
dred degrees C. In other systems the diffusion of impurities into
or out of the film at elevated temperatures give rise to substan-
tial stress effects. Even in the case of thermally grown Si02 on
Si at high temperatures, the intrinsic stresses have an important
role.
On the basis of our earlier sections we see that the funda-
mental origin of all of the strains in films lies in the constraint
that the film is being tightly bonded to the substrate and after
this bonding has taken place there is a subsequent volume change
in the film (or strictly a planar distortion). Thus, in addition
to the same assumptions made in solving the elastic problem, we
308 R.W. HOFFMAN
must evaluate the strain itself. Normally this is simplified by
assuming the substrate is rigid and that neither contraction nor
bending takes place i.n the film substrate. Under this assumption
the change in lengths of the substrate resulting from the stresses
on it is negligible, but if the film and substrate were of compa-
rable thickness, then both of these relaxation effects must be
taken into account.
Thermal Expansion Stress
Normally, the thermal expansion strain is considered to be
isotropic. In general, the coefficient of thermal expansion relates
the strain tensor to a small uniform temperature change. Thus, the
thermal expansion tensor is symmetrical since the strain tensor is.
(12) It is worth noting that since the thermal expansion of a crys-
tal must possess the symmetry of the crystal, it cannot destroy
any symmetry elements and this is why the class of a crystal does
not depend on the temperature. The thermal expansion tensor may be
referred to principal axes and thus reduce the number of coeffi-
cients to 3. For most substances the principal thermal expansion
coefficients are all positive. But in a few crystals, for instance
calcite, or silver iodide, some coefficients are negative.
We may write the nine thermal strain components in terms of
the thermal expansion tensor.
E = a /::'T
1.J 1.J
(16)
where the identical subscripts refer to the normal components and
the mixed indices to the shears. The thermal expansion tensor is
often reduced to a column matric to be consistently used with the
six single index strains Ei.
( 17)
If we consider the restrictions of crystal symmetry,(13) only
in the triclinic and monoclinic systems are the shear coefficients
other than zero. The orthorhombic system has three independent nor-
mal expansion coefficients, the tetragonal, trigonal, and hexagonal
systems have two, and the cubic system has an isotropic thermal
expansion. We must point out however, that even though the thermal
expansion may be isotropic, because the elastic constants are fourth
rank tensor properties, anisotropic stresses will be obtained even
in a cubic crystal. Only if the sample may be considered to be elas-
tically isotropic, that is in practice a polycrystalline sample with
small grain size, will the resultant thermal expansion stresses be
isotropic. Once the thermal strains are determined, the stresses are
calculated from the appropriate elastic constants. The equation
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS
(I8)
is appropriate for the rigid isotropic case and takes account of
the biaxial stress generated by the linear strain 11 = 22.
309
The sign is such that the tensile stress is positive. Obvious
extensions are needed when the temperature range considered is so
wide that the a's are no longer constant. Vook and Witt(14,74) car-
ried out calculations of the strain normal to the plane of the film
for arbitrary orientations of cubic crystals. In these calculations
which were also extended to include the strain energy it was assu-
med that the shear strains in the film were zero. For copper and
silver, the metals considered, the strain energies were found to
be in the same relative ranking as Young's modulus. Vook and Witt
point out that the observed line shifts in an x-ray diffractometer
experiment when the temperature is changed correspond directly to
the calculated thermal strain, even though other '.ontribution to
the line positions as a result of faulting may be present.
To this point we have considered the specimens to be semi-
infinite in size and edge effects have not been important.
Aleck(23) has converted the thermal stress problem in a rectangular
plate clamped along an adge to one of boundary stress conditions.
His two dimensional solutions indicate there are large shear
stresses present in the film within about two film thicknesses of
the edge, and a stress normal to the plane of the film concentra-
ted at the intersection of the edge of the film with the film sub-
strate interface. The distributions along the interface (clamped
edge) are qualitatively similar to the data in Fig. II if normali-
zed to the one-dimensional thermal stress E The large va-
riation of the normal stress 033 as a function of the
from the free surface at the edge of the film is given in Fig. 19.
Zeyfang(75) has extended Aleck's approximate solution to pla-
tes of arbitrary lengths, rather than the semi-infinite plate used
by Aleck. He also gives detailed stress, strain and displacement
distributions for plates with length to thickness ratios of 2, 5,
and IS. Compared to the semi-infinite case where there are edge
effects only, when the length of the film becomes short enough,
nonuniform stress distributions are found throughout the volume
of the film. The maximum stress in the plane of the film decreases
as the sample length decreases and actually reverses sign on the
free surface of the film for samples that are only twice as long
as their thickness. The maximum shear stress decrease with increas-
ing plate lengths and the stress normal to the plane in the center
of the interface increases with decreasing plate length. Zeyfang
310 R.W. HOFFMAN
2.
o
-2
-4
-G
-8
-10
-12
Fig. 19 Thermally generated normal stress at a free edge as a
function of distance from the free surface. After Aleck.(23)
Coordinate X3 = 10 corresponds to the interface.
points out that for plates whose length to thickness ratio is
greater than 15, it is possible to think of the distributions as
being composed of a uniform part for the central part of the sam-
ple and edge terms, but such a treatment is not valid for shorter
samples. The shape of a half film under the exaggerated thermal
expansion condition is shown in Fig. 20. Because of the singula-
rity of 033 in linear elasticity theory at the interface edge,
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 311
a cylindrical cavity was cut out and the displacements approxima-
ted by surface displacements. For such short samples it follows
that the upper suface no longer remains plane under a thermal ex-
pansion strain.
In carrying out the calculations of thermal expansion con-
tributions the questions arises as to the values of the thermal
expansion coefficient to be used. There seems to be no evidence
for abnormal expansion coefficients in deposited films. There are
cases where the total stress effects do not seem to coincide with
calculations of the thermal contributions, but these are attribu-
THERMAL EXPANSION STRAIN
----,
I
I
I
L---______ L-______ ~ ~ ____ L _ ______ O
Fig. 20 Thermal expansion distortions of short films according
to Zeyfang. (75)
312 R.W. HOFFMAN
ted to irreversible intrinsic stress changes in the film upon hea-
ting. Hence, tabulated values of the expansion coefficient seem
quite suitable in practice. Feder and Light(76) have reported an
apparatus for directly measuring the differential thermal expan-
sion between a film deposited on the substrate by observing the
optical fringe shifts corresponding to bending the sample. For the
case Ge/Ga As film couple the difference between the thermal expan-
sion coefficient is demonstrated to be less than 1 part in lOB.
One of the goals of the calculation of thermal stresses is
the prediction or explanation of properties from data on bulk,
strain-free samples. Schlotterer(77) determined the resistivity
change in epitaxial Si on spinel in terms of the piezo-resistance
coefficients for isotropic thermal stress. A second example is the
anisotropic Hall mobility found in silicon on saphire films by
Hughes and Thorsen(7B,79). After calculating the thermal expansion
stresses arising from the anisotropic thermal contraction of Al2 03
in cooling from the deposition temperature, they also use the piezo-
resistance effect to relate the Hall mobility. The approximate 10%
anisotropy of the Hall mobility iu two orthogonal directions seems
adequately described in terms of thermal strains. Hughes(BO) car-
ried out detailed calculations of the effect of stress on resisti-
vity for general orientations in several crystal systems.
Thus it appears as though both isotropic and anisotropic ther-
mal expansion strains, and their effect on transport properties
are amenable to calculation.
4. ORIGINS OF THE INTRINSIC STRESSES
The review articles by Hoffman(4,5,6) Campbell(6,7) Buckel,(B)
and Kinosita,(IO) all have the goal of trying to assimilate the
massive experimental literature on measurements of the stress and
attempting to interpret them in a few simple concepts. In addition
to the thermal stress, Buckel lists six other processes which may
produce stress in the following way.
I. Incorporation of atoms, for example, residual gases or
chemical reactions.
2. Differences of the lattice spacing of monocrystal and
substrates and the film during epitaxial growth.
3. Variation of the interatomic spacing with crystal size
4. Recrystallization processes
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 313
5. Microscopic voids and special arrangements of dislocations.
6. Phase transformations.
In addition, all of these processes may occur both during
condensation and growth of the film and under annealing conditions
afterwards. A search of the literature also indicates that most of
these models have been developed to explain the stress in metallic
films, pure metals or some simple alloys. Especially with nonmetal-
lic films, where the stoichiometry and the structure is still far
from being well established, the understanding of the origin of
the stress is now unsatisfactory. For these reasons we review the
models for the intrinsic stress very briefly, calling attention
to those cases where there is some confirmation in the literature
for non-metallic films. More importantly, the discussion may give
a physical basis for the application of these concepts to non-me-
tallic systems. Many of the models find it relatively easy to ex-
plain shrinkages and hence tensile stresses in the film, but can
explain compressive behavior only with difficulty.
Several different kinds of processes may operate in the first
classification. Nakajama and Kinosita(81) have measured the stress
in silver films deposited at 10-3 and 10-
6
Pascal and find the fol-
lowing behavior. The dependence upon the thickness is not greatly
influenced by the deposition parameters before the film becomes
continuous. The residual gas pressure during the deposition has no
appreciable influence on the stress behavior. However, after the
deposition the stress will remain at its constant tensile value
if the sample is kept in a vacuum.of 10-
6
Pascal but if kept in a
10-3 Pascal surroundings, a compressive relaxation to a smaller
value of tension in the film is found. Furthermore, this compres-
sive contribution will recover with a time constant of the order
of hours if the pressure in the system is reduced to 10-
6
Pascal.
The authors suggest the reason is the absorption of residual gas
in the film, and the stress effects correlate with changes in the
electrical resistivity.
Actually, the presence of residual gas, oxygen and water va-
por in particular, has long been suspected in the case of those
metallic films showing compression when deposited in poor vacuum,
or when annealed in air after deposition. A more direct quantita-
tive confirmation of the effect of gas incorporated in the film
during the deposition process has been forthcoming in the work
with bias sputtering which we shall treat in some detail in a la-
ter section. Ion implantation which clearly gives rise to dilation
also falls in this category.
Because of their technological importance, the growth stresses
in thermal and anodic oxide films must be included. A recent review
314 R.W. HOFFMAN
by Stringer(82) considers the stress generation and relief in
grown oxide films on metals. The sign of the stress in the oxide
layer may be predicted by the ratio of the volume per metal ion
in the oxide to the volume of metal atom in the metal or the
Pilling-Bedworth ratio. Most grown oxide layers are compressive
but MgO is correctly predicted to be tensile although actual
stresses are much less than calculated by this ratio. These dif-
ferences may arise from ion motion, particular anisotropic growths,
or, perhaps more likely, plastic flow taking place during the oxi-
dation process. A second mechanism includes the oxygen migration
down grain boundaries in the metal oxide in order to react with
metal ions diffusing through the bulk. This mechanism, illustrated
in Fig. 21, also leads to a compressive growth stress in the oxi-
de according to Rhines and Wolf. (83) The similar impurity migra-
tion down grain boundaries has been as the origin of the
compressive stress in metallic films. ,32). A similar situation

OXIDE GRA\N IOUNDARtES
NICKEL PATH
OXYGEN PATH - - - ..
OXYGEN
NlCKEL
OXIDE
HlCKEL
Fig. 21 Schematic diagram of diffusion paths. Following Rhines
and Wolf(83) and Stringer. (82)
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
315
is found in the growth of anodic oxide films, which have recently
been reviewed by Dell'Oca et al.,(84) following earlier articles
by Young(85) Vermilyea.(86) The growth of inter-metallic phases
following diffusion has just been confirmed as a stress mechanism.
( 137)
One of the best documented or1g1ns of internal stress occurs
during epitaxial growth as a result of the efforts of Van der
Merwe(87,88,89) Matthews, (90,91) Jesser and Kuhlmann-Wisdorf,(92,
93) Honjo(94) and others. An interfacial energy,(92,93) is calcu-
lated which depends on the differences of the lattice spacing of
the substrate and film, as well as the elastic constants of each
material. For small mismatches or thin films, the total energy
in the system will be minimum if the film develops a uniform elas-
tic strain. However, if the film becomes thicker or if the mismatch
is more severe, the energy of the system may be lowered by the
creation of an array of interfacial dislocations which may relieve
a large fraction of the strain in the film. A total or partial
elimination of misfit dislocations may take place as schematically
indicated in Fig. 22, where the energy gain is ~ and the critical
misfit is seen. It is worth noting that forces normal to the in-
terface contribute only about 10% of the interfacial energy. Recen-
tly the misfit dislocation energy is calculated(88) from a periodic
interfacial potential which is a generalization of a Peirls-
Nabarro model rather than the older pair-wise or elastic models.
s
A
\
\
' ....
MISFIT
MISFIT
Fig. 22 Energy gain E arising from total or partial elimination
of interfacial dislocations. Curve A is the homogeneous elastic
strain energy, and B the misfit interfacial energy. After Van der
Merwe. (87)
316 R.W. HOFFMAN
Jesser and Kuhlmann-Wilsdorf(92,93) have shown that a similar
formulation holds for films in the isolated island of growth.
The interfacial dislocation mechanism for the origin of intrinsic
stress is attractive because it may be formulated quantitatively and
compared with experiment. As a general mechanism, however, it must
be criticized on two grounds. First, the obvious one that it is not
easy to see how such a mechanism would operate for growth of poly-
crystalline films and, secondly, in this model the strain is locali-
zed and decays exponentially from the interface and thus does not
project the constant stress often experimentally found. Nevertheless,
data for 50 A alkali halide films on. glass seen to
follow this )
A stress model for heteroepitaxial films combining thermal
stress and assuming the film lattice constant is constrained to
match the substrate has been developed by Besser et al.(96) They
have examined many CVD magnetic oxides and developed a physical mo-
del in which the film stress is entirely determined by uniform mis-
fit at room temperature independent of the deposition temperature up
to a critical value and entirely by thermal expansion for larger mis-
fits. This idealized model works well when the thermal expansion
z
o
en
z

Fig.23
on ----,
AQ.a
I
I
I
I
I
I
I
I
-------- Oi
REa,oN n REGION:t
.. '.
t----ov.
4Q., (a.s-Q.,) AT 300
0
K
I
I
I
,

REGIOM n

Epitaxial misfit-thermal strain model of Besser et Al.(96)
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 317
coefficients of film and substrate are nearly the same. It takes
account of the thermal expansion in region I by using the room temr
perature lattice constants but assumes the complete relaxation of
the mismatch stress in region II.
Fig. 23 indicates these two regions, where
Ef as-af Ef
and a
U
= (a
f
-a
s
)8T
J. 1 - v f d
f
,I-V f
Carruthers(97) proposed a similar model including the effects of
plastic deformation occuring at strains larger than 0.005. He fits
the thermally-activated glide relation derived by Matthews et al.(9I)
in which the misfit strain is the driving force to the data of Bes-
ser et ale Carruthers'expression for the planar strain is
where
=
(
T T
m m
as -a
f
)(T -T )
T ssm
m
a
f
a' - a
f s
a - a
f s
T=Ts
( 12)
is
the fractional strain recovery, one finds a number of experiments
relating the lattice constant at and the lattice parameter of the
film material af evaluated at the deposition temperature T
s
'
Braginski et al., gives additional data for the magnetic
oxides.
In the third category, one finds a number of experiments re-
lating the lattice constants with the crystallite size
preting the rsults in terms of a surface energy. Cabrera
the thermodynamics following the classical paper by
Lindford and Mitchell (101) introduce the concept of
interplanar potentials which relate the surface work to macroscopic
parameters of elastic constants and thermal expansion coefficients.
Wasserman and Vermaak(I02) give recent data and references for the
surface stress as observed by electron diffraction. The surface
energy provides a recognized contribution generally com-
pressive stress in the isolated island stage of growth, however,
unless the growth of a deposit can be formulated in terms of a con-
tinual nucleation and growth framework it is difficult to see how
this category can contribute to the stresses in fhick films. It
remains then a difficult experimental area because of the vacuum
environment, but in need of future work.
318 R.W. HOFFMAN
The surface stresses in Ge, InSb and GaSb have been measured
using the bending of thin sample by Taloniand Haneman.(43). The
effects of surface stress as correlated with bending plate measu-
rements have been discussed by Hoffman.
Classification 4, recrystallization processes, contains a
number of different models that differ in their details.Klokholm
and Berry, (105) suggest that the stresses are generated "by the
annealing and constrained stinkage of disoredered material buried
behind the advancing surface of the growing film". The magnitude
of the stress is determined in this model by the amount of disor-
der at the time the atom become constrained and the subsequent
annealing during condensation process. The temperature dependence
of the intrinsic stress is determined by the kinetics and it is
held that no stress would be found in films deposited at extreme-
ly low temperature substrates, and at substrate temperatures during
deposition higher than about 1/4 of the melting temperature. This
form is approximately followed for metals. In this model, films
deposited on cold substrates would have no stress since the atomic
rearrangement is no longer possible.
The helium temperature experiments of Buckel (8) indicate that
simple metals growing in the normal crystalline phase, even though
they have very small crystals and have a high degree of disorder,
to develop large stresses when deposited at these very low tempe-
ratures. It is often stated that amorphous films have low stress.
Gallium and bismuth may be frozen in amorphous phases if the sub-
strate temperature is below about 20 K. These data as reproduced
in Fig. 24, show that for low substrate temperatures a small stress
is developed when the i&lands join at an average thickness of about
60 A but no additional stress is produced on further growth. For
higher substrate temperatures the usual tensile stress is found.
The amorphous Bi films are quite unstable, and with increasing film
thickness a spontaneous crystallization often takes place which
yields large compressive stresses. These experiments, as well as
calculations, convincingly exclude the presumption that the surface
temperature of the film is 3ubstantiantly higher than the substrate
Because of the present importance of amorphous
films, and the general behaviour that even alloys seen to grow with-
out intrinsic stress,(106) we report Buckel's result that Sb
films condensed on low temperature substrates exhibits rather high
tensile stresses. Because the Sb vapor from which the film is built
up consists almost oompletely of Sb4 molecules, Buckel has refor-
mulated the general hypothesis about amorphous films to read "amor-
phous phases with a short range order similar to tha.t of a liquid
(frozen liquids) grow without internal stress", and interprets the
statement on a microscopic void model.
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
O.l
I
1
0.7,"",
.-...
B. - CRYSTALLIZATION
,
_ 1 0
""\ ' '\ .. ft \ ... n .... ....

\
. _--
"'" TS< tSOK
-.
\
--....

\

....
....

....
....
....
\
....
....
Z
\
....
....
Q
w
\
"
...J r-
\
/" ......
I-
iP
u
Z
LA1 \
GOo,
....
..J
""
\
,
LL.
1
\
T
s
"4
0O
K
....
La.I
....
\
....
0 ....
\ ....
\
0.4
-

1
\
\

Ts:3'5K
\
\
0.6
\
1
o
1000
FILM THICKNESS (Al
Fig. 24. Stresses in depositions as measured
by Buckel(8).
319
-
-
....
.... .::li
2000
Chaudhari(107) has considered the effect of grain growth on
the stresses in films when densification occurs by the eliminRtion
of a grain boundary. If the initial grain size is very small a
tensile stress is generated and the final grain size is determined
by the minimization of the sum of the strain energy and surface
energy. When the initial grain size is above this critical size,
perhaps 20 A, no energy minimum is found and grain growth is not
restricted. In this model an initial elastic compressive strain
aids recrystallisation, whereas the experimentally found initial
tensile strain is not favorable for grain growth.
320 R.W. HOFFMAN
Wilcock et al.,(10$) have determined the stress in the early
growth stage for Ag and Au. They also consider the stress to arise
from the elimination of grain boundaries and suggest only 10% of
its boundary need be eliminated to account for the observed stress.
In a series of papers Hoffman(30,32,40) and his coworkers
have developed a different concept of a grain boundary model. In
this model this strain is generated as the adjacent surfaces of
two grains come into contact during growth. The problem may be
formulated in terms of a grain boundary potential but this is for-
mally equivalent to relating the strain energy to the difference
between the surface energy of the two crystallites and the energy
of the resultant grain boundary. In addition to values of elastic
constants and energies this model needs only a value of the final
grain size of the film for a quantitative calculation.
The grain boundary potential is demonstrated in Fig. 25. The
depth of the potential is given by 2Ysv-Ygb, where Ysv is the
i
!: 0
Z
:it
....
i
l
\II
,. Cl.
ATOMIC SEPARATION
Fig. 25 Grain boundary potential. After Doljack, Springer, and
Hoffman.
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 321
surface free energy and Ygb the grain boundary energy. As an ave-
rage grain boundary has an energy ~ Y
sv
/3, the depth of the poten-
tial is 5/3ysv at the nearest neighbor distance a. An arriving
atom may populate any position where the potential is negative
if it lies between r and a then the atom relaxes outwards ; if
it arrives at a position between a and 2a, then a contraction will
take place at the boundary. In both cases a strain energy is pro-
duced, and the minimum produced between the strain energy and po-
tential energy defines the net strain. As the potential is asym-
metric a tension in the film is produced of
<0>
where the average displacement ~ > is obtained by integrating over
the range of the potential. Good agreement has been obtained in
both the temperature dependence and the magnitude for the case of
nickel. (40)
Classification 5, voids and special dislocation arrays, beco-
mes a convenient catch-all. Voids in films have been postulated
to explain magnetic anisotropy and low values of density,(109,110)
even though the x-ray density is typicial of bulk material. With
dielectric films evaporated at angles of incidence, voids are
popular in order to explain the so-called porousity of the film.
Nevertheless, detailed experiments to show the size and shape of
such voids have usually failed. The same kind of general remark
may be made about dislocations because there is a well-defined
stress field arond a given dislocation. We may postulate a special
dislocation array in order to give rise to any stress distribution
that we desire. However, with the exception of the interfacial dis-
locations mentioned earlier, the disloeation lines found in films
by electron diffraction are generally normal to. the plane of the
film. Thus, generalized statements relative to voids and disloca-
tions, are generally of little use in detailed analysis of the
stress problem. Exceptions to this may be found in the case of
stacking faults, and other such documented defects in a particular
film in question. Abrahams et al.,(III) have proposed an oriented
array of inclined dislocations to explain the stress gradient that
curls a film upon removal from the substrate. Saito et al.,(112)
have postulated a periodic distribution of screw and edge disloca-
tions obliquely piercing the film plus added terms to have a trac-
tion-free surface. The maximum stress does not occur at the bounda-
ry of the dislocation.
Category 6, contains the expected first-order phase transfor-
mation with its atteded volume change that must take place after
the film is formed for a stress to be generated. Buckel(8) using
322 R.W. HOFFMAN
the Oswald step rule of crystal growth, suggests the film may pass
through metastable liquid phase during condensation, and the inter-
nal stresses thus arise from a different in density from the liquid
itself. Thus, one would expect that a frozen in liquid as the first
phase formed exhibits no internal stress, consistent with many re-
sults already presented for amorphous metals. As most metals melt
with a decreased density, normally a tensile stress will be obser-
ved, and this is also experimentally found. Amorphous alloy films
also show a very low stress. (106) This model has had some success
in predicting the sign of the stress but as yet has not been for-
mulated in a quantitative fashion.
Without specifically stating, our discussion so far, has led
one to believe that the intrinsic stresses would be isotropic in
the plane of the film for randomly oriented polycrystalline samples
where the elastic constants are isotropic. The experience with met-
als has indicated that this is true when the films are deposited
at normal incidence. Stress differences of perhaps 20% are found
for metals when the evaporated beam is incident at 45 with respect
to the substrate normal. (I 13) These early experiments with magne-
tic metals were concerned with the anisotropies resulting from the
anisotropic stress-. Smith et al., (110) suggested that the anisotro-
py arises during the film growth from a random nucleation in combi-
nation with some shadowing effects. This mechanism very ope-
rates for specimens deposited near grazing incidence but Hoffman
and coworkers suggested that anisotropic nucleation and growth
would allow the grain boundary mechanism to operate at smaller an-
gles of incidence. Dielectric films have much larger anisotropy ef-
fects as we shall see in a following section.
V. SUMMARY OF THE DATA
Introduction
In view of the fact that the importance of the stresses has
been known for such a long time, it is surprising that so few pa-
pers giving data exist in the literature. Even tody the pioneering
work of Turner and TrubYl(114)Heavens, (lIS) and Smith, Blackburn,
Campbell (116) Carpenter and Campbell,(95) Kinosita et
and Scheuerman(IO) contain perhaps 3/4 of the films stu-
died if we exclude the epitaxial situation. In examining even the
recent papers it is still true that improper expressions have been
used to relate the stress to the observed parameters, and sufficient
information is seldom given to separate the thermal stress from the
intrinsic stress. As it is the total stress which affects a given
property, this practice may be excused if one is interested in a
particular processing treatment, but certainly an understanding of
the origin of the intrinsic stresses is not aided by such experiments.
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 323
In order to focus our attention let us classify materials as
optical films, alkali halides, epitaxial films, and films which
perform a passivation, isolation or dielectric function. It will
be realized, of course, this is not an all inclusive way of classi-
fying non-metallic films, and furthermore, a number of redunancies
will occur when we look at an actual material. The purpose in this
classification is to see if there are generalizations we can make
within each category which prove useful in trying to interpret the
data. Table II contains an updated list following the earlier ta-
buiations of Hoffman,(4) Campbell,and Scheuerman.(lO) When only
one author made measurements for a particular it was
simple to quote a value for the stress. Now that several authors
have reported for the same material, but often carrying out the
deposition at different temperatures or different techniques, there
may be considerable disagreement as to the magnitude of the stress
and, in some cases, even the sign. Thus, the individual papers must
be consulted to see if the conditions of the reader's interest
match with those of the literature. We wish to again caution that
the intrinsic stress is indeed a structure sensitive property, and
seldom is the complete characterization of the film carried out
with sufficient detail that such a cause and effect relatioship
can be found. It is well known that there are significant effects
from the ambient atmosphere when a film is taken from vacuum sur-
roundings and that there are longer time effects associated with
diffusion of impurities along grain boundaries or through the lat-
tice. Although a few recrystallization studies have been reported,
in general stress data is not available for post-deposition treat-
ments.
We have also eliminated consideration in Table II stresses
induced in materials as the result of ion implantation as these
are not inherently a thin film effect, although measured by the
same techniques. These is no question that the incorporation of
impurity atoms during sputtering or ion beam deposition can lead
to the same kind of behavior, and some of these data are included
and will be discussed. We have also eliminated from consideration
the growth stresses in what are commonly called "oxide scales on
metals". These predominatly thermally-grown oxides, a subfield in
themselves, have been reviewed in considerable detail by Stringer.
(82) Likewise we have excluded anodic oxides.
After this rather lengthy introduction we turn now to the data
for optical films. A general review of optical films by Ritter(103)
will appear soon and laser window coatings have been reviewed by
Young.(104) The average stress is generally displayed as a func-
tion of thickness, in some cases the force per unit width or its
derivative is plotted. The average stress usually increases rapid-
ly, reaching a maximum value at a thickness of a few hundred angs-
troms, and then becomes a slowly decreasing function of the thick-
324 R.W. HOFFMAN
TABLE II
Index to Stress Literature
Film Substrate Reference
AgCI Glass 114
AgF Glass 114
AgI Glass 114
Alph* Glass )]6
AIF3 Glass 114
Al203 Al 159
Si 54, 163, 164
Silica 10
B203
Glass 116
BaF2 Glass 114
BaO Glass 116
C Glass 131
CaF2 Glass 114, 116
Mica 115
Silica 42
CdS Glass 116
CdTe Mica 165
Silica 10, 42
CeF3 Glass 1 16
Silica 42
Ce203 Glass 114
Silica 42
Chiolite Glass 114
Silica 42
Cryolite Glass ])4, lIS, 132
Silica 42
CuI Glass 114, 157
Fe3Si Fe 162
Ge Silica 10, 42
Glass 56
Mica 166
KBr Glass 95
*
Ph = phtalocyanine
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 325
TABLE II (con't)
Film Substrate Reference
KCI Glass 95
KF Glass 95
KI Glass 95
LiF Carbon on glass 134
Cellulose 135
Glass 95, 114, 116, 131
132, 134
Mica 115
MgF2 Glass 114, 116, 132, 136, Us3
Mica 115, 117, 137
Silica 42, 151
MgO Silica 10
MgPh* Glass 116
Mo03 Glass 114
NaBr Glass 95
NaCI Glass 95
NaF Glass 95, 132
PbCl2 Glass 114, 116
Silica 42
PbF2 Glass 116
PbTe Glass 138
Mica 138
RbI Glass 95
Sb203 Glass 114
Sb2S3 Glass 114
Si Sapphire 59, 78, 123, 124
Quartz 71
Si 59
Spinel 33, 77
Si3N4
Si 48, 128, 129, 158
SiO Glass 116, 122, 134, 139, 140
141, 142, 144, 145
Nickel 146
Silica 10, 42
Al 161
Glass W 133
326 R.W. HOFFMAN
TABLE II (con't)
Film Substrate Referenee
Si02(CVD) Si 67, 148, 160
Si02(Sputtered) Si 55, 67, 148, 149
164
Si02(Reactive
Evap) Si 153
Silica 10, 152
Si02(Thermal
Decomposition) Si 149
Ge 149
Si02(Thermal
Oxide) Si 1O, 51, 52, 53, 54,
67, 127, 148, 149
SiOxN
y
Glass 153
Si 13O, 158
Sn02 Glass 114
SrS04 Glass 114
TaOx
Ta 154
Ta205 Ta J55, 156
Glass 156
Te Silica 1O
Ti02 Glass 153
TlCe Silica 42
TlI Silica 42
ThF4
Silica 1O, 42, 152
(ThOF2) Silica 147
Th02 Silica 1O
ZnS Glass 114, 116
Mica lIS, 117
Silica 10, 42, 147, 152
Zr02 Silica 10
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
4 ~ ~
3
,...
"
"
.....
III
III
C
)..
'"II
2
..
0
~
III
TENSILE
II!:
I-
en
1
o
o
1000 2000 3000
fiLM THICKNESS (A)
Fig. 26 Representative data for the stress in optical films.
From Ennos. (42)
327
ness as indicated in Fig. 26 for several materials. For films
deposited in a liquid-nitrogen trapped diffusion pump system at
pressures of about 10-4 Pascal, Ennos(42) finds a significant in-
crease in tensile stress with a time constant of about 100 seconds
after the deposition is stopped. This transient very likely results
from thermal relaxation of the substrate although the author expe-
rimentally found no initial transient. Significant changes are
also found when the film is exposed to air over a period of per-
haps 30 minutes.
Of the common optical materials onlr ZnS is deposited under
compression, although Kinosita et al.,(1 7) indicate that a tensile
stress may be found for low deposition rates. Fig. 27 summarizes
some of these results for ZnS and it is interesting to point out
that for the same deposition rate, data from different laboratories
328 R.W. HOFFMAN
2
TNSILE
I
o
1000 2000
FILM TMICI<NESS tAl
I
-I COMPAUSlvt
Fig. 27 Stress in ZnS films. Tensile data from Kinosita et al.
(117) Compression from Ennos.(42)
can be obtained which are most a negative of each other. Blackburn
and Campbell(46) have found a linear increase in the compressive
stress in ZnS films with the increasing rate of deposition in the
range of 5 to 30 A/sec. The intrinsic stress in ZnS is also a ra-
pidly decreasing function as a substrate temperature is raised from
80 to 150C.
These differences point out the subtleties in the dependence
of the stress on the deposition conditions. It is known(118) that
the evaporation geometry, especially as related to the self-gette-
ring is important for determining the morphology. ED, SIMS, index
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS 329
of refraction, and density measurements have recently been made
for ZnS by Preisinger and Pulker.(119) The columnar
have a diameter which slowly increases with thickness above 500 A.
The non-stoichiometric excess of S was found to be due to S2 bound
to the surfaces of the crystallite columns.
Pulker and Jung(120) have suggested a cylindrical column model
for the structure of ZnS and other optical films based on electron
microscopy and water vapor absorption data. The observations sug-
gest that an explanation for" the stress might be sought in the imr
purity-grain boundary model.
Hill and G. R. Hoffman(122) have made an extensive evaluation
of silicon monoxide. In the first place, they find a linear force
thickness curve for constant evaporation conditions, suggesting
that some of the behavior for thin films less than 1000 A thick may
result in the difficulty in establishing steady-state conditions.
TENSILE
0.2-
0
;;-
10.
2 10.
1
10
f
N u
-
..
..
r:
-t
fI'
0

'0.1
TEMPEI\ATUf\ 12.00
0
C
en
III
11.1
Ill:
.....
OXYGEN
II)
- 0.9
-'.1.
COMPRESS IVE
Fig. 28 Stress in SiO films as a function of SiD to D arrival
Rate. After Hill and Hoffman. (122)
330
TENSILE
1.0
....
..
0.8
0.6
"
i
)..,
(1'"'<1 0."
o
o
.0.2
-0.4
-0.6
COMPRESSIVE
R.W. HOFFMAN
Fig. 29 Angle of incidence effects in SiD. After Hill and
Hoffman (122)
Furthermore, they establish that the stress is a sensitive function
of the residual gasses when the arriving silicon monoxide flus is
comparable in magnitude. That the stress was indeed a function of
the atmosphere during deposition, was strikingly demonstrated by
changing the atmosphere during deposition with the resultant change
in the sign of the stress in the material being deposited. Their
data is summarized in Fig. 28 where the stress is plotted against
the logarithm of the ratio, N, of the arrival rate of silicon mo-
noxide to residual gas molecules at the substrate. Dielectric cons-
tant measurements suggest that the tensile stress occuring when the
ratio N > 1 may be thought of as the intrinsic stress of silicon
monoxide, whereas increasing the pressure of oxygen results in a
silicon dioxide deposit. The earlier source temperature dependence
found by Novice and Priest and his co-workers may now be understood
in terms of the ratio N. Furthermore, these results confirm the ear-
lier rate and residual gas pressure measurements of Blackburn and
Campbell. (I16)
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 331
Hill and Hoffman also report that the film stress remains sta-
ble as long as the speciment is under vacuum, but on the readmission
of atmosphere the stress becomes more compressive, independent of
its initial value. The increase in compression upon exposure to am-
bient may be several times the initial compressive stress, and to
explain this fact they suggest that such films have a low density
which makes them more susceptible to oxidation.
Hill and Hoffman have investigated angle of incidence effects
where there is a dramatic increase in the tendency to form films in
tension near 45 incidence even under oxidizing conditions (Fig.29).
At the same time and enhanced compressive change results upon expo-
sure to the atmosphere. These results are in general agreement with
the earlier ones of Priest et al.(I21) in which the stress anisotro-
py was measured using two orthogonal substrates, and again emphasize
the sizeable morphological changes at modest angles of incidence.
Hodgkinson and Walker(210) have irradiated SiD films with ultra-
violet and found that prolonged irradiation in vacuum will change
the initial compression to tension, but the stress reverts back to
compression when air is admitted. Films irradiated in air also had
a substantial reduction in the compressive stress, and related chan-
ges in retractive index and UV optical absorption. The stress chan-
ges under irradiation are associated with the rearrangement of oxy-
gen already gettered during the deposition with a resultant tensile
contribution which partly counteracts the existing compression.
Although the data for other optical film is less complete, we
suggest that the purer the film the greater the tendency for tension,
and oxygen or other impurities tend to give compressive contributions.
The intrinsic stress in alkali halide films has been investiga-
ted by Carpenter and Campbell. (95) Data in the form of force and ave-
rage stress as a function of thickness are presented for many of the
alkali halides. Lithium salts show tension and potassium salts com-
pression throughout the thickness range covered. Sodium salts have
a large compression below a thi'ckness of 1000 A and a tension above.
The common features exhibited by all the curves are a stress maximum
which is usually compressive and occurs at a thickness of about 250 A
and a constant stress region for thicknesses greater than about
1000 A, with a relative insensitivity to deposition rate. LiF has
been studied by a number of investigators as summarized in Table II.
Carpenter and Campbell point out that there appear to be two diffe-
rent stress producing mechanisms and suggest a misfit model for the
first stage, even though the substrates were soda-lime glass. The
average stress as a function of lattice constant as plotted in Fig.
30 goes through zero, at the oxygen nearest-neighbor distance in the
glass. Although surface tension contributions may contribute to the
stress in this isolated island stage of growth it is not possible to
explain the systematic behavior shown in the figure on this basis.
No satisfactory explanation has been made for the intrinsic stress
in thicker films.
332
4
'"
",
3
v
-
C
I:
2
-t
IDg
rJ)
rJ)
~
t-
O
~
U.I
U>
~
-I

&a.I
>

-2
-3

Na.F
4
5
BULK
R.W. HOFFMAN
~ I
I
Ke,t-"
7

~
I
I
I
I
Fig. 30 Average stress in very thin alkali halide films correlates
with epitaxial effect on glass substrates. After Carpenter
and Campbell. (95)
Rather than to try to review the entire literature relative
to epitaxial films, we consider a few illustrative examples. When
the film is very thin or consists of small nuclei, a uniform elas-
tic strain is predicted and has been observed for metals. The in-
troduction of misfit dislocations to reduce the total energy loca-
lizes the strain at the interface. This prob lem has been treated
in various formulations by Van der Merwe and his colleagaes and
discussed earlier. However, the fact that many epitaxial films are
deposited at high temperatures and then observed at room tempera-
ture means that the thermal stresses are also often very large even
though care is usually taken to minimize the difference in expansion
coefficients. Thus, although the stress gradient near the interface
may be especially important in determining properties which are sen-
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS
333
sitive to that region it is often not the major contribution to
the total ctress in a thicker film. One should really ask the ques-
tion, "are there any data for the stress in epitaxially grown films
that require the presence of a volume intrinsic stress"? This is
a difficult experimental question to answer. First of all, the in-
trinsic stress is generally a decreasing function of the substrate
temperature and is likely to be very small at elevated temperatures.
And, secondly, relaxation processes undoubtedly operate which may
prevent an intrinsic stress, if generated, from being observed.
Dumin,(123) in the case of silicon on sapphire was not to
obtain the necessary precision. Other have
also found large compressive stresses dynes/cm
2
) for the
Si/A1203 system. Kamins and Meieran find if the epitaxial films
are electrolytically stripped from the substrates the constraint
is lifted and the strain is reduced to a level equal to the resolu-
tion limit of the diffractometer technique. In very case, thermal
expansion is sufficient to account for the compression. Dumin does
not mention any stress anisotropy, Kamins and Meieran, using a
diffractometer technique, mention but cannot measure an anisotropy,
and Hughes and Thorsen calculate the stress anisotropy resulting
from the anisotropic thermal expansion and confirm it by mobility
measurements.
In the case of silicon on spinel, larger compressive stresses
were observed. Schlotterer(77) found isotropic stresses while'Ro-
binson and Dumin(167) notice anisotropic deformations, again with
the magnitude close to that expected from a thermal origin. The
recent stress birefringence measurements of Rinehart and Logan(73)
on Alx Ga(l-x)/GaAs structures also indicate that the thermal ex-
pansion stress dominates the room temperature stress situation.
The epitaxial garnet films also respresent a well-studied sys-
tem because of their use in bubble devices. Zeyfang,(12S) using a
double crystal diffractometer concluded the stress opera-
ted in YIG on YAG.
Many examples are presented by Besser et al.,(96) clearing up
previous inconsistencies in understanding the crazing associated
with these films. Broginski et al.,(98) conclude the mismatch
stresses are of prime importance for YID films on Gd3-xDYxGaSU12
substrates as result of magnetic anisotropy field measurements.
They propose a balance between the thermal expansion stresses and
the misfit dislocations to explain the almost zero stress at large
values of x. As described earlier Carruthers(97) has extended the-
se concepts to a partial relaxation.
In summary, thermal and misfit contributions seem sufficient
to explain epitaxial stresses for the cases in the present litera-
ture.
334 R.W. HOFFMAN
In order to account for the bending that deposited epitaxial
films undergo after separation form the substrate, Abrahams et al.,
(III) had suggested an oriented array of inclined dislocations.
For films for GaAsl-
x
on GaAs, they found a correlation between
the number of inclined dislocations, the lattice mismatch, and the
amount of bending. An alternative explanation of this well-known
bending phenomenon was attributed to the misfit dislocations near
the .interface. As a stress gradient through the thickness of the
film is needed to account for the curling, the criticism of Abra-
hams et al., that misfit dislocations shoud apply only to 100 A
layers is not valid.
10
9
I
o
- I
-2 ./ THERMAL
x' OXIP
RICH
.......... SiH ... NO NHI
PIOCSS
TENSION
1
I
COMPIU'SSION
o 0.1
0.2 0.1 0.1- 0.5 0.& 0.7 0.9
0.9 1.0

hN+ "0
Fig. 31 Average stress in silicon oxynitride films as a function
of nitride composition. From Rand and Roberts. (130)
MECHANICAL PROPERTIES OF NON-MHALLIC THIN FILMS 335
We consider now the last category, namely, films used for in-
sulating, pas s ivating , or isolation purposes. This area is, of
course, of extreme importance to the technology of thin films.
We regard the electrical effects of such films beyond the scope of
this article and content ourselves with some general considerations
concerning the mechanical properties. There is a review by Pliskin
et aL (I26) on thin glass films and Dell f oca et al. consider
anodic layers. As the glass film are generally applied on heated
substrates, and glass as a material is stonger under compression,
one desires the total stress in the film to be compressive at the
operating temperature. Thermally grown Si02 has perhaps received
the most attention. But here the substrate is at high temperatures
during the formation of the film and there is only modest control
over the resultant compressive stress (rv 3 x 10
9
dyne/cm2) as the
temperature range to form the oxide is limited, and wet or dry oxi-
dation makes only a small difference. (51, 127). Most authors report
no thickness dependence in the range of 0.5 - 2 However, a
rapidly increasing comprrssion at small thicknesses is found by Lin
and Pugacz-Moraszkiewiez 53) in the range of Thermal
mismatch is recognized as the origin for thermally grown oxides,
but intrinsic mechanisms are reported for CVD Si0
2
films deposited
at somewhat lower temperatures(148) when the total stress is about
2 x 109 dyne/cm2 tension. Post deposition heat treatments change
the stress back to compression. A tensile intrinsic stress is also
found in the case of thermal decomposition of tetraethylorosilica-
tes.(I49)
More recently Si3N4(128,129) has been used as a passivation
layer for Si. In this case an extremely large intrisic stress of
> I x 10
10
dyne/cm2 tension is found coupled with thermal stresses
of perhaps 10% of the total. The stress does not change with time.
Rand and Roberts(130) report amorphous silicon oxynitride films
that range in tension of Si3N4 as the nitride concentration increa-
ses as shown in Fig. 31. These large stresses in both Si02 and
Si3N4 can damage the substrate as we shall see in the next sections.
Although the intrinsic stress rapidly decreases with increa-
sing substrate temperature in metals, we see that substantial in-
trinsic stresses are present for CVD insulation films.
We try to summarize the literature for non-metallic films as
follows
I. Although thermal contributions may be large, intrinsic
stresses are well documented in both polycrystalline and epitaxial
films.
2. Real stress contributions have been seen during the
lated island stage of growth, but these contributions should no
336 R.W. HOFFMAN
longer dominate at average film thicknesses of perhaps two hundred
angstroms and are often masked by poor thermal control in the ex-
periment.
3. There is a tendency for stress values to be tensile in
polycrystalline and independent of thickness when the deposition
parameters are well defined. This suggests that the major contri-
bution to the intrinsic stress in thick films is a volume effect
and not an interfacial one. Further, for pure films the stress
should be capable of an explanation by similar models as those de-
veloped for metals.
4. Compressional contributions are associated with impurities,
often oxygen, incorporated in the structure by design or default.
These may take place during deposition, or afterward by diffusion.
Non-stoichiometric films may also show compression.
S. The decrease in the average stress as the thickness increa-
ses may result from a relief process or a change in the intrinsic
stress mechanism. The present measurements are not sufficiently com-
plete to generalize, but on energetic grounds a relief mechanism
must ultimately operate. Cracking in film or damage to the substrate
may result.
6. Both thermal and misfit contributions operate in epitaxial
films. The intrinsic contributions are amenable to calculation,
and may be either a uniform elastic strain or localized at the sub-
strate interface.
6. CONTROL OF STRESSES, INCLUDING RELAXATION EFFECTS
One of the objectives of understanding the intrinsic stresses
in films is to be able to control the stresses and hence the pro-
perties for a given application. As we have seen, considerable pro-
gress has been made in the case of epitaxial systems. Recently for
the case of metals stress control has come about by the use of
bias sputtering. Blachman.(ISO,ISI) in the case of molybdenum and
aluminium, has found a correlation between the stress resistivity
and trapped argon content. With increasing negative bias substan-
tial amounts of gas are entrapped which reduce the tension in the
film or may even change it to compression. On the other hand, Sun
et al.,(167) for the case of sputtered tungsten, find a sizeable
compression which is not correlated with the argon concentration
In a following paper(209) they find a strong correlation with fi-
nal grain size in the W films. The structure and stress modifica-
tions by ion bombardment during deposition studied by
Maddox and his co-workers. (168,169,170) ) has used the
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 337
cantilever method to measure the stress in a number of metals
sited by low pressure triode sputtering. Date for the is
not so common, but it appears that one can change the bias to subs-
tantially modify the stresses even in these systems. We also call
attention to the UV irradiation effects mentioned earlier(210).
The use of elevated substrate temperatures to reduce the intrin-
sic stress,among other things, has been done for a long time. In ad-
dition it is sometimes possible to provide a thermal contribution
which gives cancellation with intrinsic stress mechanism by the pro-
per choice of substrate.
Post-deposition treatment may reduce the stress, as it has
been known for a some in films of metals deposited at lower sub-
strate temperatures that stress relief mechanisms operate upon
heating the substrate to elevated temperatures. (171). Resistivity
and stress changes have been correlated and activation energy mea-
surements made(172) which indicate that very likely a simple defect
Fig. 32
200
600 1000
5000
GRAIN DIAMETER
Critical stress to operate a dislocation source as a func-
tion of grain diameter(21) for several inclinations. After
Chaudhari.
338 R.W. HOFFMAN
motion at lower temperatures is followed by recrystallization.
Such irreversible changes upon initial heating are usually
followed by reversible behavior upon temperatures cycling at lower
temperatures. Chaudhari(21) has considered the mechanisms of stress
relief in polycrystalline films. Dislocation sources within a grain
at a grain boundary or at surfaces are not likely mechanisms be-
cause the stress required to operate such dislocation sources is
larger than the usual intrinsic stresses within a film. To a first
approximation the stress necessary to operate sources varies in-
versely with the diameter of the grain as shown in Fig. 32. A si-
milar result is observed for sources at a grain boundary edge,
out since the intrinsic stresses are the order of 10-2 of the shear
modulus, these p::::cesses are not normally favored. Kinetics equations
for the exponential relaxation of the stress on a grain boundary
sliding model were also derived by Chaudhary. At most grain boun-
daries in a film are perpendicular to a free surface under a planar
biaxial stress the shear stress acting on the boundary is small and
hence, the plastic strain contribution from grain boundaries sliding
is expected to be small. The annealing kinetics have been derived
for diffusional creep for the cases of volume diffusion and grain
boundary diffusion predominating. In this case the thickness of the
film matters and of course the activation energies for the two hypo-
theses are different. Although these equations were derived for the
case of a uniform stress, it is pointed out that stress
also provide an additional driving force. If a stressed film is
covered by a diffusion barrier containing a hole, the film
its elastic strains by a flow of matter between the surface and
interior of the film. Mass flow out of the film under a compressive
stress and into a film under tensile stress, leaving a hillock or
depression. Pennabacker(173) has also considered that compressiue
stresses lead to hillock growth. Chaudhari(174) has recently calcu-
lated the hillock density and growth kinetics on the basis of a
local relaxation of a compressive stress.
For the case of growing oxide films, Stringer(82) has conside-
red additional mechanisms for plastic flow. The effect of stoichio-
metry on the oxide plasicity and the reasons for suggsting plastic
flow takes place during oxidation are discussed. In addition,
vacancy injection and dislocation generation at the metal sub-
strate are examined. A more direct measurement of stress enhanced
diffusion in thin films was given by Gangulee(175) in which Be atoms
diffuse along grain boundaries of Al films and along the interface
between the film and oxidized silicon substrate at a considerably
enhanced rate when a tensile stress is applied to the plane of the
film.
Diffusion may also be a source of Takai and Francombe
(176) have measured a cantilevered deflection during the interdif-
fusion of gold and aluminium both an increase in stress correspon-
ding to the growth of intermetallic compound faces and stress re-
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
lief were noted. Tu et al.,(137) have studied the Ni-Si system
and Lau and Sun(177) the interdiffusion of Ti-Pd-Au Films.
339
Similar cases of enhanced migration are commonly found in the
recrystallization of amorphous layers of silicon which are produ-
ced by ion bombardment of an aluminium coated silicon substrate.
Silicon is transported to the surface of the thin aluminium sur-
face according to Hart, et al.(178,179) Additional low temperature
rapid migrations have also been found, and this subject has been
treated in a recent conference, (180) in which both review papers
and research reports will be found.
Helium implanted erbium films examined by Blewer and Maurin
(181) have shown dimensional expansion at the surface and micro-
scopic bubbles formed at the surface by release of helium by the
films. Additional dilation data is found in the semiconductor li-
terature.
Stress annealing data for non-metallic films is not common.
In the case of silicon monoxide Priest and Caswell(146) showed
that no stress change took place with a low temperature anneal in
vacuum. But as soon as the pressures of oxygen or water vapor ex-
ceeded about 10-
3
Pa ,rapid compressive changes took place. This
is, of course, related to the degree of oxidation as was discussed
earlier and we would anticipate that the changes often found when
optical films are exposed to the atmosphere have a related origin.
Mattox and Kominiak(168) annealed bias sputtered Corning 1720 glass
films at low temperatures. The initial compressive stress became
larger, and at temperatures above about 350
0
the films showed a
total tension. This irreversible behavior was again traced to an
oxygen deficiency.
At the present times, it is well documented that grain boundary
diffusion plays an a:lmost dominant: role in sma],l-grai.ned films,
even though both lattice and diffusion down dislocation pipes may
take place in special circumstances of large grain size, or extreme
dislocation concentrations. Nevertheless, the detailed kinetics,
especially in the case of impurity diffusion, are the subject of
present investigations. The resultant stress relaxation are less
well understood, and even the detailed mechanisms are still in
dqubt for metals, (3) and almost unknown for non-metals.
7. MODULI, FAILURE MODES AND ADHESION
We should comment on the values of elastic modulii for deposi-
ted films. Although occasionnally low values are found, as reviewed
by Hoffman,(4) most films when carrefully measured show the expec-
340 R.W. HOFFMAN
ted values of the modulii. In addition to direct deflection measu-
rements, resonance techniques(108,182,183) have been used. Spinner
(184,185) used a dynamic technique to measure the modulii of many
glasses as a function of temperature. Uozumi et al.,(186) have used
the pulse-echo technique tlf sound velocity to det"ermine the modu-
lii. In view of the present interest in amorphous solids we call
attention to the work by Chen and Wang, and emphasized by Berry
and Pritchet(182) there is a growing body of evidence that Young's
modulus for materials in the amorphous state may be characteristi-
cally 20-40 % below the value in the crystalline state. Merz et al
(211) found a 45% decrease in the shear constant of Sm2 Co17 sput-
tered alloy, but only a 7% decrease in the bulk modulus compared
to the crystalline phase. These measurements support the ideas of
Weaire et al (212) that microscopic internal movements take place
in amorphous materials under shear. We would expect similar chan-
ges in the amorphous layer induced in silicon by ion implantation.
(188). Internal friction measurements have given information as to
loss mechanisms and thermal constants(182,186).
Bunshah, (189) has reviewed the properties of evaporated thick
films and correlated the mechanical properties of thick, primarly
metallic, films were reported in the Conferences on Structure/
Property relationships in Thick Films and Bulk Coating. (190)
Kinosita(9) has reported unexpected microhardness
for LiF thin films with a Vicker-type indenter, and Winter(191) has
developed apparatus for similar measurements using both and inden-
ter and impact from spherical projectiles.
We consider now an outline of the failure modes, which may
range from a mechanical distorsion of an optical surface,(152)
through drastically changed properties, especially near the inter-
face as indicated in several examples. We concentrate here on the
obvious mechanical failures of cracking in the film or substrate
and buckling if the film has a large compressive stress.
The importance of the adhesion in preventing failure is well
known, although quantitative treatments are difficult to find.
Hunt and Gale(192) suggested a model of the adherence to stresses
in plastic film, in which the lift-off stress normal to the plane
of the film is proportional to the gradient of the tension in the
film substrate interface. The importance of the edges of the films
or other defects which may be present, is well known and can be ex-
plained by the non-uniform stresses near the edges of the film as
we mentioned earlier. Plassa(I94) has also studied the elastic ins-
tability of germanium films as they buckle form a mica substrate.
Random wrinkles are found with an isotropic intrinsic stress al-
oriented wrinkles were found earlier by Yelon and Voegeli
(194) in the case of epitaxial films and a suggestion for the sinu-
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 341
soidal wrinkle pattern in terms of a colunm instability has been
made by Plassa and Chopra(II). As the stress is generally a volume
effect, the total force that must be supported across the interfa-
ce increases as the film becomes thicker. This leads to a critical
failure thickness which is a function of the stress in the film
and the adhesion to the substrate. The buckling, of course, results
from a film under high compression whereas a cracking will be found
for films under sufficiently high tension that the fracture stress
is exceeded.
Angle of incidence effects may also produce oriented buckled
patterns observed in ZnS by Behrndt(195) which were eliminated by
rotating substrates, and we discussed earlier the extreme depen-
dence in the SiO system. Recent emphasis on deposition techniques
other than evaporation have eliminated many of these problems.
Dislocation generation and even cracking as a result of the
residual elastic strains and epitaxial films is well documented.
In the galium arsenide photodiode heteroepitaxy lattice paremeter
mismatch had to be graded at a slow rate to avoid excessive dislo-
cation generation. We referred earlier to the stress by refringence
and cracking in the magnetic bubble materials. Carruthers(97) has
treated the deformatjon and creep as well as estimating the proper-
ty changes. It was known that thin dielectric films of the order
of 1 ~ may accomodate the high stresses normall, found, thicker
films are found to crack. Matthews and Klokholm( 96) have consider-
ed the fracture of these brittle films under the influence of the
misfit stress. The cracks that form in the garnet films are per-
pendicular to the film plane. If such cracks are top!opagate
through the film, which is rigidly bonded to its substrate and
strained tension, then, the film thickness mus exceed the Griffith
crack length. If the stress exceeds the fracture strength of the
crystal then, a spontaneous fracture will take place, whereas, for
a low stress the film may strain elastically and be stable. These
results are summarized in Fig. 33 as a function of the mistfit.
Indeed, the presence of cracks in a film indicates that the density
of misfit dislocations lies below the optimum value and indicates
there are difficulties associated with the generation of these
dislocations.
Si02 or Si3N4 films are known to develop dislocations or
cracks in the underlying silicon substrates. Kato et al., Westdorp
and Schwuttke,(128) and Tamura and Sunami(129) have considered this
problem in detail. The internal stresses were measured, isothermal
annealing treatments were carried out, and the density of disloca-
tions introduced during the high temperature anneal were measured.
Dislocations in the garnet system(have been studied in a series of
papers by Matthews and Klockholm. 197)
342 R.W. HOFFMAN

10-t.
It-

u:
!!!
z:
10-
5

10-3 10.
1
ro-
I
10
THICkNESS, 't <AM)
Fig. 33. Crack generation in brittle films as a function of mis-
fit and thickness. Above B film spontaneously cracks, below C ,it
is elastically strained and above A cracks will propagate. After
Matthews and Klockholm.(196)
We finally consider adhesion. Much has been said about adhe-
sion in a qualitative way, but to date, the field still suffers
from good measurement techrriques. Campbell(7) has reviewed this
area well and annual conferences of the subject have been pub li-
shed(198) in Aspects of Adhesion. Techniques in cleaning, deposi-
ting thin reactive metal layers to ai& bonding to glass, interdif-
fusion, and grading the composition all have been used to enhance
the adhesion. Nevertheless, a detailed understanding of even the
origin of the forces is obscure, although Van de Vaals coupling is
suggested as the most important contributor with band structure
effects being important in clean metal surfaces. (199,200)
The stress distributions iy an adhesive layer have been calcu-
lated by Harrison and Harrison. 201)
We leave the discussion of the most recent measurement techni-
ques to the recent reviews by Chapman(202) and Kendall(203) and
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS 343
the forthcoming review by Weaver(204) and mention here only depo-
sition techniques which should prove useful as well as a brief re-
port of the surface analysis techniques that may provide some of
the quantitative information needed for an understanding.
!t has long been known that sputtered films generally have
better adherence to a substrate than to evaporated ones. The higher
energy of the arriving atoms allows a small penetration and /
or a cleaning of the surface. The ion plating technique as pione-
ered by Mattox carries this process even further by SUbjecting the
substrate to a flux of high energy ions before and during the film
deposition. From the point of view of adhesion, several benefits
may be obtained. The surface will be sputter cleaned and maintened
clean until the film begins to form. The high energy flux to the
substrate surface provides a high effective surface
enhancing diffusion and chemical reaction. The high defect concen-
tratiCln also provides a physical(admj.xing of the film and substra-
te material. According to Mattox 205) this process gives rise to
sizeable gas incorporation as the negative bias is increased, the
gas entrapment decreases again at biases above a few 100 volts,
presumably because the high temperature of the deposit allows the
gas to diffuse away. Densities of as much as a factor of 2 lower
than bulk densities are observed, and the growth morthology is de-
cidedly influenced. As far as the stress is concerned an increase
in intrinsic compressive film stress is generally found with in-
creasing negative bias, for metal films.
More recently experimental techniques of AES have been applied
to the adhesion question. Staoddart et al.,(206) have studied the
effect of the glow discharge on adhesion. Although no surface e-
lectrical or topographical effects were seen, gross contamin2tion
as well as gas sorption were felt to be important. Houston and
Bland(207) found that the discharge current could act as a process
control variable to indicate a clean cathode sqrface, if proper
care was taken. Westwood and Bennewitz(208) in reactively sputte-
red PTO films noted that the presence of oxygen was necessary for
good adhesion.
It is premature to generalize from the results but it appears
as though certain impurities in small concentrations at the inter-
face are beneficial and, in fact, even needed for good adhesion.
Sundahl(193) found the presence of eu and Si on the surface of
high purity Al203 as well as a small grain size in the substrate
were important factors. With the increased availability of surface
analysis techniques we expect significant progress in understan-
ding adhesion.
344 R.W. HOFFMAN
ACKNOLEDGEMENTS
I would like to dedicate this paper to my students, both past
and present, who have taught me so much about the mechanical pro-
perties. I also wish to thank P. Chaudhari, D. M. Hoffman, A. Kin-
bara, K. Kinosita, E. Klokholm , S. Mader, E. Ritter, and W. D.
Westwood for their helpful discussions and information. Mrs. M.
Young and R. Wentz have my gratitude for their faithful efforts
in production of the manuscript. The research at Case Western Re-
serve University was supported by the U. S. Atomic Energy Commis-
sion.
REFERENCES
1. See for example : E. Klokholm, J. Vac. Sci. Tech. !, 148
(1971)
2. M. H. Francombe, A. J. Noreika, W. J. Takei, and S. Y. Wu,
J. Vac. Sci. Tech. ll, 130 (1974).
3. A. Gangulee, Acta Met. ~ 177 (1974).
4. R. W. Hoffman, in "Physics of Thin Films" (G. Hass and R. E.
Thun, eds.), Vol. 3. Academic Press, New York (1966).
5. R. W. Hoffman in "Thin Films", H. G. F. Wilsdorf, ed. A. S. M.
(1964)
6. D. S. Campbell, in "Basic Problems in Thin Film Physics"
(R. Niedermayer and H. Mayer, eds), p. 223. Vandenhoeck and
Ruprecht, Gottingen (1966).
7. D. S. Campbell, in "Handbook of Thin Film Technology",
(Maissel and GIang, eds.), p. 12-1. McGraw-Hill, New York
(1970)
8. W. Buckel, J. Vac. Sci. Tech. ~ 606 (1969).
9. K. Kinosita, Thin Solid F i l m s ~ 17 (1972).
10. R. J. Scheuerman, in "Symposium on Deposited Dielectric Thin
Films", (F. Vratny, ed.), p. 561. Electrochemical Society,
New York (1969).
11. See, for example: K.. L. Chopra, "Thin film Phenomena",
McGraw-Hill, New York (1969) and, L. Maissel and M. H. Fran-
combe, "Introduction to Thin Film Physics". Gordon and Breach,
New York (1973).
12. J. F. Nye, "Physical Properties of Crystals", Oxford Univ.
Press, London (1957).
13. C. S. Smith, in "Solid State Physics", Vol. 6, F. Seitz and
D. Turnbull, eds. Academic Press, New York (1958).
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
14. R. W. Vook and F. Witt, J. Appl. Phys. 36, 2169 (1965).
IS. J. Turley and G. Sines, J. Appl. Phys. !l, 3722 (1970).
16. T. Y. Thomas, Proc. Nat. Acad. Sci. (U.S.) 235 (1966).
17. A. Brenner and S. Senderhoff, J. Res. Natl. Bur. Stand. 42,
89 (1949).
18. F. A. Doljack, AEC Tech. Rept. 76, Case Western Reserve
University, Cleveland, Ohio (1971).
345
19. E. Klockholm, in "X-ray Diffraction and Stress in Thin Films
Symposium", IBM Thomas J. Watson Research Center, Yorktown
Heights, New York (March 1969).
20. G. C. Stoney, Proc. Roy. Soc. (London) A32, 172 (1909).
21. P. Chaudhari, IBM J. Res. Develop. 12, 197 (1969).
22. H. J. Del and V. D. Frechette, J. Am: Ceram. Soc. 50, 542
(1967) .
23. J. Aleck, J. Appl. Mech. 118 (1949).
24. K. Haruta and W. J. Spencer, J. Appl. Phys. 1l, 2232 (1966).
25. R. W. Hoffman in "The use of Thin F:'lms in Physical Investi-
gations", J. C. Anderson, ed. Academic Press, New York (1966).
26. R. W. Hoffman in "Measurement Techniques for Thin Films"
Schwartz and Schwartz, eds. Electrochemical Society (1967).
27. D. G. Ashwell and E. D. Greenwood, Engineering 170, 51 (1970).
28. D. G. Bellow, G. Ford, and J. S. Kennedy, Exp. Mech. 227
(1965).
29. R. E. Rottmayer and R. W. Hoffman, J. Vac. Sci. Tech. 1, 461
(I 970).
30. R. W. Springer, AEC Tech. Rept. 79, Case Western Reserve
University, Cleveland, Ohio (1972). '
31. J. D. Finegan and R. W. Hoffman, Trans. 8th Vacuum Symposium,
p. 935. Pergamon Press (1962).
32. F. A. Doljack and R. W. Hoffman, Thin Solid 71
(1972).
33. P. H. Robinson and D. J. Durnin, J. Soc. 75
(I 968).
34. P. M. Alexander, AEC Tech. Rept. 85, Case Western Reserve
University, Cleveland, Ohio (1974).
35. K. Maki and K. Kinosita, Reported in K. Kinosita Proe. Second
Colloquium on Thin Films, p. 31 (1967).
36. E. Yoda, J. Appl. Phys. 1355 (1969).
346 R.W. HOFFMAN
37. E. Yoda, J. Appl. Phys. 1355 (1969).
38. Y. Namba, Oyo Buturi 38, 411 (1969).
39. R. E. Rottmayer, AEC Tech. Rept. 64, Case Western Reserve
University, Cleveland, Ohio
40. R. W. Springer and R. W. Hoffman, J. Vac. Sci. Tech. !Q,
238 (1973).
41. R. E. Rottmayer and R. W. Hoffman, J. Vac. Sci. Tech.
152 (I 971) .
42. A. E. Ennos, Appl. 51 (1966).
43. A. Taloni and D. Hanneman, Surface Sci. 323 (1967).
44. A. G. Blachman, Metallurg. Trans. 699 (1971).
45. K. A. Haines and B. P. Hildebrand, Appl. Optics 595 (1966).
46. R. J. Magill and T. Young. J. Vac. Sci. Tech. 47 (1967).
47. R. GIang, R. A. Hotmwood, and R. L. Rosenfeld, Rev. Sci.
Instr. 7 (1965).
48. N. N. Axelrod and H. J. Levinstein, unpublished.
49. F. P. Chiang, C. S. Faber, and F. Y. Wang, J. Appl. Phys.
42, 1422 (1971).
50. J. W. Beams, in "Structure and Properties of Thin Films",
C. A. Neogebauer, J. D. Newkirk, and D. A. Vermilyea, eds.
p. 183. Wiley, New York (1959).
51. R. J. Jaccodine and W. A. Schlegel, J. Appl. Phys.
2429 (I 966) .
52. C. H. Lane, IEEE Trans. Electron. Dev. ED-15, 998 (1968).
53. S.C.H. Lin and I. Pugacz-Muraszkiewicz, J. Appl. Phys. 43,
119 (I 972) .
54. C. W. Wilmsen, E. G. Thompson, and G. H. Meissner, IEEE
Trans. Electron. Dev. ED-19, 122 (1972).
55. P. M. Schaible and R. GIang "Symposium on Deposited Dielec-
tric Thin Films (F. Vratny, ed.), p. 577, Electrochemical
Society, New York (1969).
56. G. Zosi, Z. Angew.Phys. 24, 322 (1968).
57. G. W. Bush and H. J. Read, J. Electrochem. Soc. l!l, 289
(1964) .
58. A. Taylor, "X-ray Metallography", Wiley, New York (1961).
59. T. I. Kamins and E. S. Meieran, J. Appl. Phys. 44, 5065
(1973) .
60. B. D. Cullity, J. Appl. Phys. 35, 1915 (1964).
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
61. B. Borie, Acta Cryst. 11,542 (1960).
62. B. Borie, C. J. Sparcks, and J. V. Cathcart, Acta Met. lQ,
691 (1962).
63. C. B. McDowell and T. C. Pilkington, J. Appl. Phys. 42,
2958 (I 971) .
64. A. Authier, "Proc. of the 15th Annual Conf. on Application
of X-ray Analysis" (J. B. Newkirk and G. Mallet, eds.),
(1967) .
65. E. S. Meieran and I. A. Blech, J. Appl. Phys. 36, 3162
(1965).
66. I. A. Blech and E. S. Meieran, Appl. Phys. Letters 245
(1966).
67. G. H. Schwuttke and J. K. Howard, J. Appl. Phys. 39, 1581
(1968)
347
68. P. Penning and D. Polder, Philips Res. Rept. 419 (1961).
69. R. Zeyfang, J. Appl. Phys. 42, 1182 (1971).
70. G. A. Rozgonyi and T. J. Ciesielka, Rev. Sci. Instr. 44,
1053 (1973).
71.
72.
73.
E. P. EerNisse, J. Appl.
E. P. EerNisse, J. Appl.
F. K. Reinhart and R. A.
(1973) .
Phys. 43, 1330 (1972)
Phys. 44, 4482 (1973).
Logan, J. Appl. Phys. 44, 3171
74. R. W. Vook and F. Witt, J. Vac. Sci. Tech., 2 49 (1965).
-
75. R. Zeyfang, Solid State Electron. li, 1035 (1971).
76. R. Feder and T. B. Light, J. Appl. Phys. 43, 3114 (1972).
77. H. Schlotterer, Solid State ll, 947 (1968).
78. A. J. Hughes and A. C. Thorsen, J. Appl. Phys. 44, 2304
(1973)
79. A. C. Thorsen and A. J. Hughes, Appl. Phys. Letters ll,
579 (1972).
80. A. J. Hughes (unpublished).
81. Y. Nakajima and K. Kinosita, Thin Solid R5 (1970).
82. J. Stringer, Corrosion Sci. lQ, 513 (1970).
83. F. N. Rhines and J. S. Wolf. Reported in Ref. 82
84. C. J. Dell'Oca, D. L. Pulfrey, and L. Young, in "Physics
of Thin Films" (M. Francombe and R. W. Hoffman, eds.).,
Vol. 6. Academic Press, New York (1971).
348 R.W. HOFFMAN
85. L. Young. "Anodic Oxide Films". Academic Press. New York
(1961).
86. D. A. Vermilyea. in "Advances in Electrochemistry". Vol. 3.
Wiley (Interscience). New York (1963).
87. J. H. Van der Merwe. in "Single Crystal Films". (M. Francom--
be and H. sato. eds.). Press. Oxford (1964).
88. J. H. Van der Merwe and N. G. Van der Berg. Surface Sci.
32. I (1972).
89. J. H. Van der Merwe. Proc. 6th International Vacuum Con-
gress (to be published).
90. J. W. Matthews. in "Physics of Thin Films" (G. Hass and
R. E. Thun. eds.). Vol. 4. Academic Press. New York (1967).
91. J. W. Matthews. S. Mader. and T. B. Light. J. Appl. Phys.
3800 (1970).
92. w. A. Jesser and D. Kuhlmann-Wilsdorf. J. Appl. Phys. 38.
5128 (1967).
93. W. A. Jesser and D. Kuhlmann-Wilsdorf. Phys. Stat. Sol.
95 (J 967)
94. K. Yagi. K. Takayanagi. K. Kobayashi. and G. Honjo.
J. Cryst. Growth 2.. 84 (1971) .
95. R. Carpenter and D. S. Campbell. J. Mater. Sci. 173
(J 967) .
96. P. J. Besser. J. E. Mee. P. E. Elkins. and D. M. Heinz.
Mat. Res. 1111 (1971).
97. J. R. Carruthers. J. Cryst. 45 (1972).
19
98. A. J. Braginski. T. R. Oeffinger. W. E. Kramer. D. K. McLain.
and W. J. Takei. IEEE Trans. Mag. 300 (1972).
99. N. Cabrera. Surface Sci. 320(1964).
100. C. Herring. "Structure and Properties of Crystal Surfaces".
(R. Gomer and C. S. Smith. eds.). Univ. of Chicago Press
Chicago. Illinois (1953)
10J. R. G. Lindford and L. A. Mitchell. Surface Sci.
Y....
142
(1971) .
102. H. J. Wasserman and J. S. Vermaak. Surface Sci.
E.
168
(1972).
103. R. Ritter. in "Physics of Thin Films" (G. Hass. M. Francombe
and R.W. Hoffman. eds) Vol. 8 Academic Press. N.Y. (1975)
104. P. A. Young. Thin Solid Films 423 (1970).
105. E. Klokholm and R. Berry. J. Electrochem. Soc. 823 (1968).
MECHANICAL PROPERTIES OF NONMETALLIC THIN FILMS
106. S. Mader, J. Vac. Sci. Rech. !l, 131 (1974).
107. P. Chaudhari, J. Vac. Sci. Tech. 520 (1972).
108. J. D. Wilcox, D. S. Campbell, and J. C. Anderson, Thin
Solid Films 1, 13 (1969).
109. A. Yelon, J. R. Asik, and R. W. Hoffman, J. Appl. Phys?
33, 949 (1962).
110. D.O. Smith, M. S. Cohen, and G. P. Weiss, J. Appl. Phys.
l!, 1775 (1960).
349
111. M. S. Abrahams, L. R. Weisberg, and J. J. Tietjen, J. Appl.
Phys. 40, 3754 (1969).
112. K. Saito, R. O. Bozkurt, and T. Mura, J. Appl. Phys. 43,
182 (1972).
113. J. Reisenfeld and R. W. Hoffman, AEC Tech. Rept. 39
Case Institute of Technology, Cleveland, Ohio (1965).
114. A. F. Turner, "Thick Thin Films", Bausch & Bomb Tech. Rept.
(1951) .
115. O. S. Heavens and S. D. Smith, J. Opt. Soc. Am. 469
( 1957) .
116. H. Blackburn and D. S. Campbell, in "Trans. 8th Nat.
Vacuum Symp." p. 943. Pergamon Press, Oxford (1961).
117. K. Kinosita, K. Nakamizo, K. Maki, K. Onuki, and K. Takenchi,
J. Appl. Phys. (Suppl. 1), 340 (1965).
118. A. Barna, P. B. Barna, J. F. Pocza, and I. Pozsgai, Thin
Solid Films 201 (1970).
119. A. Preisinger and H. K. Pulker, private communication
(to be published).
120. H. K. Pulker and E. Jung, Thin Solid 57 (1971).
121. J. R. Priest, H. L. Caswell, and Y. Budo, J. Appl. Phys.
34, 347 (1963) .
122. A. E. Hill and G. R. Hoffman, Brit. J. Appl. Phys.

13 (1967) .
123. D. J. Dumin, J. Appl. Phys. 36, 2700 (1965).
124. C. Y. Ang and H. M. Manasevit, Solid State Electron. 8
994 (1965).
125. R. Zeyfang, J. Appl. Phys. 3718 (1970).
126. W. A. Pliskin, in "Physics of Thin Films" (G. Hass and
R. E. Thun, eds.), Vol. 4. Academic Press, New York (1969)
127. M. V. Whelan, A. H. Goemans, and L.M.C. Goosser:s, Appl.
Phys. Letters iQ, 262 (1967).
350 R.W. HOFFMAN
128. W. A. Westdorp and G. H. Schwuttke, in "Symposium on
Deposited Dielectric Thin Films" (F. Vratny, ed.), p. 546.
Electrochemical Society, New York (1969).
129. M. Tamura and H. Sunami, J. Appl. Phys. Japan II, 1097
(1972). -
130. M. J. T'.and and J. F. Roberts, J. Electrochem. Soc.
131 H. Blackburn and D. S. Campbell, Phil. Mag. 823 (1963).
132. Y. Doi, Machine Test. Lab. Rept. (Japan) 44 (1958).
133. D. M. Mattox and G. J. Kominiak, J. Electrochem. Soc. 120
1535 (1973).
134. D. S. Campbell, "Trans. 9th tiational Vacuum Symp." p. 29.
The Macmillan Co., New York (1962).
135. A. Catlin and W. P. Walker, J. Appl. Phys. 1!, 2135 (1960).
136. H. Schroder and G. M. Schmidt, Z. Angew. Phys. 124
(1964) .
137. J. Reisenfeld and R. W. Hoffman, AEC Tech. Rept. 39,
Case Institute of Technology, Cleveland, Ohio (1965).
138 J. D. Wilcock, Ph.D Thesis, Imperial College, London (1967).
139. L. Holland, T. Putner, and R. Ball, Brit. J. Appl. Phys.'
.!..!.' 167 (1960).
140. M. A. Novice, Brit. J. Appl. Phys. ll, 561 (1962).
141. M. A. Novice, 385 (1964).
142. J. Priest, H. L. Caswell, and Y. Budo, "Tran. 9th Nat'l
Vacuum Symp.", p. 12), The Macmillan Co., New York (1962).
143. J. D. Finegan and R. W. Hoffman, AEC Tech. Rept. IS. Case
Institute of Technology (1961).
144. J. Priest, Rev. Sci. Instr. 32,1349 (1961).
145. Y. Budo and J. Priest, Solid-State Electron. !, 159 (1963).
146. J. Priest and H. L. Caswell, "Trans. 8th Nat'l Vacuum
Symp.". p. 947. Pergamon Press, Oxford (1961).
147. R. J. J. Vac. Sci. Tech. 143 (1970).
148. H. Sunami, Y. Itoh, and K. Sato, J. Appl. Phys. 41, 5115
(1970). -
149. J. A. Aboaf, J. Electrochem. Soc. ,!l!, 1732 (1969).
ISO. A. G. Blachman, Mett. Trans.
l,
699 (1971) .
lSI. A. G. Blachman, J. Vac. Sci. Tech. 10 299 (1973).
152. R. J. Scheuerman, J. Vac. Sci. Tech.
!,
145 (1969).
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
153. W. Heitman, Appl. Optics !Q, 2685 (1971)
154. P. R. Stuart, 507 (1969).
D. A. Vermilyea, J. Electrochem. Soc.
.!lQ,
S. S. Lau and R. H. Mills, Phys. Stat. Sol.
351
345 (1963).
17, 609 (1973).
155.
156.
157. L. Gmita and E. Teneseu, Rev. Roumaine Phys. 79 (1967).
158. C. Drum and M. J. Rand, J. Appl. Phys. 39, 4458 (1968).
159. J. C. Grosspreutz, Surface Sci. 173 (1967).
160. R. Lathlaen and D. H. Diehl, J. Electrochem. Soc. 620
(1969).
161. R. Carpenter and D. S. Campbell, J. Mater. Sci. 526
(1969) .
162. C. Mai, S. Audision, and R. Riviere, C.R. Acad. Sci. Ser. B
269, 1185 (1969).
163. c. M. Drum, R. N. Tauber, J. D. P. F. Schmidt,
ABS Spring Mtg., Electrochemical Soc. Washington, D. C.
(1971) .
164. F. B. Micheletti and S. H. McFarlane, ABS Spring Mtg.,
Electrochemical Soc., Washington, D.C. (1971).
165. A.I. Vousi and L. P. Strakhov, Fiz. Tverd. Tela. 3319
(1970).
166. M. Plassa, Thin Solid Films, 2, 305 (1969).
167. R. C. Sun, T. C. Tisone, and P. D. Cruzan, J. Appl. Phys.
44, 1009 (1973).
168. D. M. Mattox and G. J. Kominiak, J. Vac. Sci. Tech.
528 (1972).
169. D. M. Mattox, 6th International Vacuum Congress, Kyoto
1974 (to be published).
170. R. D. Bland, G. J. Kominiak and D. M. Mattox, J. Vac. Sci.
Tech ..!.l, 671 (1974).
171. H. S. Story and R. W. Hoffman, Proc. Phys. Soc. B 2Q
950 (1957).
172. N. S. Rasor and R. W. Hoffman, Phys. Rev. 1555 (1955).
173. W. B. Pennebacker, J. Appl. Phys. 40, 394 (1969).
174. P. Chaudhari. Submitted to J. Appl. Phys.
175. A. Gangulee, Phil. Mag., 12, 865 (1970).
176. W. J. Takei and M. H. Francombe, Solid State Electronics
.ll, 205 (1968).
352 R.W. HOFFMAN
177. S. S. Lau and R. C. Sun, Thin Solid Films, lQ, 273 (1972).
178. R. R. Hart, D. H. Lee and O. J. Marsh, App 1. ,Phys. Letters
20, 76 (1972).
179. D. H. Lee, R. R. Hart and O. J. Marsh, Appl. Phys. letters
20, 73 (1972)
180. International Conference on Low Temperature Diffusion and
Applications to Thin Films. A. Gangulee, P.S. Ho, and
K. N. Tu, Yorktown Hts., (1974). To be published in Thin
Solid Films.
181. R. S. Blewer and J. K. Maurin, J. Nuc. Mat., 44, 260 (1972).
182. B. S. Berry and W. C. Pritchet, J. Appl. Phys. 44, 3122
(1973)
183. H. Biehl and H. H. Mende, J. Phys. Chem. Solids, 35, 37
(1974).
184.
185.
186.
S. Spinner,
S. Spinner,
K. Uozumi,
67 (1972)
J.
J.
T.
Am. Ceram. Soc., 22., 229 (1954)
Am. Ceram. Soc., 38, 113 (1955)
Nakada and A. Kinbara, Thin Solid Films,
11.,
187. H. S. Chen and T. T. Wang, J. App1. Phys. iL, 5338 (1970).
188. s. I. Tan, B. B. Berry, and L. Crowder, Appl. Phys. Letters
20, 88 (1972).
189. R. F. Bunshah, J. Vac. Sci. Tech., ll, 633 (1974).
190. Conference on Structure/Property Relationships in Thick
Films/ and Bulk Coatings, J. Vac. Sci. Tech. 11 (1974).
191. R. E. Winter, Thin Solid Films, 11., 81 (1972).
192. R. A. Hunt and B. Gale, J. Phys. D Appl. Phys., ~ 359
(1972).
193. R. C. Sundahl, J. Vac. Sci. Tech. , ,2., 181 (1972).
194. A. Yelon and O. Voegeli, in Single Crystal Films, M. Fran-
combe and H. Sato, eds., p. 321. Pergamon Press, Oxford
(1964)
195. K. H. Behrndt, J. Vac. Sci. Tech., , 63 (1965).
196. J. W. Matthews and E. Klokholm, Mat. Tes. Bull., 2, 213
( 1972)
197. J. W. Matthews, E. Klokholm. and T. S. Plasket, IBM J. Res.
Develop., 1I, 426 (1973).
198. Aspect of Adhesion, D. J. Alner, ed. U. of London Press,
London (Multiple Volumes).
MECHANICAL PROPERTIES OF NON-METALLIC THIN FILMS
199. H. Krupp, Adv. Colloid Interface Sci., !, 111 (1967).
200. J. Ferrante and J ;. Smith, Surface Sci., 38, 77 (I 973).
201. N. L. Harrison and W. J. Harrison, J. 1, 195
(1972)
202. B. N. Chapman, J. Vac. Sci. Tech. ll, 106 (1974).
203. K. Kendall, J. Adhesion, 2, 179 (1973).
204. C. Weaver, to be published.
205. D. M. Mattox, J. Vac. Sci. Tech., lQ, 47 (1973).
206. C.T.H. Stoddart, D. R. Clarke, and C. T. Robbie, J.
270 (1970).
207. J. E. Houston and R. D. Bland, J. Appl. Phys 44, 2504
(1973).
208. w. D. Westwood and C. D. Bennewitz, Private Communication
(To be published).
209. R. C. Sun, T. C. Tisone and P. D. Cruzan, Private com-
munication. (Submitted for publication in J. Appl. Phys.).
210. I. J. Hodgkinson and A. R. Walker, Thin Solid Films li,
185 (1973).
211. M. D. Merz, R. P. Allen and S. D. Dahlgren, J. Appl. Phys.
45, 4126 (1974).
353
212. D. Weaire, H. F. Ashby, J. Logan and M. J. Weins, Acta Metall
.!2.., 779 (1971).
THIN FILMS IN OPTICS
G. Baldini and L. Rigaldi
Universita degli Studi, Sassari and Gruppo Nazionale
di Struttura della Materia del C.N.R
Via Celoria, 16 Milano Italy
I . INTRODUCTION
In this paper we show how the optical response of a dielectric
thin film, or stack of thin films (multilayer), can be evaluated.
This subject has been studied by several people, particularly in
the last decade, since the use of thin films in optics has many and
varied applications. From the old antireflection film coating we go
to high reflectance mirrors, beam splitters, interference filters,
polarizers etc. Furthermore in recent ye ars, with the wide v ~
lopment of the lasers, there has been a great effort on a new field
which is known under the title of "integrated optics". Its main
goals are the design and construction of compact optical systems
having the properties of some logical circuits that, until now,
have been made with the technology of the intagrated electronics.
Examples of applications of integrated optics are : optical modu-
lators, rotators, waveguides, etc., which are useful for telecommu-
nications and computers. The field is very vast as one can imagine
from the examples listed above and therfore it is not possible to
cover it in this paper. Therefore we have confined ou r selves to
the foundations of light propagation in stratified media with par-
ticular attention to transverse and longitudinal propagation. These
two aspects have been considered at first for the single film which
will be described in the first part whereas multilayer systems will
be considered in the second part. Application of several of these
concepts can be found in the literature quoted here and, as far as
the problem of filters and polarizers is concerned, in the paper of
Prof. Pelletier.
355
356 Go BALDINI AND Lo RIGALDI
2. OPTICAL PROPERTIES OF A SINGLE THIN FILM
In this paper we will consider the propagation of light waves
in a thin film which may define as a plane layer with indefinite
extension in the Yo, y plane and finite thickness W along z or,
better, with a thickness of the order of the wavelength of light.
A substrate is also required for the practical purpose of suppor-
ting the film. Its thickness is taken as semiinfinite.
Different arrangements of a thin film can be considered as
far as light propagation is concerned. Here we will examine two
cases :
a) The indices of refraction of the film and the surrounding
media are such that total reflection does not occur (filters, anti-
reflection coating, etc.).
b) Total reflection occurs inside the film (waveguides).
A third case c) may also considered for the sake of complete-
ness but since it is not of interest in the applications which we
describe in these papers we will say only that the light may be
also totally reflected from the film as shown in Fig. 1 which illu-
strates cases a), b) and c).
2.1 Reflectance and Transmittance of a Single Film
We examine here the propagation of light when internal refle-
ction does not occur. In Fig. 2 we show the light path assuming
that the incident beams is coming from the no side.
In fig. 2 is represented a thin film where plane waves (with
amplitudes A and B) are propagating in the directions given by the
unit vectors u and v. For z > W waves propagate in the medium with
refraction index n2-whereas for z < 0 the medium is described by
no' We define : k in vacuum; ko = k no ; kl = ;
k2 = k n2 ; waves B in the direction of z(B Yj; waves in the
direction of Yj)'
Medium 0 (we use vector notation for the amplitude A and B since
later we will specify polarization). We have omitted the factor
e-iwt which can be included in the amplitude
-+ -+ o( -+ -+) -+ o( -+ -+)
Eo Bo exp 1 kouor + Ao exp 1 kovo.r
(1)
THIN FILMS IN OPTICS 357
Fig. 1
z
w
(b)
Three different arrangements for ligth propagation using
thin films. Multiple reflections have been neglected in
(a) and (b)
film
i'i =v-ix
o x
Fig. 2 Film employed in the case (a) of Fig. 1
358 G. BALDINI AND L. RIGALDI
Medium
. .
EI BI + Al
(.
HI = uIXnIB
I
+ vIXnIAI exp
Medium 2 (waves are missing)
Boundary conditions
z = 0
z = W

z X E2
- %
z X HI = z X H2
(2)
(3)
(4)
We are considering here TE waves (E.L plane of incidence) which are
also called s waves. TM waves or p polarization is analogous.
A -+ _-+ _ -+ _ -+ _-+
z.A
o
= ZBo = z.AI = z;B
I
= zB2 = 0
By using this result and by means of the following definitions
8 = knosin eo = kvsine I = kn2sine2"
b
o
knocos eo; b
2
= kn
2
cose
2
b
l

we obtain the following boundary conditions
(Ao- Bo)b
o
= (AI- BI)bl
BI exp(ibIW) + Al exp (ibIW)
[B
I
exp(ibIW) - Al exp(-ibIW)]
B2 exp(ib
2
W)
bl = b2B2 exp( ib2W)
(5)
(6)
THIN FILMS IN OPTICS 359
Let us recall that the A's and B's may be complex since b
l
is in
several cases complex.
From the boundary conditions we obtain the amplitude ratios
B2
- - t exp(io)
Bo -
(l+a
l
) (l+a2)exp(-ibIW)-(I-al)(I-a2)exp(ibIW)
(7)
(l-a
l
) (l+a
2
)exp(-ib
I
W)-(I+a
l
) (l-a
2
)exp(ib
I
W)
(l+a
l
)(I+a
2
)exp(-ib
I
W)-(I-a
l
) (l-a
2
)exp(ib
I
W)
&>
r exp(iy)
Bo
where
b
l
~ l
V-
0
b
i
h'
2
y and 0 are the phase shifts of the transmitted and reflected
beams.
In order to determine reflectance and transmittance it is re-
quired to evaluate the time average of the normal component of the
Poynting's vector
R T (8)
Since this procedure is tedious and lengthy we refer to the lite-
rature(I). We note that T is independent of the propagation direct-
ion whereas the value of R depends upon it. We may take advantage
of this fact in order to determine the optical constants of a
film. (2) As an example of the above treatment let us see the expres-
sion of T and R for a thin film of real index \! == Tht at normal in-
cidence.
R = (nl+n
o
) (nl+n2)-4nlnon2-(nl-no) (nl-n
2
)cos ~
I (nl+no)(nl+n2)+4nlnon2-(nl-no)(nl-n2)cos ~ R2
(9)
T
We easily see that R has maxima and minima when c o s ~ I ,
2n
1
W
cos ~ = I ~ = 2"m +A =
m
m=O,I,2
360
4n
1
W
cos cjJ= - 1 -+ cjJ = 1T(2m+l) -+ A = ---
2m+1
G. BALDINI AND L. RIGALDI
As an example of the optical properties of 8 thin film we shaw in
Fig. 3 the normal reflectance of two films of index nl = 1.22 and
nl = 2.0 deposited onto a substrate with n2 = 1.5.
From the above treatment we see that the optical behaviour of
a thin film is completely described when its complex index or, if
one prefers, its optical constants are known and its thickness is
given. It may occur however that both thickness and optical cons-
tants are not known, in which case one should obtain them
from experimental data such as reflectance and tranmittance. This
problem, and its solutions, are summarized in the following section.
Fig. 3
R ( ~ l r - - - - - - - - - - - - ~ - - - - ~ - - - - - - ~ - - ~
5
Reflectance at normal incidence of two films with indices
n) = 1.22 and n1 = 2.00
THIN FILMS IN OPTICS 361
2.2 Determination of the Optical Constants
The optical constants of thin films can be determined essen-
tially by two methods : (a) polarimetric ; (b) photometric. The
polarimetric method employs the ratio of the p-,and s-polarized
components of the reflected beam, = and the phase dif-
ference of the same components op- These two quantities are
usually obtained from the parameters that describe the elliptical-
ly polarized ligth reflected by the film(3). If and the pa-
rameters of the bare substrate, are also known previous measu-
rements, we may write
f(V,X, W,>..)
(10)
f(V.X, W,>..)
where the two functions of v, refractive index and X extinction co-
efficient, can be written for each \vavelength >.., whereas the thick-
ness W of the film under investigation is unknown. The procedure is
rather complicated because it is necessary to employ sets of curves
that are functions of and and to use successive approximations
to find v, X and W. Furthermore, polarizers and compensators must
be employed. If W is known, as sometimes happens, system (10) can
be solved since >.. is also known and therefore the two unknowns are
V and X.
In order to avoid the above limitations one can consider photo-
metric measurements of reflection and transmission at normal inci-
dence. With this procedure, polarizers are not required and the
scattered light, particularly important at short wavelengths, will
be less than at non-normal incidence.
Existing spectrophotometric methods are based on knowledge of
the film thickness but we wish to extend them to absorbing films
whose thickness cannot be measured directly. As a consequence, the
three unknowns V,X and W require the use of at least three indepen-
dent equations. To this purpose, we observe that three independent
measurements can be performed on a thin film supported by a semi-
infinite transparent substrate, T, R} and R2' The expression for
T, RI and R2 are those, eq(8), given in the previous section where
the unknowns are in general v, X and W.
We omit from the present review the Kramers-Kronig analysis
because it requires measurements over a wide-range of wavelengths.
The problem of the determination of optical constants has been dis-
cussed by several authors and the interested reader can find seve-
rals articles in Optics Journals.
362 G. BALDINI AND L. RIGALDI
3 . TOTAL INTERNAL REFLECTION
Let us consider now the case in which total reflection occurs
inside the film. Here we do not examine the way light has entered
the film. We can assume that plane waves are reflected back and
forth in the film whereas the two media 2 and 0 do not carry any
plane wave. The scheme of this case is given in Fig. 4.
We consider here only TE waves since the TM waves can be trea-
ted in a completely analogous way. The three refraction indices no.
n) and n2 are all real. It is also understood that the wave fronts
extend from -00 to +00 in the y direction so that = 0
Since we are dealing with TE waves it follows that Ex = E
z
= Hy=O
and then we are left with
i
k
dE
-Y
dX
(1) )
where the last two expression stem from Maxwell's equations.
Now we give the fields in the three media.
Field in medium (film)
E())= A exp i(Sx-b)z) + B) exp i(Sx + biz)
y ]
H(l)= i. = [A exp i(Sx-bz)- B) exp(Sx + biz)]
x k dZ k 1
H()= i. = .@. [ A) exp i(Sx-bz)+ B) exp(Sx+b)z)]
z k. dX k
Field in medium 0 (substrate)
(I2)
The field in this medium can be obtained by recalling that in
order to satisfy internal reflection at the substrate-film boundary
the inequali ty
must hold and from this
We choose the plus sign before the root since the field in medium
o must be finite.
We then write
THIN FILMS IN OPTICS
z
W
"2
".. ".. ".. "-"" " " ''\,. "'.. " "'.. "'..
",
0
"0
X,
Fig. 4 Film employed in the total
Light is fed into the film
E(O)=C
O
e
PoZ
exp i(Sx)
y
H(o)= ipo CoePoZexp i(Sx)
x k
H ~ o = ~ Co e
PoZ
exp i(Sx)
k
Field in medium 2 (air)
X
internal reflection mode.
from above
(13)
363
With arguments similar to those above we get, with b2 iP2'
E(2)= D2 e-P2 (z-W) exp(Sx)
y
(2) - iP2 -P2(z-W)
H = -k- D2 e exp i(Sx)
(14)
H ~ 2 = ~ D2 e-
P2
(z-W) exp i(Sx)
Fields in media 0 and 2 are called "evanescent fields". They do
not represent free radiation(4)
364
G. BALDINI AND L. RIGALDI
Boundary conditions
Let us determine the values of the amplitudes BI' Co' Do by
employing the boundary conditions at z = 0 and z = W. For z = 0,
Le.
E(I)= E(o)
y y
which amonts to
H(I)= H(o)
With simple passages we get
where
B
= A -i2<p I 0
I 1 e
<P
IO
-I
tan
(15)
bl(A
1
- B
1
) = ipoC
o
(16)
which is the well known phase loss in total internal reflection. In
the same way at the surface z = W we have :
B
- A i2<P12
I - I e
where
The range of <P
IO
and <P']2 is
7T
.2.cjllO<
2
o <
The lower limit corresponds to the critical angle whereas
responds to grazing incidence.
(I 7)
7T
"2 cor-
We summarize the above results by saying that in total inter-
nal reflection the ray undergoes a phase change - 2<p]0 at z = 0
and - 2<P12 at z = W.
3.1 Mode Equation
We recall now that the beam A entering the film has a cross
THIN FILMS IN OPTICS 365
section in the x direction which is much larger than the thickness
W of the film. Let us assume that waves A are entering the film at
z = Wand in particular let us consider the point x = x2, z = W.
In order to add strength to the waves A in the film the phase of
wave Ai at (x2'W) must be identical modulo to the phase of
the wave entering the film. Let us assume that at (xI,H) both the
entering wave and the AI wave have phase zero. At point (X2,W) the
phase of the Ai wave will be The phase of
the wave the film at (x2'W) will be S(x
2
- xI)'
The above argument leads to the following "mode equation"
(18)
Let us discuss the meaning of the mode equation. Assuming that
no> n2 (which requires nl > no> n2) we can easily see that in order
to have a waveguide mode the following condition must be met :
kno < S < knl
WhenS= knl, then 6]= = and bl = O. It follows
also that W + 00. This is not since grazing incidence is
equivalent to direct propagation along x and therefore the bounda-
ries of the film supporting this wave must be at infinite separa-
tion. When S = kno then 61 = 6
1c
(critical angle between media 0
and I), Po = 0 and = O. The mode equation (18) gives the mini-
mum thickness in order that the waveguide can support a mode of or-
der m.
I
W.

(19)
In the general case when S has intermediate values, betw"een those
considered above the physical meaning of the different modes of pro-
pagation can be grasped more firmly by writing eq. (18) in the fol-
lowing way
Where
WI 2 = 12 /b I '
WI =
(18' )
This is equivalent to say that the film thickness for given Sand
366
G. BALDINI AND L. RIGALDI
W,
", ",
---+-
Wl
______ ..1 __
Wo
m=O
I
m=3
Fig. 5 Examples of thin waveguides showing the electric field
amplitude for different propagation modes.
THIN FILMS IN OPTICS 367
m is that of the m=O mode plus the quantity mWI'
Examples are given in Fig. S.
When no' nl and n2 are known we can easily estimate both B/k
and W for a given angle of incidence. The curves W = W(S/k) with
m as a parameter are called "characteristics of the waveguide".
From the value of S/k we obtain the wave velocity in the propaga-
tion direction: Vp = C(k/S), where C is the velocity of light.
In Fig. 6 we report such curves for a GaAs film deposited on
an Irtran II substrate, according to Tien and Ulrich(S). We note
that W increases with S/k and with m as expected from the mode
equation. The curves of Fig. 6 show also the validity of eq. (18')
3.2 Power Carried in a Waveguide
Let us evaluate the power carried by a dielectric waveguide
when the waves are TE. This can be obtained by integrating from
z = -00 to Z i +00 the x component of the Poynting's vector
(c/8n)Re(E
y
H
z
) related to the field which results from the Aj and
B waves(4 ) We note that the field penetrate both media 0 2.
electric field in the thin film is given (neglecting e-
1wt
) by
E(I)= Al exp i(Sx -biz) + BI exp i(Sx + biz) (20)
y
In order to simplify the interference between waves we translate
the x axis in such a way that Ey is or, in other words, to
a point were waves Al and BI are in phase.tS)We have then
= Al exp [i(Sx - blz)]+ Al exp [i(Sx + biz)]
= 2AI cos(blz) exp(iSx)
From eq.(IO) we get
2A
l
n
l
sin6
1
cos(blz) exp(iSy)
and then
which gives for the power carried within the film
(21)
(22)
(23)
(I) l
W2
c
P x = 8lr Re (EyH
z
) dz
Wo W2
[sin(blz)cos(blz) +
I (24)
+ b
l
z] } W
o
368
Fig. 6
G. BALDINI AND L. RIGALDI
12
8
E
::1. 6
4
2
2.6 2.8 3.0 3.2
p/K
Characteristics of a thin film waveguide: thickness vs.
nl sin 8
1
according to Tien and Ulrich. (5)
(1)
In Fig. 7 we have shown the values of Ey vs.z, with the power flow
along x.
We have seen that at the boundaries the waves undergo a phase
loss so that the fields at z = Wo and z = W2 must be A cos $10
and A cos $12' respectively, with A a constant, i.e., the value
of the field at z = O. If we put b
1
W
2
= $12 then the field at z = W2
is A cos$12. By putting blW
o
= $10 + m7T we get for z =-Wothefield
+ A cos$IO if m is even and - A cos$IO if m is odd.
THIN FILMS IN OPTICS
x
Fig. 7 The diagram shows the choice of the x axis along z.
(See text).
It is straightforward to obtain :
sin
2
<jl
P (1) = -s. A 2n sine [w + 10
x 41T 1 1 1 Po
2,j,
S1n "'12
+ .
P2
(25)
369
By integrating the Poynting's vector from z = W2 to z = +m we get
the power carried by the air side
and similarly for the substrate side
cos
2
<jl12
P2
I
-Wo ( ) 2 cos
2
cp 10
p(o)= s 0 dz = -S. A n sine
x -m X 41T 1 1 1 Po
The total power is then
(26)
(27)
P = p(O)+ p(1) + p(2)= 4
c
A
2
l
n
1
sine
l
[W + _I] (28)
x x x x 1T Po 'P2
370 G. BALDINI AND L. RIGALDI
The quantity is the Poynting's vector in the x di-
rection due to the superposition of the Al and B) waves. The fac-
tor W + (I/po)+(I/P2) is the equivalent thickness Weq of the wave-
guide where the power is confined. Note that W
eq
> W since the fields
extend also into the substrate and into the air.
3.3 Experimental Setups for Coupling Waveguides to External Beams
In order to couple the light to the waveguide some device is
necessary since otherwise it is not possible to satisfy the criti-
cal angle condition. The most common way of introduction of the radia-
tion into the waveguide is by means of the evanescent field at the
base of a prism or by means of a diffraction grating.(5,6). In the
case of the prism coupler it is necessary that the index n3 of the
prism be larger than nt, the index of the film guide, if all pos-
sible modes have to be excited. We have then n3 > nI > n > n2' In
Fig. 8 is shown the scheme of a prism coupler. 0
The evanescent field is created at the gap below the base of
the prism and if the gap is a fraction of a wavelength will
be wave propagation in the film. When kn3sin83=W/vp' where vp is
the phase velocity of one of the modes in the film, the coupling
becomes efficient and the power of the laser beam is transferred to
the film via the prism.
Fig. 8
Su bst rare
Description of the prism coupler for feeding light into a
thin waveguide.
THIN FILMS IN OPTICS 371
The length of the coupling region between film and prism can-
not be arbitrary since the power flow may change direction and
leave the film entering the prism. Obviously a similar prism can
be used for extracting power from a film waveguide.
Coupling constant not too far from unity have been obtained
(88% Ulrich (7)
Another coupler is obtained from a grating deposited directly
onto the film waveguide. The coupling occurs between the horizon-
tal component of the wavevector in the outer beam and the propaga-
tion constant S of one of the allowed modes. It can be shown that
the condition for coupling is c/v = sin8 + m(Ao/d) where c is the
velocity of the guided wave, in the order of the mode, 8 the an-
gle of incidence and d the grating constant(8).
Other couplers have been examined and among them we mention
the tapered film which allows light to be both fed into the wave-
guide and to be extracted from it(9).
Goos-Haenchen effect So far we have considered a 3imple zig-zag
model with phase losses at the two boundaries where total reflec-
tion occurs. By doing so we have neglected the contribution of the
Goos-Haenchen effect which amounts to a logitudinal shift of the to-
tally reflected ray. In a recent paper(IO) it is claimed that al-
though the approach used in these papers is valid for obtaining
the mode equation, and the corresponding phase velocities, it fails
to predict the correct energy exchanges (among modes e between ad-
jacent guides) and the group velocity. For a treatment of these pro-
blems the reader is referred to the paper by Kogelnik and Weber(10).
l.;. THIN FILM MlJLTILAYERS
We are going now to study the optical behaviour of a multi-
layer stack built from several thin films. In order to illustrate
the basic concepts which are related to the use of 2x2 matrices as
proposed by Herpin and Abeles(ll) we consider normal incidence.
Oblique incidence can be easily described by means of the so called
equivalent indices. (12). In Fig: 9 is reported the scheme of the
multilayer. We omit the term e ~ w t from the equation and recall
that Nj = Vj-Xj' in general. In layer j the electric and magnetic
fields result from the superposition of a progressive wave Aj and
a regressive wave BJ are given by
E(z) Aexp[i(kNz-aJ)] + B' exp[i(-kN'z-S')]
J J J . J J (29)
H z) Nj{Ajexp[i(kNjZ-aj)] -Bjexp [i(-kNjz-Sj)]}
It is obvious that the amplitudes Aj and Bj and the phase constants
372 G. BALDINI AND L. RIGALDI
y
z
N2 ----- Nj ----- Nm Nm+l
x
o 2
j-1 m-1 m
Fig. 9 Scheme of the multilayer considered here
aj and Sj are not arbitrary but can be determined by means of, the
boundary conditions. Before employing the boundary conditions let
us define the following quantities according to Fig. 10 and using
the following notation :
A.
r J
B.
r J
(t)
E(j_l)+
exp i(a.- 2n N.z.)
J A J J
exp i(S+ 2n Nz)
J A J J
= Aj exp i(aj- NjZj_l)
E
( r ) B ( oJ. + 2,n N )
(j -I) + = j exp .L fJ /\ j Z j-I
From eqs. (30) we easily get
(t) (t).
E(j_I)+ = E
j
_
(r) _ {r) _..
E(j_I)+ - E
J
_ exp(
2n
with = Nj(zj-Zj_l)
(30)
(31 )
THIN FILMS IN OPTICS
373
Hi
i-1
Fig. 10 Light rays at the boundaries of the j-th layer.
By applying the boundary conditions at z Zj we get
E(:) + = +
r r J+ J+
(32)
N'[E(:) - = N. E(r) ]
J J- J- J+I J+ j+
We note that eqs. (32) represent a homogeneous of 2(m+l)
linear equations, with 2(m+l) unknowns. We have also = 0, i.e.
there is no reflected component from the right hand side of the
1ast surface and we may put 1 as reference. This reduces the
unknowns to 2(m+l) and therefore system (32) can be solved. By wri-
ting eqs. (30) and (31) for the j-I interface and recalling eqs 32
we have the recurrence relations for the amplitudes of the transmit-
ted and reflected fields,
(t) 1 I (t). 1 (r).
Nj_l) E
j
_ i(1-Nj/Nj_I)E
j
_
(33)
=
In particular recalling
E = t(1 +Nm+ 1 IN
m
)
that E(t)= 1 and E(r) = 0 we have
m+ m+
E(r) = 1 (I-N IN)
m- 2 m+1 m
(34)
and from this by the recurrence eqs. 33 we can determine the re-
flectance and tranamittance of the whole multilayer.
374 G. BALDINI AND L. RIGALDI
4.1 Matrix Technique
All calculations on multilayers (here we are considering iso-
tropic media) can be more easily done by employing the 2x2 matri-
ces an originally proposed by Herpin(ll). Later Abeles showedthat,
for a general multilayer where the only significant variable is z,
the fields at z = zl and z z2 can be written as
[
E(Z2)]
H(z2)
(35)
where the aij element depend only upon the optical constants and
thickness of the system between planes Z = zi and Z = z2' We note
also that the recursion eqs (33) are a linear transformation and
can then be written in the following way
E (t)
2N. 1 N. I-N.
E ~ t ----l:..!..
exp(i<pj)
r J
exp(-iq,j)
G-i)-
Nj_1+Nj 2N. 1 r
r
(36)
N. I--N.
2N. I
(r)
r J
exp(i<pj)
r
exp(-i'j)
E ~ r
EC 'J-! J-1 2Nj_1 N. l+N.
r
) J= J
where the 2x2 matrix, from now on called [Nj]' depends upon thick-
ness Wj and indices Nj-l and Nj' Futhermore the use of matrices be-
comes more meaningful if we consider the total amplitudes of the
fields E(z) and H(z). Eqs. (29) can now be written also as
N' -
J
having defined Ej E ~ ~ +
J-
It is simple to show that we can .also write
(37)
(38)
E.
J
H.
J
THIN FILMS IN OPTICS 375
where [NjJ is the 2x2 matrix of eq. (36) and [Sj] is the 2x2 matrix
of eq. (37) and therefore [M8]becomes
cos<Pj
i
sin<pj
[ r1j J
N'
J
iNjsin<Pj cos<Pj
(40)
It should be noted that the index Nj can be replaced by the so cal-
led equivalent index when considering non normal incidence. (12)
Nj + N./cos8j for p polarization and Nj +-Njcos8j for s polariza-
have seen than that by means of eq. (39), with definition
(40), the field at two boundaries are easily related to each other.
Expressed through E and H the initial conditions become
1'1. = E(t) = 1
'1Il m+ '
E (t) = N
m+ m+l
Then by applying the recurrence eq. (39) we get
(41 )
It is now possible to evaluate Rand T for any multilayer by means
of the matrix elements of [M ] by recalling that
[
nJ
E (t)
] "I mIl
m12]
J
(\
E(r)
(42)
n
l m21
m
22
0 0
which, when solved with respect
to E(t)

and E (r)
0-
becomes
E (:)
1
t
Cm
l2 +m22
/n
O) 1
1
0
I(m
11
+m
21
/no)
E(r)
1
Nm+1 J
(43)
I(mll-m21/nO) t(m
l2
-m
22
/no)J
0-
376 G. BALDINI AND L. RIGALDI
This leads to the following expressions for reflectance and trans-
mittance
R
T
mll-m21/no + Nm+l
m
12 - Nm+l
m
22/
n
o
mll+m21/no + Nm+l
m
l2 + Nm+lm22
/n
o
Re(Nm+l) 4
fto I
m
ll+
m
21/
n
o + Nm+lm22
/n
o 12
2
(44)
(45)
In concluding this section we note that a multilayer is descri-
bed by a matrix [M] whose determinant is unity. The matrix lM] is
the ordered product of individual [Mj]matrices which have real
elements along the main diagonal and purely imaginary elements on
the other diagonal when the layer is transparent. If the film is
absorbing then the elements are complex. It should also be noted
that the multilayer is fully described by the 2x2 characteristic
[ M] independently of the media which surround it (the same can
be said of any portion of the multilayer). It can be shown that any
multilayer is equivalent to a single layer if the stack is symmetric
or to a double layer otherwise. In all cases, however, one shbuld
expect strong dispersion and this is what makes the multi layers so
useful for many applications.
4.2 Some Examples
In order to show the advantages of the matrix approach let us
examine two simple examples which refer to real refractive indices.
More detailed applications to multilayer systems will be illustra-
ted by E. Pelletier in his paper(13).
Example (a). Let us consider a single selfsupporting film whose
optical thickness is nW = p(A/2) with p integer. Its characteristic
matrix M is
M = (-I)P
[ 0
1 0 j
Using this matrix it is shown by means of eq. 44 that R = 0 and
T = 1 which is the expected result for a film known to leave unaf-
fected a light beam of wavelength A.
Example (b). We examine a dielectric Fabry-Perot filter made
of a number 2p of layers with indices nL and nH' alterna ively,
THIN FILMS IN OPTICS
which we indicate with LHLH . LH substrate = (LH)P substrate. If
the individual layers have optical thickness Ao/4 their matrices
are the following :
H
377
It should be noted that the 1.M product is a diagonal matrix with
real elements
L.H
For a multilayer with p pairs the reflectance, according to eq. 44,
then becomes
where nO is the index of the free side medium (air) and ns is the
index of the substrate. As can easily be observed R + 1 at A = Ao
when nH > n
L
, and p becomes very large. This system is used for se-
veral applications such as narrow band filters, laser cavity mir-
rors, etc. As a final comment to this section we may add that the
matrix approach is very practical since it is very suitable for
computer design of optical components as will be shown by the paper
of Pelletier< 13) .
5. GENERAL FORMULATION FOR ANISOTROPIC SYSTEMS
In the previous sections we have considered only isotropic me-
dia when studying light propagation in thin films. However a great
deal of interest has arised for anisotropic materials in view of
their applications to optics and because of their applications to
thin film devices (modulators, rotators, etc.). A recent approach
to the description of the optical properties of anisotropic multi-
layers has been developped by Berreman(14) and by Vassell(15) who
have employed a 4x4 matrice technique (suggested originally by other
authors) which is a generalization on the 2x2 matrix technique des-
cribed above. Here we are only going to outline this method, refer-
ring the reader to the original papers for more details. The method
associates with each interface of the multilayer a 4-vector whose
378 G. BALDINI AND L. RIGALDI
components are the tangential projections of the electric and ma-
gnetic vectors. The 4-vectors belonging to two different layers can
be related (as it has been done previously for the tangential com-
ponents) of E and H in section 4. for isotropic media) and a 4x4
transfer matrix is established. The transfer matrix of a multilayer
results then from the ordered product of individual matrices in a
manner analogous to that of isotropic media.
5.1 4 x 4 Matrix Formalism
According to the consideration above, Maxwell's equations may
be written in the following matrix formalism
0 0 0 0 -'O/'Oz 'O/'Oy
Ex
Dx
0 0 0 'O/'Oz 0 -a/ax
Ey Dy
0 0 0 -a/'Oy a/ax 0 E
z
1 a D
z
0 'O/'Oz -fJ/'Oy 0 0 0
Hx
= cat
Bx
-'O/'Oz 0 'O/dx 0 0 0
Ry
By
'O/'Oy -fJ/'Ox 0 0 0 0 Hz
B
z
This equation may be condensed by the following expression
1 a
RG = -- c
-- c at -
(46)
(47)
where the R matrix is made of two 3X3 matrices with all zeros (2nd
and 4th quadrants) and by the 3x3 matrices ~ 1 a,d ~ 1
R l 0 Rl]
= -Rl 0
with ~ 1
curl.
Here we are interested in a stack of thin layers at oblique inciden-
ce and therefore the propagation vector along x,a = knisin6i' is a
constant whereas there is no y component. Then ~ 1 becomes
0
a
0
-a-z
!1=
a
-ia (48)
az
0
0 ia 0
THIN FILMS IN OPTICS
Continuing our analysis, which holds for the remainder of these
papert, we may write

_ =f(z)exp[i (f3x - wt)]
(49)
exp! i - wt)]
379
where r(z) and A(z) represent the amplitudes of the fields E, H, D
B along the direction of the normal to the multilayer. As
quence of this we may rewrite eq.(47) which becomes:
Rf = - i k A (50)
When neglecting non linear effects we may find a linear relation
between A and f
A. = MI' (5 I)
with the 6x6 matrix M
M
[
E P 1
e I ]J
where E, p, p', and ].1 are 3x3 matrices. If wr neglect optical rota-
tion, then p= p' = 0 and we are left with the 3X3 dielectric tensor
E and the permeability tensor].1. From eqs.(50) and (51) we get
Rf = -ikMf (52)
Now we apply definitions (48) to (52) and it is found that the 3rd
and 6th components of (52) are linear algebraic equations in the
Bix components and the remaining four linear differential equations
of the first order can be written for four components. Following
reference(14) we elirninQte r3 and r6 and one obtains
s. (53)
380 G. BALDINI AND L. RIGALDI
where
0 0 0
)
p=
0 0
-) 0
0
-)
0 0
(54)
1 0 0 0
and S is expressed through M.()4). Finally with the aid of the per-
mutation matrix Q
1 0
Q=
0 0
0 0
0
which gives
Ex
Ey
9 tjJ (55)
Hx
Hy
and
Ex
tjJ
By
Ey
Rx
eq. (53) may be written as
alP = iMtjJ
az -
(56)
where =
It can be verified that if M'
j
= M .. and p = p'= 0 (non optically
active media) then also Sij cannot give more detaizf4he-
re about the matrix and refer the reader to the literature ,15)
Let us now assume that eq.(56) has the formal solution
tjJ(z) exp[ik tjJ(ZI) (57)
THIN FILMS IN OPTICS 381
where the matrix representation of the exponential operator is the
transfer matrix U(z-z') of the layer of the thickness z-z'. The in-
verse of this operator is the characteristic matrix C known in the
literature.
G(z-z') = U (z'-z)
For the evaluation of U(z'-z) we must refer again to the literature.
Let us now consider a multilayer such as that of Fig. 9. The
profile of the field component parallel to the z plane inside the
jth layer (Zj-\ z Zj) can be written as
(58)
With this formalism we generate the 4x4 transfer matrix T for the
multilayer
T = U\,U2'" U
m
with Uj = U(Zj - Zj-1)
(59)
At this stage if the optical constants of the materials em-
ployed are known, we are able to evaluate the properties of any
stack of anisotropic layers by means of the mere use of the product
of the U matrices. The problem can be easily handled by a
and therefore one can design systems for different applications.
A few examples are found in the paper by Vassell(15). It is shown,
for instance, that by straightforward application of the 4x4 matri-
ces the waveguiding condition for a uniaxial layer are easily obtain-
ed. Other examples are found in the papaer by Berreman(14).
REFERENCES
1. L.N. Hadley and D.M. rGnnison, J. Opt, Soc. Am. 11,451 (1947).
2. G. Baldini and L. Rigaldi, J. Opt. Soc. Am. 495 (1970).
3. Sec. e.g., Meyer, Z. Physik 168, 169 (1962).
4. M. Born and E. Wolf, Principles of Optics, Pergamon Press
5. P.K. Tien and R. Ulrich, J. Opt. Soc. Am" 60, 1325 (1970).
6. P.K. Tien and R.J. Martin, Appl. Phys. Lett. 291 (1969).
7. R. Ulrich, J. Opt. Soc. Am. 1419 (1973).
382 G. BALDINI AND L. RIGALDI
8. M.L. Daicss and L. Kuhn, Appl. Phys. Lett. 16, 523 (1970).
H. Kogelnik and T. Sosnowski, Bell Syst. Tech. J. 49, 1602
(1970)
9. P.K. Tien and R.J. Martin, Appl. Phys. Lett. ~ 398 (1971).
10. H. Kogelnik and H.P. Weber, J. Opt. Soc. Am. 64, 174 (1974).
11. A. Herpin, Compt. Rend. Acad. Sci., 225, 182 (1947).
F. Abeles, Ann. Phys. (Paris), ~ 5 9 6 and 706 (1950).
12. J. MacDonald, Metal Dielectric Multilayers. A Hilger, London
(1971)
13. E. Pelletier, 'Paper in this Book
14. D.W. Berreman, J. Opt. Soc. Am. 62, 502 (1972).
15. M.O. Vassell, J. Opt. Soc. Am. 64, 166 (1974).
RADIATION EFFECTS IN THIN FILMS
A. Holmes-Siedle
J.J. Thomson Physical Laboratory
University of Reading, Reading RG 2AF, England
I . INTRODUCTION
The term "radiation effects" is usually used to describe the
secondary phenomena which occur after a high-energy photon or par-
ticle has passed through a solid lattice and disturbed it by trans-
ferring momentum to some atoms or by raising some atoms to electron-
ically excited states. ~ h i l e these primary transfers of energy are
predictable and roughly similar in form for all elements and com-
pounds, the sequelae depend very strongly on the chemical and phy-
sical structure of the solid. For thin films, we should thus observe
virtually all the effects expected in the corresponding bulk material,
to which will be added a range of effects characteristic of the thin-
film state such as surface effects, non-stoichiometry, high impurity
levels, strain and disorder. Also, since thin ,films are frequently
used in electronic devices, we must interest ourselves in effects
associated with drift or displacement of electrons, holes or ions.
This variety makes the investigation of radiation effects in thin
films a challenging task especially since the nature of thin films
eliminates several of the conventional investigative tools used for
bulk crystals. However, because of its technological importance,
the field continues to develop.
We will first discuss some radiation effects which have proved
to be dominant in thin films and will go on to describe some of the
techniques used to study these effects and some important cases of
radiation environment which may cause degradation of thin films.
We will then describe some particular thin-film devices and other
interesting interactions of radiation with thin foils or films.
383
384
Particles
High energy
photons
A. HOLMES-SIEDLE
Low energy
photons
eg Electrons uv
visible
x ray
Protons
Neutrons
Si + etc
Displacement
Increased defect concentration
Decreased carrier lifetime
Decreased carrier mobility
Decreased carrier concentration
local disorder leading to
dila tion and stress
(colour centre formation)
Long-lived
effects
Charge excitation
Alt('red population of traps
( . Colour centre formation)
Charge transport.
Charge sheet build -liP
Oxidation reduction reactions
Production of stable
free radicals
Bonding changes.
Bond scission
Cross linking
Dilatation etc
Decomposition
Gas evolution
Coloration
Ionization
'Rapid annealing' of
minority carrier
I ifetime etc
Transient
effects
Photocurrents leading
to transient voltage
changes at terminals
L atchin'] conditions in
bistable circuits
Breakdown effects
leading to abnormally
high local currents in
gas or solid state
Production of short-lived
colour centre
Fig. I A classification of radiation effects
RADIATION EFFECTS IN THIN FILMS 385
2. RADIATION EFFECTS AND POINT DEFECTS
Fig. 1 gives a summary of the wide range of effects which can
follow a primary displacement or ionisation in a solid (1). The
distinction made here between high-energy and low-energy photons
concerns whether the photon can transfer sufficient energy to the
solid to produce atomic displacements, for example by Compton scat-
tering. This requires a photon energy in the MeV range. At lower
energies, there are still two large and important groups of pheno-
mena which can occur, namely transient and permanent ionisation ef-
fects.
Point defects are involved in two ways. Firstly, the primary
vacancies and interstitial atoms may complex with impurities or ag-
gregate into groups and these structures may then be stable for
long periods at room temperature; secondly, point defects formed
in a material when it is prepared may interact with the carriers
produced by irradiation and change their charge state. Since thin
films commonly contain high concentrations of defects formed during
preparation, ionization effects will figure quite strongly in our
discussion. By the same token, radiation-induced displacement may
figure less strongly as a technological problem, in a system which
starts its life with a large number of defects although displace-
ment may be interesting scientifically, if only as a method for the
controlled change of point defect concentration in films. An exam-
ple of the latter method is ion implantation ; the range of a kilo-
volt ion is of the order of hundreds of nanometres ; we can make ma-
ny uses of this similarity between particle range and film depth.
Other chapters have described th nature of the amorphous sta-
te and some defects observed in semiconductor and dielectric films.
In this chapter, we will attempt to enlarge on the nature of point
defects which exhibit strong electronic activity. Because of the
experimental difficulties, a sound knowledge Qf the point defect
structures present in thin films has not yet been built up. We are
only at the stage of looking, in the thin films, for analogues of
those point defects which have been observed in the corresponding
bulk material. The attempt is worthwhile, however, because it is
important to control trapping in electronic thin films and success
in this will be more likely if we understand the structure of the
traps.
3. RADIATION ENVIRONMENTS
While there is not as yet as good collection of data on radia-
tion environments and their relative capacities for damage, a detai-
led description of all those of interest is outside the scope of
this review. There are some partial accounts in the specialist lite-
386
A. HOLMES-SIEDLE
rature(I)(2)(3)(4)(5)(6)(7). However, in order to determine the
precise physical problems which should concern us in the present
study, we have to define broadly the types and magnitudes of ra-
diation flux which our materials are likely to encounter and to
group or classify them for severity of effect. We thus divide the
levels of damage into three groups "ordinary" "high" and "extra
high" and, in Tables 1 and 2 the levels of severity are indicated
in these terms. The dose ranges associated with each ~ r o u p expres-
sed in units of "damage equivalent I-MeV neutrons cm- "and in
"rads" are defined at the bottom of Table 2. Usually, different
types of effect have to be considered in each range, defect-con-
trolled electrical properties being the first to appear, with op-
tical and mechanical effects appearing later. The relative magni-
tudes of the ionizing dose in rads and the degree of atomic dis-
placement in "damage-equivalent" units (see below) will vary with
the energy and type of particle. In Table 1, we attempt to show
the common associations between certain radiation sources and cer-
tain sub-systems, such as space radiation with thermal control sur-
faces or reactors wich control electronics. The diversity of mate-
rials and radiation types is obvious.
We can express the integrated damage levels received in terms
of two damage parameters - displacement and ionization effect.
For displacement effects we can crudely express the damage inflic-
ted by the various particles (electrons, protons, fusion neutrons
etc.) in terms of a 'damage-equivalent fluence". It can be said
that for many electrical degradation effects in semiconductors,
the amount of degradation per particle changes strongly with parti-
cle energy and type. For example 14 MeV neutrons generated by a
deuterium-tritium reaction cause about three times the degradation
produced by a 1 MeV neutron in a silicon junction device. The dama-
ge-equivalent fluencefor 10
24
cm-
2
of fusion neutrons is thus about
3 x 10
24
cm-
2
(] MeV-equivalent neutrons). We could equally choose
another neutron energy for expressing our damage-equivalence or
even, with limitations,another particle (space radiation calcula-
tions often express the total effect of a broad spectrum of par-
ticles in "damage-equivalent, normally incident 1 MeV electron per
cm-2" (2) (8) In Table 2, the] MeV neutron is used for comparison.
Also, in this table, a more fundamental equivalence is given, name-
ly "displacements per atom". This self-explanatory term simply ind;-
cates the number of primary atomic displacements which have occurred.
The damage-equivalent fluence, on the other hand, equates the ulti-
mate effect of that displacement on a measurable property of the
material (resistivity etc.) The latter depends on the course of ma-
ny secondary events (e.g. the number of primary point defects gene-
rated along the track of the primary knock-on atom and the subsequent
interaction of such point defects with each other or with impurities).
Thus, the units have different uses but both serve to indicate the
different physical problems we face in the three fluence ranges.
T
A
B
L
E

1

-
R
A
D
I
A
T
I
O
N

S
O
U
R
C
E
S
,

L
E
V
E
L
S

A
N
D

E
F
F
E
C
T
S

(
f
r
o
m

H
o
l
m
e
s
-
S
i
e
d
l
e
,

1
9
7
4
a
)

A
t
o
m
i
c

B
a
t
t
e
r
i
e
s

I
o
n

a
n
d

e
l
e
c
t
r
o
n

N
u
c
l
e
a
r

N
u
c
l
e
a
r

R
e
a
c
t
o
r
s

a
n
d

s
p
a
c
e

r
a
d
i
a
-
b
e
a
m

d
e
v
i
c
e
s

e
x
p
l
o
s
i
o
n
s

C
o
n
t
r
o
l
s

I
n
t
e
r
n
a
l

p
a
r
t
s

t
i
o
n

C
l
a
s
s

o
f

O
r
d
i
n
a
r
y

H
i
g
h

O
r
d
i
n
a
r
y

H
i
g
h

E
x
t
r
a

H
i
g
h

L
e
v
e
l
s

t

T
y
p
i
c
a
l

1
5

-
2

1
0
1
8

1
0
1
5

1
0
1
6
_
1
0
1
8

1
0
2
0
_
1
0
2
5

I
-
M
e
V

n
e
u
t
r
o
n

<

1
0

n
c
m

E
q
u
i
v
a
l
e
n
t

P
a
r
t
i
c
l
e

e
,

p
,

a
.

e

,

X
-
r
a
y
,

i
o
n
s

n
,

y
,

X
-
r
a
y

n
,

y

n
,

y
,

i
o
n
s

R
a
d
i
a
t
i
o
n

C
h
a
r
a
c
t
e
-
:
-
L
o
w

R
a
t
e
s

H
i
g
h

s
u
r
f
a
c
e

d
o
s
e
s

V
.

H
i
g
h

r
a
t
e
s

M
e
d
i
u
m

H
i
g
h

R
a
t
e

r
i
s
t
i
c
s

L
o
n
g

T
i
m
e
s

L
o
n
g

T
i
m
e
s

V
.

s
h
o
r
t

R
a
t
e

L
o
n
g

T
i
m
e
s

t
i
m
e
s

L
o
n
g

t
i
m
e
s

P
r
o
b
l
e
m
s

C
i
r
c
u
i
t
s

a
n
d

A
c
t
i
v
e

m
a
t
e
r
i
a
l
s

H
i
g
h

r
e
l
i
a
b
i
-
S
h
i
e
l
d
i
n
g

H
i
g
h

t
e
m
p
e
r
a
t
u
r
e
s

m
a
t
e
r
i
a
l
s

v
e
r
y

e
x
p
o
s
e
d

t
o

h
i
g
h

l
i
t
y

a
n
d

r
u
g
-
m
a
y

b
e

l
i
-
M
a
t
e
r
i
a
l

s
t
r
e
n
g
t
t

i
n
t
o
l
e
r
a
n
t

o
f

l
i
g
h
t

a
n
d

r
a
d
i
a
-
g
e
d
n
e
s
s

r
e
q
u
i

m
i
t
e
d
.

C
o
n
-
d
e
t
e
r
i
o
r
a
t
e
s

I
f
a
i
l
u
r
e

t
i
o
n

l
e
v
e
l
s

r
e
d
.

P
o
w
e
r

l
i

t
o
l

c
i
r
c
u
i
t
s

P
o
w
e
r

L
i
m
i
t
e
d

m
i
t
e
d

m
u
s
t

n
o
t

i

d
r
i
f
t

I

I

S
y
s
t
e
m
s

C
o
n
t
r
o
l
s

E
l
e
c
-
I
o
n
-
B
e
a
m

U
a
t
a

~
n
k
S

C
o
n
t
r
o
l

E
l
e
c
-
E
l
e
c
t
r
o
d
e
s

I
n
v
o
l
v
e
d

r
o
n
i
c
s
.

I
m
a
g
i
n
g

p
r
o
c
e
s
s
i
n
g

S
y
s
-
G
u
i
d
a
n
c
e

t
r
o
n
i
c
s
.

S
e
n
s
o
r
s

C
o
n
t
r
o
l

R
o
d
s

S
y
s
t
e
m
s
.

S
c
i
e
n
-
t
e
m
s

C
o
m
m
u
n
i
c
a
t
i
o
n

P
o
w
e
r

R
e
g
u
i
t
-
a
t
i
o
n

R
e
a
c
t
i
o
n

t
i
f
i
c

I
n
s
t
r
u
m
e
n
t

E
l
e
c
t
r
o
n
-
B
e
a
m

V
e
s
s
e
l

T
h
e
r
m
a
l

C
o
n
t
r
o
l

D
e
v
i
c
e
s

I
n
s
u
l
a
t
i
o
n

-
-
-
-
-
-
t

F
o
r

d
e
f
i
n
i
t
i
o
n
,

s
e
e

T
a
b
l
e

2

:
:
:
t
I

o

~

o

Z

m

"
T
1

"
T
1

m

~

Z

-
I

:
:
I
:

Z

"
T
1

r
-
s
:
:

e
n

C
o
)

c
o

'
I

T
A
B
L
E

2

-
T
Y
P
I
C
A
L

R
A
D
I
A
T
I
O
N

E
N
V
I
R
O
N
M
E
N
T
S

(
f
r
o
m

H
o
l
m
e
s
-
S
i
e
d
l
e

1
9
7
4
)

S
y
s
t
e
m

N
u
c
l
e
a
r

F
i
s
s
i
o
n

R
e
a
c
t
o
r
s

P
o
w
e
r

P
r
o
p
u
l
s
i
o
n

-
C
o
r
e

-
E
n
g
i
n
e

-
E
l
e
c
t
r
o
-
N
u
c
l
e
a
r

F
u
s
i
o
n

R
e
a
c
t
o
r
s

n
i
c
s

C
o
n
t
a
i
m
n
e
n
t

V
e
s
s
e
l

M
a
g
n
e
t
i
c

C
o
i
l
s

C
o
n
t
r
o
l

E
l
e
c
t
r
o
n
i
c
s

E
l
e
c
t
r
o
n
-
B
e
a
m

P
r
o
c
e
s
s
i
n
g

C
l
a
s
s

o
f

R
a
d
i
a
t
i
o
n

L
e
v
e
l

t

E
x
t
r
a

H
i
g
h

E
x
t
r
a

H
i
g
h

H
i
g
h

H
i
g
h

,

E
x
t
r
a

H
i
g
h

H
i
g
h

H
i
g
h

H
i
g
h

L
a
r
g
e
s
t

E
x
p
e
c
t
e
d

L
e
v
e
l
s

D
i
s
p
l
a
c
e
m
e
n
t

I
o
n
i
s
a
t
i
o
n

D
A
.
!
.
:
a
g
e

(
E
q
u
i
v
a
l
e
n
t

D
i
s
p
l
a
-
I
-
M
e
V

n
e
u
-
c
e
m
e
n
t
s

(
R
a
d
s
)

t
r
o
n
s

c
m
-
2
)

p
e
r

a
t
o
m
)

1
0
2
3

1
0
2
0

1
0
1
9

1
0
1
6

1
0
2
4

1
0
1
8

1
0
1
6

1
0
1
3

2
0
0

0
.
2

0
.
0
2

2

x

1
0
-
5

2
0
0
0

2

x

1
0
-
3

2

x

1
0
-
5

2

x

1
0
-
8

1
0
1
4

1
0
1
1

1
0
1
0

1
0
7

1
0
1
4

1
0
9

1
0
7

1
0
8

T
y
p
e
s

o
f

d
e
v
i
c
e

o
r

m
a
t
e
r
i
a
l

e
x
p
o
s
e
d

C
l
a
d
d
i
n
g
,

R
o
d
s
,

e
t
c
.

a
s

a
b
o
v
e

M
o
v
i
n
g

P
a
r
t
s

S
e
m
i
c
o
n
d
u
c
t
o
r
s

E
t
c
.

F
i
r
s
t
-
W
a
l
l

S
t
r
u
c
t
u
r
e
(
p
o
s
s
i
b
l
y

i
n
s
u
l
a
t
o
r

o
v
e
r

m
e
t
a
l
)

V
a
c
u
u
m

s
e
a
l
s

a
n
d

a
l
~

v
e
s

C
o
n
d
u
c
t
o
r
s

a
n
d

S
u
p
e
r
c
o
n
d
u
c

t
o
r
s

C
a
b
l
e

I
n
s
u
l
a
t
o
r
s

S
e
m
i
c
o
n
d
u
c
t
o
r
s

a
n
d

D
i
e
l
e
c
t
r
i
c

d
e
v
i
c
e
s

O
p
t
i
c
s

a
n
d

S
e
n
s
o
r
s

P
a
i
n
t
s
,

P
l
a
s
t
i
c
s
,

S
e
m
i
c
o
n
d
u
c
t
o
r
s

w

c
o

c
o

?
>

J
:

o

r
-
3
:

m

e
n

!
!
?

m

c

r
-
m

I
o
n
-
B
e
a
m

P
r
o
c
e
s
s
i
n
g

N
u
c
l
e
a
r

W
e
a
p
o
n
s

S
E
,
a
c
e

R
a
d
i
a
t
i
o
n

J
u
p
i
t
e
r

B
e
l
t
s

E
a
r
t
h

B
e
l
t
s

S
o
l
a
r

F
l
a
r
e
s

S
t
e
r
l
i
z
i
a
t
i
o
n

R
a
d
i
o
s
o
t
o
p
e

G
e
n
e
r
a
t
o
r
s

X
-
R
a
y
_
S
o
u
r
c
e
s

t
D
e
f
i
n
i
t
i
o
n

o
f

R
a
d
i
a
t
i
o
n

L
e
v
e
l
s

H
i
g
h

1
0
1
8

V
a
r
i
a
b
l
e

1
0
1
4

S
e
m
i
c
o
n
d
u
c
t
o
r
s

w
i
t
h

Z
.

o
f

i
o
n

O
r
d
i
n
a
r
y

1
0
1
5

:
l

x

1
0
-
6

1
0
6

S
e
m
i
c
o
n
d
u
c
t
o
r
s

T
h
e
r
m
a
l
-
C
o
n
t
r
o
l

C
o
a
t
i
n
g
s
,

S
e
n
s
o
r
s

"

1
0
1
2

2

x

1
0
-
9

1
0
7

S
e
m
i
c
o
n
d
u
c
t
o
r
s

"

1
0
1
3

:
l

x

1
0
-
8

1
0
7

T
h
e
r
m
a
l
-
C
o
n
t
r
o
l

"

1
0
1
2

2

x

1
0
-
9

1
0
3

S
e
n
s
o
r
s

"

<

1
0
1
3

<

2

x

1
0
-
8

1
0
6

F
o
o
d
s
,

s
u
r
g
i
c
a
l

m
a
t
e
r
i
n
l
s

"

<

1
0
3

<

2

x

1
0
-
1
8

1
0
5

A
s

f
o
r

S
p
a
c
e

1
0
4

R
a
d
i
a
t
i
o
n

"

0

0

I
m
a
g
i
n
g

D
e
v
i
c
e
s

e
t
c
.

L
e
v
e
l

-
2

(
1
)

!
D
n
(
n

e
m

)

D
(
r
a
d
s
)

(
2
)

E
X
T
R
A

H
I
G
H

1
0
2
0
_
1
0
2
4

1
0
1
1
_
1
9
1
5

H
I
G
H

1
0
1
5
_
1
0
2
0

1
0
6

_

1
0

I

O
R
D
I
N
A
R
Y

B
e
l
o
w

l
O
I
S

B
e
l
o
w

1
0
6

(
I
)

D
a
m
a
g
e
-
e
q
u
i
v
a
l
e
n
t

I
-
M
e
V

n
e
u
t
r
o
n
s

(
2
)

A
s
s
o
c
i
a
t
e
d

i
o
n
i
s
a
t
i
o
n

l
e
v
e
l

v
a
r
i
e
s
,

b
u
t

t
y
p
i
c
a
l
l
y

o
f

t
h
e

o
r
d
e
r

o
f

0
n

x

1
0
-
9

(
r
a
d
s
)

:
:
0

0

5
>

-
I

0

z

m

"
T
l

"
T
l

m

C
"
l

-
I

e
n

z

-
I

J
:

:
2

:
:
!
!

r

s
:

e
n

w

(
X
l

'
<
l

390 A. HOLMES-SIEDLE
Ionization effects are equally important in device degrada-
tion and optical effects. Here, we must express the total amount
of electron-hole pair production produced by the particles or pho-
tons in a convenient unit. Units of absorbed energy are frequently
used, the rad being equal to 100 erg absorbed per gram.
4. SOME EXAMPLES OF POINT DEFECTS AND TRAPPING IN SOME
THIN FILM SYSTEMS
4.1 Introduction
Examples of a well-characterised set of deep traps ,yhich act
as recombination centres in a semiconductor are the radiation-indu-
ced A and E centres in silicon which are vacancy -impurity comple-
xes produced by bombardment (9) , the iron and copper centres, which
often occur in silicon ingots as grown and the gold centre, which
can be introduced by diffusion(10.). Minority-carrier lifetime is
short in thin-film silicon (even in epitaxial silicon on sapphire)
but, it has not yet been determined whether any of the above traps
are the cause of the low lifetime. In amorphous silicon films, no
discrete levels appear but rather a continuum of states which com-
pletely alter the band structure and transport properties of the
semiconductor. Several trapping energy levels found in Si02 and
Al203 films as grown have been identified by electrical methods
tIl) but no structure has yet been assigned to them. The Si02 sys-
tem is, in fact, one of the better experimental vehicles for attem-
pting to make such assignments and we will be discussing its struc-
ture in several places in this presentation as a model for under-
standing point defect in films. Table 3 describes a set of point
defects in silica. Despite our present ignorance con-
cerning the defect structure of thin films, this pair of materials,
silicon and silicon dioxide provide about the best hopes for an
early advance in understanding such defects for several reasons :
(I) a very large amount of knowledge has been accumulated on the
properties of the bulk materials
(2) Both materials have been deposited in well-characterised thin
films
(3) Each material has been produced in a stable, amorphous thin
film form.
4.2 Observation of Point Defects in Thin-Film Insulators
We will discuss here some electrical effects of point defects
in insulators and some optical observations which may throw light
on the structure of these traps. Cathodoluminescence studies(12)
have shown the presence of three discrete defect levels in thermal-
ly grown Si02, these are possibly connected with oxygen vacancies
RADIATION EFFECTS IN THIN FILMS 391
Centre
A
I
,A2
B1 '
B2
E2'
G
C,
EI
H
D
E
Fig. 2
TABLE 3 Colour centres in quartz and a-Si02
Possible structures
associated with Al
doubly-unsatisfied Si
singly-unsatisfied Si
o vacancy, two non-
equivalent Si
void ? (sputtered Si02)
unknown
interstitial
fi
E
u
-6
..., -8
!:!'
-10
2
l/z(V
'
/ Z em-liZ)
Emax
TraE

1 .9-2.6 Hole
4.1-5.2 Electron
5.4 Electron
5.8 Electron
6.55 Unknown
7.2 Unknown
7.6 Hole
SiO
Z
(AL-)
Field dependence of current in Si3N4, Al
2
0
3
and Si0
2

Dotted part of and AI
Z
0
3
curves were probably mea-
sured during
392 A. HOLMES-SIEDLE
or impurities, luminescence occurs at 2.8, 3.3 and 4.3 eV. Photo-
depopulation studies have indicated an electron trap level about
2.1 eV below the conduction band. As will be discu_ssed later, a
deep hole-trap with a high trapping cross-section exists in thermal-
ly-grown Si02 but attempts to characterise it have so far been un-
successful. In silicon monoxide films Walley and Jonscher (13) have
identified a high concentration of traps in the region of 0.2 eV
by temperature dependence studies on the I-V characteristic of the
film. These levels allow strong Poole-Frenkel conduction, which is
not possible in silicon dioxide because of the lower density and
greater depth of electron traps in that material. Similarly high
Poole-Frenkel conduction occurs in pyrolytic silicon nitride films
(14). Most aluminium oxide films show higher conduction than ther-
mally grown silicon dioxide but the process still appears to be
barrier-limited(15)I-V curves for the three dielectric films men-
tioned above are compared in-Fig. 2. As an indication of the sen-
sitivity of thin films to the conditions of preparation, it should
be noted that amorphous pyrolytic and sputtered Al203 are found
to have higher conduction than polycrystalline pyrolytic films(16)
(17), probably due to Poole-Frenkel conduction. Several different
defect levels are detectable even in polycrystalline Al203 films
by measuring the optical release of electrons and holes. Harari
and Royce(18) find four trap levels for each carrier in the 2-5 eV
range. It is interesting that some of these energies to
optical absorption bands in bulk sapphire samples which have been
irradiated with neutrons.
These results all suggest that insulator thin films, although
frequently deposited or growing in amorphous form, contain reason-
ably well-ordered structures with at least short range order and
a continuous, near stoichiometric network with very few dangling
bonds by comparison with the chalcogenide glasses. Thus, the band
gap is maintained and likewise the very high resistivity typical
of the bulk material. However, this structure also implies that the
intrinsic defects which are present wi}l act as deep-lying traps
for electrons and holes.
Another disadvantage is that unlike the amorphous chalcogeni-
des, such a network of atoms is strongly affected by impurities
which disturb the short range order (as with Na+ in a silicate
glass). Thus, impurity atoms may also act as electrically and opti-
cally-active defect levels. Such is the case with the hydrogen a-
tom in Fig. 3, which shows models for defects in an amorphous Si02
film(19).
A few observations of transmission and reflectance of thin
oxide films have been made in order to try and identify the elec-
tri-cally-active defects. While a strong defect absorption has been
observed in one form of evaporated Al203 film(20) experimental dif-
ficulties have so far prevented any correlation of these with the
RADIATION EFFECTS IN THIN FILMS
393
Si 0
o
1.55 A
} ~
(a) (b)
5\
~
(e:) H (d)
Si H
(e)
Fig. 3 Atomic configuration in SiOZ in the vicinity of (a) a
normal oxygen site (b) an oxygen vacancy and (c-g) various
centers containing hydrogen. The plane of the paper is the
(110) plane in which we have assumed the defect atoms to
lie. The x and y axes are (Ill) and (lIZ) respectively.
The arrows in (b) show the extent of the Si displacements.
394
A. HOLMESSIEDLE
electrically-active defects in pyrolytic films. Hickmott(21) has
observed defect bands in sputtered silicon dioxide, some of which
correspond with those in bombarded bulk Si02'
4.3 Interface States
A form of defect which is often important in electronic thin
films is the interface state ; when two solids are in intimate con-
tact, it is virtually certain that not all atoms at the interface
will be satisfied chemically. The free bonds thus constitute a thin
sheet of defects which, while they may exert a strong field effect
on the surrounding material, may be impossible to detect by conven-
tional methods of measuring defects (optical, ESR etc.) The impor-
tant case of the Si-Si02 interface is described later but signi-
ficant numbers may also exist between all kinds of deposited layers,
a particular example being the layered dielectric of the MNOS memo-
ry device (11)
Fig. 4
+
,
CHARGE f
DENSITY,
f (c",-3)
METAL I
I
I
I
I
I
I
Q.
11
OXIDE SEM!CON.
Image charge in metal and semiconductor produced by a slab
of charge trapped in the insulating oxide.
RADIATION EFFECTS IN THIN FILMS 395
5. THE METAL-INSULATOR-SEMICONDUCTOR SYSTEM
Since irradiation radically affects the charge distribution
in MIS devices, we will give a brief of the effects which
charges in the insulator have on the semiconductor and metal layer.
A simple sheet of charge placed in the insulator as in Fig. 4 pro-
duces image charges in the semiconductor space-charge region and
the metal. The relative amounts of this image charge in metal and
insulator depend on the position of this slab, but charge
lity demands that the sum of the image charges equal the insulator
charge, thus,
using the notation given in the figure. The same formula applies
if, as usually occurs, a charge sheet is present at the insulator/
semiconductor interface in the form of "fast states" or if the slab
is replaced by some uneven distribution from metal to semiconductor
(see, for example (22
A particularly useful electrical measurement can indicate the
magnitude of one of the image charge sheets. This is the high-fre-
quency capacitance-voltagemeasuremen4 in which the modulation of
MIS capacitance vs. gate bias is measured, as shown in Fig. 5.
(0)
( b)
Fig. 5
"\ /AFTER
, I
h J
! '-"
I
I
I
,
,
'-----/l V FB ---'
/
/
/
I
I
/
I

/
/
/
/
/
i----/lV
r
-----I
BEFORE
IRRADIATION

Lrl.OXIDE
: SEMICONDUCTOR
+
+
Change in operating region of MIS device produced by ioni-
zing radiation.
396 A. HOLMESSIEDLE
In the simplest case, the shift of the inflection point is a direct
measure of the image charge in the semiconductor. Measurements are
relatively easily and rapidly carried out (23) and it has been
shown that they can yield interesting and meaningful results if the
proper experimental conditions are established and if care is ta-
ken in their interpre.tation. For the purposes of this measurement,
we can divide the charge-carrying states into two groups :
1) Interface states which are stationary electronic states lo-
cated right at the plane separating the semiconductor from the in-
sulator. Interface states are analogous to the fast surface states
mentioned above and can exchange charge with the semiconductor.
Because the insulator has a wide forbidden gap, the energy levels
of interface states can lie either within or outside the forbidden
gap of the semiconductor. This will determine whether or not they
change their charge state when a field is applied between metal and
semiconductor.
2) Oxide states bulk defect states, forming a "space-charge re-
gion" in the insulator film and not in communication whith the se-
miconductor surface. There are two forms of this space charge, na-
mely, mobile and immobile. The immobile species is mainly held 1n
traps that are part of the defect structure of the insulator ; the
mobile charge is mostly due to ions that are capable of migrating
through the insulator, especially during conditions of high f i ~ l d
and elevated temperature.
The effect of a buildup of oxide charge such as Q
ox
is paral-
lel translation of C-V and ID-VG characteristics along the volta-
ge axis, each by the same amount, as shown in Fig. 5. We can ex-
press this by saying that VFB' the gate voltage at which the ener-
gy bands of the silicon are not bent at the surface (flat-band con-
dition) shifts by the same amount as V
T
, the voltage "threshold" at
which the conductivity of the semiconductor channel region begins
to increase, (VT is usually defined as the gate voltage producing
a given low channel current, say ID = 10 ~ . In the case shown
in Fig. 5, that of an n-channel transistor which is normally "off"
at VG = 0, the shift in VT shown causes the transistor instead to
be "on" at this voltage. Obviously, such a, shift could cause a
circuit containing this transistor to malfunction. We shall dis-
cuss later some particular cases of degradation of circuit proper-
ties produced by irradiation. While the buildup of oxide charge
alone leads to exactly equal shifts in V
T
and V
FB
' this is not so
when new interface states are introduced. This leads to the distor-
tions of both C-V and ID-VG characteristics(24).
While the foregoing account demonstrates that the capacitance-
voltage method is one of the more useful techniques to be used in
understanding charge buildup in the MIS system, it must be remem-
bered that the precise location of the charge within the insulator
RADIATION EFFECTS IN THIN FILMS 397
is not indicated, only the value of its image in the semiconductor.
The distribution of charge density versus depth in the oxide is
usually non-uniform and other techniques must be used to determi-
ne this profile. The shift in VFB due to any buildup of charge can
be represented as
x
- pdx
t
ox
'
This means that the nearer x is to its limiting value, tox' the
larger the value of ~ F B The maximum is when the sheet is concen-
trated next to the sem1conductor.
Two other simple cases will be .constantly referred to in la-
ter sections, namely sheets of uniform density (p') and finite
thickness (d) at the two extr-emities of the oxide, i.e. up against
the metal or silicon electrodes.
The expressions for these cases are
(a) Sheet next to silicon (case for positive VI)
(b) Sheet next to metal (case for negative VI)
when d is small, this approaches
~ V F B
- ~
d
Q
ox
qN
Cox
C C
ox ox
d t
ox
Note that since a thicker oxide has a higher value of Cox. the
same charge sheet next to the silicon has a larger effect on V
FB
in a thicker oxide. A convenient formula for this dependence is
M t (11m)
t ox
2.1 X 10
10
d tox
398 A. HOLMESSIEDLE
where 6N
t
is the area density (cm-
2
) of holes trapped in the thin
charge sheet next to the silicon (not the number of traps availa-
ble). Thus, a typical oxide of tox = 0.12 will rield a shift of
-1 volt for a charge sheet consisting of 1.76 x 10 1 holes per cm
2
.
5.2 Basic Features of Radiation Sensitivity
Fig. 6 shows a model for the processes which occur in an MIS
system when it is exposed to high-energy irradiation. Holes are
trapped in the region near the silicon when the field is as shown
presumably because electrons are largely swept out of this area(25)
The result of the buildup of charge is the shift of electrical cha-
racteristics shown in Fig. 5(26). Despite a large amount of inves-
tigation, the precise details of the process are still not clear.
For example the hole trap has not been identified and some debate
still exists as to the charge transport, mechanisms involved (27)
(28)(29)(30). Not pictured here is the gradual generation of new
Si/Si02 interface states under irradiation which also
5.3 Techniques for Investigation
After some early attempts to use empirical methods to correct
the tendency of thermally-grown silicon dioxide films to build up
space-charge under irradiation (see section 8.3) more fundamen,tal
investigations have been started using physical techniques such as
photoconductivity, optical spectroscopy and ion bombardment. Two
processes in thin film MIS insulators have to be understood tho-
roughly before a proper model for space-charge buildup can be de-
veloped. The first is charge transport by electrons, holes and pos-
sibly both polarities of ion. The second is the trapping of carriers
which, since the ultimate object is to control and modify this trap-
ping, further requires a detailed knowledge of trap structure. As
will be seen, the MIS structure is very well adapted to transport
studies, since high fields can easily be obtained and transit times
are very short. On the other hand, it is very poorly adapted for
conventional studies of defect structure, such as electron-spin
resonance, diffraction and spectroscopy, since films of less than
1 micrometre, unless they contain very high concentrations of de-
fects, will not contain an absolute number of defects sufficient
for detection by such techniques. Thus, for example, colour-centre
spectra have been observed in sputtered silicon dioxide films (21)
and heavily ion-bombarded films (32) but not in simple thermally-
oxidised films. The defect concentration in the first two cases
must be greater than 10
19
cm-
3
- an unusually high density compa-
red with those normally used in colour-centre studies (say 1017cm-3).
Some of the methods and their results will be described.
Williams (33) , Goodman (34) Snow et al(35) and Powell(36) have
characterised the barriers between silicon dioxide, silicon and
various electrode metals, using the threshold we elength for pho-
RADIATION EFFECTS IN THIN FILMS
____ ____ ______ ___
E
p-TYPE
SILICON
SUBSTRATE
FERM;.;..I:.....-_
LEVEL
INCIDENT
RADIATION

INCIDENT
RADIATION
E (si)
c
399
Fig. 6 Mechanism of Space Charge Trapping in Single Oxide Film
(after Gwyn(29.
toemission as a measure of the energy. These studies thereby esta-
blished the first estimates of the band-gap energy for thermal si-
licon dioxide, since the sum of the barrier energy for holes and
that for electron emission from the same material should equal the
band-to-band transition necessary for electron-hole conduction. Pho-
toinjection currents due to electrons could be measured with ease,
while those due to holes were at the limit of detection. Two impor-
tant facts concerning transport could be elicited. The mobility of
electrons was surprisingly high for a material presumed to be amor-
phous (about 30 cm
2
volt-1sec-1as compared with 1 cm
2
volt_l sec-
1
for amorphous germanium) ; the mobility of holes was vanishingly
small, and certainly several orders of magnitude lower than that
for electrons.
Another form of injection into MIS insulators is by production
400
A. HOLMES-SIEDLE
of hot carriers in the semiconductor which can then drift into the
insulator. Nicollian, using a high-frequency, 50-volt bias across
the semiconductor space-charge region, could produce avalanche con-
ditions which led to electren injection but no appreciable injection
of holes into the oxide. However, Werwey(37), using a DC avalanche
at a junction edge, claims to have also detected hole currents in
oxide films. In all these forms of injection, charging effects are
observed after currents have flowed for some time, which implies
that the oxide contains a limited number of traps for electrons and
holes. However, these studies have not proceeded far enough to give
any clues as to the structure of the defects involved, especially
in the case of the hard-to-obtain holes. In the case of the elec-
tron traps only, the energy levels of the traps have been estima-
ted, using a measurement of the threshold of light energy at which
the trapped electrons can be de-trapped to produce a photo-depopu-
lation current. The trap depths found in this way vary from 0.4 to
2 eV (33)
Experiments by Zaininger(38),Harari and Royce(18) and Emms ,
Holmes-Siedle, Groomhridge and Bosnell(30) have also shown that
UV light can be used to annihilate, or "photo-anneal", the radia-
tion-induced positive space-charge. Photoinjection has been used
in another way by Peel and co-workers(39) to observe a barrier
lowering in silicon dioxide films after the buildup of a radiation-
induced positive charge.
Since the process which creates space-charge in high-energy
electron or gamma ray irradiation is though to involve electron-
hole pair creation, it is of major interest to discover what hap-
pens when light with only just enough energy to create the pair is
used, rather than paPticles with many thousands of times the ener-
gy, as with the former type of high-energy irradiation.
Other basic information on oxide structure can also be obtai -
ned by such means. For example, it is a well-known technique to use
measurements of the threshold of intrinsic photo-conductivity to
determine the true optical band-gap. This experiment has been per-
f9rmed recently with silicon dioxide films. Di Stefano and Eastman
(40) found that the onset of conduction began sharply at slightly
below 9.0 eV. Powell (41) has confirmed this finding and it is con-
firmed indirectly by measurments of external photoemission currents
vs. wavelength in very thin oxide films, by Di Stefano and Eastman
(40). The best figure for conductivity band-gap of amorphous, ther-
mally-grown silicon dioxide films is 8.9 eV. It is interesting that
there seems to be little difference between the optical properties
of thermally-grown silicon dioxide films and those of bulk amorphous
silicon dioxide, such as synthetic fused silica (42). Furthermore,
the optical properties of the latter two materials has been sholYn
to exhibit surprisingly little difference from crystalline silicon
RADIATION EFFECTS IN THIN FILMS 401
dioxide(43). The importance of this general finding lies in its
bearing on the trapping model which we adopt for silicon dioxide
films. Firstly, we should not assume that, as for many amorphous
compounds, a high density of localised states exists near the con-
ductivity edge; secondly, we can perhaps borrow more freely than
in many other cases on the models used to eXRlain the defect struc-
ture of bulk silicas. Powell and Derbenwick( 2),Holmes-Siedle and
Groombridge(30) and Emms et al(30) have demonstrated the uSe of
irradiation with band-gap light to produce positive space-charge
buildup in silicon dioxide films.
Normally, the trapped space-charge in irradiated MOS capaci-
tor samples is extremely stable at room temperature under various
ambient and bias conditions, showing a reduction in oxide charge
of not more than a few percent within a period of several months.
However, annealing of an irradiated MOS sample free of interface
states in an inert gas at elevated temperatures results in an ap-
proximately parallel positive shift of the C-V characteristic of
the sample, indicating a reduction of the positive charge in the
oxide. The major changes take place between ISOoC and 300C and
by 4 ~ C all of the radiation effects in the oxide appear to have
been annealed out(63)(3S). Danchenko and co-workers have drawn two
conclusions from a careful study of the temperature dependence of
annealing of the positive space-charge. The activation energy is
in the region of I.S eV and seems to corresP9nd to the tunnelling
of electrons from the silicon interface into the space-charge re-
gion over a barrier of this height.
o. RADIATION SENSITIVITY IN SEMICONDUCTOR FILMS
The majority of the semiconductor devices used today employ
the p-n junction in silicon as the basic operating element. How-
ever, the fact that most of the operating principles involved
in p-n juntion devices require high crystallinity, low defect con-
centration, low doping levels and surface passivity make silicon
devices difficult to employ in a high energy radiation environment.
Because of this fact, and because of the desire to cut cost, much
work has been done recently towards the development of thin-film
semiconductor devices in which the key subelements operate satis-
factorily in thin film form, despite the presence of high defect
density (polycrystalline films) or absence of crystalline order.
Typical products of this research are the cadmium selenide thin-
film solar cell and the amorphous chalcogenide switching device.
Before these developments, other semiconductor films were already
in use, mainly in the field of large area photosensors (such as the
selenium films in vidicon TV camera tubes). Table 4 shows a survey
list of the most common types of semiconductor device and notes
which forms are usually made from bulk crystalline semiconductor
and which from thin semiconductor films (I). IX' signs indicate
402 A. HOLMES-SIEDLE
TABLE 4 DISPLACEMENT EFFECTS IN SEMICONDUCTOR DEVICES . PRIMARY
AND SECONDARY FAILURE MODES A d ~ p t e d from Holmes-Siedle(l
2 3 4 5 6 7
Usual form Life Car- Mobr Trap Other
of semicon- time rier lity ping
ductor Reduc Remo De-
Device Function tion val crea
l'le
Bulk Thin key to Symbols
p-n JUNCTION DEVICES
(a) Low Reverse Fields
Bipolar transistors & SCR 0
MIS (MOS) Field-Effect
Transistor 0
Variable Threshold Trans. 0
Junction FET 0
Rectifying/Blocking Diode 0
Tunnel Diode 0
Schottky Barrier Diode
Junction Photosensor
Optoisolator
Junction Electrolumines-
cent Diode
MIS Electroluminescent
Diode
Solar Cell
(b) Avalanche Devices
Zener and IMPATT' diode
Surface-Controlled Ava-
lanche Diode
(c) Other
Charge-coupled Device
Hall-Effect Device
OTHER DEVICE
Transferred Electron
Device
Photoconductive photo
sensor
Storage Photosensor
Mechanical Tranducer
Ovonic Threshold Switch
Ovonic Memory Cell
Amorphous Tunnel Triode
Photostructural Switch
Cold-Cathode Electron Emit-
ter
o
o
o
o
o
o
o
o
o
o
o
o
o
o
Film XX = Primary Failure Mode
X = Secondary Failure Mode
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
XX X
- XX
- XX
- XX
XX X
XX
XX
XX
XX
X
X
X
X
X
X
X
XX
XX
XX
X
X
XX
X
X
X
X
xx
X
X
X
XX
X
X
xx
xx
X
xx
X
X
X
X
X
X
X
X
xx
XX
XX
XX
RADIATION EFFECTS IN THIN FILMS 403
whether the basic operating principle of the device is likely to
be affected by displacement damage in the semiconductor material
and the particular effect in the material which leads to degrada-
tion of device action. With increasing particle fluence, the ear-
liest affected are those devices which require a high minority car-
rier lifetime, such as solar cells, bipolar transistors and silicon
controlled rectifiers. Field-effect transistor are affected only at
much higher fluence levels when resistivity and carrier mobility
begin to change. Also, a few devices are sensitive to the slow emp-
tying of shallow defect trapping levels which may produce an in-
convenient 'tail' on the falling edge of a square electrical pulse
emitted by the device. This would increase the dark current of an
imaging device, for example. Only at much higher fluence levels
would "other" properties like the electromechanical constant, utili-
zed in a mechanical transducer, or the hopping conductivity of an
amorphous semiconductor, be affected significantly. Thus, the rea-
der can choose a semiconductor device principle which, ideally,
should be radiation tolerant by noting whether the "Xli signs clus-
ter to the left or the right hand side of the table.
It is often found that, if a satisfactory device can be fabri-
cated from thin films, the effect of radiation on the device action
is less, presumably because the device has been designed to tolera-
te the considerable concentration of defects introduced in the fabri-
cation. The solar cell is an example ; thin-film cadmium sulphide
cells are very radiation-tolerant compared with silicon cells but
cannot be fabricated with such high initial conversion efficiencies
as silicon cells.
The ionization effects of most interest within semiconductor
materials are non-destructive and consist of the photocurrents pro-
duced when carriers are generated, especially if generated near p-n
juntion field regions. Diffusion, drift and recombination are the
important controlling parameters. Material parameters such as trap
concentration radically affect recombination and drift mobility.
Even in devices of simple geometry, models for these processes are
still difficult to construct because of the very complex gradations
of impurity concentration present in a junction device. Thin-film
p-n junctions will respond less strongly than diffused p-n junctions
of the same area because the collection volume for carriers in the
thin film will be much smaller. Against this must be weighed the
fact that the high concentration of defects present may lead to high
photoconductivity. However, on the whole it can be said that poly-
crystalline thin film semiconductors should show a lower overall
response to irradiation than bulk semiconductors because of size
effects and recombination effec.ts. The case of the amorphous film
can only be properly discussed with specific device principles in
view (see next section).
404 A. HOLMESSIEDLE
7. SOME IMPORTANT RADIATION EFFECTS IN THIN FILM DEVICES AND
MATERIALS
A few examples are given here of the impact of radiation ef-
fects on the working of several important thin-film systems. The
first two are not strictly thin-film devices but contain a thin
film layer with a critical function as a part of a crystalline se-
miconductor device.
A typical circuit configuration for MOS devices is shown in
Fig. 7(a) namely the complementary-symmetry MOS inverter (a "CMOS
logic gate") and the geometry of an integrated form of this logical
inverter shown in Fig. 7(b). The semiconductor is crystalline sili-
con. The voltages and fields across the gate insulators are shown
for the two logic states of this digital device. There may be many
thousands of such pairs interconnected on a single integrated cir-
cuit chip, only a very few being accessible via the external con-
nections. The effect of irradiation on the switching characteris-
tics of the device is shown in Fig. a-b. It can be seen that,
under irradiation, not only do the input-output voltage characteris-
tics change but the current drain (and hence power consumed) can
also increase or decrease, depending on the logic state while under
irradiation. Initially, degradation is gradual as the speed of ope-
ration of the circuit and sensitivity to spurious input signals
("noise immunity") deteriorate; finally, a condition is reached
in which the circuit completely ceases to perform its logical func-
tion. Examination of the circuit will show that the function of the
n-channel device is the one most affected by a given degree of space-
charge buildup. It is obvious that, for a large-scale logic circuit,
a careful circuit analysis is required before one can predict the
effect of the individual changes descr.ibed on the overall operation
of the circuit. (See for example Dennehy et al(44. The "duty cycle"
of the device while being irradiated must also be considered ; very
often, the logical input signal levels are alternating between "}"
and "0" (say between + 5 V and 0 V) while irradiation is proceeding
in this case, some "radiation-induced annealing" can occur during
the time when each gate insulator is under zero field (45) and the
net shifts of each transistor lie between the "100% off" <;md "100%
on" 7alues.
Burghard and Gwyn(46) have made a careful comparison of some
circuits with precisely the same logical function, made by diffe-
rent manufacturers to the same electrical specification. These dif-
ferences stem almost entirely from differences in fabrication of the
gate insulator. As the next section will indicate, it is not yet
clear why a material like silicon dioxide which, when thermally
grown, has good stoichiometry, low impurity content and few dang-
ling bonds, should show this great sensitivity to the fabrication
technique and research is still proceeding on the topic.
RADIATION EFFECTS IN THIN FILMS
405
1.2
;(
E 1.0
'0
-.0.8
!-
Z
0.6
a::

c.J 0.4
Z
0.2
o
o
0 10
>
w
8
I-
....
a 6
>
I-
4
I-

a 2
Va
ORADS
Ix 104
3x 10
4
'0
o RADS
I _ 104 --t-+-.I..l
30104 ---1_

GATE VOLTAGE, VG
(a)
12 6 Va
fo
1.2. RADS
'\ 3.7. III
10
o
;( S ---... -:.::; -'to!!.:-:}
E -----_, --'l.. "
>
w 0 ,."
II - 4 " :
!- !:;.
,
g G a:: 3 I
a:: I
I- =>
c.J
:= 4 Z 2
is ;;:
2 l5 I
GATE VOLTAGE vG
(b)
Fig. 7 (a) Two CMOS inverters in cascade
(b) Representation of First Inverter Pair in CMOS Network
in (a). Physical Structure, Bias conditions and electric
Fields in Insulator.
(from Poch and Holmes Siedle(47
Planar bipolar 'transistors employ thermally-grown Si0
2
films
for a different but critical purpose, namely to prevent leakage
and surface recombination effects where the junctions meet the sur-
face. The oxide is affected by radiation in approximately the same
manner as described for MIS devices, despite the radically diffe-
rent field conditions. Since silicon bipolar transistors are used
extensively in space and missiles, considerable effort has been
spent on analyzing the problem introduced at the junction by the
accumulation of positive charge in the overlaying oxide. Fig. 9
shows the regions of oxide important for degradation of gain.
While little headway has been made in finding passivation proces-
ses which will reduce the effect in junction devices, a clearer
understanding is emerging of the damage mechanisms involved, and
a method has been developed for distinguishing between and predic-
ting the two kinds of damage effects possible in bipolar transis-
tors, namely, damage to the crystalline silicon lattice, which re-
duces minority carrier lifetime. and changes in the oxide layer,
which increases the surface recombination velocity. In studying the
406
A. HOLMES-SIEDLE
VDD
+10V
+10V +10V
PON


+5 X 10'V/cm 0 V/cm -=
NOTE: FIELD VALUES ARE THE VOLUME AVERAGE
VALUES ACROSS A TYPICAL MOS GATE INSU-
LATOR OF THICKNESS 0.2 MICROMETERS.
"1"
INPUT
+10V
"0"
OUTPUT
OV
"1"
INPUT
+10V
GROUND
OV
p-CHANNEL, "OFF" n-CHANNEL. "ON"
SUBSTRATE: n-SILICON
p DIFFUSED
n REGIONS

mnmm 2
THIN
INSULATORS
1::::::::::::::::::1 METAL
(a)
(b)
Fig. 8 (a) Change in CMOS inverter Transfer Characteristic. Gate
bias during irradiation -10 Volts for the p-channel, 0 Volts
for the n-channel ; V
DD
is 10 Volts.
(b) Change in CMOS inverter Transfer characteristic. Gate
bias during irradiation 0 Volts for the p-channel and +10
Volts for the n-channel ; V
DD
is 10 Volts.
(from Poch and
RADIATION EFFECTS IN THIN FILMS 407
Fig. 9
JUNCTION-FIELD SURFACE-RECOM-
BINATION REGION SHOWING RADIATION-
INDUCED HOLE CHARGE TRAPPED IN OXIDE
BASE
COLLECTOR
FIELD-FREE BULK-RECOMBINATION
HAVING RECOMBINATION
STATISTICS SIMILAR TO BULK
SILICON
JUNCTION-FIELD BULK-RECOM-
BINATION REGION HAVING RECOM-
BINATION STATISTICS DIFFERENT
FROM FIELD-FREE BULK SILICON
Cross section of bipolar transistor showing region invol-
ved in gain loss due to the generation of oxide change or
interface states (from Holmes-Siedle and Zaininger(26
..

<3
Cl:
o

(,) 0.01

W
C)
Cl
2i
Cl
o
__ __
0.1 1.0 10
COLLECTOR CURRENT (MAl
Fig. 10 Form of an engineering specification for the worst case of
surface damage factor as a function of collector
current in 2N916 transistors of a given manufacturer ac-
tually a specification of 1 percent probability of excee-
ding the given level of beta change (from Holmes-Siedle
and Zaininger (47.
408 A. HOLMESSIEDLE
effect of radiation on tranststpr forward gain, the author and
co-workers (see for example t
45
) found that the combined effect
of bulk and ionization damage on transistor gain could be des-
cribed as a combination of surface and bulk damage factors as fol-
lows
~ i (Total) = ~ i (Bulk) + (Surface)
where S is the forward gain of the transistor. In other words,
when damage was expressed in this form, the two types of damage
were additive. While the bulk factor could be predicted fairly well
by means of an analytical expression, the value of the surface da-
mage factor could not be predicted for a given transistor struc-
ture. The effect arose from the oxide and, apparently, was deter-
mined in magnitude by several uncontrolled variables in oxide pro-
cessing which gave great scatter of gain effects from batch to
batch of devices or even within a batch.
Units with identical electrical specifications but from diffe-
rent manufacturers (i.e., different processing) vary even more ra-
dically than the samples shown. Thus, in order to qualify devices
for use in a radiation environment radiation tests are required
for every type of device used, these results then being converted
by some valid technique into predictions of performance degradation
in a given operational situation. For such analysis, the surface
damage factor ~ I / S ) s is a useful tool, especially, since it is
found that the damage is very dependent on emitter injection level.
The dependence is regular and can sometimes be cautiously extrapo-
lated. Also, the radiation-induced loss of gain, analyzed statis-
tically using ~ I / S ) s as a measure of damage, was found to have a
frequency distribution fitting the normal, bell-shaped curve. Thus,
a statistical prediction of the probability of meeting a particu-
lar level of degradation in a given number of devices can be deve-
loped in a form which can be used by circuit engineers who must pro-
vide sufficient allowances in their designs for transistor degra-
dation. Shown in Fig. 10 for the 2N916 transistor are anticipated
values of the surface damage factor as a function of collector cur-
rent at five different radiation dose levels. Knowing the expected
dose and collector current, a worst case anticipated value of ~ I / S ) s
and hence end-of-mission beta can be calculated simply and routine-
ly : This is an example of the practical application of research
on thin dielectric film structures under irradiation. Even before
a means of suppressing the effects of radiation has been found, an
understanding of the effects can be used to develop predictive for-
mulae and preventive measures (see for example (45.
An early application for glassy semiconductors was as thermo-
meters in a nuclear reactor core(48). Thus, for certain devices
RADIATION EFFECTS IN THIN FILMS 409
the theoretical prediction seems to hold true-namely that the glas-
sy semiconductors in bulk form maintain their properties well abo-
ve the level at which the transport properties of silicon or gal-
lium arsenide begin to change. The threshold and memory devices
have also been tested in reactors and under pulsed X rays. X ray
dose rates as i ~ as 1.8xIO
ll
rads-
1
and reactor neutron f1uences
of 1.2xI0
17
n cm- did not change the switching characteristics of
the thin-film "STAG" materials with graphite e1ectrodes(49). Smith
and his co-workers (1972) tested a more advanced memory device and
found no loss of function after reactor irradiation to l016n cm-
2
accompanied by 3 x l07rad of gamma irradiation. These results are
generally confirmed by the work of Nicolaides and Doremus(SO) and
of Flanagan and Wyatt(51) on similar chalcogenide devices. Thus,
the particular charge injection processes used in the monostab1e
and bistable chalcogenide switches also appear not to be affected
by either transient ionization or the impact of neutrons on the
network. However, Chen and Bai1ey(52) believe that some memory de-
vices will be affected at a f1uence about 10
18
n cm-
2
, since they
have found that a bulk sample of crystalline "memory" material,
GeO.08 ASO.50 TeO.42, can be rendered amorphous by a fluence of
3 x IOlBn cm-
2
in a reactor. Bistable devices made from such mate-
rial hold information in the form of thin crystalline filaments
running between the e1ectrodes(53). Even partial disordering of
these filaments could reduce or extinguish the high conductivity
of these paths and hence erase the memory effect. Hench et al(54)
find that the dielectric loss in V20S-P20S glasses is altered after
I017cm_2 reactor neutrons while Adler et a1(5S) find that the resis-
tivity of amorphous silicon films at a fluence of 8 x IOlBcm-2 re-
actor neutrons decreases by a factor of 4. This change is not ne-
gligible but is much less than would occur with crystalline sili-
con and, moreover, would probably be tolerated as a change in the
"off" resistance of a monostab1e device. However, it should be
borne in mind that the switching device principles described repre-
sent only two of many possible principles which can be developed
in future from the amorphous semiconductor field and are somewhat
limited in their applications. For example, device principles in
which local application of energy produces changes in structure
will be much more susceptible to the intense local disturbance pro-
duced by heavy particle bombardment. Thus most memory principles
may be less intrinsically radiation resistant than monostable prin-
ciples. Clearly we will have to be cautious of all lightly trigge-
red memory principles, employing what Grigorivici(S6) calls "less
drastic structural changes than crystallization" such as polariza-
tion or minor changes in short-range order. Olley and Yoffe(S7)
have demonstrated that ion bombardment can introduce strong damage
effects in chalcogenides at low temperature, creating local concen-
trations of "dangling bonds" which are not present in the original
evaporated amorphous film. Thus it should be expected that, since
many different forms of disorder are possible in the glassy semi-
410 A. HOLMES-SIEDLE
conductors, particle irradiation can still effect strong changes
in devices. formed from them. It is seen that statements, often ma-
de, that disordered materials cannot be further damaged is unfoun-
ded as a generalization.
The use of ion implantation to modify or create thin surface
layers is only starting to be investigated for a few cases. While
ion beams have been widely used to place chemical dopant atoms at
a desired depth in semiconductors and other materials (58) , the ac-
companying heavy displacement damage has been exploited in a few
cases. For example, optical waveguides have been formed in silica
surfaces by creating displacement damage in the surface with a beam
of ions(59). Also, one approach to producing radiation-tolerance
in MOS devices is to ion-bombard the top half of the oxide, produ-
cing electron trapping centres(31)(30).
A few cases of radiation effects in protective coatings should
be noted. In space vehicles, evaporated, grown or deposited dielec-
tric films are frequently used to maintain the desired radiative
equilibrium, between the vehicle, the sun and deep space. For this,
a high emissivity in the IR and a low absorptivity in the visible
is required. For long missions near the Van Allen particle trapping
regions, high fluences of low-energy protons and electrons are ac-
cumulated by the outer surfaces of a spacecraft, leading to colour-
centre formation ; most space vehicles have thus suffered gradual
increases in temperature as the solar absorptivity of the thermal-
control surfaces has increased (3). Some nuclear fuel granules are
encapsulated in pyrolytically-deposited ceramic coating to prevent
the creation of radioactive dust. These may rupture through accu-
mulated radiation damage, accompanied by helium gas generation from
i_mplanted alpha-particles and transmutation of elements.
Finally, some useful scientific information can be obtained
from observing microscopically the effects of particles in thin
foils. Lead iodide foils are used for studying nuclear fission re-
coil processes, since the tracks formed are visible(60). In many
materials, the region around the track of a heavy ion becomes more
susceptible to etching because of heavy ionisation or displacement
damage and the track can be made visible by this etching. The same
phenomenon has been observed in thermal silicon dioxide films in
which the etch rate can be significantly increased by kilovolt e-
lectron bombardment. The phenomenon can be used to define patterns
electron bombardment is also observed to produce .local densifica-
tion and stress relief in the film(61).
Another interesting case of visualisation of defects is that
of the cubic voids observed in alkali halide foils when irradiated
for some length of time in the electron microscope beam. The nor-
mal production of colour centres by ionisation must occur and be
followed by the agglomeration of vacancies
(62
).
RADIATION EFFECTS IN THIN FILMS
411
8. CONCLUSIONS
This brief survey has shown that several of the radiation ef-
fects already observed in the bulk state of matter are expressed
in modified and striking forms in the thin-film state. Such effects
are : charge trapping, leading to space-charge buildup ; mobility
changes and photo-conductivity ; densification, leading to changes
in etch rates and built-in strain ; amorphisation and/or changes
in resistivity which may inhibit the processes essential for the
action of a device. One effect which is significant only in the
thin film state is the radiation-induced generation of interface
states.
The primitive state of understanding of the changes which ir-
radiation produces in thin films reflects the general state of un-
derstanding of thin films on the atomic scale. However, the delete-
rious effects of radiation are sufficiently important, especially
in dielectrics, that work will continue to improve this knowledge.
Also, since other thin films are tolerant to degradation, thin film
devices may be developed further for use in radiation environments.
Finally, the technique of ion bombardment is so well suited for
probing and modifying thin films that the study of the interaction
of thin films with radiation is sure to continue for some time for
this reason alone.
REFERENCES
1. A.G. Holmes-Siedle,Reports of Progress in.Physics, 37 (6), 669
(1974)
2. W.L. Brown, J.D. Gabbe and W. Rosenzweig, Bell Syst. Tech. 42
1505 (1963).
3. W.S. West, W.J. Poch, A.G. Holmes-Siedle and D. Carroll
NASA TR-R-371 (Washington NASA 1971).
4. A.G. Holmes-Siedle, Nature 251, 191 (1974).
5. A. Charlesby, Atomic Radiation and Polymers (Oxford Pergamon)
p. 52 (1960).
6. S. Glasstone, The Effects of Nuclear Weapons (US Atomic.
Energy Commission: Washington DC 1962).
7. G.M. McCracken and S. Blow, UKAEA Culham Report No.CLM-RI20
(1972)
8. W.J. Poch and A.G. Holmes-Siedle IEEE Trans. Nucl. Sci. NS IS
(6) 213 (1968).
412 A. HOLMES-SIEDLE
9. J.W. Corbett, "Electron Radiation Damage in Semiconductors
and Metals "New York, Academic Press, (1966).
10. S. Braun and H.G. Grimmeiss, J. Appl. Phys. ~ 2658 (1974).
11. P. Balk, "Solid -State Devices, 1973" Conference Series n 19
(Institute of Physics, London) (1974).
12. P.J. Mitchell and DeNure, Solid State Electronics ~ 825
(1973) .
13. P.A. Walley and A.K. Jonscher, Thin Solid Films 1, 367 (1967).
14. S.M. Sze, J. Appl. P h y s . ~ 2951 (1967).
15. R.H. Walden, J. Appl. Phys. 43,1178 (1972).
16. T. Tsujide and K. Iida, Japan J. Appl. Phys. ll, 600 (1972).
17. C.A.T. Salama, J. Electrochem; Soc. 112, 913 (1970).
18. E. Harari and B.S.H. Royce. IEEE Trans. Nucl. Sci. NS-20 (6)
280 (I 973) .
19. A.J. Bennett, and L.M. Roth, J. Phys. Chem. Solids 32, 1251
(1971)
20. E. Harari and A.G. Holmes-Siedle, Bull. Am. Phys. Soc., Ser II,
.!i. 500 (J 97 1) .
21. T.W. Hickmott, J. Appl. Phys. 42 (6), 2543 (1971).
22. A.S. Grove, "Physics and Technology of Semiconductor Devices"
Wiley, N.Y. (1967).
23. K.H. Zaininger, RCA Review 27, 341 (1966).
24. K.H. Zaininger, IEEE Trans. Nucl. Sci. NS-13 (6) 237 (1966).
25. P.J. Mitchell, IEEE Trans. Electron. D. v ED-14 (11) 764 (1967).
26. A.G. Holmes-Siedle and K.H. Zaininger IEEE Traris. on Reliabi-
lity, ll, 34 (1968).
27. B. Andre, J. Buxo, DEsteve and H. Martinot. Solid State Elec-
tronics 12, 123 (1969).
28. H.L. Hugues, R.D. Baxter and B. Phillips, IEEE Trans. Nucl.
Sci. NS-19 (6) 256 (1972).
RADIATION EFFECTS IN THIN FILMS
413
29. C.W. Gwyn, J. App. Phys. 40 (12), 4886 (1969).
30. C. Emms, A.G. Holmes-Siedle, I. Groombridge and J.R. Bosnell,
IEEE Trans. Nucl. Sci., to be published.
31. H.L. Hughes, Proc. IEEE Reliability Physicso Las Vegas ~ r h
1 971 ) .
32. T.J. Russell, T. Pandolfi and B.S.H. Royce, Colour Centre
Conference, Japan (1974).
33. R. Williams, Phys. Rev. 140 (2A) , 569 (1965).
34. A. Goodman, Phys. Rev. 152 (2), 780 (1966).
35. E.H. Snow, A.S. Grove and D.J. Fitzgerald, Proc. IEEE, 55
1168 (1967).
36. R.J. Powell, J. Appl. Phys. i!,.,(6), 2424(1970).
37. J.F. Verwey, Appl. Phys. Lett. 43, 2273 (1972).
38. K.H. Zaininger, Appl. Phys. Letters ~ 140 (1966).
39. J.L. Peel, R.A. Kjar and R.C.Eden, Appl. Phys. Lett. 1I, 3
(1970) .
40. T.H. Di Stefano and D.E. Eastman, Solid State Commun. 2,
2259 (I 971) .
41. R.J. Powell, (Private Communication).
42. R.J. Powell and Derbenwick, IEEE Trans. Nucl. Sci. NS-18 (6)
99 (1970).
43. H.R. Philipp, J. Phys. Chem. Solids B, 1935 (1971).
44. W.J. Dennehy, A:G. Holmes-Siedle and K.H. Zaininger , RCA
Review 30, (4) 668 (1969).
45. A.G. Holmes-Siedle and W.l. Poch, Brit. Interplan. Soc. 24
273 (1971).
46. R.A. Burghard and C.W. Gwyn, IEEE Trans. Nucl. Sci. NS-20
(6) 300 (1973).
47. W.J. Poch and A.G. A.G. Holmes-Siedle IEEE Trans. Nucl. Sci.
NS-17 (6) 33 (1970).
414 A. HOLMESSIEDLE
48. J.T. Edmond, J.C. Male and P.F. Chester, J. Phys. E. Sci.
Instrum. l ' 373 (1968).
49. S.R. Ovshinsky, E.J. Evans, D.C. Nelson and H. Fritzsche
IEEE Trans. Nucl. Sci. NS-15 (6), 311 (1968).
50. R.G. Nicolaides and L.W. Doremus, J. Solids 8-10
(1972).
51. T.M. Flannagan and M.E. Wyatt, J. Noncryst. Solids 2 , 229
(1970).
52. C.W. Chen and D.M. Bailey, Radiation Effects lQ, 65 (1971).
53. J. Bosnell and C.B. Thomas, Solid. St. Electron. 1261
(1972)
54. L.L. Hench, J. Noncryst. Solids 229 (1970).
55. D. Adler, H.K. Bowen, L.P.C. Ferrard, D.D. Merchant, R.N. Singh
and J.A. Sauvage J. Noncryst. Solids 8-10, 844 (1972).
56. R. Grogorivici, Thin Solid Films, 153 (1972).
57. J.A. Olley and A.D. Yoffe, J. Noncryst. Solids 8-10, 850
(1972)
58. G. Dearnaley, J.H. Freemann, R.S. Nelson and J. Stephen, Ion
Implantation (Amsterdam, North Holland) (1973).
59. R.D. Standley, W.M. Gibson and J.W. Rogers, Appl. Optics 11
(6) 1313 (1972).
60. Chadderton, "Fission Damage in Crystals" (Methuen, London)
(1969) .
61. R.A. Sigsbee and R.H. Wilson, Appl, Phys. Lett. 23 (10)
541 (1973).
62. D.J. Stir1and, "The use of Thin Films in Scientific Investi-
gations" (J.C. Anderson Ed. Academic Press. N.Y.), p. 172
(1966).
63. V. Danchenko, U.D. Desai and S.S. Brashears, J. Appl. Phys.
39, 2417 (1968).
Applications
THIN FILM APPLICATIONS IN MICROELECTRONICS
v. Le Goascoz
L.E.T.I. Centre d'Etudes Nucleaires de Grenoble
B.P. 85, Centre de Tri, 38041 Grenoble, FRANCE
1. INTRODUCTION
The purpose of microelectronics is to associate together several
physical phenomena to realize a function, a logical or analogical
function. Let me mention for example, diode phenomena, injection of
current between each side of a barrier, field effect in the bulk of
a semiconductor (the field effect transistor), field effect near by
surfaces of a semiconductor (the metal oxyde s'emiconductor field
effect transistor), conduction in insulating layers etc The well
known bipolar transistor is an example of such device as well as the
MOS field effect transistor which are examples of combination of
several and different physical phenomena to obtain a particular de-
vice.
In the former case, the bipolar device, main physical pheno-
mena are bulk phenomena. All parameters we need to realize a device
or to compute a circuit are bulk parameters and in this case thin
film insulating layers are not of main importance. In the other
hand, in the later case with MOS devices, surface phenomena are very
important; physical parameters are surface parameters and we need
thin film layers to realize such a devices not only from the point
of view of physics but also for the technological aspect.
For these different reasons, we have chosen MOS Structure to
illustrate the use of non metallic thin films in microelectronics.
But we cannot forget other important and numerous devices which
use thin films at any time of the making. These layers can be only
a step in the technological process but they can be an active area
of the device. Let me mention some devices as microwave device
family, piezoelectric devices used in conjunction with MOS circuit,
417
418
v. Le GOASCOZ
Aluminum
N Si
substrate
Fig. 1
Schematic diagram of a MOS transistor.
THIN FILM APPLICATIONS IN MICROELECTRONICS 419
Josephson devices, photodetector devices, light emitting diodes
etc ...
Technological aspect is a very important one in the use of thin
insulating films. In this case these thin films do not serve in the
active part of the device and in many case these layers are removed
from the device when this one is completely achieved. Such layers
are important especially by their mechanical and chemical properties
and not at all for their electrical characteristics.
Electrical characteristics can be ranged in insHlating proper-
ties and conducting properties. In the first case insulating proper-
ties are used in a device when a field effect is needed to control,
for instance, a voltage level inside the device. These layers can
be used in an active part of a device but they only participate in the
device for their insulating properties. Therefore the pain proper-
ty we are searching in such a layer is insulating property. It is
difficult to obtain good layers and the most important parameters we
must control in the making of these layers are stability and repro-
ductibility of layers.
We can also use these thin films in an active part of a device
for their own conduction properties. Conduction is very weak and the
film can be considered as an insulator. Such a property is used in
MNOS memory devices where differential conduction between two layers
is used to inject charges inside insulators, and to trap carriers at
the interface between insulators. Reproducibility of layers and a
good knowledge of physical aspects of conduction are the more im-
portant features to be considered in this case.
To appreciate importance of thin insulating film in microelec-
tronic it is necessary to draw up list of main steps of technology
used to realize MOS Field effect transitors and large scale inte-
gration circuits. Before technology, a short review on MOS transis-
tor is necessary. The Fig. 1 shews diagrams of a MOS Transistor.
A MOS transistor is a surface device. A silicon surface poten-
tial controlled by a insulated Gate, allows current to flow at the
silicon surface between two diodes drain and source. The former with
a reverse bias and the later with a forward bias. If the silicon
surface is not inverted by the action of gate potential, no current
can flow. In the case of N silicon substrate if negative voltage
above treshold voltage is put on the gate, silicon surface is in-
verted and a current can flow between drain and source in the channel.
Inversion of the channel is obtained by surface field supplied by the
gate electrode. This electrode is completely insulated from the
channel by an insulator or combination of insulators as can be seen on
the Fig. I. The action of the field decreases with distance and to
allow a good control of surface potential it is necessary to use
420
V. Le GOASCOZ

I
{2000A
r\'\
N TYPE SILICON
'1 Souru
Sou ,.,e
cuul D ,-ain
r: ,_ ( M .... "')
",1\

f Gift & anJ ,01-& OlCiJe.
(t"J MIII.ld
Alumi Ilium aft"
(4
th
Mot. K)
Fig. 2 Main different steps of the bulk MOS technology.
THIN FILM APPLICATIONS IN' MICROELECTRONICS
SiJiGor\
pt/
N
.J>iodU
SO,.O"
N't"/P Diode.

,
SappJ.i,..
Si 0,.

Si Oa.
rY ..........
:"':':/:"':';:':.:
":':':'
Si 0
4
: 0,1"1'
1+*+< .d+t+}pb+;&
p. p.
I
Fig. 3 The different steps of the silicon on sapphire CMOS
technology .
421
T
E
C
H
N
O
L
O
G
Y

N
U
M
B
E
R

O
F

P
H
O
T
O
M
A
S
K
S

S
T
A
N
D
A
R
D

P

S
O
S
M
O
S

'
5

D
E
P
L
E
T
I
O
N

-
E
N
H
A
N
C
E
M
E
N
T

6

S
E
L
E
C
T
I
V
E

P

C
H
A
N
N
E
L

I
O
N

I
M
P
L
A
N
T
A
T
I
O
N

C
.
M
O
S
/
S
O
S

6

N
+

D
I
F
F
U
S
I
O
N

(

D
E
P
L
E

T
I
O
N

-
N

C
H
A
N
N
E

L

)

C
.

M
O
S

/

S
O

S

7

P

W
E
L
L

(

E
N
H
A
N
C
E
M
E
N
T
)

N
+

D
I
F
F
U
S
I
O
N

c
.

M
O
S

/

S
O

S

6

N
+

D
I
F
F
U
S
I
O
N

(

D
E
E
P

D
E
P
L
E
T
I
O
N
)

_
.
_
.
-
F
i
g
.

4

C
o
m
p
a
r
i
s
o
n

o
f

t
h
e

c
o
m
p
l
e
x
i
t
i
e
s

o
f

d
i
f
f
e
r
e
n
t

s
i
l
i
c
o
n

o
n

s
a
p
p
h
i
r
e

t
e
c
h
n
o
l
o
g
i
e
s
.

t

<

r

C
D

G
)

o

8

N

THIN FILM APPLICATIONS IN MICROELECTRONICS 423
insulating thin layers in the range of 1000 A. In the other hand,
it is necessary to take contact on diode to apply voltages. Aluminium
contact pads allowing device mounting in a package is deposited on
a field oxide with thickness in the range of 1.2 ~
To realize how it is possible to use non metallic thin film in
microelectronic, two reviews on MOS bulk technology and on CMOS on
sapphire technology will be done. The Fig. 2 shows main steps of
MOS process on bulk silicon. We can observe that two kinds of thin
film material is used in such case. Organic layers as photoresist
which allow to describe a circuit pattern and inorganic materials as
silicon oxide, silicon nitride or aluminium. The range of thickness
can be from 1000 A to 15 000 A.
The first step of the process consist in the cleaning of a N
type silicon wafer and then in oxidization of silicon to obtain a
Si02 12 000 A thickness layer. This layer is called field oxide or
thick oxide, it is used in the process but do not participate in
the device function. By means of photoresist pattern we open in the
field oxide windows allowing boron diffusion inside silicon. This
step is the making of source drains diodes. The next step consists in
opening of gate location, in cleaning of silicon surface and in
oxidizing of silicon for instance with dry or wet oxygen at a tempe-
rature near by 1000C. Si02 thickness obtained is around 1200 A.
Then we must open holes trough field oxide to take contacts on drain
source diodes. The last step is aluminium deposit and etching. Alu-
minium thickness is in the range of 15000 A. With this technology,
only bulk silicon wafer is not a thin film.
On the Fig. 3, we can see the most sophisticated CMOS techno-
logy : the SOS CMOS technolgy : the silicon on sapphire complemen-
tary MOS technology. In this device, all the layers are thin films,
even silicon thickness is in the range of 0.8 ~ m The most important
layer finds in the process is photoresist layer used as mask. This
CMOS Technology is the more complicated and use 6 or 7 photomasks.
The Fig. 4 shows complexity of different technology using silicon on
sapphire substrate ; and Fig. 5 explains main advantages of one of
them, the SOS MOS depletion enhencement technology.
2. DIELECTRICS IN THE MOS TECHNOLOGY
Several dielectric layers can be used to achieve a MOS device.
Each step of the technology needs one or several insulating layers
to process the device. For example to etch silicon nitride we must
use a Si02 mask because adhesion of photoresist on silicon nitride
is not good and the Si02 mask is obtained by a photoresist process.
We shall concentrate now on thin insulating films extensively used
in MOS technology. Several other dielectric layers exist and can
be used but only in the research field because all their properties
are not known, and especially their stability with the time.
424 V. Le GOASCOZ
DIELECTRIC ISOLATION
- DECREASE OF PARASITIC CAPACITANCES
- NO PROBLEM OF FIELD OXIDE THRESHOLD
- NO BULK EFFECT ON THRESHOLD VOLTAGES
- NO PUNCH THROUGH EFFECT
- HIGR TECHNOLOGICAL YIELD
DEPLETION ENHANCEMENT
- A SINGLE SUPPLY VOLTAGE
- SMALL SENSITIVITY TO SUPPLY VOLTAGE SHIFT
- LOW POWER CONSUMPTION
- TTL COMPATIBILITY
- WELL SUITED FOR ANALOG DEVICES
- HIGH DENSITY OF INTEGRATION
Fig. 5. Advantages of the SOS MOS depletion enhancement technology
THIN FILM APPLICATIONS IN MICROELECTRONICS
425
2.1 - Dielectrics:
- : Thermally grown Si02 can be obtained
by different ways according to properties and thickness to reach.
The steam oxygen growth is used for thick layers because growth
rate is very important (90 A/minute with 1150C). 12000 A field
oxide in bulk MaS Technology is achieved with steam oxide.
The wet and dry oxygen is used for thin layer and es-
pecially for gate layers. lhe growth rate is slow and can be for
example 40 A/minute with dry oxygen. Typically Si02 thickness of
o
1200 A 1S used to real1ze gate oX1de.
-
: In this case, Si02 layer is used as field oxide
layer in silicon on sapphire MaS technology and as a passivating
layer at the end of the process. To prevent holes and defects
across the layer, we must deposit more tahn 5000 A of Si02.
- is use4 as a gate insula-
ting layer over thin Si02 or as a passivating layer to prevent pollu-
tion by moistures and against radiation damages. In the former case
thickness is in the range of 500 A and in the later case in the range
of 5000 A.
- Silicon : It is of two types : epi silicon CVD deposited
or on substrate (sapphire) ,or polycrystalline silicon CVD
deposited over an oxide as a metal gate instead of aluminium. In the
former case, epi silicon thickness can be in the range of one micron
in the SOS technology and up to ten microns in the bulk technology.
In the later case polycrystalline thickness is in the range of
5000 A up to one micron.
2.2 - Main properties :
-
- insulating support for metal stripe line connecting between
them each part of a circuit.
- mask against species diffusion such as boron, phosphorous
or arsenic as well as against ion implantation.
- protective layers against pollutions, radiation damage,
short circuits.
The main chemical property of these
insulating thin films used 1n the MaS technology is the specific
426 v. Le GOASCOZ
1000 (A)
Si 02 thickness
Si O
2
growth with dry 02
fig 6
o 10 20 3 40
Fig. 6. Si0
2
growth curves for dry oxygen thermal oxidation.
THIN FILM APPLICATIONS IN MICROELECTRONICS 427
etching of these layers versus several acid or basic solutions.
For example silicon is etched by a solution of KOH or by a mixture
of N03H, FH, CH3COOH; Si02 is etched by FH (10% in water); Si3N4 is
etched by P04H3 (180C) and lightly by FH (10% in water) and Al is
etched by P04H3 (room temperature). Sometimes we can combine the
etch rate of two layers to obtain bent etching.
- : The insultating property is the main
feature of these layers and it is extensively used in the technology
and in the devices. In the MOS devices, gate is insulated from sili-
con surface by a dielectric and surface potential is controlled by
the electric field induced by gate potential. No current flows across
gate in the common MOS device.
The insulating properties of non metallic thin films is used
to allow aluminium stripe lines to run over the circuit without
short circuits. In this case insulation is achieved by the Si0
2
field oxide.
Conducting properties of these insulators are used in some par-
ticular devices such as MNOS memory device and MOS Tunnel devices.
- Silicon dioKide material : Si02 is the more important dielec-
tric Thermally grown Si02 is well
known. Numerous workers have contributed to the knowledge of silicon-
silicon dioxide interface(l) (2). This interface is very important
for the quality of the device, and specially for obtaining high
transconductance values of the FET MOS. Si02 can be obtained by dif-
ferent ways for the different uses. Quality of Si02 depends on the
making,and wanted thickness range indicates the kind of oxidation
process, that is the technology. Very thick layers are obtained by
steam oxidation and thin layers by wet or dry oxidation
Fig. 6 shows Si02 growth curves with dry oxygen and for diffe-
rent oxidation temperatures. Temperature must choosen to obtain
an oxidation time suitable for a good oxidation reprocibility.
We give some examples of oxidation time, temperature, and process.
Thickness 1000 A with wet 02 temperature 1000C time 12'
Thickness 100 K with dry 02 temperature 800C time 60'
Thickness
.
750C time 20' 30 A w1th dry 02 temperature
Thickness 20 A with dry 02 temperature 700C time 13'
Thickness 17 K with dry 02 temperature 680C time 10'
Thickness 12000 A with steam 02 temperature 1150
0
C time 140'
Main problem with Si02 layer is oxide charges introduced in the
layer in the thermal oxidation. These charges can be reduced by high
temperature annealing in nitrogen after oxidation, or after metal-
lisation by low temperature (400C, 500C) annealing in nitrogen or
water vapour.
428
10
17
o
concentration At, 3
cm
Aluminum diffusion
anneal ing 500C 20'
Silicon
5 0 1000
V. Le GOASCOZ
e
1500 (AJ
Fig. 7. Aluminium diffusion profile in Si0
2
-Si system as obtained
from ion microprobe measurements.
THIN FILM APPLICATIONS IN MICROELECTRONICS 429
Other problems can appear in the SiOZ layers and the more im-
portant is the presence in the layer of some species like Na+ or H+.
These ionized species are very mobile in the electric field and they
can drift toward silicon surface or gate electrode involving a shift
of threshold voltage.
We have seen to reduce surface states, low temperature anneal-
ing after metallisation gives good results and significant decrease
of positive exide charges. However it is necessary to be careful
with annealing.
We show in Fig. 7 aluminium profile in SiOZ after annealing
obtained with ion microprobe. We can see diffusion of aluminium in-
side Si02. With a 600 X SiOZ layer we obtained z. 1018 At/cm3 of
aluminium at the SiOZ-Si interface which is very important. With
gate oxide layer of 1Z00 A, surface concentration is very weak, but
in the case of future technology for i n s t n c ~ MICROMOS technology
where gate thickness is in the range of 100 A, these results must
be taken into account to prevent short circuits across gate. In the
case of MICROMOS technology, either we must reduce annealing tempera-
ture or we must use another metal as molybdenum or polycrystalline
silicon.
Presence of H+ species in SiOZ layers can be reduced py suitable
annealing at high temperature with nitrogen whereas presence of Na+
species is very difficult to avoid ;to minimize pollution by Na+ we
must use very pure water and to decrease influence of Na+ ions we
can prepare SiOZ layer with 02-HCl mixture. Cl- ions have the pro-
perty to trap Na+ ions and to' increase breakdown strength(5).
Now we can realize very good SiOZ layer in wet or steam oxygen
without Na+ pollution. We generate steam not by boiling water but
HZO synthesis with very pure Oz and HZ in a reactor. Results obtain-
ed with this process are similar to results achieved with dry oxygen.
We have some number of surface state, similar breakdown strength,
no shift of threshold voltage under field and temperature. However
the microbreakdowns observed with dry oxygen grown layers are not
seen with these layers.
Another point of main importance is the capability of dry oxy-
gen grown SiOZ to trap holes. Even when composition is stochiometric,
Si02 may exhibit hole trapping due to non-bonding p-valence bant6)
levels. The origin of this seems to be the presence of 0-- ions
In the other hand oxide grown with water does not exhibit this
feature and unlike dry oxide, seems to trap electroas(7).
The creation of fast states and positive oxide charges under
positive and negative bias (slow trappinft instability) has been
further investigated in several studies( , 9,5). At the present
430 V. Le GOASCOZ
time, the best process to decrease and sometime to avoid Na+ drift
of threshold voltage is to use phosphosilicate glass passivation.
Addition of P20S inside Si02 layer allows gettering of Na+ by phos-
phosilicate glass. With these layers we obtain an increase of di-
electric constant from 3.82 to but dielectric strengh is as
good as with as-grown Si02 films.
With the silicon on sapphire technology we must use deposited
Si0
2
to obtain thick insulating layer allowing passage of metal
stripe lines. Deposited Si02 is achieved at low temperature (4S0C)
with 02, SiH4 reactions. Quality and density of these layers are
not the same than thermally grown Si02, but despite of some defects
as holes across layers and fast etch rate, bulk technology and SOS
MaS technology use extensively deposited Si02 as field oxide layer,
mask oxide, passivating layers. Quality can be improve by particular
annealing in order to obtain densification of the layer. Quality of
deposited Si02 after densification is quite similar with thermally
grown Si02 (etch rate has almost the same value).
- Silicon nitride material
one of the several materials intro-
duced, eight years ago in the MaS technology with great interest.
However problems connected with the use of deposited Si3N4 arised
rapidly. Silicon nitride deposited directly on silicon exhibit very
mportant shift of flat band voltage due to injection of charges
directly from silicon into trap levels inside To avoid this
effect it is necessary to put between silicon surface and silicon
nitride layer, a thermally grown Si02 layer to prevent charges
injection into silicon nitride traps. These so called MNOS structures
are similar to MaS structures and we can realize MaS integrated
circuits using this double layer as insulating layer of the gate.
However great care must be taken with this technology and especially
with Si02 layer quality to prevent threshold instability due to
hole injection in Si0
2
layer. These holes choosed up by Si02-Si3N4
interface and threshold voltage shift can be very important.
However this charge injection and the shift of the flat band voltage
have been used to realize a memory, the so called MIII2S memory.
In this case, Si02 layer is very thin and we favoured electron in-
jection across Si0
2
layer to obtain flat band voltage shift. In the
last part we shall speak about these devices.
Deposited silicon nitride layer can be achieved by different
methods but the most widely used is the chemical vapour deposition
of Si3ij"4 after reaction in vapour phase of SiH4(or SiCI4) with NH3
with N2 or H2 gas carrier. Deposit temperature can be in the range
of 6S0C to 100QoC.Quality and characteristics depend on both tempe-
rature and SiH4 over NH3 ratio. The index of refraction increases
when temperature decreases and decreases when SiH4/NH3 decreases.
Values can be 1.98 at 8S0C and SiH4/NH3=1/100 and 2.03 at 770 and
SiH4/NH3=1/70. In the same way the dielectric constant depends
THIN FILM APPLICATIONS IN MICROELECTRONICS
18
10
17
10
A luminum diffusion inside Si3N4
versus annealing temperature
Silicon
- - - - - ~ ~ - - - P - - - - ~ - - ~ - - - - ~ - - ~ - - - - ~ - - ~ ~
100
200 300 400 ~ o
431
Fig. 8. Aluminium diffusion profile in Si3N4 -Si system as obtained
from ion microprobe measurements.
V.LeGOASCOZ
on the technological parameters. We observe a decrease when tempera-
ture decreases and an increase when SiH4 over NH1 ratio increases.
The values are in the range of 6.S to 7.4(10,11,12).
In the other hand, the dielectric strength is high and quite
similar to Si02 dielectric strength (10
7
V/cm)(13,IO). The dielectric
strength is better with HZ deposited than with N2 deposited silicon
nitride and with a very low SiH4 over NH3 ratio, the layer has the
lowest density of pinholes(12).
From the point of view of the conduction across the layer
which is of great interest for the stability of MNOS structures and
for MIII2S memory devices, the current transport across silicon
nitride is bulk controlled. The conduction due to Poole-Frenkel
effect is the most probable(13) at high field. As we have seen with
previous parameters, silicon nitride conductivity is strongly depend-
ent on technological parameters. The conductivity increases when
temperature decreases and when SiH4 over NH3 ratio increases(14).
We can minimize the conductivity by deposition between 800C and
900C and with SiH4 over NH3 ratio in the range of 1/100.
For devices as MIII2S memory transistors, it is very important
to obtain silicon nitride layer with poor conductivity to avoid
charge injection into the silicon nitride at the metal silicon
nitride interface. According to technological process, pollutton
by different species can appear at this interface, lowering barrier
and allowing easy injection from the metal. Pollution is mainly due
to aluminium diffusion inside silicon nitride during annealing. In
the case of aluminium diffusion, we can see on Fig. 8 aluminium
profile obtained with ionic microprobe in silicon nitride layer
for different annealing temperatures. As we have seen previously
for silicon dioxide, even at low temperature (400C) an important
part of metal can diffuse inside the insulator ~ d for SOOC this
diffusion can reach silicon interface with a doping level closed to
1019At/cm3
What is the impact of these resultsoon physics and on thin film
technology? For layer ticker than 1000 A, aluminium does not reach
silicon interface and no effect can be noted. But with MIII2S memory
devices, where conductivity is of main importance to exhibit memory
phenomena, this diffusion is of main importance. Effectively l u ~
minium diffusion changes the conductivity of the layers. Fig. 9
shows current versus voltage curves for two devices. The former is
annealed at 400C and the later at SOOC. We observe the important
changes of conductivity. The other curves are related to different
measuring temperatures. These results show how conduction phenomena
and injection phenomena can be modified by technological process.
THIN FILM APPLICATIONS IN MICROELECTRONICS
-,
10
10
.,Jt.
10
_I
10
Fig. 9.
E
F
DIAL 2b measured at
-36C
OC
--------- +80C
-- 4b ----- -36C
OC
-------- + 8o"C
25
30 35
I(V) characteristics of Si3N4 layers for different
annealing temperatures.
433
434 V. Le GOASCOZ
3. THIN FILM APPLICATIONS IN DEVICES
Up to now we have shown why we need thin insulating films in
microelectronic and what kind of dielectric films are used in the
technology.
This part will be devoted to the use of these thin films in
devices. We have seen that non metallic thin films have several
characteristics : chemical characteristics, mechanical characteris-
tics and at last electrical characteristics. These later characte-
ristics are extensively llsed in microelectronic devices. To illus-
trate the use of these layers we have choosen three kinds of devices.
MOS tunnel diode and FAMOS device to illustrate insulating property
and MI
I
I
2
S device to show how it is possible to use very weak con-
ductivity in insulating film to obtain memory phenomena.
3.1 - MOS tunnel diode
In thisodevice, the insulating property of a very thin Si02
layer 50 A) and the tunnel conductivity across this layer are
used in the same time. Fig. 10 shows the structure: metal, Si02,
N type silicon. If we apply a negative bias to the metal versus
the silicon bulk, the silicon will be in depletion. Minority carriers
coming from silicon bulk toward surface are drained off by tunnel
current Ivm' Therefore silicon surface cannot be in inversion be-
cause carriers generated in silicon are not sufficient to accumulate
at the interface(15, 16).
We note that the aluminium Fermi level is just facing the
forbidden silicon band gap, therefore, no current can flow this
way.
If we generate numerous carriers in silicon, for example by
light illumination, excess carriers will be at the silicon surface
and the surface potential will increase. Therefore metal Fermi
level will be facing the silicon conduction band and an important
current will flow between aluminium and silicon. In figure lOa we
can see MOS tunnel structure in the deep depletion mode. Minority
carrier generation is not sufficient to invert silicon surface and
to allow metal Fermi level to rise up to the silicon conduction
band. In this case no Imc current can flow from metal to silicon.
In figure lOb numerous minority carriers are available, due to
light generation. In this case, minority tunnel current Ivm across
Si0
2
from silicon valence band is less important than optical gene-
ration current. Therefore minority carrier accumulation occurs at
the interface, involving metal Fermi level rising up to the silicon
conduction band. The position of metal Fermi level determines current
flow and this position is due to silicon surface potential depending
on optically generated minority carriers.
THIN FILM APPLICATIONS IN MICROELECTRONICS
Aluminu
Aluminu
Sili con
Ec
-----====-Ef
Ev
deep depletion
with illumination
Ec

---Ef
Ev
depletion
Fig. 10. MOS tunnel diode structure with and without light
i 11 uminat ion.
435
436
SOURCE
floating
polysilicon
Si02
I
V. Le GOASCOZ
DRAIN
V<O
I
I
".-
I' " I
/
N type
breackdown zone
--r /
space charge region
Fig. 11. Vertical section of a FAMOS structure under bias.
THIN FILM APPLICATIONS IN MICROELECTRONICS 437
This device has an internal gain. We have measured values in
the range of 30. The thickness of SiOZ layer must be less than
50 A to allow significant current to flow by tunnel mode. The main
interests of these structures are a very good sensitivity of the
device to low lighting and a low value at the current flowing
across the device. Then performances of these devices are strongly
reduced by interface states at the SiOZ-silicon interface. Improve-
ment of technology is still necessary to obtain better results.
These devices can be used as single low level photodetector
as well as in photodetector matrix. This structure can also be used
for tunnel spectrometry by scanning the metal Fermi level over the
band gap to determine interface state location. At last, instead
of generate minority carriers by optical means, it is possible to
supply minority carriers by peripherical diode.
3.Z - FAMOS devices
Injection of carriers into Sib
Z
layer of a MOS or
transistor from an avalanche diode has been previously reported
by many workers(17,I8,19,ZO). In these cases charge injection was
produced by silicon surface breakdown due to large ac signal on
the metal gate. This injection produces a shift of the flat band
voltage specially with p substrate. Hot carriers can be produced
by another way. When we put a diode in breakdown mode hot carriers
are present at the silicon surface and can be injected in or across
the SiOZ gate. This effect has been extensively used by Frohman-
Bentchowsky(ZI) from Intel to realize a memory device.
Figure II shows a vertical section of FAMOS structure under
bias (FAMOS = floating gate avalanche injection MOS). This device
is similar to MOS transistor with the difference that the metal
gate is completely insulated by a SiOZ layer underlaying the gate.
This floating gate is made of polycrystalline silicon. When suffi-
cient voltage is applied to the drain diode to cause breakdown,
then hot carriers are injected across gate insulator and are col-
lected by the gate metal. With N type substrate, electrons are in-
jected in the gate insulator and the gate metal is charged negative-
ly by electrons. If the number of charges is sufficient, the field
produced can invert the silicon surface and current can flow from
drain to source. At the present time, we have no means to discharge
electrically this structure and we need U.V. light or X ray to
excite carriers and to discharge insulated metal gate. Retention
time of charges on insulated metal gate can be important (up to
10 years).
438
V. Le GOASCOZ
Aluminum
N Si
substrate
AI
J
SiJN4
(1
1 d
2
n:
Sj07
f ..
q-.,
i d,
Sj N
I
Fig. 12. Vertical section of a M1112S transistor.
THIN FILM APPLICATIONS IN MICROELECTRONICS
Fig. 13.
MNOS Memory on silicon sapphire
N channel
100ms
-,20
6
5
4
3
2'
-1
"Ws Si02 = 30A
SiJN4 = 450 A
fig 13
g
Memory cycle obtained with MNOS device in the 50S
technology.
439
440 V. Le GOASCOZ
3.3. - MI
I
I
2
S memory devices
MI
I
I
2
S device (22) is a particular device because it uses
the insultating properties of silicon dioxide and silicon nitride
films to control silicon surface potential and in the same time it
uses conduction properties of these two materials to change the
threshold voltage of the transistor. Change of threshold voltage
is achieved by charge injection inside the gate insulator.
Figure 12 shows a vertical section of a M I I I ~ transistor.
The gate insulator consists of two insulators. A very thin Si0
4
layer is closed to silicon and a silicon nitride film is depos1ted
over this first layer. The thickness of the Si02 layer can be in
the range of 30 A and the thickness of Si3N4 is in the range of
500 A. The conductivity of.Si02 and Si3N4 layer is related to cur-
rent lox and In respectively, flowing across layers under bias.
For Si02 thickness in the range of 30 A to 50 A, the conduc-
tion mechanism is a Fowler-Nordheim tunneling and for silicon
nitride layer, the Poole-Frenkel effect is the most important at
high field. When applying a voltage Vg to the MIII2S structure,
currents lox and In flow across layer. The magnitude of these cur-
rents is determined by electric field distribution :
Fields and currents are not identical in each layer, and charges
02 accumulate at the Si02-Si3N4 interface. The continuity of cur-
rent at the interface gives relations :
d02 _
dt - lox - In
Accumulation of charges at the Si02-Si3N4 interface gives a shift
of the flat band voltage
d2
~ V F B =-0
2
T
Figure 13 shows an example of memory cycle obtained with N channel
MIII2S device. Technological threshold is + 1.2 volts; with a
positive pulse (+25 V, 100 ms) we reach a threshold of + 7 volts
and then with a negative pulse we go back to technological threshold.
The thickness of the Si02 layer is very important in these devices
because injection and rate retention time depend strongly of Si02
thickness.
THIN FILM APPLICATIONS IN MICROELECTRONICS 441
With thickness in the range of 30 A the injection time is about
100 ms and with 17 A the injection time can be 100 ns. In the former
case, conduction across SiOZ is due to a Fowler-Nordheim process
whereas in the later case direct tunneling from silicon to silicon
nitride traps allows fast injection. Simultaneously the retention
time varies. With thick SiO
Z
layers, retention times of more than
ten years can be expected, whereas with very thin layers we have
only retention times of few weeks.
The main difficulty encountered with these devices is the need
to apply positive voltage to inject charges and then negative volt-
ages to discharge the structure. Because the need of these two volt-
ages, it is difficult to realize on bulk silicon a complex device
using MIIIZS components. With the use of silicon on sapphire techno-
logy, these difficulties are removed since each transistor can be
insulated from the others by silicon etching. This technology is
now in development in several laboratories.
4. CONCLUSION
This paper presents a brief survey of some thin insulating films
used in microelectronic. Our discussion was only concerned by insu-
lators extensively used in microelectronic industry. However in
research field, numerous materials are in use to study their pro-
perties. These layers as those used for example in Josephson devices
will be extensively used in few years but at the present time techno-
logy and properties are not well known.
Advanced devices in microelectronic are generally obtained with
well known materials except when a new material, with its particular
property, is an important and active part of the device. It is the
case of MIIIZS devices where very thin SiO
Z
layers was necessary to
obtain memory phenomena.
We have seen that several kinds of non metallic thin films are
used in microelectronic, either organic or inorganic material.
With the last and most sophisticated technology, the SOS MOS process,
all the layers have a thickness less than one micron. In the future
new devices are in the spotlight as micro MOS devices where all sizes
will be divided by a factor of ten : Channel length in the range of
one micron instead of ten micron ; gate thickness in the range of
100 A instead of 1000 A. The need of such thin layers indicates that
the technological process has to be changed. Etching process of Si02
or silicon will be obtained with new methods, for example by ion
plasma etching and thin Si0
2
layers will be obtained at lower tempe-
rature.
With the technology in use in industry, very large integration
circuits are realized. At this time we can have on the same chip
442
V.LeGOASCOZ
more than 10 000 transistors in a circuit. With the new technology
in development, complexity of 50 000 transistors and more on the
same chip is expected.
REFERENCES
1. A.G. Revesz and K.H. Zaininger, RCA Rev. 29, 22 (1968).
2. A.G. Reve&z, K.H. Zaininger and S.A. Evans, J. Phys. Chem.
Solids 7.8, 197 (1967).
3. E.H. Snow and B.E. Deal, J. Electrochem. Soc. ~ 263 (1966).
4. E.H. Snow, k.S. Grove, B.E. Deal and C.T. Sach, J.A.P. 36,
1664.
5. C.M. Osburn and E.J. Weitzman. J. Electrochem. Soc. ~ 603
(1972).
C.M. Osburn and D.W. Ormond, J. Electrochem. Soc. ~ 597
(1972).
C.M. Osbura and S.l. Raider. J. Electrochem. Soc., 120, (1973).
C.M. Osburn and J. Chou, J. Electrochem. Soc., 120, (1973).
6. T.M.UiStefano Appl. Phys. Lett. ~ 280 (1971).
J. A. P. 44, 527 (1973).
7. E.H. Nicollian, C.N. Berglund, P.F. Schmidt and J.M. Andrews,
J. A. P. 42, 5654. (1971).
8. P. Rossel, H. Martinot and D. Esteve, Solid State Electron.
11, 425 (1970).
9. A. Goetzberger, A.D. Lopez and R.J. Strain, J. Electrochem.
Soc. 120, 90 (1973).
10. G.A. Brown, J.C. Robinette and H.G. Carlson, J. Electrochem.
Soc. ~ 948 (1968).
11. V.Y. Doo, D.R. Kerr and D.R. Nichols, J. Electrochem. Soc.
~ 61 (1968).
12. M.T. Duffy and W. Kern, RCA Rev. 1l, 742. (1970).
13. S.M. Sze, J. Appl. Phys. 38, 2951 (1967).
THIN FILM APPLICATIONS IN MICROELECTRONICS 443
14. F.A. Sewell, E.T. Lewis and H.A.R. Wegener, Technical Report
AFAL-TR-70-148 (1970).
15. R.A. Clarcke and J. Shewchun, Solid State Electron. ~ 957
(1971)
M.A. Green and J. Shevchun, Solid State Electron. lL, 349
(1974).
16. V. Le Goascoz, J. Borel and A. Payo-Casares, ESDERC, Nottingham
(1974).
17. E.H. Nicollian and A. Goetzberger, IEEE Trans. Electron.
Devices ~ 686 (1968).
18. E.H. Nicollian, A. Goetzbergerand C.R. Berglund, Appl. Phys.
Lett. ~ 174 (I 969) .
19. Poirier and Olivier, Appl. Phys. Lett. ~ 364 (1969).
20. J.F. VERWEY, Appl. Phys. Lett. ~ 270 (1969).
21. D. Frohman-Bentchkowsky,Appl. Phys. Lett. ~ 8 (1971).
22. J.T. Wallmark and J.H. Scott, RCA Rev. lQ, 335 (1969).
E.C. Ross and T. Wallmark, RCA Rev. 30, 366 (1969).
E.C. Ross, M.A. Goodman and M.T. Duffy, RCA Rev. ll, 467(1970).
G. Dorda and M. Pulver, Phys. Status Solidi l, 71 (1970).
K.I. Lundstrom and C.M. Svensson, J. Appl. Phys. ~ 5045
(1972); Electron. Lett. ~ 645 (1970).
23. V. Le Goascoz and J. Borel, ESDERC, Munich (197J).
P. Gentil, V. Le Goascoz and J. Borel. Onde Electrique Nov.
(1972) .
J. Borel, E. Mackowiak and V. Le Goascoz, Colloque Memoires,
Paris (1973).
APPLICATION OF THIN NON-METALLIC FILMS IN OPTICS
E. Pelletier
Centre d'Etude des Couches Minces (Associe au C.N.R.S.)
St Jerome - 130013 MARSEILLE - France
1. INTRODUCTION
The optical properties of a thin film are
- the reflectance R (or R')
- the transmittance T
- the corresponding phase changes ~ r (or ~ r ) . ~ t
- the absorption is defined as : A = 1 - R - T or A' = J - R' - T
These measurements can give information about structure and
properties of the material of which the layer is composed. A very
connnon application of thin films is to modify the optical proper-
ties of a surface, for example to obtain an high reflectance or a
low reflectance without modification of the geometrical properties.
If the desired result can be obtained with only one layer, then no
complex problems of analysis or thechnique need to be faced, however
if a single layer does not achieve the desired results, we may be
able to conceive a multilayer system which will acheive it. This
poses a series of experimental and computational problems which are
not all completely resolved. We will discuss these problems in this
article.
2. PROBLEMS IN THIN FILM OPTICAL FILTERS
2.1 Multilayer Calculations.
It is easy to compute the optical properties of a single idea-
lized layer of which the index is n (complex) and the thickness is
e, deposited on a substrate. The computation of the optical proper-
ties of a multilayer stack (design equation: (ni,ei) i = 1,P, sub-
trate : no' incident medium: np+l) is tedious but it can be easily
445
446 E. PELLETIER
carried out with a digital computer, even if the ni values are
complex. So that the tI analysis" of any complicated design can be
obtained in few seconds if we assume perfectly flat, parallel homo-
geneous layers. However, this is not the main problem encountered
in the application of dielectric films. We may have need of a fil-
ter with a given profile of R versus wavelength. The main problem
is to determine the number of layers P, the indices and the thick-
ness of each one of the component layers to obtain the required
properties. We do not have a logical set of rules so that our "syn-
thesis" must proceed by trial and error. The calculations start from
a proposed design, and then adjustments may be made until a satis-
factory solution is found.
2.2 Problems in the Preparation of Optical Coatings
The calculations to be described permit us to determine the
design of a dielectric stack having prescribed optical properties.
We have to consider the following parameters :
1) Thickness - The finished product will correspond to the de-
sign specifications only if the three following conditions are ful-
filled :
a) Uniformity. Control of the uniformity to layer thickness
over the area of the substrate is an important factor. A lack df
uniformity causes a shift of characteristic wavelength over the
surface of the filter. The effects of lack of uniformity are depen-
dent upon the "spatial wavelength" A of the defect.
If A A then we must use "couches de passage" (defect of unifor--
mi ty of index).
If A A then we must use calculations carried out with the rules
of optical geometry (aberrations).
Between the two domains is a scattering domain.
b) Accuracy. The second factor is ,control of the overall thick-
ness of each layer. This creates a major practical problem, namely
the precise control of the optical thickness of the layer during
deposition. The best method is to use a thickness monitoring device
which responds directly to changes of optical thickness. Even if
the refractive index differs appreciably from the specification,
the difference can be partially compensated by an appropriate and
automatic adjustment of the geometrical thickness. The optical mo-
nitoring of layers of thickness p Ao/4 (p integer) is very easy,
because the optical properties T(Ao) are very peculiar. However,
the monitoring of stacks made of unequal layers is much more diffi-
cult.
c) Preparation. The basic manufacturing technique for the cons-
truction of thin films filters is that of vacuum deposition. With
this process the two above problems can be resolved and the number
APPLICATION OF THIN NON-METALLIC FILMS IN OPTICS
of available materials is not large. Dielectric materials are of
particular interest because of their low absorption.
447
2) Refractive index - The refractive index depends on the che-
mical composition of the material. The number of materials found
to be suitable for optical coatings is limited, so the number of
input values of index should be minimised in the design of filters.
This limitation can be partially resolved in two ways :
- The co-evaporation to two dielectric materials mixed in various
proportions to provide refractive index values ranging between
those of pure materials.
- A mathematical equivalence : A symetrieal stack made of two ma-
terials nl, n2 is equivalent to a single layer of equivalent index
n (nl < n < n2)
The judicious design of the thicknesses and_of the total thickness
gives a stack with the desired values of nand e. This statment is
valid at a given wavelength and remains good enough in a broad band
if the total thickness is smaller than the wavelength. The main in-
convenience of this method is a large increase of the number of
layers of the filter.
3) Environmental resistance. Probably the most important as-
pect of the environmental performance of the filter is its resis-
tance to humidity, but the resistance to other agent such as tem-
perature, shock and corrosive fluids, may all be important. All
these difficulties are dependent on the specified application, but
we are often thereby confined to a limited set of available materials.
3. MAIN APPLICATIONS : CLASSICAL FILTERS
In this paper it is not possible to give a detailed study of
various thin film optical devices. For more details see (1).
I) Antireflection coatings
- single layer antireflection
- multilayer antireflection
2) High-reflectance mirror coatings
- Multilayer dielectric stack Ao/4
- Extending the high reflectance zones
3) Edge filters
- Symmetrical multi layers and the Herpin index
- Extending the transmission zone
4) Band pass filters
- Broad-band pass
448 E. PELLETIER
- Narrow band filters f Fabry Perot and Double half-wave)
5) Polarizing beam splitter
The main part of these filters are obtained with quarter wa-
ve stacks Ao/4 made of alternate high index and low index layers.
This is for several reasons :
- very little absorption
- the optical properties are stationary in the neighbourhood
of AO' and of a = 0, so main part of these filters can be illumi-
nated with a convergent beam.
- optical monitoring is convenient
- the layers deposited can be almost uniform and homogeneous.
4. SYNTHESIS OF MULTI-LAYERS WITH PRESCRIBED OPTICAL PROPERTIES
4.1 The Problem
We can increase the field of application of thin films consi-
derably with stacks made of layers having unequal thicknesses. In
order to obtain prescribed optical properties, what are the succes-
sive indices and thicknesses making an optimum design ? The synthe-
sis may be performed by successive iterations and the free parame-
ters which are adjusted can be :
- the indices of each layer
- the optical thicknesses
or both together. These specifications can be used to study the dif-
ferent methods of synthesis, which are summarized in the following
table.

thicknesses
niei
Ao/4 unequal
indices stacks thicknesses
n'
1
available
dielectric classical free parameters
materials filters e'
1
intermediate POHLACK's free parameters
in#ces relations ni and ei
APPLICATION OF THIN NON-METALLIC FILMS IN OPTICS 449
4.2 Synthesis with Constraint over the Thicknesses
This method involves the case in which the optical thicknes-
ses of the stack are equal to Ao/4. For a stack made of Players
we can write :
liT ~ Ao + AlcOS IT Ao/A + .. + Ap cos P IT Ao/A
The identification of this polynomial with an analogous development
a
o
' aI, ~ giving the Fourier coefficients of the prescribed
optical propert1es, can permit us to determine the number of layers.
In fact, the application of this method is very limited because the
constraints over the optical thicknesses, which must be kept equal,
is too restrictive. In addition, the values of the refractive in-
dices found from these equations are between 1 and about 6 and the prac-
tical realisation of these designs is impossible exactly in this
form. Analogous calculations can be carried out for angles of inci-
dence other than normal ; in this case, the effective phase thicknes-
ses are not stationary and the refractive indices should be repla-
ced by the more general optical admittance (ni cos 8i or ni/cos 8i)
according to the polarisation.
4.3 Synthesis without Constraints
This calculation starts from a proposed design and adjustments
may be made until a satisfactory solution is found. At the begining,
we may define a suitable criterion for comparing together the quali-
ty of anyone design; so a "merit coefficient", f permits us to com-
pute the distance between curves showing the optical properties. The
problem lies in finding the values of the free parameters (that is
to say the design) corresponding to the minimum of the distance f.
The vast volume and complexity of these calculations overload
the capacity of available computers. The best design found under
these conditions is the "nearest" to the starting value. It is com-
pletely impossible to study all the designs completely. For example,
for the function f( ni .. ei"')' if P = 10, the calculations must
be performed with N values of nand E values of e. The number of de-
signs is (N.E)P. Assume that N = 10, E = 10 and that the computation
of f for one wavelength is about 10-6 second, then the complete stu-
dy of the whole solutions is made in 10
14
second ~ 3.106 years!
This is not the only difficulty : it is also not possible to add suc-
cessive layers to a refined design. The best solution with q layers
does not have the same structure as the best solution with p + q
layers.
4.4. Synthesis with Available Materials
A method for simplifying the synthesis problem is to reduce
450 E.PELLETIER
arbitrarily the number of parameters. In addition, to simplify the
experimental production of the coatings, only common evaporation ma-
terials are used, the only variables in the calculation being the
mechanical thicknesses of the films. A wide variety of filtering
properties can be obtained by the use of only a limited set of ma-
terials. But these restrictions can seriously prevent the determi-
nation of an optimum design.
4.5. Some results
The methods of synthesis are not perfect but we are now able
to use them to find theoretical designs for numerous problems in
the design of filters for normal incidence or oblique incidence
We can say that the large part of these problems can be resolved
with less than twenty layers, made up of four or five materials.
However the methods of calculation are still very empirical. Des-
pite these difficulties, very beautiful results can be shown; for
example in the case of band pass filters in the infra-red.
(Dobrowolski). The free parameters are thicknesses and indices.
Another peculiar example concerns the design of a beam split-
ter for the P polarisation (8 = 55) for several visible and infra-
red wavelengths. This design is computed by assuming that the com-
ponent materials are Zinc Sulfide and cryolite. This result permits
the importance of the dispersion of the refractive indices v r ~ u s
the wavelength. This factor cannot be neglected if the optical pro-
perties of the design may be determined with good accuracy. It will
always be necessary if the allowed tolerances are small.
5. PROBLEMS IN THE PREPARATION OF OPTICAL COATINGS
5.1 Readily available materials
It is sufficient to quote a list'of available dielectric ma-
terials which are commonly used. The spectral region of transparen-
cy and the refractive indices are the two main points. For an accu-
rate synthesis, and above all for precise monitoring, an exact know-
ledge of the values of the refractive index of a thin film is neces-
sary. Sophisticated methods of determination may be used to obtain
this result (for example nand k of ZnS in the visible).
5.2 Preparation of a Layer of Uniform Thickness
The uniformity of a layer depends on the source-to-substrate
distance. The layers obtained on a rotating substrate are much more
uniform in thickness than on a fixed substrate. With appropriate ad-
justments, we can experimentally determine the best disposition for
a given evaporation geometry. A very good uniformity is of the ut-
most importance for some kinds of filter such as narrow band pass
APPLICATION OF THIN NON-METALLIC FILMS IN OPTICS 451
filters. On the other hand, for achromatic coatings, a perfect uni-
formity in thickness is not absolutely necessary (However do not
forget 1/Ir)'
5.3 Monitoring of the Deposited Thickness
For more details see(2).
- quartz crystal method
- ionisation gauge
- OPTICAL MONITORING(3)
Thickness monitoring devices respond directly to changes of
optical thickness. A difference between the effective and the spe-
cified values of refractive index is compensated by an automatic
adjustment of the geometrical thickness : the optical thickness is
>"0/ 4
a) Principle: As the thickness of a dielectric film deposited
on glass increases, the reflectance or transmittance of the coated
surface has a maximum or a minimum value (turning point) at the de-
sign wavelength >"0' whenever the optical thickness of the film pas-
ses through a multiple of >"0/4.
b) The location of extremums:. In order to increase the accu-
racy of the location of the turning points, two very different me-
thods have been developed :
- Zero of dT/de. In practice the voltage signal from the detector
T, may be recorded and the transmittance is differentiated with
respect to time. An extremum is recognized by the zero value of
dT/de and this is the signal for the end of deposition.
- "Maximetre" of Giacomo-Jacquinot .. Essentially, the transmittance
(or reflectance) of the coating is differentiated with respect to
wavelength. An extremum is recognized by the zero value of
dT/d>". It is very important to repeat that the zero of dT/de should
not be confused with the zero of dT/d>". A complete comparison of
the two monitoring processes is necessary.
c) Monitoring of dielectric stacks
- Classical filters: The optical thicknesses of the component layers
may be equal to >"0/4. The optical monitoring of >"0 is very easy and.
a knowledge of the value of the refractive indices is not necessary
The whole stack can be monitored on the same substrate and for each
lay, we must stop deposition as soon as the zero point of dT/de (or
dT/d>") is reached.
- Unequal-layer stack: The experimental production of no quarter
452 E. PELLETIER
wave layers requires a "control programme" that must be followed
step by step to obtain the required optical properties. If the de-
sign is n1e], nZez, n3e3' we must first compute the evolution of
the spectral profile T{X) during the deposition of layer number one
{i.e. the thickness increases from zero to n1e1).Now we have toas-
sume that layer number one is monitored. A new calculation
of the optical properties of stack {glass + n1el + nZeZ) gives
the wavelengths for direct optical monitoring Th1S process is
continued for each layer; a trace of the evolution of the 'spectral
profile is made and we can then choose for each one the type of
differential and the monitoring wavelength giving the best accura-
cy (Note that oT/oe = 0 and oT/oX= 0 must not be confused). The va-
lues of the refractive indices must be known with a large accuracy
in order to perform these calculations.
5.4 The Importance of the Choice of Monitoring Process
Over all the methods, the ones which depend directly on op-
tical properties are usually preferred over those which are less
direct.
a) Direct and indirect monitoring: The optical monitoring of
the whole stack on only one monitoring plate can be difficult to
perform. It seems easier to use several monitoring plates. Under
these conditions, we monitor only a small number of component la-
yers on each monitoring plate, the work piece being as close as pos-
sible to the monitoring plate and receiving the whole stack (indi-
rect monitoring). Unfortunately the best accuracy is obtained when
the monitoring is performed on the whole piece itself (direct moni-
toring) because the thickness deposited at the same time on the two
different plates are never exactly equal. (The angular distribution
of the evaporant beam from crucibles varies with time). Sometimes
the work piece cannot be used to perform the monitoring (metallic
substrate, geometrical size, etc.) and, the next best method is to
use only one monitoring plate.
b) Relation between production errors and the monitoring pro-
cess used: Sensitivity is not the lone criterion in a comparison of
optical monitoring processes. We can demonstrate the need for a
"stable" monitoring process in which the cumulative effect of er-
rors in successive layers is considerably reduced. Under these con-
ditions, whatever small errors occur in the early layers of the
stack can be corrected. This notion will be developed in a particu-
lar example : take the filter : glass H L H L H L H L ZH L H L H L
H L H air with the notations H high index layer of optical thickness
ness A/4 (zinc sulfide)
A/4 (cryolite) A = 5461 X.
A 10% error in the optical
a filter of design :
L low index layer of optical thickness
thickness of the spacer layer results in
APPLICATION OF THIN NONMETALLIC FILMS IN OPTICS
453
glass H L H L H L H L (2.2H) L H L H L H L H air of which the peak
transmission is in error by about 220 A. However, this error will
have no marked effect on the optical properties of the completed
filter if direct optical monitoring of (aT/ae)A = 0 is used. For
layer number 9 aTg/aeg 0 but the next layer is then corrected
such that aTIO/aeIO = 0 (different from A/4) thus obtaining a fil-
ter of design: H L H L H L H L (2.2H)(0.66 L) H L H L H L Hair.
The error is thereby perfectly corrected and the filter obtained is
centred to better than 0.1 A. Under these conditions, the usefulness
of a "stable" monitoring process is obvious.
5. CONDITIONS UNDER WHICH FILTERS MAY BE USED AND MANUFACTURING
TOLERANCES
5.1 Performance Specification
To simplify as far as possible the problems of synthesis, the
calculation of the optical properties are made for only one single
angle of incidence 0
0
, It is important to verify that the optical
properties do remain stationary in the neighbourhood of 0
0
, In an
optical arrangement the energy carried by a beam of light is direc-
tly proportional to the aperture. The performance of a filter is
therefore dependent on the conditions under which it is used. Kn
this respect the case of narrow band interference filters is parti-
cularly relevant and we give one single example of this.
Consider a filter of design M7 8L M7 illuminated by a parallel
beam of light having the following performance :
o
Maximum transmittance : 0.86, Peak wavelength: 5623 A, Band-
width = 14 A. If the filter is illuminated by a beam of aperture
f/2.8 the performance will be only : Tm = 0.31 Am = 5608 1
= 47 A. As soon as the aperture of the beam is greater than f/IO,
a considerable degradation of the performance of a narrow band fil-
ter may be observed. In addition, in the calculation of the optical
properties of a filter the component layers are assumed to have
ne parallel surfaces. In fact, the layers have microdefects which
are important in the case of narrow band interference filters.
se cause both a reduction of peak transmission accompanied by a
broadening of the pass band and a loss of luminous energy by scat-
tering.
Thus to compare theoretical and experimental performances it is
best to illuminate only a small area of the filter.
Furthermore, the peak wavelength of a filter depends on
ambient temperature, the variation being of the order of 0.2 A per
degree C.
454 E.PELLETIER
The conditions under which narrow band interference filters
should be used are relatively restricted. This is not usually the
case for other types of filter although the possibility of such res-
trictions should always be borne in mind.
5.2 Improvment of Manufacturing Precision
Taking account of all the above elementary precautions, the
next step is the comparison of calculated and experimental perfor-
mance. If there is considerable difference between these it is of
course necessary to investigate the causes so that the manufactu-
ring process may be improved.
a) Defects in the monitoring system - During deposition, the
observed values of Ti (or dTi/dA)Ai will be in good agreement with
predicted values if and only if the following conditions are ful-
filled :
- the indices of refraction of the deposited layers are exac-
tly known
- each of the layers in the stack has exactly the required
thickness.
Computer simulation 4as been found to be a most powerful me-
thod for the analysis of possible defects in the monitoring method
and has also yielded much information on the magnitude of refracti-
ve index and thickness errors which may be expected in practice.
Let us consider for example a filter which consists of only
three layers. The required thicknesses are ej, e2, e3. If we use
refractive indices nt, n2' n3,we find that the correct wavelengths
for use with the Maximetre are At, A2, A3' Now let us suppose that
in practice the actual values of refractive index are slightly dif-
ferent, say, NI, N2, N3, with Nj ~ nl, N2 ~ n2, N3 ~ n3' If the
layers of the completed filter had nevertheless the correct values
el, e2, e3, then the spectral profile of the filter, T(Niei) would
be very nearly the same as the calculated profile T(Oiei). However,
the errors in the values of refractive index affect the accuracy
of the monitoring process and therefore, the thicknesses of the de-
posited layers. Using control wavelength, AI' A2, A
3
, perfect mo-
nitoring, that is to say monitoring in which there are no other
errors, would give thicknesses EI, E2' E
3
, such that (dTi/dA)A.=O
i.e. EI is such that for the combination (glass + NjEI) we h v ~
(dTI/dA) = 0 for the wavelength AI. E2 is such that for (glass +
NIEI + N2E2) we have (dT2/dA)A2 = 0 and so on. The filter obtained
under these conditions can have a sepctral profile T(NiEi) very dif-
ferent from that expected, T(niei). Not only are these errors in
the values of refractive index but much more important there are
considerable errors in the layer thicknesses.
APPLICATION OF THIN NONMETALLIC FILMS IN OPTICS 455
Computer simulation is an excellent tool in the determination
of precise values of refractive index. It permit one to take ac-
count of errors which are introduced when deposition is not termi-
nated exactly at the correct point when a zero of dT/dA is obser-
ved. It has been used to demonstrate the importance of choosing a
"stable" monitoring process in which the cumulative effect of er-
rors in the thickness of layers already deposited on the thicknes-
ses of succeeding layers is not at all catastrophic. Under these con-
ditions the manufactured filters will have acceptable performance
regardless of those errors which may exist in the thicknesses of
the individual layers. For the best monitoring procedure it is ne-
cessary to choose for each layer that monitoring method (e.g.
dT/dA or dT/de) which is optimal, taking account of both the sensi-
tivityand the "stability" of all possibilities.
b) Study of a completed experimental filter - It is intersting
to determine the errors which have been made in an experimental
filter in order to avoid making the same error in subsequent fil-
ters. In order to do that it is necessary to calculate the thick-
nesses which have actually been deposited. This is a synthesis pro-
blem : what are the thicknesses of the component layers knowing the
spectral profile of the filter ? Without additional information this
problem cannot be simply resolved and synthesis programmes can give
several solutions. Computer simulation of monitoring together with
the study of measurements obtained during deposition of the filter
are therefore most useful. Under these conditions the synthesis
problem can be easily resolved if the number of unknown parameters
is small. If, during the deposition, a rapid measurement of the fil-
ter could be made with good precision, it would be possible to check
step by step, the performance of the stack and to make slight ad-
justments of the control programme which could minimise the effects
of any error on the monitoring of the following layers.
In addition, synthesis calculations can be performed actually
during deposition to calculate the optimum thicknesses for the sub-
sequent layers of the filter taking account of the errors which
have already been made. All this can be obtained with a real time
computer. This method is difficult to put into operation. However
it does represent a necessary step if complex filters are to be ma-
de successfully in industry.
c) Spectral profile - Among the earliest control techniques,
the observation with the naked eye of the colour of the coated
glass (in reflection) was already a global evaluation of spectral
profile. This method should 0e perfected so that it can be used
with any kind of stack. With the help of a rapid scanning monochro-
mator one could measure over a large spectral range the transmit-
tance (or reflectance) of a stack during deposition. In this way
one would be replacing a very precise measurement at one or two
wavelengths with a less precise control of the global spectral
456 E. PELLETIER
response. The usefulness of this method is especially clear for
filters which are to be used over a large spectral range. Because
of its complexity, this method is not in current use.
5.3 Manufacturing Tolerances
Taking account of the required optical properties (and of the
tolerances in these optical properties) it is intersting to be able
to put a figure on the accuracy necessary in obtaining the correct
values of both refractive index and thickness for each layer du-
ring the production of a filter. This calculation is far from be-
ing evident. Nevertheless it is necessary to complete our synthe-
sis methods so that we can give to the manufacturer designs which
can be easily produced. The synthesis problem must therefore be
enlarged by the introduction"of this "new dimension". In this way
one would have the maximum chance of success for industrial pro-
duction.
At the start of the section on synthesis we defined a "coef-
ficient of merit" representing the distance between the ideal op-
tical properties and those of the filter calculated at each itera-
tion.
For Tperfect the best solution obtained, f = E 0
This same definition can be used to express the distance between
the optical properties of the perfect filter and those of the fil-
ter which is actually produced
f = II Tperfect - Texp II
If the parameters ( ei, ni' ) are independent, the pro-
duction tolerances will be large if the partial derivat\ves
The minimum f( ni ei ) is large and deep. In general the-
se solutions are those which have the best chance of being found by
our synthesis methods.
In fact, to obtain good precision it is necessary to use di-
rect monitoring method. But in this case the errors in the optical
thicknesses of successive layers are not independent. We have just
seen that they perturb the control of subsequent layers (unless
one has a real time computer). Optical monitoring has the advan-
tage of measuring the optical thickness niei: an error in ni is
partially compensated by an error in ei' But the control of the
following layers risks being seriously perturbed. Thus the tole-
APPLICATION OF THIN NON-METALLIC FILMS IN OPTICS
457
rances necessary for the ith layer can be very tight even if the
derivative of/oei = 0 ; they depend on the monitoring method cho-
sen (optical monitoring, direct or semi-direct, detection of the
zeros of oT/oe or of oT/oA , .... choice of the wavelength for each
layer ) and on the errors which have occured in earlier layers.
Further, the accuracy of the measurement of the optical thicknes-
ses depends on the monitoring method chosen for each layer. Nume-
rical calculation of manufacturing tolerances can only be carried
out in several particular cases. It is quite impossible, at the
present stage of development, to give general conclusions.
6. CONCLUSION
To day, in the industrial sphere, an extensive range of fil -
ters with very diverse optical properties is produced. Without
doubt, in the years to come, one will see spectacular progress in
this field. This will certainly pose a whole series of completely
new problems. The development of filter design synthesis must take
account of the problem of manufacture. Only a limited number of ma-
terials are available and also the conditions which are necessary
to obtain resistant layers (heated substrates) are difficult to
reconcile with accurate monitoring. The indices of refraction of
component layers can vary easily by several percents as a function
of parameters which are difficult to control : pressure, tempera-
ture, and so on. Mixed layers are a fortiori more difficult to pro-
duce with precision. The control of optical thickness becomes dif-
ficult when the required accuracy is of order of 1%. In any case
we can never avoid some imprecision. In addition, absorption, struc-
tural irregularities are never strictly zero. The manufactured pro-
duct can differ considerably from that projected. In such cases
one must simply look for the causes and it is thus that one makes
progress.
ACKNOWLEDGMENTS
The author wish to thank Dr A. Holmes Siedle and Dr H.A. Mac-
Leod for their help in the translation of this work.
REFERENCES
1. H.A. MacLeod "Thin Film optical filters " Adam Hilger Ltd
London (1969).
2. "Compte rendu des journees de controle continu de deposition
sous vide" Geneve 21-22 juin 1971
3. "Optical Monitoring" Le Vide I, 157 (I 972).
SOME APPLICATIONS OF NON-METALLIC THIN FILMS
M. H. Francombe
Westinghouse Research Laboratories
Pittsburgh, Pennsylvania 15235 (U.S.A)
1. INTRODUCTION
Thin films of elemental semiconductors such as Si, and of pas-
sive dielectrics, such as Si02 and Si3N4, have already found wides-
pread applications in solid state component technology and in high-
density microelectronics. As preparative techniques develop, films
of more complex semiconducting and insulating materials are recei-
ving increased study and are gradually being incorporated in a va-
riety of novel devices, both active. Many of these de-
cannot yet be considered as part of the standard range of mi-
croelectronic components, and in fact their hybrid organization,
and final structural form and function are still the subject of
speculation and experimental development. The numerous innovations
occuring in thin film devices are amply illustrated by recent work
in fields such as microwave diodes and trans:i:stors, thin film tran-
sistors, infrared detectors, magnetic bubble memories, photovo1taic
solar cells, microwave acoustics, capacitors and solid state ima-
ging and display systems.
In this lecture we shall consider some examples of recent ap-
plications of thin films, in which much of materials and fabriea-
tion technology is still undergoing research and development.
The special features of thin film preparative approaches being used
for these devices will be discussed briefly and the problems for 0P-
timizing film quality and the device structural form in relation to
its required function will be outlined. There are, in fact, numerous
examples of applications in various stages of development from which
we could choose, but time and space does not permit us to cover mo-
re than a representative selection. In the following sections we
shall discuss thin film aspects in the development of the GaAs field
459
460 M.H. FRANCOMBE
effect transistor, infrared detectors magnetic bubble and ferro-
electric memories, capacitors and some Iaagneto-optic and electro-
optic structures suitable for use in integrated optic elements,
displays or imaging systems. For a more complete review of work on
these and other film applications the interested reader is refer-
red to journals and texts dealing with thin film and solid state
device topics [e.g., References (1)-(3)] and to special transac-
tions of the IEEE covering microwave devices,solid state imaging
and microwave acoustics.
2. THE GaAs FIELD-EFFECT TRANSISTOR
2.1 General Considerations
Historically, thin films have played an important role in the
development of transistors. This is especially true of the thin
film transistor (TFT) of the Weimer type (4) which utilizes a thin,
polycrystalline high-mobility semiconductor film, usually of CdS
or Te,together with an insulated metal gate in a field-effect struc-
ture. It also is true of conventional bipolar and field-effect tran-
sistors based upon epitaxial Ge, Si or GaAs structures. As the need
for higher frequency signal generation and amplification atincrea-
sed power levels has .grown, special requirements for shallow diffu-
sed regions or for thin epitaxial films (especially in Si and GaAs)
have emerged. For bipolar transistors operating at GHz frequencies,
base regions of submicron thicknesses are required, and defect-
free material relatively free of traps is mandatory in order t.O
keep the transit time for minority carriers short.
To achieve high-frequency and high-pmver operation, semicon-
ductors posseesing high mobility and large energy gap are prefer-
red, and thus recent work in this field has been directed towards
compound semiconductors such as GaAs or. InP. (5) Unfortunately,
despite improvements in crystal growth techniques it has proven
very difficult to reduce the traps in such materials to densities
small compared with the minority carrier density, and as a result,
operating frequencies in minority carrier devices have been disap-
pointingly low. Three-terminal devices of the FET type, in which
the current flow may be due to either minority or majority carriers,
possess several attractive advantages. In these devices current flow
between the source and drain contacts (see Fig. 1) is modulated
by a voltage applied to the gate. The active region of the device
is a thin film ~ 1 micron) which is electrically isolated from
its substrate either by making the substrate from insulating
material or by using material of opposite electrical type.
APPLICATIONS OF NONMETALLIC THIN FILMS
461
SOURCE
o
(0)
(b)
Fig. I Simplified view of a junction FET mounted on an insulating
substrate (a) depletion layer profile for operation in the line-
ar region; (b) depletion layer profile for operation in saturation.
In silicon integrated circuits used for memory or logic appli-
cations, FET devices with insulated gates, MOSFET's, built either
on silicon substrates of opposite conductivity type 'or on sapphire,
find widespread use. In these FET's the source and drain electro-
des constitute junctions (e.g., with p-type regions diffused in an
n-type layer) and the device is made to conduct by applying a gate
potential of sign and magnitude suitable to introduce an inver-
sion region or channel under the gate insulator. The channel con-
462 M.H. FRANCOMBE
ductance between source and drain then occurs at the silicon sur-
face via transport of minority carriers. In the junction-, or
Schottky-barrier FET, the source and drain electrodes constitute
ohmic contacts to the active film region, and with no gate poten-
tial applied the channel region conducts via transport of majori-
ty carriers. As reverse bias is applied to the gate junction or
Schottky barrier the channel is depleted of majority carriers, and
with increasing gate bias the conducting region is progressively
constricted and eventually "pinched-off" [Fig. l(b)J.
In majority carrier devices of the junction FET or Schottky-
barrier FET type reasonably high levels of defects and traps can
be tolerated. However, it is desirable to use semiconductors with
high carrier mobility and saturated drift velocity. Also, for high-
temperature or high-power operation both the energy gap and thermal
conductivity should be high. With the exception of the thermal con-
ductivity, which is about three times that of Si, bulk n-type GaAs
satisfies these requirements admirably. The problem in flaking an
FET device is to duplicate the desirable bulk GaAs properties in a
thin electrically isolated thin film.
2.2. Electrical Properties
The requirements placed upon the fabrication technology for
a GaAs microwave FET may best be appreciated by considering the fac-
tors influencing the low- and high-frequency operation of the devi-
ce. Following the treatment given by Hower et al. (6) for a junc-
tion device as depicted in Fig. I, we may use the gradual-channel
solution approach of Shockley to describe the form of the I-V cha-
racteristics for the FET. For a small potential difference bet-
ween source and drain the depletion layer will extend uniformly un-
der the biased gate [Fig. l(a)J. However, as the drain current,
In, is increased, a significant potential drop builds up along the
channel and the depletion layer varies in thickness [Fig. l(b)J.
The gradual channel solution can be used to predict the I-V charac-
teristics if the drain-to-gate bias does not exceed Uo=qNna2/2KEo'
a the bias, where Nn is film
a LS the fLlm thLckness, q LS 1.602 x 10-
1
9 C, Eo LS 8.85 x 10 14
F/cm, and K is the relative dielectric constant. ('" 12). The resul-
tant I-V characteristics are shown in Fig. 2, and are obtained by
plotting the normalized current derived from a function which
is the result of integrating Ohm s law over the entire length of
the channel. The current is plotted as a function of normalized
drain voltage Vn/Uo for different values of normalized gate volta-
ge, n = (VB - VG)/U
o
(where VG is the gate voltage and VB the junc-
tion built-in-voltage). The normalizing current, 1
0
, is related to
the device parameters by
10 = G
o
U
o
/3 (I)
APPLICATIONS OF NON-METALLIC THIN FILMS
1.0
0.5
/
/
0.5
/
/
/
/
/
/
1.0
463
..,,=0
..,,=
Ve-VG
uo
'0.2
0.4
0.6
0.8
1.0
Vo IUO
Fig. 2 Common source current-voltage characteristics of an FET
based on the gradual channel solution. Dashed line indicates the
onset of saturation for drain-to-gate bias sufficient to cause
pinch-off.
where
(2)
~ is the mobility and Z/L the geometrical width-to-length ratio ;
Go is the unmodulated conductance of the material under the gate.
'If the gate voltage is kept constant and the drain voltage
increased, a point is reached when the potential drop from drain
to gate is U
o
This condition is termed "saturation" and occurs
along the line where
(3)
plotted as the dashed line in Fig. 2. The drain current in satura-
tion varies only with gate voltage according to the relation
ID /10 = I - 3n + 2n
3
/
2
sat
(4)
464 M.H. FRANCOMBE
When n = 1 the drain current is reduced to zero and the entire
channel is in the pinched-off condition. The corresponding value
of "pinch-off" voltage is Vp = U
o
- VB.
Two small-signal quantities of interest are the source-drain
conductance in the linear range
(5)
s
!-L.-...... I. -:--L--- ..... I Ld-l 0
(0)
Cgd Rd
G
o ~ ~ I ~ ~ ~ J
( b)
Fig. 3 (a) Active portion of the FET in saturation. (b) Circuit
model of the FET.
APPLICATIONS OF NONMETALLIC THIN FILMS 465
and the transconductance in saturation. This transconductance is
defined as gm = aID/avD, and for VD VD
sat
is also equal to
Go (1 - Iri).
With reference to the above relationships and to the circuit
parameters shown in Fig. 3 we can describe some of the quantities
used to define the frequency limits of performance. Figure 3(a)
depicts the neutral and depleted regions of the device for termi-
nal bias values corresponding to the conditions of saturation. Fi-
gure 3(b} shows the corresponding lumped-element circuit model
where the RS and RD are the source and drain extrinsic resistances
and include the effect of spacings LS and L
D

An internal cut-off frequency for the device can be defined
in terms of the time for the gate capacitance, C
gs
' to be changed
through the "channel resistance," rc. The correspoudiug rac!ian
frequency is Wo = l/rcCgs. If the effect of the parasitic resis-
tive element, R
s
, is considered, the frequency is modified to a
new value,
(6)
where G
s
= l/Rs and gc = l/rc. For the ideal case of Rs = RD = 0
it may be shown that gc = 3 gm and the cut-off frequency is given
approximately as
(7)
Other high-frequency performance criteria often quoted are the
unilateral gain, U, and the maximum frequency of oscillation, f
max

The gain may be expressed as
U
g'
m
(8)
where the prime denotes the inclusion of the effect of the parasi-
tic element Rs and x = w/wl ; W is the operating The
unilateral gain decreases as the square of the frequency (f2), i.e.,
by 6 dB/octave (see Fig. 4). Setting U = 1 gives the maximum fre-
quency of oscillation.
(9)
To obtain high values of U and fmax' devices with small gate lengths
(and hence small C
gs
) and also low values of parasitic resistance
are required.
466
25
M.H. FRANCOMBE
Run 26-14
Ios=17mA, Vos=+6V
Coax package
Fig. 4 Unilateral gain, U, and maximum available gain Gmax for
GaAs FET, "TFE-3" geometry in coaxial package.
2.3 Device Fabrication Aspects
Turning now to the aspects of FET fabrication, we discuss
briefly some thin film problems vital to the optimization of de-
vice parameters. The GaAs epitaxial layer usually is grown on a
semi-insulating GaAs substrate by the well-known open tube vapor
transport technique developed by Knight et a1.(7) using high puri-
ty AsC13, Ga and H2' Using this method, epitaxial films with net
donor concentrations in the 1014/ cm3 range with Hall mobilities
of 8300-8400 cm
2
/V-sec at 300
0
K have been achieved. Films grown
by liquid-phase epitaxy show even higher mobi11t1es but are diffi-
cult to produce with the thickness uniformity and surface smooth-
ness needed for FET fabrication. Accurate control of GaAs film
thickness and of epitaxial quality are important. The thickness,
APPLICATIONS OF NONMETALLIC THIN FILMS 467
in relation to doping concentration, must be tailored so that pinch-
off can be achieved at gate voltages not exceeding breakdown for the
Schottky barrier. This breakdown voltage decreases with increasing
carrier concentration of the GaAs and is lowered locally by the pre-
sence of defects in the epitaxial film.
r
~ -Type GaAs Epitaxial Film
J NOlll3x I015cm-3, 11111 .51'-
Semi -lnSUlatinQ
GaAs Substrate
(p =IOLI09ncm)
CD DEPOSIT n-TYPE
EPITAXIAL FILM
o EVAPORATE SOURCE,
DRAIN AND BACKSIDE
METAL
g,solation
f5\ ETCH SOURCE - DRAIN
\V "ISOLATION"
Gote Metal
(2) EVAPORATE GATE
METAL
~ MASK FOR SOURCE-
DRAIN METALlZATION
Drain Contacts
".dl
o LIFT SOURCE -DRAIN
MASK, ALLOY CONTACTS
o MASK FOR GATE
METALIZATION
. Source
Gale Droin
o LIFT GATE MASK
Fig. 5 Process steps for fabrication of a Schottky-barrier-gate
GaAs FET.
468 M.H. FRANCOMBE
Figure S illustrates schematically the main processing steps
involved in fabricating a GaAs FET device. Assuming a gate length
of L = S ~ m the epitaxial thickness is choosen to be I.S ~ m . If a
pinch-off voltage of 6V is desired, then the donor concentration,
ND = 3 x 10IS/cm
3
The process steps in Fig. S are self-explana-
tory and simply depict the use of photolithographic masking steps
to incorporate ohmic source and drain contacts and Schottky bar-
rier gate in the device. The ohmic contacts are produced by eva-
porating a mixed metal film, consisting typically of Au-Ge (9S/S)
or Ag-In-Ge (90/S/S). Alloy formation with the GaAs surface, lea-
ding to the production of ohmic source and drain contacts, is ob-
tained by annealing in an inert gas atmosphere at temperatures
ranging (depending upon the composition used) from 4S0 to 600C.
The final steps outlined in Fig. S illustrate the formation of the
Schottky barrier gate contact (usually aluminium) by evaporation.
The process steps are subject to many variations and these involve
changes both in the methods of forming the contacts and in the geo-
metry of the device. Thus, for the Schottky-barrier contact other
metals or alloys, such as Pt-Ni, have been used, while to obtain
reduced contact resistance a surface n+ layer is produced (e.g.,
by ion implantation) interposed between the GaAs layer and the me-
tal alloy contacts. Accurate registration of the gate relative to
the source and drain contacts is difficult to achieve with small
spacing using the two-mask procedure shown in Fig. S. To obtain
better control of registration, schemes such as those using self-
alignment of the gate in a gap defined by the inner edges of the
source and drain contacts have been proposed. Figure 6 shows a
photograph of a final device illustrating the geometry of the con-
tact pads chosen to optimize high-frequency performance.
3. INFRARED DETECTORS
Considerable work has been done over the past three decades
in the field of infrared sensitive, photoconductive films, espe-
cially those based upon the lead salts ~ b S PbSe and PbTe. More
recently, the need in military applications for infrared detectors
of high sensitivity and low noise to be used for surveillance and
imaging has prompted increased research on single-crystal materials,
in many cases in thin film form. The response characteristics of
such detectors are specially tailored to fit certain "spectral win-
dows," i.e., infrared wavelength ranges within which attenuation of
radiation in the atmosphere is relatively small. Two such windows
lie in the tanges 3-S ~ m and 3-14 ~ m .
Infrared detectors are chosen on the basis of their energy
gaps, EG, and reference to the semiconductor and infrared detector
literature shows that the pure compounds available do not provide
a good match to the spectral windows mentioned above. To obtain a
APPLICATIONS OF NONMETALLIC THIN FILMS
469
DRAIN CONTACT
SOURCE CONTACT
Fig. 6 Surface photograph of Schottky-barrier-gate GaAs FET
good match suitably chosen pure compounds are alloyed together to
give pseudo-binary solutions of the type In(As,Sb), (Pb,Sn)Te,
(Hg,Cd)Te, Etc. By adjusting the composition, materials with peak
response at any desired wavelength can be obtained. Thus, the
alloy PbO.81SnO.19Te provides a peak response at approximately
II ~ m (Fig. 7) while lnAsO.9SbO.1 peaks at about 4 ~ m 1m most al-
loy systems of this type the value of energy gap EG varies roughly
linearly with composition. However, the system. lnAsxSbl-
x
is a
notable exception (Fig. 8) and displays a shallow minimum in the
band gap value at about x = 40%. It is interesting to note that,
in principle, detectors matching both of the spectral windows
referred to above can be produced from compositions in this sin-
gle alloy system.
3.1 Detector Parameters
The performance of infrared detectors is characterized by a
number of important parameters (8) such as the noise equivalent
470
M.H. FRANCOMBE
100r-------------------------------,
1.5 2 5 15 20
WAVELENGTH (JLl
Fig. 7 Responsivity spectra of a PbO.81SnO.19Te diode at 77K
before and after application of a selenium antireflection coating.
power, detectivity, responsivity and speed of response. The con-
ditions of measurement for these parameters are standardized in
terms of the temperature of the radiating source, the modulating
frequency (frequency at which the radiation incident on the detec-
tor is chopped) and the amplifier bandwidth. For long IR work a
blackbody temperature of SOOoK usually is used, while to minimize
the noise variations within the frequency interval over which noise
is measured a narrow amplifier bandwidth ~ S Hz) is employed.
The noise equivalent power (NEP) may be written as follows :
NEP (SOOOK, 900 Hz, S Hz) (10)
APPLICATIONS OF NON-METALLIC THIN FILMS
Curve 658394-A
0.4
3
/
0.3
/
4
/
/
E

/300
o
K
/
5
:l.
c.
/
..c
'"
0,
t!)
0.2
/
c:
"C
/ '" c:
7
a;
'"
/
>
<Xl
'"
/
;:
.,-
,;
10
;'
0.1 -------------
15
InSb 0.2 0.4 0.6 0.8 InAs
x
Fig. 8 Variation of bandgap and equivalent cutoff wavelength
with composition in system InAsxSb
x
_
I
'
471
The quantltles in parentheses refer to the blackbody tempera-
ture, modulating frequency and amplifier bandwidth, respectively,
Sand N represent signal and noise under the conditions of measure-
ments, A is the detector area and P
D
the radiant power density rea-
ching the detector from the black body.
The reciprocal of NET is the detectivity, D, and is often
normalized to an amplifier bandwidth of I Hz and detector area of
I cm
2
This yields the parameter D*, expressed as
* D
(II)
472 M.H.FRANCOMBE
The normalization is based upon evidence that noise varies as
the square root of the amplifier bandwidth and that D varies in-
versely as the square root of the detector area. Unless otherwise
stated D* is given for a field of view of 2 ~ sr and a background
temperature of 300
o
K.
Thin film detectors will vary in structure depending upon the
proposed mode of operation, and before proceeding further with our
discussion of detector parameters, it will be useful to describe
these modes. In general, three types of structure have been consi-
dered, i.e., photoconductors, photovoltaic and MOS photodiode.
For operation in the photoconductive mode, semiconductors of high
purity possessing low majority carrier concentrations are prefer-
red ~ 1014/ cm
3
). For use in the photovoltaic or MOS photodiode
mode, material requirements are less stringent and, for example,
majority carrier densities on each side of a p-n junction should
be in the 1017-10
18
range. The MOS photodiode detector is an in-
teresting special case of the photovoltaic mode and involves the
use of an optically transparent electroded capacitor structure on
the semiconductor. In the case of n-type InSb this leads to an in-
verted p-type region adjacent to the oxide interface and hence to
an induced p-n junction. Since much of the recent and current re-
search on thin film detectors is aimed towards photovoltaic strue-
tures, we may illustrate the other important detector parameters
by reference to junction detectors.
A typical junction detector in (Pb,Sn)Te would comprise a
bulk p-type crystal with a shallow n-type region formed at the
surface by diffusion. Photons absorbed in the surface region gene-
rate carriers which diffuse towards the p-n junction. In the open
c i x c ~ i t case they produce a change in junction potential as a re-
sult of an increase in minority carrier density on both sides of
the junction. If the junction is shorted externally a current
flows tending to bring the minority carrier distribution back to
its equilibrium state. If reverse bias is applied to the junction
the photocurrent adds to the normal reverse diode current. For
high sensitivity as many as possible of the photogenerated carriers
should reach the junction (high efficiency), and the incremental
junction resistance should be maximized.
For a diode detector the current It can be expressed as the
sum of the photocurrent, I
p
' and a current, I(V), which flows in
the absence of incident radiation as a result of the applied vol-
tage V.
It Ip + I(V) (12)
APPLICATIONS OF NONMETALLIC THIN FILMS
473
in an ideal diode in which all of the current is due to injec-
tion
I (V) Is[exp(qV/kT) - 1] (13)
where Is is the diode saturation current. The photocurrent Ip may
be expressed as
(14)
where N is the number of incident photons per second and n is the
quantum efficiency (i.e., the number of light-generated carriers
crossing the junction per incident photon.
For small voltages I(V) can be assumed linear and is expressed
as (I/R)V. With the diode operated as a photovoltaic detector,
It 0, then
V R Ip (15)
where V is the open-circuit photovoltage, P
A
the power of incident
photons (PA = NE
A
). and EA is the energy per photon at wavelength,
A.
The voltage responsivity Rv.A at wavelength A can be expressed
as
Rv,A = V/P
A
= n q R/EA (16)
The noise voltage of the open circuit junction is given by the
Johnson (or thermal) noise of the incremental diode resistance.
-2
V = 4 kTR f:,.f. (17)
If the l/f noise due to surface and contact contribution, which
predominates at low frequencies, is neglected, the noise equiva-
lent power may be derived for a 1 Hz bandwidth by using Eqs. (15)
and (17). Then the detectivity for an area A can be expressed as
1/2 %
~ = n q (AR) /2 EA (kT)
In addition to detectivity and noise parameters another impor-
tant criterion in the characterization of diode performance is res-
ponse speed. This can be influenced by both the effective life-time
of photoexcited carriers and by the effective time constant of the
474 M.H. FRANCOMBE
junction. The second effect depends on the magnitude of diode re-
sistance, R and capacitance, C. In turn, the capacitance is influ-
enced by factors such as carrier concentration in the p and n re-
gions, dielectric constant E, and magnitude of reverse bias. In ge-
neral, fast response is favored by short carrier lifetime and a
low RC product. However, in adjusting parameters (such as R) for
fast response it is important to avoid sacrificing detectivity.
3.2. Preparation and Structure of Thin Film Junction Detectors
The fact that the active region used in an infrared detector
is very shallow (a few microns at most) implies that thin film
geometries and processing approaches should be ideally suited for
such devices. The increasing role of thin films in this area is
effectively illustrated in particular by recent work on
(Pb,Sn)Te(9,10) and (Hg,Cd)Te(ll) detectors. Thin films of these
compounds (usually in epitaxial form) have been produced by a va-
riety of techniques, e.g., vacuum sublimation, chemical vapor de-
position, liquid phase epitaxy, vacuum evaporation and sputtering.
To a large extent the motivation in this work has arisen from a
need for large-area detector arrays with uniform characteristics.
Much of the earlier research on (Pb,Sn)Te and (Hg,Cd)Te
detectors involved the use of bulk crystals in which junctions
were formed by diffusion. More recently, studies by Holloway and
coworkers(9) have shown that device-quality epitaxial layers of
PbTe and (Pb,Sn)Te could be grown on foreign substrates such as
BaF2, and that photovoltaic diodes could be formed in these either
by proton bombardment(9a) (to produce shallow n-type regions in
p-type films) or by the application of rectifying metal contact
layers (9b) (using for example evaporated films of Pb) to produce a
Schottky barrier. Both proton bombardment induced- and Schottky
barrier junctions in PbTe epitaxial films have yielded high peak
detectivities, with values of Dt (5.5 11m, 600, I) of about
6 x lOll cmHzlk W-
I
. However, Schottky junctions on films of the
pseudo-binary (Pb,Sn)Te compositions tend to be leaky with rather
low resistances.
A desirable aim for both the (Pb,Sn)Te and (Hg,Cd)Te film
materials is to be able to fabricate p-n junctions by processes
such as diffusion or successive epitaxial growth. Unfortunately,
epitaxial films grown on foreign substrates often contain numerous
grain boundaries or crystallographic faults, and attempts at dif-
fusion lead to the localized penetration through the entire film thick-
ness. This can be avoided by selecting a substrate such as PbTe
or CdTe for film growth which is miscible with the epitaxial compo-
sition. The resulting film then possesses essentially a bulk
APPLICATIONS OF NON-METALLIC THIN FILMS 475
monocrystalline structure. Successive deposition approaches can be
used providing that growth conditions can be established for the
controlled deposition of p and n layers. Recent work by Krikorian
et al. (lO)and by Cohen-Solal et al.(ll) indicates that this can be
achieved through growth of the layers by cathodic sputtering.
In the case of (Pb,Sn)Te the electrical type of both bulk
crystals and films may be controlled by adjusting the metal-tellu-
rium stoichiometric balance. This can be illustrated with reference
to Fig. 9, which shows a magnified region of the phase diagram in
tne vicinity of a compound Pb
O
8SnO 2Te. Compositions within the
enclosed single-phase region wh1ch are Te-rich are p-type while tho-
se which are metal-rich are n-type. It is clear from Fig. 9 that the
Liquidus
Metal
Saturated
Solidus
Excess Metal
..
Composition
Te Saturated
Solidus
Fig. 9 Magnified view of phase diagram for approximate composition
Pb
O

S
Sn
O

2
Te in vicinity of stoichiometric compound.
476 M.H. FRANCOMBE
the type and carrier concentration of a (Pb,Sn)Te crystal or film
can be adjusted by annealing in either a metal- or tellurima-rich
atmosphere at an appropriately chosen temperature. However, in the
case of thin films it is virtually impossible to control selecti-
vely the type of a surface region in this way since rapid diffu-
sion causes the whole film to convert. The preferred approach is
to grow the p and n layers successively from the vapor phase, and
the studies of Krikorian et al.(10) indicate that this should be
possible simply by adjusting the substrate temperature and depo-
sition rate. As shown in Fig. 10, which is based upon results
for epitaxial layers produced by triode sputtering from a target
of composition Pb
O
SSnO 2Te, low deposition rates and low substra-
te temperatures favor tne growth of n-type films while, high rates
350
U
2...
ct
300 E
{!
CD
1;
...
"til
.c
250
:::J
VI
200
0
Fig. 10
films.
Epitaxial
Films
Polycrystals
Polycrystals
2
Growth Rate (lJIlll hr )
3
Epitaxial-polycrystalline phase diagram for Pb.
S
Sn.
2
Te
APPLICATIONS OF NON-METALLIC THIN FILMS
and elevated temperatures lead to the formation of p-type films.
No data on p-n junctions produced by this approach are yet avai-
lable.
477
Cohen Solal et al.(II) have been successful in growing p-n
hetero- and homojunctions in the solid solution series CdxHgl_xTe
by triode sputtering in mercury vapor from alloy cathodes in a
vacuum system evacuated by a mercury diffusion pump. Due to the
high partial pressure of Hg in the system, loss of mercury from
the alloy film at elevated growth temperatures could be avoided
and epitaxial layers with low x values were successfully grown on
CdTe substrates. The deposited films were made p and n type by
impurity doping with metals such as gold which were incorporated
in the growing layers by using an auxiliary metal sputtering source.
Table 1 shows some typical results obtained for two compositions
with peak spectral response occurring at 6.5 and 11.5 vm, re-
spectively.
TABLE I
PROPERTIES OF FAR I.R SPUTTERED PHOTODIODES AT 77K
/..
Rd
11
D*
p
HzJz
-I
micron ohm percent cm. W
6.5 1,000.000 35 - 50
/"6.5
10
10
11.5 1,500.000 55 - 60
/"10.6
4
x
IO
IO
4. THIN-FILM CAPACITORS
Thin film insulator materials have found many useful applica-
tions in integrated circuits as dielectric layers in devices and
as protective or passivating coatings. One of the more obvious
advantages derives from the fact that due to their small thickness,
high values of capacitance can in principle be achieved in very
small volumes. Numerous materials, processed by several prepara-
tive techniques, have been examined and some have found application.
Those studied include SiO, SiOZ and Al
Z
0
3
formed by evaporation,
sputtering, chemical vapor deposition and glow-discharge chemical
decomposition, anodic oxides such as TaZ05 formed by anodization
of sputtered metal films, polymer layers such as parylene or poly-
tetrafluorethylene (PTFE) deposited by uv polymerization or sput-
tering, and high permittivity mixed oxide dielectrics such as
ferroelectric BaTi0
3
and Bi4Ti30lZ grown by flash- or multi-source
478 M.H. FRANCOMBE
evaporation or by sputtering.
For thin film capacitors two types of structure usually are
sought, both capable of yielding relatively high specific capaci-
tance. One of these utilizes thin amorphous films, for example of
SiO or SiO
Z
, which can be made free from cracks or pinholes, and
where a high capacitance (together with reasonably high voltage
breakdown) can be obtained by using small thickness. Capacitance
values can be further enhanced (to about 0.1 as in the case
of tantalum capacitors by using a dielectric such as TaZOS (E
T
0 =
ZI, cf. E
S
'
O
= S). This choice would be especially appropriate
az
S
where the metal, tantalum, is used for resistors in other parts
of the thin film circuit.(IZ) The .second structure utilizes poly-
crystalline films of a high-permittivity mixed oxide such as BaTi03'
Here, the high permittivity, typical of the bulk dielectric, can
only be achieved by developing a relatively well-crystallized film
structure, and this in turn requires the use of high growth tempera-
tures. This is illustrated by Fig. II which shows the variation of
dielectric constant with substrate temperature for flash-evaporated
films of BaTi0
3
at different positions on a resistively hea-
ted platinum substrate. Similar results have been obtained for sput-
tered films of SrTi0
3
. The resulting film structure contains grain
boundaries, and to avoid low-voltage breakdown the films must be
made thick (> I thus sacrificing capacitance.
t-
z
<t
f-
<J1
Z
0
U
u
!r
f-
U
w
-l
w
a
600
500
400
300
200
100
I. 6 /-L
0.4/-L
o 100 200 300 400 500 600 700
DEPOSITION TEMPERATURE (Oe)
Fig. II Dependence of dielectric permittivity on deposition tem-
perature for flash evaporated BaTi0
3
films.
APPLICATIONS OF NONMETALLIC THIN FILMS
479
Since in monolithic silicon analog integrated circuits there
is little need for high capacitance, satisfactory low-value capa-
citors can be produced from simple dielectrics such as SiO, SiO
Z
or anodic Taz05. In hybrid microelectronics, especially in those
circuits operating at lower frequency, chip capacitors are found
to be both satisfactory and economical. It appears, therefore,
that the main area for technological expansion in the thin film
capacitor field will be in high-value discrete capacitors, using
thin organic films such as parylene or PTFE, or in microwave inte-
grated circuits used in multi-module systems such as phased-array
radars where high-value bypass capacitors are needed. In the lat-
ter case capacitances of the order of 1 are desirable to-
gether with very low dispersion characteristics.
High specific capacitance values have been obtatned with a
number of ferroelectric materials in thin film form, 13) as indi-
cated in Table Z. Unfortunately, with some of these materials the
dielectric loss is high, and dispersion effects occur at higher
frequencies. Figure 1Z shows dispersion effects observed in vacuum
evaporated PbTi03 films at frequencies in the kHz range. Presuma-
bly, the large changes with frequency in capacitance and dissipa-
tion factor arise from chemical inhomogeneity, perhaps due to
chemical reduction of the dielectric film surface. Supporting evi-
dence for this model is obtained by assuming a film structure com-
prising high and low conductivity regions in series. From calcula-
tions based on the assumed values of capacitance and resistance
shown in Fig. 1Z the frequency behaviour found with the PbTi03
films can be closely duplicated.
TABLE Z
DIELECTRIC PROPERTIES OF HIGH PERMITTIVITY MIXED OXIDE CAPACITOR
FILMS PREPARED BY VACUUM DEPOSITION
Material Method
Deposition
t
CiA
D(%)
Temp.
(OC)

BaTi03 FE 550 O.Z Z.Z5 500 4
BaTi03 EB >600 0.46 1.58 8Z0 4. 1
BaTi03 RF 720 1.3 0.49 700 3.0
PbTi03 EB 700 I. 81 3.7
SrTi03 RF 500 0.Z4 0.75 ZOO 1.0
Bi4Ti3]2
RF 550 0.50 0.4 ZOO O.Z
480
M.H. FRANCOMBE
22 -<>0 EXPERIMENTAL-CAPACITANCE
--G-EXPERIMENTAL- DISSIPATION FACTOR
-CALCULATED-CAPACITANCE
F
"''''''''I()o .... 0 -DIS SI PATIO N FACT OR
18
C1
15.8nF
50
40


30
a:
0
I-
u
it
z
20 0

Q.
VI
VI
10 0
FREOUENCY (Hz)
Fig. 12 Equivalent circuit fit for polycrystalline PbTi03 films
Films of SrTi03 and Bi4Ti3012 prepared by rf sputtering are
found to have somewhat lower loss and better high frequency charac-
teristics, while still offering high specific capacitance. In the
case of the SrTi03 films pulse measurements have shown that no sig-
nificant loss occurs at frequencies up to I GHz, while for sputte-
red films of Bi4Ti3012 similar data were obtained for frequencies
as high as 4 GHz. Both materials would be excellent candidates for
bypass capacitors.
5. THIN FILM MEMORIES
Probably the best known application of thin films to memories
is that involving the use of thin magnetic permalloy (NiSIFeI9)
films. Considerable research and development effort has been 1n-
vested in this area over the past 15 years in the hope of develop-
ing fast, high-density memories based upon thin film matrices
which would out-perform the existing ferrite core memories. While
permalloy film memories have found some application in special
systems, they have not achieved the multi-megabit scale with ac-
ceptable performance justifying replacement of core memories.
This was attributable to problems of obtaining reproducibility,
uniformity and disturb insensitivity in the permalloy matrices,
and also to the fact that more attractive memory structures using
high-reliability, stable semiconductor devices and also a highly
stable magnetic alternative utilizing single-crystal rather than
APPLICATIONS OF NON-METALLIC THIN FILMS 481
polycrystalline media have recently become available. In this sec-
tion we shall discuss briefly the characteristics of such magnetic
bubble memories fabricated in single-cry tal, magnetic oxide, film
media and some more speculative memory work based upon ferroelec-
tric thin film.
5.1 Magnetic Bubble Memories
In magnetic media possessing large values of saturation magne-
tization Ms (e.g., permalloy) reduction to a thin-film geometry gi-
ves rise to a very high demagnetizing field (4rrMs) perpendicular to
the film surface. As a result, the magnetization is constrained by
shape anisotropy effects to lie in the plane of the film. However,
in media with relatively low Ms values, other competing magnetic
anisotropy terms (e.g., crystalline, growth-induced or magnetostric-
tive effects) may possess energy exceeding that of the shape aniso-
tropy, and allow the formation of a magnetic easy axis perpendicu-
lar to the film surface. This situation holds true in the case of
many optically transparent magnetic oxides such as the orthoferri-
tes (e.g., YFe03) or garnets. In the demagnetized condition a thin
crystal is divided up into strip domains (Fig. 13), in which the
magnetization directions are antiparallel and normal to the large
surface. Optical contrast between the domains is obtained by vir-
tue of the Faraday rotation effect.
Fig. 13 Strip domains (left) seen by means of Faraday rotation of
polarized light represent regions magnetized alternately inward and
outward in a 55 ~ thick platelet of Ytterbium orthoferrite. Domains
are typically 90 ~ in width. When a 50 Oe bias field is applied
perpendicular to the orthoferrite sheet (right) a cylindrical do-
main or "bubble" precisely 35 ~ in diameter is formed for each pro-
perly oriented single-walled strip domain.
482 M.H. FRANCOMBE
In practice, single-crystal magnetic films several microns in
thickness, with domain structure of the type shown in Fig. 13, may
be grown epitaxially on a supporting substrate of non-magnetic
garnet such as Gd3GaS012 using chemical vapor deposition (CVD) or
liquid phase epitaxy growth techniques. The garnet composi-
tion must be optimized to obtain a good lattice and thermal expan-
sion match with the substrate. In the case of the LPE approach,
which appears to be the best growth technique evolved to date,
films of excellent quality with uniform thick:ress are currently
achieved by spinning the substrate about an axis normal to its
plane while suspending it in the molten garnet solution. A typical
garnet composition grown by this method, possessing excellent mag-
netic characteritics for high-speed memory application, is
Yl.03Gdl.29YbO.6SAlO.7Fe4.3012
If a bias field HA is applied normal to the plane of the mag-
netic oxide film, those strip domains whose magnetization is op-
posed to the field shrink in size and eventually change shape to
a cylindrical geometry of the type shown in Fig. 13 (right).
These cylindrical domains are called bubble domains, and their size
and stability are dependent upon relative contributions respective-
ly from the applied field, the domain wall and magnetostatic ener-
gies. The theory of the domain structure has been developed by
Thiele, (14) while the general technology of magnetic bubble media
and their applications to memory and logic devices have been exten-
sively explor.ed by Bobeck (15) and other worker"s.
HBIAS
+


Fig. 14 Geometry of a cylindrical domain in a uniaxial platelet.
The domain wall width is assumed much less than 2r.
APPLICATIONS OF NON-METALLIC THIN FILMS 483
The theory of Thiele leads to the following conclusions : (I)
The diameter of the smallest stable domain realizable is approxi-
mately 4t. The term t is a material length defined as 0w/4TIMs.
(2) The media thickness h which yields the smallest bubble diameter
is TIt. (3) A bias field HA = (O.3)4TIM
s
is needed to support the mi-
nimum diameter bubble domain. Figure 14 shows the geometry for a
cylindrical bubble domain in a film of thickness h. In Fig. 15 the
dependence of domain diameter on thickness is illustrated. The
upper curve defines the bubble collapse. Between the curves stable
bubbles with diameters inversely proportional to the applied field
can exist.
The utility of magnetic bubbles for memory devices arises from
the fact that they can be moved in directions parallel to the film
surface by the application of a field gradient. Several methods
of applying gradients have been proposed. In one, forming the basis
of a shift register, an array of photolithographically printed con-
ductors is used to step bubbles at a high data rate by sequentially
300
100
d ~ 30
10
Radially
Unstable
Region
3
0.3 1
Elliptically
Unstable Region
3 10
h/t
30
Fig. 15 Dependence of the normalized diameter dlh on the normalized
thickness hit. The upper curve defines the bubble to strip instabi-
lity and the lower curve bubble collapse.
484
IN PLANE
FIELD
M.H. FRANCOM BE
tAAA
Fig. 16 Step by step bubble propagation in chevron circuit caused
by clockwise rotation of in-plane field.
applied current pulses. The bubbles can also be moved using a super-
imposed array of magnetic permalloy elements. When a field rotates
in the plane of these elements, travening positive and negative
magnetic poles are generated at their extremities, as shown for
example in Fig. 16. These poles selectively attract and repel the
bubble so as to control its motion. Several element geometries
have been studied for such permalloy circuits, but the chevron de-
sign shown in Fig. 16 is the most widely used.
The generation and detection of bubbles requires specially
designed circuit elements. One method of generation is to provide
for expansion of a bubble and its separation into subsidiary domains.
Several methods of detection have been proposed including Hall effect
detectors, magneto-optic detectors employing the Faraday rotation
effect and magnetoresistive detection using thin permalloy elements.
The magnetoresistive detector seems to be the most generally used,
and its sensitivity may be optimized by using a special expander
circuit to increase the size of the bubble just prior to detection.
APPLICATIONS OF NONMETALLIC THIN FILMS 485
Much research is still in progress with the aim of achiEving
higher densities and data rates with magnetic bubbles. Several
properties must be optimized for this purpose. Among the key para-
meters are coercivity H
c
, which should be lower than I Oe, bubble
mobility, which should be in excess of 100 cm/sec Oe, and bubble
diameter, which should be small, consistent with the highest resolu-
tion drive-circuit elements obtainable by lithographic procedures.
Since electron beam lithography is capable of submicron resolution,
this implies the ability to handle small bubbles in close spaced
geometries which should lead to the evolution of very high density
data stores. Amorphous alloy media of the Gd-Co and Gd-Fe type
have recently been discovered which display bubble domain behavior,
with bubbles in the micron size range. These should provide good
candidates for high density information processing.
5.2 Ferroelectric Memories
The charge storage properties of ferroelectrics which derive
from the P-E hysteresis loop characteristic, offer memory storage
potentialities analogous to those available, and fully exploited,
with ferrimagnetic metrials. Unfortunately, several practical pro-
blems have been encountered over the years which have prevented the
realization of those potentialities. In fact, these arise from the
intrinsic tendency for ferroelectrics not to possess a well-defined
threshold field for switching, a factor resulting in considerable
leakage of stored charge under the influence of adjacent disturb
fields. In addition, serious difficulties were encountered with
control of material parameters and in devising access techniques
which could place ferroelectric memories on an economic footing com-
petitive with that for magnetic core systems.
Given the situation that cheap, low-coercivity, square-loop
ferroelectrics are available with reproducible'properties, the cor-
responding availability of the same materials in thin film forms
inunediately offers considerable further advantages. Reduction if!
material thickness would enable switching voltages to be reduced,
while large-area thin-film ferroelectrics would permit the use of
high-resolution photolithographic methods in making thin film ad-
dress conductors and electrodes. A recent evaluation by Chapman(16)
and further results published by Mehta(17) indicate that develop-
ments with ferroelectric films (specifically with the PZBFN-65
ceramics*) and in ,the areas of CRT, phosphor, photoconductor and
*
These are slip-cast foils of nominal
325
Zr
O.27)03
were also prepared oy rf
of the same
486 M.H. FRANCOMBE
high-resolution optics technology have now reached a level where a
new type of fast-access large-capacity ferroelectric memory becomes
feasible.
A large storage capacity memory with ultimate
excess of 10
9
bits can in principle be built from an assembly of
photoconductor-ferroelectric chips. Each chip in the array is ad-
dressed by a flying light spot about 20 in diameter, produced
by focusing the 10 mil spots on a CRT phosphor screen. The photo-
conductor provides a light-controlled division of the address vol-
tage applied to the PC/FE sandwich, and essentially supplies an
artifical 'threshold' condition for switching. The chip comprises
the layered structure--transparent electrode/photoconductor/ferro-
electric/electroded substrate--and might be addressed as shown
schematically in Fig. 17. In his assessment of feasibility, Chap-
man assumed a switching voltage for a PZBFN-65 ferroelctric layer
4 thick of 20 volts, giving a switching time of 2 Allow-
ing for a complete cycle time of 2.5 and a total applied vol-
tage pulse of 30 volts, 2/3 of this applied pulse would have to
appear across the ferroelectric in 0.5 Given the specific
capacitance for the ferroelectric, time constant and voltage divi-
sion considerations now lead to values for the light and dark
resistivities of the photoconductor. It appears that these photo-
conductor properties, including the desired rate of photocurrent
decay, estimated from the mobility, can be supplied by several
available materials including CdS, CdSe and ZnSe. The general con-
clusion of this study, which is too detailed for reproduction here,
TRANSPARENT
ELECTRODES
I MM DIAMETER
(= 1000 MICRONS)
FE/PC
CONDUCTIVE
SUBSTRATE
LIGHT BEAM
,'---DIAMETER = 25 TO 75 MICRONS
Fig. 17 Test circuit for evaluating storage characteristics of
photoconductor-ferroelectric memory chip.
APPLICATIONS OF NONMETALLIC THIN FILMS
was that all aspects of the proposed memory system except the
storage medium could be engineered with existing technology.
487
In a subsequent treatment Mehta(17) carried out experimental
tests on a storage medium comprising the PZBFN-65 ferroelectric
ceramic at a thickness of 2-2.5 in coniuntion with a CdSe pho-
toconductor layer 0.5 Light address was accomplished
using illumination form a He-Ne laser (6328 A) covering a spot size
of about 25 to 75 (Fig. 17). To separate the small signal gene-
rated in switching the illuminated spot, it was necessary to com-
pensate for the output produced by partial switching of the non-
illuminated area. This was achieved in a test situation by using
equivalent transparent electrode areas of 1 mm diameter for the
dark and illuminated areas and employing the differential sensing
scheme indicated in Fig. 17. The feasibility of addressing the
medium using a 40 diameter spot focused from a 400 spot on a
CRT was demonstrated successfully. The ability of the ferroelectric
Q
Charge
Switched
"One"

T,seC -+
"Zero"
Q

"One"
"Zero"
Fig. 18 The ability to store 'one' and 'zero' bits was determined
with the indicated voltage pulses. Spontaneous depoling causes the
'one' to decrease and the 'zero' to increase as the time interval
between pulses increases.
488 M.H. FRANCOMBE
films to retain their stored charge was also tested under various
experimental conditions,(16) and was found to be satisfactory.
Typical data for stored 'one' and 'zero' states as a function
of the time interval between address pulses are shown in Fig. 18.
These tests revealed certain shortcomings in the CdSe photo-
conductor and in the CRT mode of address. Low dark resistivity of
the CdSe layer caused a considerable amount of charge to be swit-
ched in unaddressed spots and made the device unusually disturb
sensitive. Also, the low sensitivity of the photoconductor and its
long decay time limited cycle times to about 2 S ~ sec. Variations
occurred between the 'one' and 'zero' signal levels, which appeared
attributable to non-uniformities in the device or in the CRT light
inten.sity.
6. MAGNETO-OPTIC AND ELECTRO-OPTIC DEVICES
The interaction of light with thin films forms the basis of
many useful and commercially profitable communication and display
devices. In the previous section we considered examples of such
interactions in the field of high-density memories. There are in
fact numerous other examples in which the modulation of light by
transparent thin film media can be used in integrated optic cir7
cuits, as well as magneto, electro- and acousto-optic displays. In
some such displays intensity modulation is achieved by variable
scattering in a thin liquid-crystal layer, or light is generated
at varying levels of intensity by an electroluminescent thin film.
Space does not permit us to describe more than two types of devices
in this wide class.
Integrated optic circuits offer exciting possibilities in the
communication field because of the high speed and density at which
information can be transmitted. The four basic elements of such
circuits require the means for (a) light generation, (b) light pro-
pagation and deflection in thin-film wave guides, (c) light modula-
tion via e.g. magneto- or electro-optic switching effects, and
(d) light detection. Research is still progressing with elements
for all of these purposes. One obvious candidate for the modulation
of light being propagated in a thin film wave guide is the magneto-
optic Faraday rotation phenomenon in transparent magnetic films.
High-quality garnet films have already been found suitable for use
as single-crystal wave guides. It has been shown recently by Tien
and his coworkers (I8) that by inducing a magnetic anisotropy axis
to lie parallel with a garnet film surface a superimposed rf signal
can be used to modulate the intensity of light traversing the film.
A view of the experimental arrangement is shown in Fig. 19. The gar-
net film comprises a thin circular disc of composition Y3Ga i I Sc 4
Fe
3
.
S
0
12
grown epitaxially on a Gd
3
Ga
S
O
l2
substrate. Ligfit lS c o ~ ~
APPLICATIONS OF NONMETALLIC THIN FILMS 489
.y---.a_.:......
y)
'.'
Fig. 19 EXperimental arrangement of a magneto-optic switch or
modulator .
pled into and out of the film by means of titanium oxide prisms,
and the modulation signal is applied with a superimposed serpen-
tine shape thin film coil which is fabricated by standard photo-
lithographic techniques. Typically, the magnetization is constrai-
ned by a dc field to lie at 45 to the light propagation direction.
The TE (transverse electric) and TM (transverse magnetic) waveguide
modes propagating through the film will be angularly split on arri-
ving at the second birefringent prism. The component of magnetiza-
tion along the propagation direction causes conversion of the TM
into the TE mode and thus influences the relative intensities of
the split beam components . Superposition of the rf field causes high
frequency modulation of this magnetization component, and conse-
quently modulates the intensities of the TM and TE beams. Modulation
at frequencies higher than 80 MHz has been obtained with applied
fields as small as 0.2 Oe. The frequency p p e r ~ presently to be
limited by the response characteristics of available detectors. This
approach offers a novel means of transmitting electrical signals via
light beams in thin films.
Modulation of light intensity can also be produced by any of
the well-known electro-optic effects, but until recently the only
film media available for this purpose were liquid crystals. The
phenomenon chosen for intensity modulation will depend upon whether
the light is being propagated in a waveguide film structure parallel
to the surface as in integrated optics, or through the thickness of
the film as in a large-area optical display or memory. In the first
case, the path length over which the intensity is modulated can be
chosen arbitrarily, depending upon the magnitude of the electro-
490
M.H. FRANCOMBE
optic effect used. In the second, the path length is of necessity
limited to a few microns by practical considerations such as film
growth rate, transparency and adhesion to the substrate. The elec-
tro-optic effect should therefore be strong and preferably indepen-
dent of optical path length. One obvious effect would be based
on extinction of polarized light due to field-induced rotation of
the optic axis of a single-crystal film. Among a small class of
ferroelectric structures currently under investigation, bismuth ti-
tanate Bi4Ti3012 offers an unusual extinction mode of contrast
which fulfills this need. The polarization axis in Bi
4
Ti
J
O
l2
lies
in the a-c plane of the monoclinic (pseudo-orthorhomb1c) structure
at about 4
0
to the a-axis, giving rise to a small polarization
component (4 along the c-axis and a large component
(50 uC/cm) along the a-axis. Reversal of the c component of P
leads to a rotation of the optical indicatrix major axis (in a
a-c plane) from +25
0
to -25
0
relative to the c-axis.(19). With
the placed between crossed polarizers; near-optimum con-
trast can be obtained. Unfortunately, the natural growth habit
of Bi4Ti3012 (platelets with the c-axis normal to the large face)
makes it difficult to produce large a-c face areas suitable for
the display by mechanical sectioning. A solution to the problem
has been found in the author's laboratory, in which large (ap-
proximately 1 cm
2
) a-c crystal faces have been grown by epitaxial
deposition of the compound onto single crystal of MgO and MgA1204'
(13,20) ,
The as-grown epitaxial bismuth titanate films deposited onto
(110) MgAl
2
0
4
initially multi-domain, with antiparallelism of
the a and c potarization components. To pole and subsequently
switch a large-area film in situ on the substrate for display pur-
poses, a test structure consisting of an interdigitated (IDT) elec-
trode array incorporated with a 45
0
electrode scheme was employed.
This structure comprises a set of electrode fingers at the substra-
te-film interface, followed by a titanate laver 10 thick and fi-
nally an upper set of electrode fingers connected (in common with
the underlying electrodes) to peripheral contact pads. The prin-
ciple of operation of the test structure can be explained from the
unit cell enclosed in the dotted lines, as illustrated in Fig. 20.
Each unit cell consists of two outer electrodes with one center
electrode which divides the active area into two regions. To ope-
rate the structure, the outer electrodes are grounded. A field
high enough to switch both a and c polarization components is then
applied to all the center electrodes, resulting in a poled struc-
ture in the two active regions of each cell. The resulting pola-
rization directions are electrically opposed but the antiparallel
orientations of the indicatrix are optically equivalent (see primed
letters). Switching of a particular cell is effected by addressing
its center electrode with an opposite lower field sufficient to
reverse only the low coercive force c-axis polarization compo-
APPLICATIONS OF NON-METALLIC THIN FILMS
c a
MgAI 204
Substrate
491
Fig. 20 Schematic diagram of a unit cell of the IDT structure used
with a bismuth titanate epitaxial film, showing the tilt of the opti-
cal indicatrix caused by switching the c component of P
s
'
nents* in each region. This reversal produces the same tilt in both
of the optical indicatrices (as shown dashed in Fig. 20) and thus
a cooperative change in optical transmission.
To evaluate the contrast obtainable in lO']1m layers, prelimi-
nary electro-optic switching experiments were carried out using
this test structure, with all the center electrodes connected toge-
ther. Results of measurements with a photodetector of the contrast
ratio, based on the ratios of the light intensity transmitted at
different stages of c-axis polarization reversal to the residual
intensity at extinction, are shown in Fig. 21. The contrast ratio
achieves a maximum of 7/1 at an effective field of about 60 kV/cm,
and then gradually decreases with further increase in field. This
decrease is due to the progressive reversal of the a-axis compo-
*
The coercive fields required to switch the c and a components of
polarization in these films separately are respectively about 5 and
80 kV/cm .
492
10
8
o

0:::
6
....

(,.,
4
2
\
\
(a)
..... x
--- p
" s
\
\
yl
a
XI
C /y
......
I
......
I
.....
....
a
I
....
P
I
........
S
I
.....
I
X
I
(b) yl
60 a>
Switching Field, kv/cm
M.H. FRANCOMBE
/I
E (reversing
field,
y
C
\
\
XI
\
--- \
....
.... "
\
a
\
\
(c)
\ yl
100 120 141
Fig. 21 Contrast ratio vs. switching field plot for 45 address of
bismuth titanate films. Inserted also are sketches showing the P
s
and optical indicatrix settings, (a) at low reversing field, (b) at
the field for maximum contrast, and (c) at high reversing field.
nent of polarization , which tilts the optical indicatrix back into
antiparallelism with the original extinction setting. The gradual
variation of the contrast ratio with switching field indicates that
the epitaxial bismuth titanate film also has a grey scale capabi-
lity. It can be used for optical display and memory application in
optical, electron-beam and matrix addressing modes by special de-
signs of the electrode cell structure incorporated in the film.
APPLICATIONS OF NONMETALLIC THIN FILMS
REFERENCES
I. "Physics of Thin Films" (eds. G. Hass, M. H. Francombe and
R. W. Hoffman), Vol. 1-7, Academic Press, New York.
493
2. "Application of Thin Proceedings of International Conf.
on Thin Films, Venice, May 15-18, 1972 ; Thin Solid Films
1-443 (1972)
3. "Handbook of Thin Film Technology," (eds. L. 1. Maissel and
R. GIang), McGraw-Hill Book Co., New York (1970),
4. P. K. Weimer, "Physics of Thin Films," (eds. G. Hass and
R. E. Thun), Vol. 2, p. 148, Academic Press, New York (1964).
5. "Gall ium Arsenide and Related Compounds," Proceeding of the
Third International Aachen, Germany, October, 1970;
The Institute of Physics, London (1971)'
6. P. L. Hower, W. W. Cooper, B. R. Cairns, R. D. Fairman and
D. A. Tremere, in "Semiconductors and Semimetals" (eds.
R.K. Willardson and A. C. Beer), Vol. 7, Applications and
Devices, Part B, p. 147, Academic Press, New York (1971)'
7. J. R. Knight, P. Effer and P. R. Evans, Electron.
178 (1965)'
8. "Semiconductors and Semimetals," (eds. R. K. Willardson and
A.C. Beer), Vol. 5, "Infrared Detectors," Academic Press,
New York (1970)'
9. (a) E. M. Logothetis, H. Holloway, A.J. Varga and W.J. Johnson,
Appl. Phys. 411 (1972)'
(b) E. M. Logothetis, H. Holloway, A. J. Varga and E. Wilkes,
Appl. Phys. Letters 318 (1971)'
10. E. Krikorian, R. T"ongo and M. Crisp, "Pbl-xSnxTe Variable Band
Gap Alloy System Study," First Interim Report. Contract No.
F33615-72-C-1042, May 1972'
II. G. Cohen-Solal, C. Sella, D. Imhoff and A. Zozime, Sixth Inter-
national Vacuum Congress - Kyoto, March 25-29, 1974. To be
published in Japan J. Appl. Phys.
12. "Thin Film Technology," Robert W. Perry, Peter M. Hull and
Murray T. Harris, D. Van Nostrand Co., Inc., Princeton, New
Jersey (1968).
494 M.H. FRANCOMBE
13. M. H. Francombe, Thin Solid Films
12.,
413 (I 972) .
14. A. A. Thiele, Bell System Tech. J. 48, 3287 (1969) .
15. A. H. Bobeck, J. Vac. Sci. Technol.
~
] 145 (1972) .
16. D. W. Chapman, J. Vac. Sci. Technol.
~
425 Ci 972) .
17. R. R. Mehta, J. Appl. Phys.
~
1842 (1971) .
18. P. K. Tien, R. J. Martin, R. Wolfe, R. C. LeCraw and S. L.
Blank, Appl. Phys. Letters ~ 394 (1972).
]9. S. E. Cummins and L. E. Cross, J. Appl. Phys. ~ 2268 (1968).
20. M. H. Francombe, Ferroelectrics l, 199 (1972).
PARTICIPANTS
DIRECTOR
C.H.S. DUPUY
Departement de Physique des
Materiaux
Universite Claude Bernard -
Lyon I
43 Bd du 11 novembre 1918
69621 Villeurbanne FRANCE
COMMITTEE:
F. ABELES
Laboratoire d'Optique des
Solides
Universite Paris VI
Tour 13 - 12
4 Place Jussieu
75230 Paris Cedex 05 FRANCE
P. CAMAGNI
Centre Commun des Recherches
Euratom
Division de Physique
EURATOM ISPRA (Varese) ITALIE
H. CURIEN
Delegue General a la Recherche
Scientifique et Technique
Ministere de l'Industrie et
de la Recherche
35 Rue St Dominique
75007 Paris FRANCE
495
G. DEARNALEY
Theoretical Physics Division
UKAEA Research Group
AERE Harwell
Berkshire ENGLAND
M. H. FRANCOMBE
Westinghouse Research Laboratories
Research and Development Center
Beulah Road
Pittsburg, PA 15"235 USA
A.K. JONSCHER
Department of Physics
Chelsea College
University of London
Pulton Place
London SW 6 5PR ENGLAND
R. NIEDERMAYER
Ruhr UniverSitat
Institut fur Experimentalphysik
Arbeitsgruppe 4
463 Bochum GERMANY
LECTURERS:
Prof. D.S. CAMPBELL
Dept. of Electronic and Electri-
cal Engineering
University of Technology
Loughborough
Leicestershire LE 11 ENGLAND
496
Dr. D. GREENWOOD
H.H. Wills Physics Laboratory
University of Bristol
Tyndall Avenue
Bristol BS 8 ENGLAND
Dr. R.M. HILL
Dept. of Physics
Chelsea College
University of London
Pulton Place
London SW 6 5PR ENGLAND
Dr. A.G. HOLMES-SIEDLE
J.J. Thomson Physical Laboratory
Whiteknights .
Reading RG6 2AF ENGLAND
Prof. R. NIEDERMAYER
(cf. Committee)
Prof. H. RAETHER
Universit!t Hamburg
Institut fur Angewandte Physik
2 Hamburg 35
Jungiusstrasse 11 GERMANY
Porf. G. BALDINI
PARTICIPANTS
Prof. S.J. FONASH
Engineering Science Major
The Pennsylvania State University
231 A Sackett Building
University Park, PA 16802 USA
Dr. A. CACHARD
Departement de Physique des
Materiaux
Universite Claude Bernard Lyon I
43 Bd du 11 novembre 1918
69621 Villeurbanne FRANCE
Dr. E. PELLETIER
Universite de Provence
Laboratoire d'Optique
13, rue Henri Poincare
13391, Marseille Cedex 4
Dr. V. LE GOASCOZ
L.E.T.r.
FRANCE
Centre d'Etudes Nucleaires de
Grenoble
B.P. 85 Centre de Tri
38041 Grenoble FRANCE
PARTICIPANTS:
Universita degli Studi di Milano M. ADAMOV
"Instituto di Scienze fisiche Boris Kidric Institute
Aldo Pontremoli" Belgrade P.O. Box 522 YOUGOSLAVIE
20133 Milano
Via Celoria 16 ITALY
Prof. H.K. HENISCH
Materials Research Laboratory
Pennsylvania State University
University Park, PA 16802 USA
Prof. R.W. HOFFMAN
Dept. of Physics
Case Western Reserve University
Cleveland, Ohio 44106 USA
Dr. M.H. FRANCOMBE
(cf. Committee)
ARCONADA
Dept. Heliphysique
Universite de Provence -
Centre St Jerome
13 Rue H. Poincare
13013 Marseille FRANCE
J. AVARITSIOTIS
Loughborough University of
Technology
Physics Dept.
Leicestershire ENGLAND
PARTICIPANTS 497
BANGERT M. CELASCO
Institut fur Physik der Technische I.E.N.G.F.
Hochschule Wien Corso d'Azeguo 42
1040 Karlsplatz 13 Wien AUSTRIA 10125 Torino ITALIA
B. BANYAI
Institute of Physical Chemistry
Str. Galati 31 Sec. 2
Bucharest ROUMANIE
M. BEN MALEK
Lab. Chimie Nucleaire
Institut de Physique Nucleaire
Universite Claude Bernard Lyon I
69621 Villeurbanne FRANCE
T. BOTILA
Institute of Physics
Bodul Pacii
Bucharest ROUMANIE
P. BRICK
Chelsea College
Physics Dept. Lab.
Pulton Place
London SW 6 ENGLAND
J. CARDOSO
Institut fur Angewandte Physik
2 Hamburg
36 Jungiusstrasse 11 GERMANY
D. CARLES
Faculte des Sciences de Rouen
76130 Mont St Aignan FRANCE
A.N. CASPERD
Newcastle upon Tyne Polytechnic
Dept. Physics Research
Elison Building - Elison Place
Newcastle upon Tyne NEI 855
ENGLAND
G. CEASAR
Dept. of Chemistry
University of Rochester
River Station
Rochester, N.Y. USA
T. CHAKUPURAKAL
Celestijnenlaan 200 D 3030
Heverlee
Dept. of Physics
University of Leuven BELGIQUE
A. De CHATEAU THIERRY
LN.S.T.N.
B.P. 6
91 Gif sur Yvette FRANCE
P. CHEYSSAC
Lab. Electro Optique
Faculte des Sciences
Pare Valrose
06 Nice FRANCE
A. CHOUDRY
Physics Dept.
University of Rhode Island
Kingson, R.I. 02881 USA
T. CLAESON
Dept. of Physics
Chalmers University of Technology
Fack S. 40220 Gothenburg 5
SUEDE
G. COHEN SOLAL
Lab. Physique du Solide
1 Place Briand
92190 Bellevue FRANCE
Le CONTELLEC
C.N.E.T.
Route de Tregastel
22301 Lannion FRANCE
H. COSTEANU
Institute of Physical Chemistry
Str. Galati 31 Sec. 2
Bucharest ROUMANIE
498
M. CROSET
Lab. Central de Recherches
Thomson C S F
Corbevi11e 91 Orsay FRANCE
B. DELAUNAY
D P H NBC C E N Sac1ay B.P.2
Gif sur Yvette FRANCE
S. FEIERABEND
Ruhr Universitat Bochum Inst.
fur Experimental Physik
Arbeitsgruppe 4
463 Bochum Post Fach 2148 GERMANY
J P. FILLARD
Laboratoire de Physique des
Solides II U.S.T.L.
Place Batai110n
34000 Montpe11ier FRAl{CE
H. FRELLER
SIEMENS AG.Z F AFT E 2 F T 4
85 Nurnberg Katzwangerstrasse
GERMANY
C. FURTADO
.Lab. de Physica da Universidade
COIMBRA PORTUGAL
M. GED
C.N.E.T.
196 Rue de Paris
92220 Bagneux FRANCE
A. GLADIEUX
INS T N B.P. 6
Gif sur Yvette FRANCE
A. GUSTINETTI
Institute of Physics
University of Pavie
Pavie ITALY
K. HARTIG
Ruhr Universitat Bochum Inst.
fur Experimental Physik
Arbeitsgruppe 4
463 Bochum Post Fach 2148
GERMANY
PARTICIPANTS
K. HECQ
Universit de Mons
23 Av. Maistrian
7000 Mons BELGIQUE
C. HOLM
Norvegian Defense Research
Establishment P.O. Box 25
2007 Kje11er NORVEGE
K. LYNN
University of Utah
Salt Lake City 2006 MEB
Materials Science and Engineering
Utah 84112 USA
U. KICHKO
Ruhr Universitat
Bochum Institut fUr Experimental
Physik
Arbeitsgruppe 4
463 Bochum Post Fach 2148
GERMANY
M. KLEITZ
Ecole E1ectrochimie - Laboratoire
Cintique E1ectrochimie
Minra1e
E.N.S.E.E.G. B.P. 44
38401 St Martin d H ~ r e s FRANCE
R. KOWALCZYK
03 839 Warszawa UI. Grochowska
320
Lab. of Physik Polish Optical
Work POLOGNE
C. KURL (Mrs.)
SIEMENS Zentra1e Forschung und
Entwick1ung FL Ha1b1eiter-
systeme
D 8000 MUnchen Post Fack 801709
RFA
B. SAINT MARTIN LAVILLE
Lab. Physique App1ique -
Couches Minces
61 Rue A. Camus
68 Mu1house
FRANCE
PARTICIPANTS
1. 1. MAC WALTER
Imperial College of London
Dept. of Electrical Engineering
London SW 7 ENGLAND
MAZETTI
lEN G F
Croso d'Azeguo 42
10125 Torino ITALIE
V. NAROO (Mrs.)
Imperial College of London
Dept. of Electrical Engineering
London SW 7 ENGLAND
T. NENADOVIC
Boris Kidric Institute Belgrade
P.O. Box 522
Belgrade YOUGOSLAVIE
R. NICOLAIDES
Pica tinny Arsenal Building 350
Dover, N.J. 07801 USA
G. OLIVE
B.C. Research 3650 Wesbrook CRES
Vancouver BC VGT lW5 CANADA
A.E. OWEN
University of Edinburg
School of Science and Engineering
King's Building, Mayfield Rd.
Edinburgh EH9 3JL ENGLAND
CONTRERAS J. PEDRAZA
Universite de Provence
Laboratoire d'Optique
13 Rue H. Poincare
13397 Marseille Cedex 4 FRANCE
PELOZZI
Laboratory of Special Materials
University of Parme
Parme ITALIE
A. DE POLIGNAC (Mrs.)
Lab. Electronique
Faculte des Sciences de Reims
Moulin de la Housse
51100 Reims FRANCE
E. PRAVECZKI
University of Bristol
499
H.H. Wills Physics Laboratory
Royal Fort Tyndall Avenue
Bristol BS 8 ITL ENGLAND
PUYCHEVRIER (Mrs.)
C.N.E.T.
196 Rue de Paris
92220 Bagneux FRANCE
H.F. RAGAIE
LETI CEN Grenoble
B.P. 85 Centre de Tri
38041 Grenoble Cedex 4 FRANCE
H.S. REEHAL
University of Bradford
Physics Dept.
Bradford 7 Yorkshire ENGLAND
1. RIGALDI
Institute di Fisicia -
Gruppo Solidi
Via Celoria 16
Milano ITALIE
A. ROGER
Dept. Physique des Materiaux
Universite Claude Bernard LYON I
43 Bd du 11 Novembre 1918
69621 Villeurbanne FRANCE
N. ROMEO
Institute of Physics
University of Parme
Parme ITALIE
H.W. RUDOLF
Universitat Munchen Lehrstuhl
Rollwagen 8
Schelling Strasse 4 Munchen
GERMANY
SCHATT SCHNEIDER
Institut fur Angewandte Physik
der Technischen
Hochschule Wien 1040 Karlplatz
13 Wien AUTRICHE
500 PARTICIPANTS
P.E. SCHMIDT D.T. VIGREN
I.V.I.C. Freie Universitat Berlin 33
Apartado 1827 Caracas VENEZUELA Institute fur Theorestishe Physik
L. SELLA
Lab. Micro Electronica
OLIVETTI
10015 Ivrea TO ITALIE
L. DE SRAZER
Wes Arnin Allee 3 GERMANY
VUILLERMOZ
C.N.E.T.
Route de Tregastel
22301 Lannion FRANCE
Solid State Physics WAGENDRISTEL
Centre for Laser Studies Institut fur Angewandte Physik
University of Southern California der Technischen Hoschschule
Los Angeles, CA 90007 USA Wien 1040 Karlplatz
13 Wien AUTRICHE
J. SIEJKA
r o u p ~ de Physique des Solides
E.N.S. Tour 23
2 Place Jussieu
75221 Paris Cedex 05 FRANCE
W.R. SMITH
Pennsylvania University
University Park, PA 16802 USA
A. STEPANESCU (Mrs.)
LE.N.G.F.
Corso d'Azeguo 42
10125 Torino ITALIE
A.C. TSUI
Physics Dept.
University of Bradford
Bradford 7 Yorkshire ENGLAND
P.C. THACKRAY
348 Benedum Hall
University of Pittsburgh
Pittsburg, PA 15621 USA
B. VIDAL
Lab. d'Optique
Faculte des Sciences de
St Jerome
13013 Marseille FRANCE
F. WEHKING
Ruhr Universitat Bochum
Postfach 2148
Bochum GERMANY
J. WEINER
Imperial College of London
Dept. of Electrical Engineering
London SW 7 ENGLAND
R. WENZ
Case Western Reserve University
Dept. of Physics
Cleveland, Ohio 44106 USA
Y. YODOGAWA
Dept. of Electrical Engineering
Kyoto Technical University
Matsugasaki Sakyo Kyoto
Japan 606 JAPAN
INDEX
Absorption coefficients, 190
Activation energy, 199, 221
Adhesion of films
failure of, 342
preventing film failure,
340
sputtering and, 343
Alkali halide films, stress
in, 331
Alkali halide foils, voids
in, 410
Amorphous films
dielectric polarization,
225
electronic conduction, 248
for capacitors, 478
Poole-Frenkel effect in, 221
radiation effects, 392
reflectivity of, 190
Anistropic materials, optical
properties, 371
Anodisation, 31
Anticlastic bending, 290
Anti-reflection coatings, 4,
356, 447
Atomic diffusion on
surfaces, 51
Band pass filters, 447
Beam splitters, polarizing,
448
Behaviour of thin films, 2
Bending,
anticlastic and synclastic,
290, 291
Bending (cont'd):
linear theory, 291
stresses involved, 280, 318
origin of, 334
Bipolar transistors, radiation
effects in, 405, 407, 408
Bismuth titanate films, optical
devices from, 490
Boundary conditions
optical properties and, 373
reflection and, 364
Boundary monomer concentration,
56, 67, 69
evaluation of, 63
CMOS inverters, 404
Cadmium films in memory devices,
486
Cantilever plate method of
measuring stress, 289
experimental results, 295
Capacitance methods of thickness
measurement, 174, 180
Capacitors, thin film, 477
Capture rate, 51, 65, 73
Chemical properties of
dielectrics, 425
Chemical reduction plating, 40
Clusters
501
coalescence, 70, 76
decay of,
rate of, 56
relation to substrate and
film, 50
direct impingement on, 70, 78
502
Clusters (cont'd):
growth rate, 72
model, 59
shape of, 62
size of, 66, 85
Cluster scattering, 99, 102,
104
Composition of thin films,
analysis by nuclear
reactions, 148
analysis of thicker films,
143
ion beam analysis, 142
backscattering, 142,
147, 155
in-depth, 146
quantitative, 145, 151
X-ray emission, 154
MeV energy range ion beam
analysis, 142
photoelectron spectroscopy,
158
secondary ion mass
spectrometry, 155
X-ray spectroscopy, 157
Concentration profile
determination, 146,
152, 153
Conductivity, 189, 203
injection limited, 211
space charge limited, 212
Crazing in films, 333
Crucibles for evaporating
sources, 16
Crystalline films
activation energy, 199
characteristics, 203
defect density, 221
dielectric polarization, 225
electronic transport
properties
doping, 203
injection limited
conduction, 211
physics of, 215
reflectivity of, 190
stress in, 316
control of, 338
origins of, 318
Crystalline films (cont'd).
thermal expansion of, 308
thermal switching in, 266
transport processes
INDEX
Kubo-Greenwood formula, 196
mobility, 201
Decay rates of clusters, 56, 63
Definitions of thick and thin
films, 1, 6
Deposition of thin films, 3
chemical methods, 30
anodisation, 31
decomposition, 36
nitriding, 36
oxidation, 35
polymerisation, 35
reduction, 36, 40
thermal growth, 30
vapour phase growth, 34
control of thickness, 11
electroplating, 37
evaporation methods, 10, 178
electron-beam heating, 17 ,
reactive, 18
sources, 14
growth rate, 11
ion beam, 26, 27
stress and, 323
ion plating, 29
methods, 9
summary of, 44
physical methods, 10
plasma reactions, 28
polyDi:ers, 43
rate of
measurement, 178
solution deposition, 42
sputtering, 19
adhesion and, 343
Getter and bias methods, 24
glow discharge, 22
principles, 19
reactive, 23
R. F., 26, 27
stress and, 323
stress control and, 336
triode method, 25
INDEX
Deposition of thin films
(cont'd) :
stress and, 323, 327, 328,
330,332
control of, 336
distribution of, 280
temperature stabilization
of substrate, 293,
299
thermal stress and, 306
Dielectrics
in MOS technology, 423
properties, 425
silicon and silicon
oxides, 425, 427
Dielectric characterisation, 4
Dielectric constants, 220
Dieletric properties of thin
films, 225-252
boundary conditions and, 231
dipole-like phenomena, 246,
248, 249
interface effects, 229, 247
band bending at, 235
low frequency capacitance
at, 240
ions as carriers, 233, 234
Maxwell-Wagner model, 238,
239, 240, 241, 244, 250
polarization and effective
polarization, 225
characteristics, 246
effective currents,
231, 23.2
general solutions, 232
Dielectric relaxation time,
266
Diffusion source of
stress, 338
Diffusion equation, 54
solutions of, 65
Diffusion potentials, 229
Diode infrared detectors, 472
Diode phenomena, 417
'Dirty' films, 22
Doping, 203
Edge filters, 447
Elastic moduli, 339
503
Elastic problem, 274
thermal expansion and, 296
Electric flux configuration, 226
Electrical characteristics of
films, 419
Electrical conduction, 3
Electrical properties, 3
of FET devices, 462
of MOS devices, 427
Electroforming, 40
Electroless deposition, 40
Electron-beam heating, 17
Electronic transport properties,
236
activation energy, 199
continuity, 206
density of ionised donors, 209
doping, 203
excitation, 205, 209
experimental results, 190
hopping, 248
injection limited conduction, 211
ionic defect hopping, 249
Kubo-Greenwood formula, 196,
197, 209
localisation and disorder, 195
mott hopping, 205
Poole-Frenkel and Poole
behaviour, 208, 213
single carrier control, 213
space charge control, 212
specific processes, 203
temperature and, 192
thermopower, 200
Electro-optic devices, 488
Electroplating, 37
Epitaxial films
in junction detectors, 474
stresses in, 332
control of, 336
Evaporation energy, 60
Failure of films, 273
adhesion and, 340, 342
FAMOS devices, 436, 437
Ferroelectric memories, 485
Field effect transistors, 460,
461, 462
504
Field effect transistors
(cont'd):
electrical properties, 462
fabrication of, 466
Filament
definition, 261
thermal switching and, 262
Filters, 356, 447
band pass, 447
conditions for use, 453
dielectric stacks, 451
edge, 447
manufacturing tolerances,
456
performance specification,
453
problems in, 445
Floating gate avalanche
injection MOS, 436, 437
Foils, particles in, 410
Friedel sum rule, 99
Gallium-arsenic field effect
transistor, 460
Gallium films, stress in, 334
Garnet films, stress in, 341
Gaseous anodisation, 28, 29,
33
Germanium, reflectiv,ity of,
190
Glow discharge condition, 20
Gold films
growth of, 77
thickness measurement, 177
Goos-Haenchen effect, 371
Grain boundary diffusion, 339
Grain boundary potential, 320
Grain growth and stress, 319
Graphite films for thermal
switching, 255
Growth of thin films, 3, 49-91
boundary monomer
concentration, 56, 67, 69
evaluation of, 63
capture rate, 51, 65, 73
characterisation, 81
clusters,
coalescence, 70, 71, 76
decay rate of, 56
INDEX
Growth of thin films (cont'd):
clusters (cont'd):
direct impingement on, 70, 78
growth rates, 72
model, ~
size, 66, 85
coalescence of nuclei, 70, 76,
81
decay rates, 56, 63
diffusion equation, 54
solution of, 65
equations, 70
computer calculations, 77
solution of, 73
evaluation of time constants, 54
kinetics, 50
monomer concentration, 78
prediction of mode of, 85
rate of twin formation, 69
rates, 72
connection with material
constants, 50
evaporation, 10, 11
liquid phase epitaxy, 42
of polymer films, 43
sputtering, 22
with anodisation, 32
with vapour phase reactions,
34
stresses in, 313, 314, 315, 319,
320
Hamiltonian, one electron, 93
Helium implanted erbium films
stress in, 339
Hydrophilic films, 42
Infrared detectors, 468
parameters, 469
structure, 472
Injection limited conduction, 211
Insulation devices, 419, 427
point defects in, 390
Integrated optics, 355, 488
Interface, 229
metal-insulator-semiconductor
system, 396
polarization, 229, 235, 240,
247, 250
INDEX
Interface (cont'd):
radiation effects in, 394
stresses at, 287, 288,
304, 309, 321
impurities and, 343
origin of, 315, 316
Ion beam analysis
backscattering, 142, 147,
155
in-depth, 146
MeV energy range,
composition deter-
mination of, 142
of thick films, 143
quantitative, 145, 151
Ion beam measurement of film
thickness, 177
Ion implantation, 410
Ion induced X-ray emission,
154
Ion mass spectrometry,
secondary, 155
Ion plating, 29
Ionic defect hopping, 249
Ionic transport in thin
films, 219-224
drift current, 221
impurity, 219, 222
mobility, 222
Ionization gauge ratemeter,
179, 180
Junction detectors,
preparation and
structure, 474
Kubo-Greenwood formula, 196,
197, 201, 209
Lead salts in infrared
detectors, 468
Light
interaction with thin
films, 488
modulation of intensity,
489
Light waves
propagation in thin
films, 356
Lithium films
microhardness, 340
stresses in, 331
Localization and disorder in
transport, 195
Logical inverters, 404
permalloy films, 480
Magneto-optic devices, 488
Mechanical properties of thin
films,
adhesion, 340
anisotropic stresses, 292
edge effects, 285, 303
elastic problem, thermal
expansion and, 296
for dielectrics, 425
non-metallic, 273-353
elasticity, 274
forces acting on film and
substrate, 277, 278
stresses (see Stresses)
thermal effects, 294
thermal expansion, 287
thermal stresses, 306-312
deposition and, 306
505
Memory devices, 409, 440, 461, 480
magnetic buble, 481
Memory transistors, silicon
nitride layers in, 432
Metal films, thickness
reproducibility, 176
Metal foil structures for
film preparation, 14, 15
Metal-insulator semiconductor
system
hot carriers, 400
radiation effects, 395
Metal-thin film-metal structure,
dielectric properties, 226
Metal-thin film-semiconduction
structure
dielectric properties, 226
Microelectronics
CMOS technology, 421
dielectrics, 425
properties, 425
MOS technology, 420
506
Microelectronics (cont'd):
thin films in, 417-443,
(see also Specific
devices)
Mirror coatings, high
reflectance, 447
Modulators, 377
Molybdenum films, dielectric
properties, 245
Monomers
concentration, 56, 67, 69,
78, 85
evaluation, 63
condensation by direct
impingement, 70
MOSFETS, 461
MOS depletion enhancement
technology, 424
MOS floating gate avalanche
injection, 436, 437
MOS technology, 420
complementary, 421
dielectrics in, 423
silicon nitride films in,
430
silicon on sapphire, 423
MOS tunnel diode, 434
Mott hopping, 205
Multicomponent glasses, 253
thermal switching in, 265
Naturally occuring films, 2
Non-metallic thin films
anistropic stresses, 292
applications of, 459-494,
(see also Separate
devices)
edge effects, 285, 303
elastic problem, 274
forces acting on, 277, 278
mechanical properties of,
273-353
stresses in
anisotropic, 292
at interface, 287, 288,
303, 304, 309, 321
impurities and, 343
components of, 306
compression and, 331
INDEX
Non-metallic thin films (cont'd):
stresses in (cont'd):
control of, 336
deposition and, 323, 327
328, 330, 332
distribution of, 278, 279, 280
edge effects, 285, 303, 310
measurements, 288
mechanism, 306
origin, 312
thermal effects, 294
thermal expansion in, 287
thermal stresses in, 306-312
deposition and, 306
Nuclear reactions, analysis by,
148
Optical coatings
preparation of, 446, 450
monitoring, 452
reactive index, 455
spectral profiles, 455
Optical constants, deterimination
of, 361
Optical filters, 4
Optical properties of thin films,
400, 445
anisotropic, 377
multilayer, 371
matrix technique, 374
single, 356
Optical thin films, 355-382
applications, 445-457
prpblems, 445
determination of constants, 361
stress in, 323
synthesis of multilayer with
prescribed properties, 448
Optic circuits, integrated, 488
Oxidation in preparation, 35
Oxide capacitor films, 477
Oxide films
as thickness guide, 167
growth stresses, 313, 314
space charge trapping, 399
stress in, control of, 338
transmission and reflectance,
392
INDEX
Oxygen transport in anodic
aluminium oxide, 148,
152
Passivation layers, stresses
in, 335
Phase transformations,
stresses and, 321
Photoelectron spectroscopy,
158
Photometric determination
of optical constants,
361
Photometric methods of film
thickness measurement.,
168
Photons, passing through
solid lattices, 383, 385
Physico-chemical analysis of
thin films, 141-161
Plasmons, 138
Polarimetric methods of film
thickness measurement,
169
Polarizing beam splitters,
448
Polarization. 225
characterist":'cs of
materials, 246
dipole-like phenomena, 246,
248, 249
in optic devices, 490, 491,
492
interface effects, 229,
235, 240, 247, 250
Polymer films, growth of, 43
Polymerisation in deposition,
35
Poole-Frankel effect, 208,
213, 221, 392
Preparation of thin films,
2, 9
(see also Deposition)
methods, 9
Properties of thin films
electronic transport
(see Electronic
transport properties)
relation to thickness, 163
Protective coatings, 477
radiation effects in, 410
507
Quartz, colour centres in, 391
Quartz crystal measurement
systems of thickness, 173
Radiation effects in thin films,
383-414
basic features of sensitivity,
398
classification of, 384
damage equivalent units, 386
displacement, 386
environments, 385, 388
in MOS devices, 404
in metal-insulator-semiconductor
system, 395
in protective coatings, 410
interface states, 394
in transistors, 405, 407, 408
investigation techniques, 398
ionization 386, 403
point defects and, 385, 390
sources and levels, 387
surface damage factor, 408
trapping, 390
Re-crystallised materials,
reflectivity of, 190
Reflectance, 445
in multilayer thin films, 376
Reflection in films, 356
total internal, 362
Refreactive i.ndex of films. 447
Research policy, 5
Rotators, 377
Schottly-barrier-gate FET, 467,
468
Secondary ion beam mass
spectroscopy, 155
measurement of film thicl< . .ness
by, 177
quantitative, 156
Semiconductors, 3, 459
activation energy, 199
amorphous, 96
surface states, 106, 113
wave functions, 105, 110
508
Semiconductors (cont'd):
cluster scattering, 99,
102, 104
compound, 460
covalently bonded, 112
density of states, 215
displacement effects in,
402
electronic states of, 93
nature of, 105
one dimensional, 108, 109
energy band structure, 94,
97, 104, 109, 194
failure modes, 402
germanium and silicon
polymorphs, 104
interface with metal, 116
ionization in, 403
lifetime, electronic
theory, 264
multiple scattering approach
to density of state,
97, 104
one electron Hamiltonian, 93
orbitals, 94, 95, 105, 112
physical properties of,
190, 191
radiation effects, 385
trapping, 390
radiation sensitivity, 401
relaxation, electron theory
for, 266
stresses in, 305
stresses build-up in, 304
surface states, 106, 113
effects of, 114
localised, 114
occupation, 115
X-ray spectroscopy, 157
Silicon in MOS tunnel diode,
434
Silicon dioxide
atomic configuration, 393
colour centres in, 391
in dielectrics, 427
trapping, 429
Silicon dioxide films,
optical properties, 400
radiation effects, 390
radiation effects,
investigation of, 398
INDEX
Silicon films
ion transport in, 222, 223
stress in, 333, 334, 335, 341
Silicon monoxide films, stress
in, 329, 339
Silicon nitride films, in MOS
technology, 430
Silicon on sapphire complementary
KOS technology, 423
Silicon on sapphire technologies,
421, 422, 423, 430
Silicon oxide
growth curves, 426
in thin film capacitors, 477
measurement of thickness, ~ 7
Silicon semiconductors. radiation
sensitivity, 401
Single carrier control, 213
Space charge control, 212
Stresses, 274, 275
anisotropic, 292
as interface, 287, 288, 303,
304, 309, 321
impurities and, 343
components of, 306
compression and, 331
control of, 336
post-deposition, 337
deposition, and, 323, 327,
328, 330, 332
distribution within film, 278,
279, 280
edge effects, 285, 303, 310
from bending, 280
growth, 313
index to literature, 324
measurements,
cantilever plate method, 289
indirect methods, 306
X-ray methods, 302
witb measurement of mass
change, 305
mechanisms of, 306
origin of, 312
differences of lattice
spacing, 312, 317
diffusion, 338
grain growth and, 319
incorporation of atoms, 313
phase transformation, 321
INDEX
Stresses (cont'd):
origin of (cont'd):
recrystallization
processes, 318, 339
voids and dislocation
arrays, 321
summary of data of, 322
thermal, 306-312, 322, 332
depostion and, 306
Structure of thin films,
123-:140
amorphous state, 124
crystalline, 124
electron diffraction of,
126, 138
electron microscopy, 130
lattice planes, 131, 132
orientation, 126
roughness of surface,
136, 137, 138
size of crystals, 129
determination, 123
distortion of, 195
geometry of, 123
ideal characterisitcs, 123
inelastic scattering, 134
schematic, 124
surface, 133, 135
roughness of, 135
thickness, 123, 124, 141
Substrate
bending of, 280, 282, 289
during deposition, 282
failures in, 273
in stress control, 337
plastic deformation in,_ 285
relation to growth rate and
cluster decay, 50
stresses, 274, 275, 289
distribution of, 280
temperature history of, 294
temperature stabilization
during deposition,
293, 299
thermal stress and, 306,
307
Substrate-film composite
edge effects, 286, 303
forces acting on, 277, 278
Surface of thin films, 133, 135
atomic diffusion on, 51
energy, 60
plasmons, 138
roughness of, 135
Switching devices, 409
Talystep, 170
509
Tantalum for capacitors, 477, 479
Tellurium in junction detectors,
475
Thermal effects, 294
Thermal expansion, 287
stress and, 308
Thermal expansion tensor, 308
Thermal stress, 306-312, 322, 332
deposition and, 306
Thermal switching
critique of models, 267
current distribution, 261, 262,
263
electronic model, 264, 267
nature of the ON-state, 264
'electrothermal', 254
experimental problems, 269
'forming', 268
heating effects, 254
minimum holding current, 269
non-thermal (electronic)
models, 253
phenomenological
characteristics, 255
polarity effects, 268
primary chqracteristics, 255,
256, 257
relation between ambient
temperature and voltage, 260
secondary characteristics, 255,
258, 259
switch structure, 255, 256
thermal model, 261, 267
Thermal theory for uniform
structures, 261
Thermometers, 408
Thermopower, 200
Thickness of thin films, 123,
124, 141
for optical coatings, 446, 450,
454
510
Thickness of thin films
(cont'd):
measurement of, 163-185
capacitance methods,
174, 180
chemical techniques,
177
direct methods, 164
electrical methods,
173, 179
interference methods, 164
ionisation gauge systems,
179
magnetic methods, 177
mechanical methods, 169,
178
methods, 163
micro-balances, 170
oxide on metal as
guide, 167
photometric methods, 168
polarimetric methods, 169
probe systems, 169
quartz crystal systems,
173, 179
radioactive techniques,
177
rate control systems, 178
resistance methods, 176
rotating systems, 171
summary of methods , 181,
182
optical applications and,
449
relation to properties, 163
Thiele theory, 483
Threshold 253-272
thermal models, 253
thermal theory, 261
Transistors
bipolar, radiation effects,
405, 407, 408
field effect, 460, 461, 462
electrical properties,
462
fabrication of, 466
GaAs field-effects, 460
MOS, 419
diagram of, 418
Transistors (cont'd):
memory, 440
INDEX
silicon nitride layers in,
432
radiation effects in, 405, 407,
408
Transport, electronic (see
Electronic transport)
ionic (see Ionic transport)
Transport across films, 4
Transport properties, electronic
(see Electronic transport
properties)
Twin formation, rate of, 69
Uniform structures
electronic theory for, 264
thermal theory for, 261
Waveguides, 366, 367, 410
characteristics, 368
coupling to external beams, 370
Wigner delay time, 98, 99
X-ray measurements of stress, 302
X-ray spectroscopy of films, 157
Ytterbium orthoferrate films in
magnetic bubble devices,
481
Zinc sulphide films
deposition and stress, 327
failure of, 341
structure of, 329