E L E C T R O N I C ENGINEERING No.

19

Metal-Semiconductor
Contacts
SECOND EDITION

E. H. RHODERICK
and
R. H . WILLIAMS
in mouern semiconauctor lecnnoiogy, contacts Between semiconducting
devices and the metal conductors that connect them with the rest of the
system are of fundamental importance. This bodk deals with the basic
science of such contacts, and discusses the electrical properties that are
relevant to semiconductor technology.
Topics covered include the mechanism of formation oi Schottky
barriers, the current/voltage relationship, the capacitance of rectifying
contacts, and practical methods of fabricating contacts. The practical
implications are emphasized wherever they are relevant to device technol­
ogy, though there is no treatment of devices themselves.
The main difference between the first and second editions is the greatly
expanded treatment of the physics of Schottky-barrier formation, together
with some refinements in the discussion of electrical characteristics and
of the effect of deep levels.
The book is aimed at semiconductor technologists ana at physicists
engaged in research on semiconductor interfaces.
E. H . Rhoderick is Emeritus Professor of Solid State Electronics, UMIST.
R. H . Williams is Head of Department of Physics, University College,
Cardiff.
Monographs in Electrical and Electronic Engineering
General Editors: P. Hammond and R. L . Grimsdale
The theory of linear induction machinery (1980)
Michel Poloujadoff
Energy methods in electromagnetism (1981)
P. Hammond
Low-noise electrical motors (1981)
S. J. Yang
Superconducting rotating electrical machines (1983)
J. R. Bumby
Stepping motors and their microprocessor controls (1984)
T. Kenjo
Machinery noise measurement (1985)
S.J. Yang and A. J. Ellison
Permanent-magnet and brushless DC motors (198S)
T. Kenjo and S. Nagamori
Metal-semiconductor contacts. Second edition (1988)
E. H . Rhoderick and R. H . Williams H ^ ' ' o-i9-859335-x
Monographs in Electrical and Electronic Engineering 19

Series Editors: P. Hammond and R. L . Grimsdale

Monographs in Electrical and Electronic Engineering

10. The theory of linear induction machinery (1980)

M . Poloujadoff
12. Energy methods in electromagnetism (1981) P. Hammond
13. Low-noise electrical motors (1981) S. J. Yang
15. Superconducting rotating electrical machines (1983) J. R. Bumby
16. Stepping motors and their microprocessor controls (1984) T. Kenjo
17. Machinery noise measurement (1985) S. J. Yang and A . J. Ellison
18. Permanent-magnet and brushless DC motors (1985) T. Kenjo and
S. Nagamori
19. Metal-semiconductor contacts. Second edition (1988)
E . H . Rhoderick and R. H . Williams
Metal-Semiconductor
Contacts

Second Edition

E. H. Rhoderick
Emeritus Professor of Solid-State Electronics
The University of iVlanchester Institute of
Science and Technology

and
R. H. Williams
Professor of Physics
University College
Cardiff

n A R F . N D O N PRFSS • O X F O R D 198 8
 Oxford University Press, Walton Street, Oxford 0x2 6DP
Oxford New York Toronto
Delhi Bombay Calcutta Madras Karachi
PetalingJaya Singapore Hong Kong Tokyo
Nairobi Dar es Salaam Cape Town
Melbourne Auckland
and associated companies in
Berlin Ibadan

Oxford is a trade mark of Oxford University Press

Published in the United States

by Oxford University Press, New York

First edition © E. H. Rhoderick, 1978

Second edition© E. H. Rhoderick and R. H. Williams, 1988

All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise, without
the prior permission of Oxford University Press

This book is sold subject to the condition that it shall not, by way
of trade or otherwise, be lent, re-sold, hired out, or otherwise circulated
without the publisher's prior consent in any form of binding or cover
other than that in which it is published and without a similar condition
including this condition being imposed on the subsequent purchaser

British Library Cataloguing in Publication Data

Rhoderick, E. H. (Emlyn Huw).
Metal-semiconductor contacts.—2nd ed.
1. Semiconductors. Interfaces with metals
1. rule II. Williams, R. H. Ш. Series
537.6'22
ISBN 019 859335 Xp/b
ISBN 0 19 859336 8 h/b

Library of Congress Cataloging in Publication Data

Data Available.
fISBN 019859335 Xp/b]
[ISBN 0 19 859336 8 h/bj

Typeset by Cotswold Typesetting Ltd, Cheltenham

Printed and bound in Great Britain by
Biddies Ltd, Guildford and King's Lynn
Preface to Second Edition

Ten years have gone by since the appearance of the first edition of this
book. During this time there has been intense activity in the field of semi­
conductor contacts, both with regard to the fundamental physics and in
the applications and technology. Although there is very little in the first
edition that needs correction, the immense progress that has been made
calls out for a new edition. Most of the recent developments have been in
the basic physics of barrier formation, and herein lies the most significant
difference between the two editions. But we have to admit that we are
still some way from a complete understanding of all the factors that
determine the height of a Schottky barrier and from being able to
provide a reliable prescription for making a contact with specified
electrical properties. A l l we have attempted to do is to guide the reader
through the welter of publications that have recently appeared, with the
hope that eventually some light will appear at the end of the tunnel. If our
discussion of the basic physics of barrier formation appears less than
conclusive, we can but plead that this is in many ways a fair reflection of
the current situation.
We have benefited from many fruitful discussions with colleagues in
the Department of Electrical Engineering and Electronics at U M I S T and
in the Physics Department of University College, Cardiff, and we wish to
thank D r G . W. Rubloff of I B M , D r D . C. Northrop of UMIST, and
Professor W. E . Spear of Dundee for reading through and commenting
on parts of the manuscript. Our very special thanks are due to Dorothy
Denton, Jeanne Carter, Gail Jelhngs, and Carol-Lyim May for their
uncanny skill in reducing the chaos of our manuscript to perfect order,
and to Bob Watkins for help with the illustrations. Finally, we must thank
our wives, Mary and Gill, for their patience and forbearance in putting
up with hours of neglect.
Manchester E . H . R.
Cardiff R. H . W.
September 1987
Contents

List of symbols xi
1. S U R F A C E S , I N T E R F A C E S , A N D
SCHOTTKY BARRIERS 1
1.1 Historical 1
1.2 Preliminaries—some surface properties of solids 2
1.2.1 Clean and real surfaces 3
1.2.2 Electron states in solids and at surfaces 5
1.2.3 The work function of a solid 10
1.3 The formation of a Schottky barrier 11
1.3.1 The Schottky-Mott theory 11
1.3.2 The effect of surface (or interface) states 15
1.4 Generalized analysis of the Bardeen model 17
1.4.1 The flat-band barrier height with a continuous 17
distribution of surface states
1.4.2 The flat-band barrier height with two bands 20
of surface states
1.4.3 The field dependence of the barrier height 23
1.4.4 p-type semiconductors 25
1.4.5 The bias dependence of the barrier height 26
1.4.6 The penetration of the field into the metal 27
1.5 Intimate contacts 28
1.5.1 General 28
1.5.2 Metal-induced gap states 30
1.5.3 Defects at interfaces 33
1.6 Image-force lowering of the barrier 35
1.7 Methods of measurement of barrier heights 38
1.7.1 From current/voltage characteristics 38
1.7.2 From photoelectric measurements 41
1.7.3 From capacitance measurements 42
1.7.4 From photoelectron emission spectroscopy 43
1.8 Electronegativity and work function 47

2. E X P E R I M E N T A L S T U D I E S O F M E T A L S
ON SEMICONDUCTORS 49
2.1 Metals on silicon and germanium 49
viii CONTENTS

2.1.1 Clean silicon 49

2.1.2 Etched and oxidized silicon surfaces 53
2.1.3 Germanium 57
2.1.4 Discussion 57
2.1.4.1 Lack of reproducibility 57
2.1.4.2 Etched surfaces 60
2.1.4.3 Intimate contacts 61
2.2 Metals on III-V semiconductors 63
2.2.1 General 63
2.2.2 Ultra-thin metal layers on gallium arsenide 65
and indium phosphide
2.2.2.1 GalHum arsenide 65
2.2.2.2 Indium phosphide 68
2.2.3 Thick metal contacts on gallium arsenide and 69
indium phosphide
2.2.3.1 Clean gallium arsenide 69
2.2.3.2 Oxidized gallium arsenide 72
2.2.3.3 Indium phosphide 74
2.2.4 Discussion—metals on gallium arsenide and 76
indium phosphide
2.2.5 Metal contacts to other III-V semiconductors 79
2.2.5.1 Binary compounds 79
2.2.5.2 Ternary alloys 80
2.3 Metals on other semiconductors 82
2.4 Summary and conclusions 84

3. CURRENT-TRANSPORT MECHANISMS 89
3.1 Introduction 89
3.2 Emission over the barrier 90
3.2.1 The two basic mechanisms 90
3.2.2 The diffusion theory 92
3.2.3 The thermionic-emission theory 94
3.2.4 The effect of the image force on the 98
current/voltage relationship
3.2.5 The combined thermionic-emission/diffusion 100
theory
3.2.6 Hot-electron effects 104
3.2.7 Refinements of the thermionic-emission theory 105
3.2.8 Comparison with experiment 107
3.3 Tunnelling through the barrier 109
3.3.1 Field and thermionic-field emission 109
 CONTENTS ix

3.3.2 Ohmic contacts 116

3.4 Recombination in the depletion region 118
3.5 Hole injection 121
3.5.1 Hole inj ection in plane contacts 121
3.5.2 Hole injection in point contacts 123
3.6 Reverse characteristics 124
3.6.1 Field dependence of the barrier height 124
3.6.2 The effect of tunnelling 126
3.6.3 Generation in the depletion region 132
3.7 Transient effects 132
3.8 The effect of an interfacial layer 133
3.9 The To'effect 139
3.10 Numerical analysis of current flow 139

4. T H E C A P A C I T A N C E O F A
SCHOTTKY BARRIER 141
4.1 The capacitance of an ideal diode under reverse bias 141
4.1.1 The general case 141
4.1.2 The case in which the minority carriers are 144
negligible
4.1.3 The effect of minority carriers 147
4.2 The effect of an insulating layer with interface states 150
4.2.1 Very thin interfacial layers 150
4.2.2 Thick interfacial layers 153
4.2.3 The general case 154
4.2.4 Interface state spectroscopy 155
4.3 Non-uniform donor distribution 158
4.4 C/V methods of measuring dopant distributions 159
4.5 The effect of deep traps 162
4.5.1 The population of deep traps under reverse bias 163
4.5.2 The contribution of traps to the capacitance 167
under reverse bias
4.5.3 Transient measurements 174
4.5.4 The effect of light 177
4.6 The capacitance under forward bias 179
4.6.1 The diffusion capacitance 179
4.6.2 The effect of traps 179
X CONTENTS

5. P R A C T I C A L C O N T A C T S
5.1 Methods of manufacture
5.1.1 Point contacts
5.1.2 Evaporated contacts
5.1.3 Sputtered contacts
5.1.4 Chemical deposition
5.2 The effects of heat treatment
5.2.1 Contacts to silicon
5.2.2 Contacts to compound semiconductors
5.3 Silicides 193
5.3.1 Mechanism of formation 1^4
5.3.2 The sequence of silicide phase formation 196
5.3.3 The abruptness of the interface 198
5.3.4 Barrier heights 199
5.4 Control of barrier heights 202
5.5 Ohmic contacts 204
APPENDIX A The depletion approximation 211
APPENDIX B Exact analysis of the electric field in a Schottky 216
barrier
APPENDIX C Comparison of Schottky diodes and 219
p-n junctions
C . l Current-transport mechanisms 219
C.2 Hole injection 221
C.3 Minority-carrier storage 222
APPENDIXD The hole quasi-Fermi level 224
APPENDIXE Contacts to amorphous semiconductors 226
E . l Introduction 226
E.2 Schottky barriers in a-Si:H 227
E.3 Capacitance measurements 228
E.4 Current/voltage characteristics 229

REFERENCES 233

INDEX 249
List of symbols

a = {q^NJlE^kT)''^ Ieqn(3.7)] energy of bottom of

A Richardson constant conduction band in
corresponding to free-electron semiconductor
mass E, Fermi energy
A* Richardson constant Ef Fermi level in metal
corresponding to effective £f Fermi level in semiconductor
mass in semiconductor [eqn £g energy gap of semiconductor
(3.11)1 Ei intrinsic Fermi level of
A** value oiA* corrected for semiconductor
quantum-mechanical £. energy of trap
reflection and phonon energy of top of valence band
backscattering electron demarcation level
b temperature coefficient of ^^o [eqn(4.10e)]
= a„v Eap hole demarcation level [eqn
(4.10c)]
C differential capacitance per £o defined by eqn (3.27)
unit area( = d2/dr) defined by eqn (3.23)
Q capacitance per unit area £i = E, + (kT/2q)ln(a/a„)
associated with depletion (Section 4.5.1)
region S' electric-field strength in
Co. capacitance per unit area semiconductor
associated with interfacial s; electric-field strength in
oxide layer interfacial layer
Q capacitance per unit area ^max maximum electric-field
associated with interface states strength in Schottky barrier
Q capacitance per unit area / probability of occupation of
associated with deep traps trap
diffusion constant for probability of occupation of
electrons in n-type interface state
semiconductor m Dawson's integral
A diffusion constant for ( = exp(-A:2)/5exp(y^)dy)
electrons in p-type h Planck's constant
semiconductor n Planck's constant divided by
diffusion constant for holes
A density of surface states i =7-1
(eV-' / current
probability per unit time of a J current density
trap emitting an electron Jo reverse saturation current
probability per unit time of a density
trap emitting a hole J. current density due to
electron energy electrons
xii LIST OF SYMBOLS
current density due to holes
4 current density due to
thermionic emission
J. current density due to
recombination in depletion
region
h current density due to
generation in depletion region
k Boltzmann's constant or wave
vector of electron
I mean free path of electrons
L thickness of quasi-neutral
region
diffusion length for electrons
m free-electron mass
m* effective mass of electrons in
semiconductor
n density of electrons in
conduction band of
semiconductor or ideality
factor [eqn (3.14)]
n, intrinsic electron
concentration
K effective density of states in
conduction band of
semiconductor
K donor density
N. trap density
K effective density of states in
valence band of
semiconductor
p density of holes in valence
band of semiconductor
Po equilibrium density of holes at
edge of depletion region
1 magnitude of electronic charge
charge per unit area due to
uncompensated donors
a charge per unit area due to
holes
charge per unit area on surface
of metal
Gss charge per unit area due to
surface states
'•o Thomas-Fermi screening
distance
area of contact
T absolute temperature
u = qip/kT (Appendix B)
value of u at surface
V mean thermal velocity of
electrons or holes
Vd diffusion velocity [eqn (3.18)]
recombination velocity
(Section 3.2.5)
V applied bias (positive for
forward bias)
V. diffusion voltage or 'band
bending'
diffusion voltage at zero bias
drop in potential across
interfacicJ layer
K reverse bias ( = — F)
w width of depletion region
position of maximum of
barrier
y length defined in Fig. 4.12
(=w-A)
a = 6£/(£i + qdD,) (eqn (1.18)]
a' empirical quantity denoting
dependence of on
(Section 3.6.1)
= (d^,/aF) [eqn (3.12)]
y = £/(£i + 9(5A)[eqn(1.6)]
hole-injection ratio
6 thickness of interfacial layer
permittivity of interfacial layer
(=eireo)
permittivity of semiconductor
(=esreo)
e; effective permittivity of
semiconductor for image force
permittivity of free space
quasi-Fermi level (or imref)
for electrons
quasi-Fermi level (or imref)
for holes
I length defined in Fig. 4.12
mobility of electrons
V frequency of light wave
= K-Ei
p surface-dipole moment per
unit area
capture cross-section for
electrons
capture cross-section for holes
 LIST OF SYMBOLS xiii

T time constant of interface flat-band barrier height for

'm states due to tunnelling to and p-type semiconductor
from metal 4. effective barrier height
TCBH time constant of interface
states due to S-R-H effective barrier height for
recombination with zero bias
semiconductor bands work function of metal
mean time between collisions work function of
for electrons semiconductor
T, recombination time in the 4o neutral level for surface states
depletion region (Section 1.2.2)
^ch mean time between collisions lowering of barrier due to
for holes image force
Tre recombination time for image-force lowering for zero
electrons bias
Trh recombination time for holes Xs electron affinity of
f^b height of Schottky barrier semiconductor
(measured from Ef) Xm electronegativity of metal
barrier height for zero bias V electrostatic potential
barrier height for zero electric angular frequency of bias
field (flat-band barrier height) modulation
<l>l flat-band barrier height for angular frequency of test
n-type semiconductor signal

Note: MKS units are used throughout. A l l energies are measured in eV,
so the magnitude of the potential energy of an electron associated with its
electrostatic potential is equal to the potential measured in volts.
The density of surface states is sometimes measured in J~' m~^. If is
expressed in eV"' and D'^ in J"' m~^, then = qD^.
 I Surfaces, interfaces, and Schottky barriers

1.1 Historical

Our knowledge of metal-semiconductor contacts goes back more than a

hundred years to the early work of Braun (1874), who discovered the
asymmetric nature of electrical conduction between metal contacts and
semiconductors such as copper and iron sulphide. Although the rectifica­
tion mechanism was not understood, contacts between metal points and
metallic sulphides were used extensively as detectors in early experiments
on radio, and it seems likely that Lodge's 'coherer' (1890) must have
rehed for its action on the conduction properties of metal particles
separated by oxide films. In 1906 Pickard took out a patent for a point-
contact detector using silicon, and in 1907 Pierce published rectification
characteristics of diodes made by sputtering metals onto a variety of
semiconductors. The rapid growth of broadcasting in the 1920s owed
much to the 'cat's-whisker' rectifier which consisted of a tungsten point in
contact with a crystal, usually of lead sulphide. The first copper oxide
plate rectifiers appeared at about the same time (Grondahl, 1926,1933).
The first step towards understanding the rectifying acdon of metal-
semiconductor contacts was taken in 1931, when Schottky, Stormer, and
Waibel showed that if a current flows then the potential drop occurs
almost entirely at the contact, thereby implying the existence of some sort
of potential barrier. By this time quantum mechanics was firmly
established and in 1932 Wilson and others tried to explain the rectifying
action in terms of quantum-mechanical tuimelling of electrons through a
barrier, but it was soon realized that this mechanism predicted the wrong
direction of easy current flow. In 1938 Schottky and, independently, Mott
pointed out that the observed direction of rectification could be explained
by supposing that electrons passed over a potential barrier through the
normal processes of drift and diffusion. According to Mott (1938), the
potential barrier arose because of a difference between the work
functions of the metal and semiconductor; he supposed that the barrier
region was devoid of charged impurities so that the electric field was
constant and the electrostatic potential varied linearly with distance as the
metal was approached. In contrast, Schottky (1939) supposed that the
barrier region contained a constant density of charged impurities so that
the electric field increased linearly and the electrostatic potential
quadratically, in accordance with Poisson's equation, as the metal was

1
 SURFACES, INTERFACES, AND SCHOTTKY BARRIERS

approached. Similar ideas about the role of the space charge in

determining the shape of the barrier were advanced by Davydov (1939,
1941) in the USSR.
A significant advance in our understanding of metal-semiconductor
contacts came during the Second World War as a result of the use of
silicon and germanium point-contact rectifiers in microwave radar. This
advance was considerably helped by developments in semiconductor
physics. Perhaps the most important contribution during this period was
Bethe's thermionic-emission theory (1942), according to which the
current is determined by the process of emission of electrons into the
metal, rather than by drift and diffusion in the semiconductor as was
supposed by Mott and Schottky.
After 1945, work on metal-semiconductor contacts was stimulated by
the intense activity in the field of semiconductor physics which lead up to
the invention of the point-contact transistor, and attention was mainly
focused on point-contacts as injectors of minority carriers. The eventual
demise of the point-contact transistor switched attention towards
extended-area contacts.
The realization that evaporation of metal films in a high vacuum system
produced contacts which were much more stable and reproducible than
point-contacts triggered off a great flurry of activity in the 1950s and
1960s and laid the foundation for our present extensive knowledge of the
subject. This activity was inspired to a considerable extent by the great
importance of contacts in semiconductor technology and was largely
associated with the theoretical work of Bardeen, Crowell, and Sze, and
the experimental work of Goodman, Archer and Atalla, and Kahng,
Mead, and Cowley.
During the past twenty years components based on Schottky barriers
have been increasingly used in microelectronics, and research activity has
continued with the aim of obtaining a full understanding of the physics of
barrier formation and of current transport across metal-semiconductor
interfaces. The armoury of techniques available has grown to include
modern surface-science analytical methods, which have been applied to
probe the detailed microscopic interactions at interfaces between metals
and semiconductors. Theoretical methods of establishing the electronic
properties of solids and of boundaries between them have also advanced
in parallel, and computers have increasingly been used to analyse the
current transport processes.

1.2 Preliminaries—some surface properties of solids

A n overview of the surface properties of solids will be presented in this

section; for more detailed information relating to surfaces the reader is

2
 PRELIMINARIES-SOME SURFACE PROPERTIES OF SOLIDS

eferred to articles by Inglesfield (1984), Williams, Srivastava, and

McGovern (1984), and to the book by Prutton (1983).

1.2.1 Clean and real surfaces

The atoms in a crystalline solid are arranged in a well-ordered structure

so that the net force acting on each atom is zero. Figure 1.1 shows the bulk
crystallographic structure of silicon; here each atom is surrounded by four
others in a tetrahedral configuration. Except at imperfections such as
vacancies, substitutional or interstitial impurities, or dislocations, this
configuration extends throughout a single crystal. The situation at the
surface of a crystalline solid is different from that prevaihng in the bulk in
many important ways. First of all, because of the absence of neighbouring
atoms, the equilibrium positions adopted by the surface atoms differ from
those corresponding to the perfect crystal lattice, and the surface is said to
be relaxed or reconstructed.
(Ill)

Fig. 1.1 The silicon structure, showing dangling bonds on the (111) face. (From Prutton,
1983. Copyright Oxford University Press.)

Relaxation refers to the situation where the surface retains the

symmetry of the bulk in a plane parallel to the surface, but the spacing
and arrangement of atoms is different from that in the bulk in the
direction perpendicular to it. A n example of a relaxed surface is shown in
Fig. 1.2 for the case of the gallium arsenide (110) surface. This surface
structure has been determined following low-energy electron diffraction

3
 SURFACES, INTERFACES, AND SCHOTTKY BARRIERS

Fig. 1.2 (a) Schematic illustration of the (110) surface geometry for III-V semiconductor
compounds, (b) The corresponding surface unit cell. The values of the parameters for i
GaAs are given in Table 1. (After Mailhiot, Duke, and Chadi, 1985. Copyright Elsevier
S.P.)

Table 1.1 Values (in A) of the parameters shown in Fig. 1.2(a)

for GaAs(llO)

5.6537 0.69 4.518 1.442 3.339

Studies (Mailhiot, Duke, and Chadi, 1985). Here the surface gallium
atoms are displaced towards the bulk and the surface arsenic atoms away
from the bulk, but the surface unit mesh is independent of these .
movements. Table 1.1 presents the structural parameters for this surface.
The surface is said to be reconstructed when the symmetry in the plane \
of surface is not the same as that of the bulk. Both the (100) and (111)
clean surfaces of silicon are naturally reconstructed. In fact several
different reconstructions are observed on the (111) surface, each one
being associated with different preparation conditions. Reconstruction
and relaxation of surfaces often involve atoms not just in the surface
atomic layer but in the second and third layers as well. The depth over \
which this occurs is referred to as the 'selvedge' region.
Surface reconstructions are usually denoted according to the
convention of Wood (1964). A n {m x n) unit means that the mesh is m
times larger than that associated with the underlying bulk structure in one
particular crystallographic direction, and n times larger in another. In
some cases the unit mesh is centred and is referred to as c(m x «) (the i
GaAs(OOl) c(2 X 8) structure is an example of this kind].
The second important difference between the bulk and the surface of a
solid is associated with the chemical nature of the surface. The structure

4
 PRELIMINARIES-SOME SURFACE PROPERTIES OF SOLIDS

... tj-ated in Fig. 1.2 refers to a surface which is atomically clean, i.e. the
atomic composition of the surface is the same as that in the bulk. These
1 an surfaces can only be generated in certain ideal conditions, for
example by fracture or cleavage of a crystal in ultra-high vacuum. Thus, if
a crystal of silicon is cleaved in a vacuum of 10" Torr to reveal a clean
n i l ) surface, then that surface can be retained for several hours before a
monolayer of impurity atoms from the surrounding vacuum environment
is formed on it. However, cleavage under ordinary atmospheric
conditions leads to the formation of an adsorbed layer of atoms from the
environment in a fraction of a second. Thus, under normal atmospheric
conditions the surfaces of metals and semiconductors are not clean but
are covered by one or more atomic layers of contaminant, usually carbon,
hydrocarbon, or oxygen. Nearly always these 'real' as opposed to 'clean'
surfaces are disordered, i.e. they lack periodicity both parallel and
perpendicular to the surface.
The surfaces of solids are therefore complex regions where the
chemistry and crystallography may be quite different from the bulk. Not
surprisingly, the distributions of electrons and their associated energy
levels also generally differ at the surface from those pertaining to the bulk,
as will be discussed in Section 1.2.2. Since these surfaces are crucial in the
formation of metal-semiconductor interfaces and Schottky barriers, it is
of some importance to study and to understand them to the fullest
possible extent. Regretfully, our understanding of surfaces, and of surface
structures in particular, is still rather primitive due to the experimental
problems of probing layers whose thickness is only of the order of a few
atomic dimensions.

1.2.2 Electron states in solids and at surfaces

In a crystalline solid the electrons are influenced by the periodic

potential associated with the arrangement of atoms, and as a result gaps
occur in the allowed energy spectrum of electrons. The valence band of
silicon, for example, is composed of the outermost atomic 3s^ and 3p^
electrons and is separated from the conduction band by the energy gap,
which has a magnitude of 1.1 eV. The valence electrons are in energy
states between the valence band maximum, taken as the zero of energy,
and —14 eV. The density of these states as a function of energy is shown
in Fig. 1.3(a). This distribution of states has been calculated by
Chelikowsky and Cohen (1974) and may be readily measured by
techniques such as photoelectron spectroscopy (see Section 1.7.4). The
procedure used to calculate the density of states also allows a contour
map of the distribution of valence electrons in silicon to be obtained. The
resulting charge distribution is illustrated in Fig. 1.4, where it may be
seen that a high density of charge exists between the silicon atoms in the

5
 Density of states

Distance -

(a) (b)

Fig. 1.3 (a) Schematic illustration of the density of states for silicon, (b) Band structure
plot as a function of distance, showing surface parameters.

Si (111) surface
total valence charge

Fig. 1.4 Total valence charge density for an ideal unrelaxed Si(lll) surface (theoretical).
The contours are plotted in the (110) plane with shaded circles representing the atomic
cores. (After Cohen, 1980,1982. Copyright Academic Press.)

6
 bulk; this of course is associated with the bonding electrons. Normally it
is a diagram of electron energy levels as a function of distance into the
semiconductor that is used in dealing with semiconductors, as shown in
Fig. 1.3(b).
At the surface of a sohd the bulk periodic potential is abruptly
terminated so that the conditions which give rise to band gaps in the bulk
no longer prevail at the surface. This means that 'surface states' may exist
whose wave-functions correspond to imaginary values of the wave vector
k and decay exponentially with distance from the surface into the bulk.
They often, but not always, have energies within the forbidden band gap.
Surface states of this kind, which exist on a perfect free surface of an
ideal sohd, are sometimes referred to as 'intrinsic' states. They were first
discussed by Tamm (1932) and Shockley (1939).
^It is sometimes helpful to look at surface states from the point of view
of chemical bonds. Consider the sihcon lattice illustrated in Fig. 1.1. In
the bulk the atoms are bound to their neighbours by covalent bonds,
each involving two electrons. A t the surface the atoms have neighbours
on one side only and, on the vacuum side, the valence electrons have no
partners with which to form covalent bonds. Each surface atom therefore
has associated with it an unpaired electron in a localized orbital which is
directed away from the surface. Such an orbital is often spoken of as a
'dangling' bond; it can either give up an electron and act as a donor or
accept another and act as an acceptor. According to this simple model,
there should be twice as many surface states as there are surface atoms,
and surface charge neutrality should correspond to half of the surface
states being occupied. Surface states on ordered crystals may be readily
calculated with modern computational methods and a calculated charge
distribution for an unreconstructed silicon (111) surface is illustrated in
Fig. 1.5 (after Cohen, 1980, 1982). These charge contours correspond to
electrons occupying energy states in the band gap and clearly have a
'dangling bond' character.
Surface states are regularly seen in experiments, and often form two-
dunensional bands which may overlap the valence and conduction bands.
Such overlapping surface states would differ from the valence and
conduction states in that they would be localized near the surface rather
than extending through the crystal. The detailed form of the surface
states, of course, depends on the location of the atoms at the surface and
is significantly influenced by relaxation or reconstruction of the surface.
The states measured on clean cleaved sihcon (111) surfaces by Uhrberg
et al. (1982) are quite different from the theoretical states appropriate to
the unreconstructed surface illustrated in Figs. 1.1 and 1.5.
It is apparent that surface states on semiconductors may exist within
the fundamental band gap as discrete energy states or as a continuum.
The wave functions which constitute the surface states are drawn from

7
 Si (111) surface

Fig. 1.5 Theoretical charge density contours calculated for dangling bond surface states
on Si(lll) unrelaxed surfaces. (After Cohen, 1980, 1982. Copyright Academic Press.)
those states which would constitute the valence and conduction bands of
an infinite solid, so that the densities of states of valence and conduction
bands at the surface are diminished. It follows that the full complement
of electrons necessary to make the surface as a whole electrically neutral
can only be accommodated if the surface state bands are partially filled.
Consider a continuum of surface states extending through the energy
gap. These states will be occupied by electrons in accordance with a
Fermi-Dirac distribution, and their charge state will depend on the
position of the Fermi level Ep. It is often a good enough approximation to
assume that the Fermi-Dirac distribution is a step function, i.e. that all
the states below E^ are occupied and all those above Ep are empty, as
would be the case at the absolute zero of temperature. It is possible to
define a neutral level ^Q, which is the position that the Fermi level must
assume if the surface is electrically neutral. If states below are empty
the surface has a net positive charge, while if states above ^^e full the
surface has a net negative charge. The states below are sometimes
described as donor-like (positive when empty) and the levels above as
acceptor-hke (negative when filled). If the surface states are in the form
of more than one band rather than a single continuum, then may lie in
one of the bands or in gaps in betwen bands.

8
 Clean cleaved (111) surfaces of silicon and germanium have a high
density of surface states. The states comprise two bands, one of which
Ues entirely below the top of the valence band while the other lies within
the band gap as illustrated in Fig. 1.6(a) (after Hansson et ah, 1983 and
NichoUs, Martensson, and Hansson, 1986). A t absolute zero the first
band would be completely filled (assuming £ p lies above the top of the
valence band), while the second band would be empty in p-type material.
For this situation lies between the occupied and unoccupied bands of
surface states. If the Fermi level at the surface does not coincide with the
neutral level, there will be a net charge at the surface, and this will
produce an electric field in the semiconductor which causes bending of
the energy bands. If the surface charge is negative the bands will bend
upwards towards the surface; this is the situation illustrated in Fig. 1.6(a)
for an n-type sihcon crystal. The majority carrier concentration in the
region of the crystal over which the bands are bent is less than that in the
interior of the semiconductor and the surface region is said to be
depleted. If, on the other hand, the surface charge is positive (for the
n-type case) the bands will bend downwards towards the surface, the
majority carrier concentration will be increased, and the near surface
region of the semiconductor is said to be accumulated.
In contrast to silicon, no surface states are observed in the band-gap of
step-free clean cleaved (110) surfaces of most of the III-V semi­
conductors, such as GaAs. Theoretical considerations show that states
should exist in the gap if the (110) surfaces were not relaxed; however,
when the surface relaxation described in Section 1.2.1 is taken into
consideration, the calculations show that the occupied states lie below
the top of the valence band whereas the empty ones lie above the bottom
of the conduction band, as illustrated in Fig. 1.6(b) (Chadi, 1979;

Empty r
states L.

Fig. 1.6 Schematic illustration of surface states on (a) the Si(lll) cleaved surface, and (b)
the GaAs(llO) cleaved surface.

9
 Chelikowsky and Cohen, 1979). The surface states have no net charge,
and the bands remain flat up to the surface. Gallium phosphide is an
exception; empty states do appear in the band gap of cleaved surfaces in
this case. Imperfections such as steps may also lead to locahzed energy
levels in the band gap at the surface.

1.2.3 The work function of a solid

The work function ^„ of a metal is the amount of energy required to raise

an electron from the Fermi level to a state of rest outside the surface of
the metal (the so-called vacuum level). If the work function is calculated
using quantum mechanics (see, for example, Seitz, 1940; Inglesfield,
1984) it is found to consist of two parts: a volume contribution (which
represents the energy of an electron due to the periodic potential of the
crystal and the interaction of the electron with other electrons) and a
surface contribution (due to the possible existence of a dipole layer at the
surface). In general the electron charge distribution around the atoms at
the metal surface is not symmetrically disposed in relation to the nucleus,
so that the centres of positive and negative charge do not coincide,
resulting in a dipole layer. H^^ikewise, the relaxation or reconstruction
associated with a surface can also lead to dipole layers. If the resulting
dipole layer has an electric dipole movement p per unit area, there will
be an electrostatic potential difference of magnitude P/EQ between the
vacuum and the interior of the metal. The change in energy qp/sa of an
electron due to this change in electrostatic potential constitutes part of
the work function. Clearly, any modification in the surface electron
charge distribution, for example by adsorption of gas atoms on a clean
metal surface, will lead to a change in the dipole layer and hence in (^^.
Different crystallographic faces of the same crystal may also have
different values of due to surface dipoles of unequal magnitudes. For
example, the work functions of the tungsten (110), (100), and (111)
surfaces have been reported as 5.25, 4.63, and 4.47 eV, respectively
(Strayer, Mackie, and Swanson, 1973).
The work function of a semiconductor is the difference in energy
between the Fermi level and the vacuum level and is also composed of
bulk and surface contributions. It may seem strange that the work
function is defined in this way when there are usually no allowed energy
levels within the semiconductor at the Fermi level, but it must be
remembered that the work function is a statistical concept and represents
the weighted average of the energies necessary to remove an electron
, from the valence and conduction bands, respectively.
Another important surface parameter for a semiconductor is the
electron affinity Xs illustrated in Fig. 1.3(b). This is the difference in
energy between an electron at rest outside the surface and an electron at

10
the bottom of the conduction band just inside the surface. The electron
affinity is sensitive to surface dipoles in an analogous manner to the work
function. If the bands are flat (i.e. there is no electric field inside the
semiconductor), the work function and the electron affinity are related by

<l>. = Xs+^ (1.1)

where ^ is the energy difference between the Fermi level and the bottom
of the conduction band. Finally, the ionization energy / is defined by the
relationship.

/ is the minimum energy needed to remove an electron from the valence

band.

1.3 The formation of a Schottky barrier

1.3.1 The Schottky-Mott theory

To see how a Schottky barrier may form when a metal comes into
contact with a semiconductor, suppose that the metal and semiconductor
are both electrically neutral and separated from each other. The energy-
band diagram is shown in Fig. 1.7(a) for an n-type semiconductor with a
work function less than that of the metal; this is the most important case
m practice, and we suppose that there are no surface states present. If the
metal and semiconductor are connected electrically by a wire, electrons
pass from the semiconductor into the metal and the two Fermi levels are
forced into coincidence as shown in Fig. 1.7(b). The energies of electrons
at rest outside the surfaces of the two solids are no longer the same, and
there is an electric field in the gap directed from right to left. There must
be a negative charge on the surface of the metal balanced by a positive
charge in the semiconductor. The charge on the surface of the metal
consists simply of extra conduction electrons contained within the
Thomas-Fermi screening distance (~ 0.5 A ) . Since the semiconductor is
n-type, the positive charge will be provided by conduction electrons
receding from the surface, leaving uncompensated positive donor ions in
a region depleted of electrons. Because the donor concentration is many
orders of magnitude less than the concentration of electrons in the metal,
the uncompensated donors occupy a layer of appreciable thickness w,
comparable to the width of the depletion region in a p - n junction, and
the bands in the semiconductor are bent upwards as shown in Fig. 1.7(b).
The difference Fj between the electrostatic potentials outside the
surfaces of the metal and semiconductor is given by Fj = dS"^, where 6 is
their separation and S'-, the field in the gap. If the metal and semi­
conductor approach each other, I^ must tend to zero if is to remain

11
 IV
•Ec
oooo

Metal
(a)

(c) (d)
Fig. 1.7 Formation of a barrier between a metal and a semiconductor (a) neutral and
isolated, (b) electrically connected, (c) separated by a narrow gap, (d) in perfect contact. O
denotes electron in conduction band; + denotes donor ion.

finite [Fig. 1.7(c)] and, when they finally touch [Fig. 1.7(d)], the barrier
due to the vacuum disappears altogether and we are left with an ideal
metal-semiconductor contact. It is clear from the fact that tends to
zero that the height of the barrier measured relative to the Fermi level
is given by
(1.2)
In most practical metal-semiconductor contacts, the ideal situation
shown in Fig. 1.7(d) is never reached because there is usually a thin
insulating layer of oxide, about 10-20 A thick, on the surface of the
semiconductor. Such an insulating film is often referred to as an
interfacial layer. A practical contact is therefore more like that shown in
Fig. 1.7(c); however, the barrier presented to electrons by the oxide layer
is usually so narrow that electrons can tunnel through it quite easily, and
Fig. 1.7(c) is almost indistinguishable from Fig. 1.7(d) as far as the

12
 conduction electrons are concerned. Moreover, the potential drop Fj in
the oxide film is so small that eqn (1.2) is still a very good approximation.
Although it is usually attributed to Schottky, eqn (1.2) was first stated
implicitly by Mott (1938) and will be referred to as the Schottky-Mott
limit. In obtaining it a number of important assumptions have been
made, namely that the surface dipole contributions to and Xs do not
change when the metal and semiconductor are brought into contact (or,
at least, that the difference between them does not change), that there are
no localized states on the surface of the semiconductor, and that there is
perfect contact between the semiconductor and the metal, i.e. no
intervening layer. The fact that eqn (1.2) is not obeyed in practice shows
that one or more of these assumptions are not valid, and we shall discuss
them later. It should also be emphasized that, contrary to what is
sometimes stated, the Schottky-Mott theory does imply the existence of
a dipole layer at the interface.
The shape of the potential barrier depends on the charge distribution
within the depletion region. If the bottom of the conduction band is
raised by about 2kT/q above its position in the bulk, the electron density
is reduced by an order of magnitude, and between this plane and the
metal-semiconductor interface the space charge is due entirely to the
uncompensated donors. If we neglect the bending of the bands in the
transition region where the electron density is less than the donor
concentration but has not fallen by an order of magnitude (the so-called
depletion approximation), the shape of the barrier will be determined
entirely by the spatial distribution of the donors.
In the model first put forward by Schottky (1938) and elaborated by
Schottky and Spenke (1939), the semiconductor is assumed to be
homogeneous right up to the boundary with the metal, so that the
uncompensated donors give rise to a uniform space charge in the
depletion region. The electric-field strength therefore increases linearly
with distance from the edge of the depletion region in accordance with
Gauss's theorem, and the electrostatic potential increases quadratically
[see Fig. 1.8(a)]. The resulting paraboUc barrier is known as a Schottky
barrier.
A somewhat different model was put forward by Mott (1938), who
assumed that the semiconductor had a thin layer devoid of donors
immediately next to the metal. The electric-field strength would be
constant throughout such a layer giving rise to an electrostatic potential
which increases quadratically at first and then linearly, as shown in Fig.
1.8(b). Such a barrier is known as a Mott barrier. It is rarely encountered
in practice.
The foregoing description applies to an n-type semiconductor with
work function less than the work function (j>^ of the metal. It will be
seen later that such a contact behaves as a rectifier. If a similar argument

13
 (a) (b)
Fig. 1.8 (a) Schottky barrier, (b) Mott barrier, (i) charge density, (ii) electric-field
strength (iii) electrostatic potential (V). For an n-type semiconductor, ^ is negative and the
electron potential energy (- qtp) positive.

is developed for the case when is greater than one obtains a band
diagram of the form shown in Fig. 1.9(b). Clearly, if such a contact is
biased so that electrons flow from the semiconductors to the metal, they
encounter no barrier. If it is biased so that electrons flow in the reverse
direction, the comparatively high concentration of electrons in the region
where the semiconductor bands are bent downwards (usually referred to
as an accumulation region) behaves like a cathode which is easily capable
of providing a copious supply of electrons. The current is then
determined by the bulk resistance of the semiconductor. Such a contact
is termed an ohmic contact. This type of contact has a sufficiently low
resistance for the current to be determined by the resistance of the bulk
semiconductor rather than by the properties of the contact.
In a p-type semiconductor for which exceeds <^^, we obtain the band
diagram shown in Fig. 1.9(c), which also represents an ohmic contact.
The case of a p-type semiconductor for which exceeds ^„ is shown in
Fig. 1.9(d). Bearing in mind that holes have difficulty in going underneath
a barrier, one sees that Fig. 1.9(d) is the p-type analogue of Fig. 1.9(a)
and gives rise to rectification.
Figure 1.9(b) and (c) are very uncommon in practice, and the majority
of metal-semiconductor combinations form rectifying or 'blocking'

14
 (a) (b)

(c) (d)

Fig. 1.9 Barriers for semiconductors of different types and work functions, n-type: (a)
K > (*s (rectifying); (b) (/S„ < (ohmic). p-type: (c) > (ohmic); (d) ^„ < <j>^ (rectifying).

contacts. Unless the contrary is clearly stated, all subsequent discussions

will centre round the case of n-type semiconductors with > ^„ which is
the most important case in practice.

1.3.2 The effect of surface (or interface) states

Even if the assumption about the constancy of the surface dipole layer is
incorrect, the barrier height should still depend on the metal work
function (j)^ if the simple Schottky-Mott theory [eqn (1.2)] is valid. But
experimentally it is found that the barrier iieight is a less sensitive
function of ^„ than eqn (1.2) would suggest and that, under certain
circumstances, ^t, may be almost independent of the choice of metal.
A n explanation of this weak dependence on was put forward by
Bardeen (1947), who suggested that the discrepancy may be due to the
effect of surface states. Suppose that the metal and semiconductor
remain separated by a thin insulating layer as shown in Fig. 1.10 and that
there is a continuous distribution of surface states present at the interface
between the semiconductor and the insulator, characterized by a neutral
level (j>Q. Although the term 'surface states' is generally used to describe

15
 Insulating film

Semiconductor

Fig. 1.10 Metal-semiconductor contact with surface states.

these states it would perhaps be more appropriate to refer to them as

'interface states', in order to distinguish them from the states which are
present on the free semiconductor surface. Throughout the remainder of
the book we shall use the terms 'surface states' and 'interface states'
somewhat indiscriminately, but the precise meaning should be clear from
the context. In the absence of surface states, the negative charge on
the surface of the metal must be equal and opposite to the positive charge
i2d due to the uncompensated donors because the junction as a whole is
electrically neutral. In the presence of surface states, the neutraUty
condition becomes Qm + Qd + Gss - 0' where <2ss is the charge in the
surface states. The occupancy of the surface states is determined by the
Fermi level, which is constant throughout the barrier region in the
absence of applied bias, and for most purposes it is good enough to use
the 'absolute-zero' approximation in which the states are supposed to be
filled up to the Fermi level and empty above it. If the neutral level
happens to be above the Fermi level Ep as shown in Fig. 1.10, the surface
states contain a net positive charge and <2d must therefore be smaller
than if surface states were absent. This means that the width w of the
depletion region will be correspondingly reduced, and the amount of
band bending [proportional to according to eqn (A.l)] will also be
decreased. The barrier height is equal to the diffusion voltage or band
bending Fjo plus i [see Fig. 1.7(d)] so will be reduced. The reduction
in has the effect of pushing towards Ep, i.e. it tends to reduce the
positive charge in the surface states. On the other hand, if happens to
be below Ep, is negative and must be greater than if surface states
were absent. This means that w and will both be increased and will
be pulled up towards Ep.

16
 The surface states therefore behave like a negative-feedback loop, the
error signal of which is the deviation of fro"! Ep. The 'gain' in this
feedback loop is proportional to the density of surface states per unit of
energy, since this is what determines Q^^ for a given departure of from
Ep. If the density of surface states becomes very large, the error signal
will be very small and ^g " ^f- It is usual in the literature to measure <J>Q
from the top of the valence band, in which case the barrier height will be
given by

This will be called the Bardeen limit. The barrier height is said to be
'pinned' by the high density of surface states.
A n alternative way of looking at the effect of surface states is to regard
them as screening the semiconductor from the electric field in the
insulating layer, so that the amount of charge in the depletion region (and
therefore the barrier height) is independent of the work function of the
metal.

1.4 Generalized analysis of the Bardeen model

The model described in Section 1.3.2, in which it is supposed that the

semiconductor and metal are separated by a thin insulating layer and that
there are localized states at the insulator-semiconductor interface, has
the merits of being easy to analyse and of corresponding to the practical
case of a semiconductor covered with a thin oxide film. It is over­
simplified because it ignores any possible modification of the surface-
dipole contributions to the work functions of the metal and semi­
conductor when they come into contact with the insulating layer, and
also assumes that the interface states can be represented by point charges
whereas, in practice, they are smeared out over a distane of 5-10 A. It
has, however, been widely used to interpret experimental studies of
Schottky barriers and will for this reason be analysed in some detail.
Furthermore, the mechanism whereby interface states can affect the
barrier height as incorporated in the Bardeen theory is fundamentally the
same as that implicit in other, more comphcated, models of interface
states. The Bardeen model therefore gives a good insight into the general
features of interface state behaviour.

1.4.1. The flat-band barrier height with a continuous distribution

of surface states
I
A band diagram of a metal-semiconductor contact as postulated by
Bardeen is shown in Fig. 1.11(a). The 'vacuum level' for the insulating
17
 Conduction Vacuum
band of level
insulator

(b)

Fig. 1.11 Metal-semiconductor contact with insulating interfacial layer (a) with arbitrary
bias and (b) flat-band case.

interfacial layer may be visualized by supposing that the contact is

bounded by a flat surface perpendicular to the plane of the junction (i.e.
parallel to the plane of the figure). The vacuum level is then the energy of
an electron at rest outside this surface. The insulator is represented as
having well-defined conduction and valence bands although, for a film of
about 10 A thick, the band structure may be somewhat different from
that of the bulk. Figure 1.11(a) shows the situation when a forward bias V
is applied between semiconductor and metal. A current will, of course,
flow because of the forward bias, but this will not affect the charges or
energy relationships within the system if one assumes the vahdity of the
depletion approximation. The barrier height ^ i , is defined as the
difference in energy between the bottom of the conduction band at the
boundary of the semiconductor and the Fermi level in the metal. This is
because, if the interfacial layer is very thin, electrons can readily tuimel
through it, so the barrier to electron flow is determined by the maximum
height of the bottom of the conduction band in the semiconductor.
There are three distinct sources of charge in the system. The first
resides on the surface of the metal, the second is due to the
uncompensated donor ions in the depletion region, and the third Q^^ is
due to electrons in the surface (or interface) states. The term 'interface
states' tends to be used when the states are at the interface between two
sohds, in contrast with surface states at the free surface of a solid. is
determined by the electric-field strength in the insulating layer, by the
width of the depletion region and the density of donors, while Q^^ is
determined by the density of interface states and by their occupation

18
probability. Because there is no electric field in either the metal or the
semiconductor at some distance from the junction,! electrical neutraUty
requires that Qm + Qd + Qss = 0. and in addition the various field
strengths and charge densities are related by Gauss's theorem. A
complete analysis of the problem involves a lot of tedious algebra, but it
can be simplified by supposing that sufficient bias is apphed to the
system to make the depletion region disappear as in Fig. 1.11(b). This is
known as the 'flat-band' situation, and the simpHfication arises because
now there is no electric field in the semiconductor and Qa = 0.
Alternatively, the flat-band situation may be visualized by imagining that
no bias is applied but the bulk doping is adjusted so that the distance of
the Fermi level below the conduction band in the bulk is the same as at
the surface.
If the insulating layer is very thin, say less than 10 A as in a good
Schottky diode, the interface states are closely coupled by tunnelling to
the conduction-band states in the metal and their population is
determined by the metal Fermi level. It is a good approximation to use
the zero-temperature Umit for the Fermi-Dirac distribution, i.e. to
suppose that the states are filled up to Ef and empty above it. The
validity of this approximation has been discussed by Crowell and
Roberts (1969). Their conclusion is that the approximation is a good one
provided the density of states does not change very much over an mterval
of kT/q. Hence, by definition of the neutral level,
Q:. = qD,(^l + <^,-E^) (1.3)
where is the 'flat-band barrier height', is measured from the top of
the valence band, and is the density of interface states per unit area
per eV (assumed constant). Primed quantities refer to the flat-band
situation, and all charges refer to unit area. There is no electric field in
the semiconductor or in the metal, and the field strength in the insulator
is related to the charges on the surface of the metal and in the interface
states by Gauss's theorem, so that

where £i ( = Cireg) is the permittivity of the interfacial layer, and the drop
in potential across the layer is given by t

t There must, of course, be a weak electric field in the bulk of the semiconductor if a
current is flowing, but under forward bias this will be in the opposite direction to the field
in the depletion region. Between these two regions there must be a plane where the electric
field vanishes, and we may consider this plane as defining the edge of the depletion regioii.
The field in the metal is always negligible.
fAll energies are measured in electron-volts, so the drop in electrostatic potential is
numerically equal to the fall in the vacuum level across the interfacial layer.

19
The positive direction of <^,' is taken to be from right to left in Fig. 1.11. If
we assume that the dipole contributions to and Xs are unchanged by
the presence of the insulating film, we have from Fig. 1.11(b) that

't>^= v, +X. +i>l

so that

or, making use of eqn (1.3),

<t>><l>m-X.-^{<t>l + <l>o- E,) (1.4a)

I.e.
>i:-y(f<.-z.)+(i-)')(£,-« (1.5)
where

Equation (1.5) was first derived by Cowley and Sze (1965) as an

approximation to the zero-bias case, but they seem to have been unaware
ttiat the approximation they made in obtaining it was equivalent to
neglecting Q j . It may be observed that tends to the Schottky-Mott
limit ^„ - %5 as Z)s — 0 and to the Bardeen limit Eg - as A -* °°. Taking
20 A as an upper hmit for 5 in a reasonably good Schottky diode, and
assuming Ej to be 3 x 10~" F m " ' (corresponding to SiOz), we find that
must approach 10''' eV~" m"^ to make y significantly less than unity.
It should be emphasized that eqn (1.5) was obtained by assuming that,
under the apphcation of bias, the occupation of the interface states is
determined by the Fermi level in the metal, which is only true for very
thin interfacial layers. However, if the flat-band situation is visualized as
arising because the bulk doping level is adjusted so that E^ — Ep = (j>l,
no apphcation of bias is necessary and there is a single Fermi level
throughout, in which case the problem does not arise.

1.4.2 The flat-band barrier height with two bands of surface states

The surface states considered in Section 1.4.1 were assumed to have a

uniform distribution throughout the energy gap. Recent experimental
results for metals on compound semiconductors indicate that it is also
necessary to consider the situation where the surface states are discrete
rather than continuous. Consider, therefore, the situation where a narrow

20
 a4

a4

Fig. 1.12 Donor and acceptor surface states in the band gap of a semiconductor.

band of donor states lies below, and separated from, a narrow band of
acceptor states, as shown in Fig. 1.12. This is a situation which often
occurs in practice. Assume also that the acceptor band has an energy
width of with the bottom of the band at an energy above E^.
Likewise the upper edge of the donor band is at an energy E^ above E^
and its width is A^. The charge in these bands is determined by the Fermi
level position, and if Ep lies between E^ + E^ and Ey + E^ the donor
band is occupied and the acceptor band empty in the zero temperature
hmit, so that the net interface charge Q^^ = 0.
Consider again the situation where a bias is applied to generate the flat
bands shown in Fig. 1.11(b). If there is no net charge hi the surface states,
S'l = 0 so that

r.= '^m-Xs (1-7)

and
Ep-Ey = E^-<^°, = E^ + Xs-^r„.
Therefore, provided E^< E^ + Xs~ ^m"^^a> the Fermi level will he
between Ey + Eg and Ey + E^ and eqn (1.7) will apply. This condition
may be rewritten as
E^ + Xs-E.<<^m<E, + X.-E, (1.8)
and in this region (denoted by C D in Fig. 1.13) the behaviour
corresponds to the Mott-Schottky limit.
Now suppose i^m^ Eg+ Xs~ Eg, so that electrons can be transferred
from the donor states to the metal. The donor states are no longer
neutral and £ ^ lies at an energy E^ - above E^, or E^ - E^ + below
Eg. Suppose that within the donor band there is a constant density of
states Z),j so that

21
 F

E
Eg-£d+Ad
D ^^^^^^
Eg-Ed
Slope Vd

E,-E,-A, B_-Tr— ^
/1 Slope r, 1

a / I 1
t ^n.^ 1 1
+ + + +
I I I
I +
>
2«
Fig. 1.13 Plot of Schottky barrier height against metal work function for donor and
acceptor surface states. AB: donors and acceptors both full; BC: donors full, acceptors
partly full; CD: donors full, acceptors empty; DE: donors partly full, acceptors empty; EF:
donors and acceptors both empty.

A s before, Qss= ^1^.'= ^KV-J^, where V'= <l>^- Xri>l and the primes
denote the flat-band situation. Hence

so that

yd(?i.- X.)+ (1 - y d ) ( ^ g - ^d) (1-9)

where

This equation may be compared with eqn (1.5), and it may also be seen
that eqn (1.7) is recovered for the case where D^g = 0. Again D^^ must
approach 1 0 " eV m~^ to make Yd significantly less than unity. Equation
(1.9) remains valid as increases until Ej, coincides with Ey+ Eg — A^.
Hence, in the plot of ^ against ^„ shown in Fig. 1.13, the region D E with
slope yg will continue until
<l>l=E^-{Eg-Ag) (1.11)

22
or
(1.12)
Yd
At this point (E) the donor states are all empty and the surface state
charge is given by Q^^ = qD^^Ag. Beyond this point the interface state
charge remains constant and
o_ , ^ gAdAd6

so that will again increase with a slope of unity, as shown by the region
E F in Fig. 1.13.
For the case where ^„ is small, charge may be transferred from the
metal into the acceptor states. In this case Q^^ and S'-, are both negative,
and by analogy with the above argument it may be shown that for

E, + Xs-E,>^^>E, + Xs-E,-^

the versus relationship will have a slope of

where is the density of acceptor states (region BC). For values of ^„

less than E^ + x^ - E^ - (A^/y^), Q^^ is constant and

as represented by A B . Hence we see that there are two pinning positions,

one at the centre of the acceptor band and one at the centre of the donor
band. For positions of the Fermi level outside either band there is no
pinning and the slope of the graph of ^g versus graph should be unity,
in accordance with the Schottky-Mott theory.

1.4.3 The field dependence of the barrier height

If there is no interfacial layer, the barrier height is independent of any

electric field which may exist inside the semiconductor. However, when
there is an interfacial layer present, an electric field in the semiconductor
changes the potential V, across the layer and so modifies the barrier
height. A n electric field normally exists within a Schottky barrier, and it
is important to know how this affects the height of the barrier.
If we revert to Fig. 1.11(a) where the bands are bent, there is now an

23
electric field in the semiconductor due to the uncompensated donors,
all the charges are modified, and the electric field in the insulator is
changed.
If we assume a continuous distribution of interface states with a
constant density D^, the charge in the interface states is given by

= Q;,+ 9Z),(^b-<^g), (1.15)

and Gauss's theorem now tells us that

where fs (= E^^EQ) is the permittivity of the semiconductor and S"^^^ is the

value of at the top of the barrier.
As before, K = dS", and = K + ;f s + <!^b> so that

i ^ = 5 ; l „ , - X s - - ( a s + fs^.ax)- (1-16)

Combining eqn (1.16) with eqns (1.4) and (1.15) gives

9b- 9b {9b~ 9b) ©max

or
(1.17)
where

a = (1.18)

In other words, the barrier height is reduced from the flat-band value by
an amount which is proportional to the maximum electric field in the
semiconductor. This result is true for any value of apphed bias, provided
that the density of interface states remains constant over the range of
energy involved and as long as their occupation is determined by Ef.
This can only be expected to be true if the reduction in (^^, is less than
about 0.1 eV.
There are other possible causes of field dependence of the barrier
height besides the presence of an interfacial layer. Apart from the image-
force lowering of the barrier to be discussed in Section 1.6, there is also a
possible effect due to the penetration of the wave functions of the metal
electrons into the semiconductor, which is discussed in Section 1.5. The
field dependence of the barrier height is very important in connection

24
with the reverse current/voltage characteristics, as will be discussed in
Section 3.6.1.

1.4.4 p-type semiconductors

The case of a p-type semiconductor for which (f)^ exceeds ^„ is analogous

to the n-type case discussed in Section 1.4.1. Reference to Fig. 1.14 shows
that the flat-band barrier height is given by

where

E\

The positive direction of is taken to be from right to left as before.

Equation (1.4a) is now replaced by

Kv = Eg + - <l>m - {'t>hv - M
SO that

rbp=Y(E,+ Xs-K) + a - Y ) K (1.19)

where Y is given by eqn (1.6). Combining eqns (1.19) and (1.5) gives
^L+K-E^ (1.20)
where is the flat-band barrier height between the same metal and the
same semiconductor doped to make it n-type. [Equation (1.5) refers to
the n-type case and so gives Equation (1.20) is only vaHd if <5, Cj,
and A are the same in each case, so that y is unaltered. This is a

Eg

I • <t>a
i
6

Fig. 1.14 p-type semiconductor contact with interfacial layer (flat-band case).

25
reasonable assumption if the semiconductor surface is prepared in the
same way in both cases, since the surface properties are not influenced
by the bulk impurities.
As with eqn (1.17), the field dependence of the barrier can be shown to
be given by
(l>l,= r,,+a^^,, (1.21)
where a is defined by eqn (1.18). Since the positive direction of ^ is
taken to be from right to left, S"^^^ is negative for a Schottky barrier on
p-type material. It is therefore more appropriate to write eqn (1.21) in the
form

i>,, = i>l,-a\^^J (1.21a)

which is of the same form as eqn (1.17).

1.4.5 The bias dependence of the barrier height

Equation (1.17) is not very useful as it stands because the electric field in
the barrier is not usually known explicitly. For most purposes it is more
desirable to know how the barrier height varies with bias voltage and
doping level, so that we need to express S'^^^^ in terms of these
parameters. This is easily done if we adopt the depletion approximation,
according to which the electron density in the conduction band falls
abruptly from its bulk value to a value which is negUgible compared with
the donor density A^^. This approximation is equivalent to supposing that
the charge density rises abruptly from zero to the value qN^ at the edge
of the depletion region. If A^^ is constant, the electric-field strength will
increase linearly with distance from the edge of the depletion region in
accordance with Gauss's theorem as in Fig. 1.8(a). The field strength at
the surface will be given by .^n,ax= q^iWk^, where w is the width of the
depletion region, and the average field strength will be i ^ „ a x - The
difference in potential across the depletion region (i.e. the diffusion
potential) will be given by

2qNg
so that
^r^.= (2qN,V,/E,y^\ (1.22)
If the effect of the transition region in which the electron density falls
gradually from its bulk value to a value negligible compared with
(sometimes incorrectly referred to as the 'reserve' layer) is properly taken
into account (see Appendix B), eqn (1.22) must be modified to

26
 1/2
1/2 kT
^ , 3 x = (2qN,/e,) (1.23)

A n expression for the barrier height in terms of the diffusion potential

can be obtained by combining eqn (1.23) with eqns (1.5) and (1.17) to
give
1/2
kT
<t>l-a(2qNJe>,'^' (1.24a)
9J
1/2
1/2 kT
V,- . (1.24b)
? J
Finally, an explicit expression for ^ as a function of the bias voltage V
can be obtained by inserting Vg = ^ - V- ^ into eqn (1.24a) and solving
the resulting quadratic to give
1/2
kT
(1.25)

where ^1 = la^qNg/e^. The negative sign must be taken in front of the

radical because ^ is obviously less than ^g. The zero-bias barrier height
^0 is obtained by putting F = 0 in eqn (1.25).
Equation (1.24) shows that ^ should decrease with increasing
because of the increasing field in the barrier. For an interfacial layer not
more than 20 A thick, a is essentially equal to dejunless exceeds
about 10'^ eV~' m~^, so that a would not exceed about 60 A if £j« e / 3 .
For Vg ~ 0.5 V and moderate doping (Ng < 10^^ m"^), the zero-bias
lowering of the barrier due to the electric field should not exceed about
0.02 V. The effect would be exaggerated by heavier doping or a thicker
interfacial layer, but would be reduced by a high surface-state density. It
may be much more pronounced under reverse bias because of the
increase in F j .

1.4.6 The penetration of the field into the metal

Various authors (Crowell, Shore, and L a Bate, 1965; Perlman, 1969;

Kumar, 1970; Duke and Mailhiot, 1985) have taken into account the fact
that the charge on the surface of the metal is only confined within the
Thomas-Fermi screening distance r,, of the surface (see, for example,
Mott and Jones, 1936). A s a result, the electric field penetrates slightly
into the metal and there is a difference in potential between the surface
and the interior of the metal of magnitude (2m''o/£o' where is the
charge on the metal. In the case of an ideal metal-semiconductor contact
with no interfacial layer and no interface states, it is easy to show that the

27
effect of this is to reduce the height of the barrier by a fraction IE^TQ/EQW,

where w is the width of the depletion layer. For most metals, is about
0.5 A and w is usually at least 1000 A unless the donor density is very
high, so for an ideal contact the effect of penetration of the field into the
metal is in most cases negligible.
If there is an insulating layer between metal and semiconductor, the
effect of field penetration, although still small, may be comparable with
the drop in potential across the layer and may increase the field
dependence of the barrier height. Eimers and Stevens (1971) have shown
that the effect is simply to replace dhy 6 + {r^EjE^) in the expression for
Y [eqn (1.6)] and a [eqn (1.18)]. For chemically poUshed semiconductor
surfaces, 6 will normally be at least 20 times r^, but the factor EJEQ may
be approximately equal to four. The effect of field penetration is
comparable with the effect of the uncertainty in d and EJ.
Taking the field penetration into the metal into account is equivalent to
making a partial calculation of the modification in the surface-dipole
contribution to the metal's work function when it is brought into contact
with the semiconductor. There are other contributions which have so far
been ignored (see Section 1.5). There is little point in considering just
one of these contributions in isolation, particularly when its effect is so
small, so the effect of field penetration into the metal will henceforth be
disregarded. One must also be wary about using macroscopic concepts
such as dielectric constant on an atomic scale. The dielectric constant of
a material determines the average electric field over a distance of several
lattice spacings, and one should not suppose, as is usually done, that the
permittivity changes abruptly from EQ to E^ at the metal-semiconductor
boundary.

1.5 Intimate contacts

1.5.1 General

The situation considered in Section 1.4 refers to interfaces where the

metal and semiconductor are separated by an insulating layer. The
insulating layer is thin, enough for electrons to tunnel through it quite
easily, but is thick enough to ensure that the interface states are, to a
good approximation, a property of the insulator-semiconductor
interface. The semiconductor surface is thus decoupled from the metal.
Oxide layers, for example SiOj on silicon, often perform this function.
Metal contacts may also be deposited on atomically clean semi­
conductor surfaces by making use of ultra-high vacuum systems which
are now routinely available. Surfaces may be cleaned by heating the
semiconductor, by sputtering followed by annealing, or a fresh surface

28
may be prepared in situ by molecular beam epitaxial growth. Clean semi­
conductor surfaces are also often prepared by cleaving crystals in ultra­
high vacuum. The disadvantage of the latter method is that it is restricted
to a few crystal planes, the (111) planes for silicon and germanium, and
the (110) planes for most of the III-V semiconductors. Contacts
deposited on atomically clean semiconductor surfaces in an ultra-high
vacuum system are free from an insulating interfacial layer and are
usually referred to as intimate contacts. Intimate interfaces are more
complex than those considered in Section 1.4 and the metal-
semiconductor system ideally needs to be dealt with as a single entity.
The electronic structure at the interface may be influenced by a range of
important factors.

1. Even if the interface is atomically abrupt and perfectly ordered, it is

most likely that the surface structures associated with the clean
semiconductor and clean metal will change when the interface is
formed. A knowledge of this structure is required in order to carry out
a reahstic calculation of the interface electronic structure. The surface
dipole contributions to the work function of the metal and to the
electron affinity of the semiconductor change in an unknown way
when an interface is formed.
2. The deposited metal film may be epitaxial and ordered and the lattice
constant may be well matched to that of the semiconductor. It is even
possible to grow ordered epitaxial layers when the lattice constant of
, the metal does not match that of the semiconductor; in this case the
metal film will be strained or there will be many mis-fit dislocations. In
most cases, though, films deposited on semiconductors are poly-
crystalline, with the precise form of the polycrystalhne layer
depending on the details of the deposition conditions. The electronic
structure of the interface may be dependent on the structural nature of
the metal overlayer.
3. Many intimate interfaces between metals and semiconductors are not
atomically abrupt, even when the semiconductor surface is maintained
at room temperature during deposition. Atoms from the metal may
diffuse into the semiconductor and adopt interstitial or substitutional
positions. This may have the effect of changing the semiconductor
doping near the interface. Likewise, atoms from the semiconductor
may diffuse into the metal contact.
4. Many metals, when deposited on clean semiconductor surfaces, lead
to strong chemical reactions, resulting in a layer of a compound at the
interface. The nature of this reacted layer depends on many factors. It
depends on thermodynamic considerations, i.e. on whether the metal
and semiconductor can react, and it also depends on kinetic

29
 considerations and the precise conditions under which the interface
was formed. The surface stoichiometry of the clean semiconductor is
known to influence these reactions. The electronic properties of the
interface will be sensitive to the nature and order of this layer.
5. It is possible for the wave functions of those electrons in the metal with
energies corresponding to the forbidden gap in the semiconductor to
penetrate into the semiconductor in the form of exponentially damped
evanescent waves as pointed out by Heine (1965, 1972). This
represents a transfer of charge from the metal to the semiconductor
and the states are often referred to as metal-induced gap states, or
MIGS. These are considered further in Section 1.5.2.
It is difficult to establish the precise crytallographic structure, degree of
order, and abruptness of interfaces when the film deposited on the semi­
conductor is thick. Using modern surface science techniques, such as
low-energy electron diffraction, it is possible to probe the structure of
thin films of monolayer dimensions, and electron spectroscopic
techniques such as Auger and photoelectron spectroscopy enable
chemical reactions and interdiffusion to be investigated when the metal
contact is only a few Angstroms in thickness. Angle-resolved photo-
emission even allows the occupied interface states to be studied, but
again only for very thin metal overlayers. There is no certainty that the
interface remams the same as the film thickness increases so that,
although surface science techniques are extremely valuable, the results
they provide do not necessarily describe interfaces between semi­
conductors and thick metal contacts particularly well.

1.5.2 Metal-induced gap states

Heine, in 1965, pointed out that any intrinsic electron states which may
be present on a free semiconductor surface will be replaced by metal-
induced gap states (MIGS) when a metal is deposited on that surface.
These M I G S are associated with the tails of the conduction electron
wave functions in the metal which tunnel into the band gap of the semi­
conductor at the interface, with an attenuation length of the order of a
few Angstroms. Heine discussed the effect of these tails on the Schottky
barrier heights in a qualitative fashion. Cohen and his co-workers (Louie
and Cohen, 1976; Louie, Chehkowsky, and Cohen, 1977; Ihm, Louie,
and Cohen, 1978) have carried out detailed theoretical studies of M I G S
at interfaces between aluminium and the semiconductors Si, Ge, GaAs,
ZnSe, and ZnS. These semiconductors range from highly covalent (Si) to
highly ionic (ZnS). In these calculations the metal contact was simulated
by a 'jellium' model, i.e. it was assumed that the positive ion cores in the
metal can be approximated by a uniform positive background, with the

30
density of aluminium. They then calculated the densities of states in the
metal, at the interface, and in the semiconductor. These are shown in Fig.
1.15 for the case of aluminium on the silicon (111) surface. The diagram
reveals the density of electron states as a function of energy for three
atomic layers of the semiconductor (regions I V - V I ) and for the
equivalent of three atomic layers in the metal (regions I-III), the interface
being located at the boundary between regions III and IV. Well into the
semiconductor, the density of states reflects that of bulk silicon and the
band gap is well defined. WeU into the metal, the density of states is free-
electron like, with states occupied to Ep. However, at the interface extra
states are seen in the band gap around Ep and these extend a few atomic
layers into the semiconductor. The presence of metal-induced gap states
(MIGS) is clearly visible in region IV. Interface states may also be seen

-14-12-10 -8 -6 -4 -2 0 2 4
Energy (eV)

Fig. 1.15 Local density of states on either side of the junction for A l on a Si(lll) surface.
(After Louie and Cohen, 1976. Copyright American Institute of Physics.)

31
lower in the valence band. It is possible to calculate the charge distribu­
tion in the energy range 0 < £ ' < E g , and this is shown in Fig. 1.16,
averaged parallel to the interface, as a function of distance into the semi­
conductor, with z = 0 at the edge of the jellium core. It may be seen that
p(z) penetrates into the semiconductor for several atomic layers. The
penetration of charge into the semiconductor is also shown for
aluminium on GaAs, ZnSe, and ZnS; the larger the band gap of the semi­
conductor, the smaller the penetration depth, as would be expected.

1.0

0 5 10 15
Z (Atomic units)
Fig. 1.16 Ciiarge distribution of the penetrating tails of the metal-induced gap states in
the semiconductor band gap, as a function of distance from the interface. The unit of
distance is the Bohr radius (0.53 A). (After Louie, Chelikowsky, and Cohen, 1977.
Copyright American Institute of Physics.)

Cohen and co-workers, then considered the formation of Schottky

barriers at these interfaces by determining the position of the conduction
band minimum in the bulk relative to the Fermi level at the interface. The
values obtained are shown in the Table 1.2. The resuUs of Cohen et al.
also suggest that ^ is much less dependent on for the covalent
materials such as Si and GaAs than for the ionic ones hke ZnSe and ZnS.
It has been pointed out by Yndurain (1971) and by Tejedor, Flores,

Table 1.2 Schottky barrier heights evaluated

theoretically for Al on a number of n-type semi­
conductors by Louie, Chelikowsky, and Cohen
(1977)

Al-Si 0.6 + 0.1

Al-GaAs 0.8 + 0.2
Al-ZnSe 0.2 + 0.2
Al-ZnS 0.5 ±0.2

32
and Louis (1977) that the MIGS at the interface are derived from the
valence and conduction bands of the semiconductor, and one can
therefore use the concept of a neutral level as in the Bardeen model.
According to Tersoff (1984) this neutral level must fall at or near the
energy where the M I G S wave functions cross over from being largely
valence-band-derived to being conduction-band-derived. Tersoff further
argued that E^ should be near the centre of the semiconductor energy
gap, and that one should use the indirect gap rather than the direct gap.
The Fermi level should be pinned close to E^ as in the Bardeen theory,
and according to this simple theory the barrier heights should be
independent of the metal, though Tersoff (1985) has recently refined his
theory to accomodate a weak dependence on the metal.
The calculations of Louie et al. (1976, 1977) indicate that MIGS may
be present at metal-semiconductor interfaces in densities sufficient to
cause strong pinning of the Fermi level. Their existence in such high
densities has been called into question by Duke and Mailhiot (1985) and
experimental evidence for their existence is still lacking. In particular, the
theory does not take into account the atomic structure and bonding at
the interface.
It is worth remarking that the term 'metal-induced gap states' is slightly
unfortunate, and that 'metal-perturbed gap states' might be a more
appropriate description. If the metal were gradually removed from the
semiconductor, the MIGS would change smoothly into the intrinsic states
(if there are any) on the free surface of the semiconductor. The effect of
the metal is to perturb these intrinsic states by changing the matching
conditions at the surface.

1.5.3 Defects at interfaces

The termination of the bulk periodic potential of a sohd at its surface

leads to surface states whose wave-functions decay exponentially into the
bulk, as discussed in Section 1.2.2. These surface states lead to energy
bands which have a two-dimensional character in the plane of the
surface. However, for a surface which is imperfect, defects such as steps
or vacancies may lead to additional localized states. The energy levels
associated with these defects may lie in the band gap of the semi­
conductor, where they may act like surface states and lead to pinning of
the Fermi level.
Consider again the clean (110) surface of GaAs discussed in Section
1.2.2. On account of the surface relaxation there are no intrinsic surface
states in the fundamental band gap provided the cleaved surface is highly
perfect. However, the existence of a high step density resulting from a
poor cleavage leads to acceptor-hke states which have energies close to
the middle of the band gap (Huijser, van Laar, and van Rooy, 1977). For

33
n-type crystals, electrons from the conduction band may be trapped in
these acceptor defect states, leading to a depletion layer in the semi­
conductor. Even if these defects are located in the surface layer of the
semiconductor, the centre of charge associated with them will extend
some distance below the surface. There may also be other types of
defects, such as vacancies and atoms on incorrect sites (for example Ga
on A s sites in GaAs) which are close to the surface. When a metal makes
contact with such a semiconductor surface, these defects can act as
interface states as in the Bardeen model and lead to pinning of the Fermi
level.
By analogy with the generalized Bardeen model considered in Section
1.4 it may be shown that defect interface densities of the order of 10'^
eV~' m~^ are needed for strong pinning of the Fermi level to be
observed. Detailed models have been constructed by Zur, M c G i l l , and
Smith (1983), Duke and Mailhiot (1985), and Palau, Ismail, and Lassa-
batere (1985). The influence of discrete surface states on Fermi level
pinning has already been described in Section 1.4.2 and the same treat­
ment may be applied for defect levels at the interface.
It has been proposed that Schottky barrier heights at metal-
semiconductor interfaces are nearly always determined by defects at the
interface (Spicer et al. 1980). In the so-called 'unified defect model' it is
assumed that defects are generated near the semiconductor surface when
the metal contact is deposited on that surface. These defects in turn lead
to pianing of the Fermi level. Calculations of the energy levels associated
with anion vacancies (Daw and Smith, 1981) and anti-site defects such as
cations on anion sites (Allen and Dow, 1981) lead to a reasonable
description of Schottky barrier heights for gold on a number of III-V
semiconductors and alloys. Moreover, experiments have shown that
intimate metal-semiconductor interfaces are often not perfect, with
chemical reactions and interdiffusion of metal and semiconductor being
common occurences. However, highly perfect interfaces may be
prepared, for example by molecular beam epitaxy, and it is not clear
whether the model should apply to these cases. We return to consider
these aspects in more detail in Chapter 2.
It should be emphasized that for both metal-induced gap states and the
unified defect model the fundamental pinning mechanism is essentially
the same as that postulated by Bardeen, but with the layer of semi­
conductor between the surface of the metal and the centre of charge of
the interface states playing the role of the insulating layer in the Bardeen
model. This separation between the metal and the interface states is
essential if the interface states are to have any effect on the barrier
height; what is important is that they should be able to generate a dipole
layer. For this reason, the Bardeen model provides a useful qualitative
picture of the pinning action of both M I G S and Spicer-type defects.

34
1.6 Image-force lowering of the barrier
Before we compare the predictions of theory and experiment, we must
take into account the image force between an electron and the surface of
the metal. In doing so, we shall assume there is no interfacial layer
present.
When an electron approaches a metal, the requirement that the electric
field must be perpendicular to the surface enables the electric field to be
calculated as if there were a positive charge of magnitude q located at the
mirror-image of the electron with respect to the surface of the metal.
Therefore, when the electron is at a distance x from the surface of the
metal it experiences a force q^/AnE\(lxf = q^/16ne'^x^ attracting it
towards the surface of the metal. Because of this attractive force the
electron has a negative potential energy -qV^ relative to that of an
electron at infinity, where

q
2
\67ie',X

Following Sze, CroweU, and Kahng (1964), we have written e[ for the
permittivity of the semiconductor because the electron approaches the
metal with the thermal velocity (~ 10^ m s~'), and one might expect that
there is not enough time for the semiconductor to become fully polarized
by the electric field, so that e[ should be the high-frequency rather than
the static permittivity.
The image potential energy has to be added to the potential energy due
to the Schottky barrier, as shown in Fig. 1.17. Since the image potential is

• Image potential energy

• Schottky barrier

Fig. 1.17 Image-force lowering of barrier.

35
only important near to the surface, it is a very good approximation to
regard the field due to the Schottky barrier as constant with the value
S'^^^. The maximum potential energy occurs at a position where the
resultant electric field vanishes; i.e. where the field due to the image force
is equal and opposite to the field in the depletion region, or

- , , 2 ® max •
l6nE,X^

The maximum potential energy in the barrier is lowered by an amount

= A a x + ^ , = 2A:n,^max

as a result of the image force. Hence

Substituting for S'^^^ from eqn (1.23) and remembering that

VA = <k'~ V~ ^ , we obtain
kT
(1.26a)
^n\E[fE,

and
1/4 -1/4
q£s kT
(1.26b)
9 I
Sze, Crowell, and Kahng (1964) have shown from photoelectric
measurements of the barrier height of silicon Schottky diodes under
reverse bias that the experimental data can be well explained by taking E'^
equal to (12.0 ± 0.5)£o which is indistinguishable from the static value
£ 5 = 1 1 . 7 EQ. This is understandable because, for the diodes under
consideration, the maxiriium value of x^ (corresponding to zero bias)
was about 50 A , and the time taken for an electron to travel this distance
with the thermal velocity is about 5 x 10"'" s. The permittivity of silicon
remains constant up to a frequency (~ 3 x 10"* Hz) corresponding to the
band gap and, since the inverse of this is nearly an order of magnitude
shorter than the electron's transit time, the silicon should become fully
polarized. With a polar compound like gallium arsenide, however, there
is a small decrease in permittivity above the 'reststrahlung' frequency
(~ 7 X 10^2Hz), so that E', should be slightly less than E,. Rideout and
Crowell (1970) have used the values E,, = 12.5 and E',, = 11.0 for gallium
arsenide. Figure 1.18 shows values of the image-force lowering A^^,^ and
of the position of the maximum for various values of the diffusion

36
 \—I I i I I III 1 1 II II III 1—I I II I III r

Fig. 1.18 Image-force lowering (A^bi) and distance of maximum of barrier from metal
{x„) as a function of band bending for various combinations of and £5^.

potential Vg, donor density Ng, and dielectric constant e^^. For this
purpose has been taken equal to e^,.
The effect of the image force is that the barrier which an electron has
to surmount in passing from the metal into the semiconductor is lowered
by an amount zl^bi- The image-force lowering differs from the other
contributions to in that it arises from the field produced by the
particular electron under consideration and is absent if there is no
electron in the conduction band near the top of the barrier. On the other
hand, contributions to from the work-function difference, surface-
state charge, and so on, are present whether or not there is an electron
near the top of the barrier. We shall use to denote the barrier height
arising from the latter causes, and will denote the image-force lowering
exphcitly by the quantity A^^,-^. Measurements of the barrier height which
depend on the movement of conduction electrons from metal to
semiconductor or vice versa yield the quantity ^i,-At^^^, whereas
measurements which depend on the space charge in the depletion region
(e.g. capacitance measurements) give without the effect of the image
force.
Like electrons, holes are attracted to the metal by an image force.
However, one must remember that the energy of a hole is measured
downwards from the top of the valence band, so that the effect of the
image force is to bend the valence band upwards near to the surface of
the metal as shown in Fig. 1.19. There is therefore no maximum in the top

37
 Ec

Fig. 1.19 Image-force effects in conduction and valence bands.

of the valence band as there is in the conduction band, and the energy
gap is reduced close to the surface of the metal due to the effect of the
image force.
The theory presented above holds only for an ideal junction. If there is
an insulating layer between the metal and semiconductor, the potential
due to the image force is much more complicated. We discuss this briefly
in Section 3.8, where we consider the effect of an interfacial layer on the
transport properties.

1.7 Methods of measurement of barrier heights

1.7.1 From current/voltage characteristics

It will be shown in Chapter 3 that good Schottky diodes made from high-
mobility semiconductors possess J/V characteristics given by the
thermionic-emission theory, provided the forward bias is not too large.
According to this theory.
/ = /o{exp(?F/A:r) - 1 j (1.27)
where J is the current density (current/unit area) and
/o=.A**rexp{-g(^b-^<2ibi)MT}.
For convenience, we shall call ^ 5 - J^bi the effective barrier height (j)^.
A** is the Richardson constant modified to take into account the
effective mass of electrons in the semiconductor, quantum-mechanical
reflection of those electrons which are able to negotiate the barrier, and
phonon scattering of electrons between the top of the barrier (as
determined by the image force) and the surface of the metal. The factors
which determine A** are discussed in Section 3.2.7.
It will be seen in Chapter 3 that in practice diodes never satisfy the
ideal equation (1.27) exactly, but can be more closely described by the
modified equation

38
 7 = 7ocxp(qV/nkT)[l - e x p ( - qV/kT)} (1.28)
where n (which may depend on temperature) is approximately
independent of V and is greater than unity. There are many possible
reasons why n should exceed unity, the most common being a bias
dependence of ^ and -d^bi- For values of V greater than 3kT/q, eqn
(1.28) can be written in the simpler form
J=Jo exp{qV/nkT) (1.28a)
so that a plot of In / against V in the forward direction should give a
straight line except for the region where V< 3kT/q. In practice one
always measures the total current I rather than the current density / , and
the determination of J necessitates an accurate measurement of the area
S. The advantage of retaining the more exact form of eqn (1.28) is that a
plot of ln[//{ 1 - exp( - qV/kT)]] against V should give a straight line for
all values of V, not only for the region V< 3kT/q but for negative (i.e.
reverse) values as well. Figure 3.5 (p. 108) shows such a plot for a diode
consisting of an epitaxial film of aluminium on gallium arsenide prepared
by molecular beam epitaxy (Missous and Rhoderick, 1986). It is clear
that this diode is very close to ideal, the plot being linear over the whole
range from - 1 V to +0.5 V, with an n value of 1.01. From such a graph
it is possible to determine IQ with great precision. In general the diode
will be less ideal, and the plot will show noticeable deviations from
hnearity. The determination of barrier heights from I/V characteristics is
only reliable if one can be confident that the current is determined by the
thermionic emission theory. For this to be so, the forward portion of the
characteristic must be a good straight line over more than one decade,
with a low value of n, say n < 1.1. For large values of n, or non-linear
characteristics, the diode is far from ideal, probably due to the presence
of a thick interfacial layer, or to recombination in the depletion region,
and the current will not be determined by the thermionic emission
theory.
Provided the characteristic is sufficiently close to ideal to enable a
reliable value of IQ (and hence JQ) to be determined, the barrier height
can be deduced from the data in two ways:
1. If A** is known, the value of JQ immediately gives = ^ ~ A ^ - ^ . A**
is often not known with any great precision, but since an error of a
factor of 3inA** gives rise to an error of about kT/q in an impre­
cise value can usually be used unless a very accurate measurement of
is required. The value of the barrier height found by extrapolating a
logarithmic plot of In[//{1 - exp( - qV/ kT)}] to F = 0 is the value of
for zero bias [see eqn (3.12)], which we shall write as
^eo^ho - (^<*bi)o. where (A^^i)o is the image-force lowering given by
eqn (1.26a) with F = 0.

39
2. If A ** is not known, one may measure the I/V characteristic over a
range of temperatures and hence find as a function of T. A plot of
ln(jQ/T^) against T " ' should give a straight line of slope - q^^o/k and
intercept on the vertical axis equal to In A**. The barrier height is
generally a decreasing function of temperature because the expansion
of the lattice causes a change in the band-gap and other parameters
which determine ^bo- To a first approximation we may write
^eo(^) = ^eo(O) - bT, in which case the slope of the plot of ln(jQ/T^)
against r ~ ' is - q^eo(Oyk and the intercept is In A**+ (qb/k). This
method therefore gives the barrier height at absolute zero.

A problem arises if it is necessary to extract rehable values of when

the sample has a large series resistance. In this case the region over which
the plot of In / versus V is linear may be small, and accurate extrapola­
tion to zero voltage may be difficult. A n elegant way of circumventing
this problem for the case where thermionic emission dominates has been
presented by Norde (1979). In this method use is made of the function

(1.29)
SA**T^

Suppose the diode has a series resistance R. Then, assuming « = 1,

9 (V-IR)\ -1
/ = / n exp
kT

where / Q = SA**T'^exp(- q<j)JkT). We neglect the voltage dependence of

(j)^. For the situation where the voltage across the diode is greater than
3kT/q, the above equation may be combined with eqn (1.29) to give

F(V)=^,+ IR-^- (1-30)

For small values of V (but exceeding 3kT/q) the term IR is neghgible, so

that a plot of F[V) versus Vyields a straight line with a slope of - i and
the intercept for F = 0 yields though the linear portion is not usually
sufficiently extensive to allow to be determined accurately. O n the
other hand, for large values of V the current is determined almost
entirely by R so that / = V/R and
V
(1.31)
RSA**T'

This function, for large V, approaches a straight line of slope + i. Thus

the slope of the graph of F{V) versus V should change from - i to + T
and F ( F ) should pass through a minimum somewhere between the two

40
limits. Norde showed that the value of F(V) corresponding to this
minimum satisfied the relationship
V kT
<f>. = F(V^..) + -Y-— (1-32)

where Fj„jn is the vohage corresponding to the minimum and

V kT
f(K„i„) = ^ - - l n
z q SA**T\ •
Thus, using measured values of I^in and V^^„ it is a straightforward
matter to derive the value of
A further difficulty occurs if there is a component of current arising
from the recombination of electrons and holes in the depletion region.
As will be seen in Section 3.4, such a component can be approximately
represented by
J , - ' J,oCxp(qV/nkT)
with « « 2. Recombination current is most important at low forward
voltages and, if ignored, can lead to a false value of JQ. Several labora­
tories (e.g. McLean, Dharmadasa, and Williams, 1986) have developed
computer programs which use a curve-fitting technique to obtain values
of the parameters JQ, J^Q, and R which best represent the experimental
I/V characteristics, thus enabling reliable values of to be obtained.
McLean (1986) has discussed the effect of recombination current and of
a bias dependence of on the Norde method.

1.7.2 From photoelectric measurements

If radiation with quantum energy exceeding ~ ^^bi is incident on a

metal in contact with a semiconductor, electrons excited from the Fermi
level of the metal will have sufficient energy to cross into the semi­
conductor, and a photovoltage will be developed causing a current to
flow in an external circuit. According to Fowler (1931), the photocurrent
y per absorbed photon of energy hv is given by

71
(y/T')^B e - +
~6

where 5 is a constant and ju= h(v- v^ykT. The threshold energy hv^ is
equal to ^t, - J^bi- For ^ 5, y is proportional to [h{v — VQ)}^, S O a plot
of y'''^ against hv should give a straight line, the intercept of which on the
hv axis is equal to hv^. This again gives the barrier height reduced by the
image-force term A^^,,. De Sousa Pires and his co-workers have
considered the errors that can result if the linear fit to the data is not
41
made for sufficiently large values of n that //^/2 is much greater than
jrV6 (de Sousa Pires, 1978; de Sousa Pires, Donoval, and Tove, 1982,
1984). In addition, the quantum energy hv must be less than if band-
to-band excitation in the semiconductor is to be avoided, so the range of
hv over which a linear dependence of y'^^ on hv can be expected may be
quite small. If the ultimate accuracy is desired, a computer fit to the full
expression should be made. A detailed discussion of the photoelectric
method has been given by Anderson, CroweU, and Kao (1975).

1.7.3 From capacitance measurements

It will be seen in Chapter 4 that, subject to certain precautions being

taken in the measurement, and provided the diode is nearly ideal and the
semiconductor has a uniform donor concentration, the differential
capacitance C{ — AQ/AV) under reverse bias F, is given for non-
degenerate semiconductors by
-1/2
' kT

where S is the area of the contact and ^ is the difference in energy

between the Fermi level and the bottom of the conduction band in the
bulk semiconductor. Hence

9 /
If is independent of (i.e. if there is no appreciable interfacial layer),
a plot of against V, should give a straight line with an intercept - F,
on the horizontal axis equal to - ( ^ - ^ - kT/q). The barrier height is
then given by
kT
<h-V, + ^ + — . (1.33)

If, however, there is an appreciable interfacial layer, so that (f)^ depends

on V, it has been shown by Cowley (1966) that the intercept of the
against V, plot when inserted into eqn (1.33) yields the quantity
+ (^i/4), where (l>l is the flat-band barrier height and = 2a^qNg/e^
(see Sections 1.4.5 and 4.2). Since in all practical cases is less than
by about two orders of magnitude, the C/V method essentially gives the
flat-band barrier height. Furthermore, because the differential
capacitance is determined by the width of the depletion region which
depends only on the diffusion voltage and the donor density, the barrier
height given by eqn (1.33) does not include the image-force lowering
J^bi- A comprehensive discussion of the various sources of error in the
42
determination of barrier heights from CIV characteristics has been given
by Goodman (1963). In the case of low barrier heights (less than, say,
0.5 V), the parallel conductance of the diode may be so large that it is not
possible to make capacitance measurements unless the diode is cooled.
Computer programs for the numerical determination of barrier heights
from JIV and CIV characteristics and by the photoelectric method have
been pubhshed by Nguyen et al. (1975). The rehability of the three
methods can be good in the sense that measurements on a number of
diodes manufactured simultaneously by evaporation of the metal onto a
single semiconductor slice are generally in good agreement; for example.
Fig. 1.20 shows plots of C~^ against V, for gold-silicon diodes made by
simultaneous evaporation on to a single slice of silicon. However,
agreement between measurements on diodes made on separate shces at
different times is usually less good and reflects the reproducibility of the
fabrication procedure. Generally, for diodes which have been made on
separate occasions but using the same fabrication procedure, measure­
ments made by any one of the three methods can be reproducible to
about 2%. If there is a pronounced bias dependence of the barrier height,
the capacitance method is hable to give results which differ significantly
from those obtained by the JIV or photoelectric methods. This is the case
when there is a fairly thick interfacial film, as was pointed out by Cowley
(1966) and by Card and Rhoderick (1971a).

1.7.4 Fromphotoelectron emission spectroscopy

Photoelectron spectroscopy is one of the most important techniques

available to study sohd surfaces. Photoelectrons are emitted from the

1 1 1 p A, / / '
so­

fa 20-
o

1 A 1 1 1
-0.8 -0.4 0 0.4 0.8 1.2
Reverse bias V,(V)
Fig. 1.20 Plots of C"^ against V, for five gold-silicon Schottky diodes made simultane­
ously by evaporation onto a single slice of silicon (Turner and Rhoderick, 1968).

43
solid, as a result of light falling on it, via the classic 'photoelectric effect';
for this to happen the photon energy associated with the radiation must
exceed the work function of the solid. Values of metal work functions
and ionization energies of semiconductors have been obtained by
measuring the photon energy corresponding to the onset of
photoemission for a range of solids.
Present-day photoelectron spectrometers measure the energy
distribution of the emitted electrons, rather than the total photocurrent.
The spectrometer gives a measure of the number of electrons emitted at
each value of kinetic energy, from zero up to some maximum value which
is determined by the energy of the incident photons. In many
spectrometers the photons are provided by an X-ray source and the
method is then referred to as X-ray photoelectron spectroscopy, or X P S .
(It is also often referred to as ESCA—electron spectroscopy for chemical
analysis.) Other spectrometers make use of ultraviolet radiation to excite
the electrons; the method is then called ultraviolet photoelectron
spectroscopy, or UPS. During recent years extensive use has been made
of 'synchroton radiation' to provide photons with energies bridging the
gap between UPS and X P S . Photoelectron spectroscopy with synchro­
tron radiation is generally called 'soft XPS'. Discussions of various light
sources and electron spectrometers have been presented by Williams,
Srivastava, and McGovern (1980), and by Prutton (1983).
A typical 'photoelectron spectrum' and the way it originates is shown
schematically in Fig. 1.21. The incident radiation penetrates several
hundreds of Angstroms into the sohd and leads to the photoexcitation of
electrons from the valence band and from the core levels into empty
states. Many of these electrons move towards the surface and some of
them escape into the vacuum. However, electrons photoexcited at depths
greater than 20 or 30 A generally collide with other electrons before
reaching the surface; they lose energy in the process and arrive at the
surface without sufficient energy to escape. Electrons photoexcited
within 20 A or so of the surface can escape without loss of energy. The
broad high-energy group of electrons in Fig. 1.21 originates in the
valence band, and the sharper peak is composed of electrons
photoexcited from the core levels of atoms near the surface. The 'core
level' photoemission peaks are used to identify atoms near the surface of
solids, since each element has a set of well-defined and well-catalogued
core level binding energies.
Consider photoemission due to radiation of quantum energy hv from an
n-type semiconductor with band bending near the surface, due either to
surface states on a free surface or to a metal overlayer, as illustrated in Fig.
1.21. The photoelectrons which are emitted without loss of energy
originate within a distance which is small compared with the width of the

44
space charge region. It may be readily seen that the emitted electrons cover
a range of kinetic energies from zero to hv - I, so that the ionization energy
can be established from the maximum electron energy. To obtain the
barrier height, Eg - (E^ - E^^), it is necessary to establish the position of
the Fermi level with respect to E^^ (the valence band edge at the surface).
This may be achieved by measuring a spectrum for a metal in contact with
the semiconductor; often one simply measures the emission from a metallic
sample holder. Those electrons now emerging with the largest kinetic
energies originate at the Fermi level of the metal, which is the same as the

45
rermi level ot the semiconductor. Thus, Ep in Fig. 1.21 may be readily
established, enabling the barrier height E^—(Ep—Ey^) to be deduced from
the difference between the maximum energies of the photoelectrons from
the metal and semiconductor, respectively.
Suppose the Schottky barrier in Fig. 1.21 is generated by a metal on
the semiconductor surface. In the analysis thus far, we have ignored any
photoemission originating in the metal overlayer itself; this is only vahd
for very small metal coverages, less than a monolayer or so. For metal
coverages of more than a few Angstroms, photoemission from the metal
overlayer tends to mask that from the substrate and it then becomes
difficult to establish E p — a c c u r a t e l y . Photoelectron spectroscopy is
therefore only suitable for measuring barrier heights for ultra-low metal
coverages. It should also be appreciated that, since this method
essentially measures the position of the valence band at the surface (E^^),
it can only give the true barrier height if the energy gap E^ has the same
value at the surface as in the bulk. This may not be true if there is
considerable intermixing at the interface.
In many cases photoemission from surface states makes it difficult to
determine precisely those electrons which originated at the valence band
edge. In such instances it is common to investigate Schottky barrier
formation by observing shifts in the kinetic energies of peaks originating
from core levels. Consider the case where the bands on an n-type semi­
conductor are flat right up to the surface before any metal is deposited;
this is the case for the clean cleaved (110) galhum arsenide surface
already discussed in Section 1.2.2. In this case, it is straightforward to
measure the binding energy E\ of a core level with respect to the Fermi
level at the surface. When a metal is deposited and the barrier forms, the
separation of Ep and the core level at the surface decreases. This leads to
a decrease in the value of E\ to as shown in Fig. 1.21, and the
Schottky barrier is given by E\—Ei + ^. The method has been used
extensively to study how the barrier forms as the thickness of the metal
layer is increased from zero to a few tens of Angstroms. In the analysis it
has been assumed that the kinetic energy of the core level emission is
changed only due to the shift of the Fermi level at the surface of the semi­
conductor. Unfortunately, this assumption is not always correct and
further changes may be caused by chemical reactions between the metal
and the semiconductor. These are generally referred to as 'chemical
shifts' and are very important in the study of chemical reactions at
interfaces. These, and other complications, mean that photoelectron
spectroscopy cannot be apphed universally to measure Schottky barriers,
and considerable experience is required to avoid the many pitfalls
associated with the interpretation of data obtained using the technique.
The accuracy of barrier determination from photoemission measure­
ments is rather limited, of the order of ± 0.1 eV in favourable cases.

46
1.8 Electronegativity and work function

In considering the formation of Schottky barriers in terms of the

Schottky-Mott and the Bardeen theories in Section 1.3, the electron
affinity of the semiconductor and the work function of the metal were
important parameters. A s outlined in Section 1.2.3, the work function of
a metal is the energy required to remove an electron from the Fermi level
to a state of rest outside the surface, and it consists of a volume term
and a surface dipole term D^, so that ^^ = ^^ + D^. Similarly the electron
affinity of a semiconductor contains a surface dipole term D^. The sur­
face dipole terms are governed by the way the electronic charge is distri­
buted at the surfaces of solids, and are clearly influenced by relaxation
and reconstruction of clean surfaces. A s pointed out in Section 1.2.3,
they also depend on the particular surface, differing by 0.78 eV for the
(110) and (111) surfaces of tungsten (Strayer, Mackie, and Swanson,
1973). In measuring the work function of sohds it is not possible, unfor­
tunately, to estabhsh the volume and surface contributions separately.
When a metal and a semiconductor make intimate contact, the precise
atomic positions and the charge distributions at the soUd surfaces in
contact change from those characteristics of clean surfaces, and and
change in an unknown way. The Schottky-Mott theory cannot there­
fore be expected to account well for barrier heights at such interfaces
since it assumes that and Xs remain unchanged when contact is made
(or at least that their difference remains unchanged). In an attempt to
overcome this problem it is common to use the electronegativity (xj) of a
metal rather than its work function. Pauling (1932, 1960) described the
electronegativity as 'the power of an atom in a molecule to attract
electrons to itself and developed a table for the elements which is known
as the Pauling electronegativity scale. Values of the electronegativity
calculated by Pauling are shown, for a selection of elements, in Table 1.3.
Though, in general, elements with high electronegativities tend to form
solids with large work functions, a precise relationship between them
cannot be defined. A n evaluation by Gordy and Thomas (1956) suggests
that ^„ may be roughly linearly related to Xm via an equation of the type
= AXrn'^ B, where A and B are constants. In their analysis Gordy and
Thomas suggested that the value of A is around 2.27. Many groups have
made use of the electronegativity scale rather than work functions in
their attempts to understand the variation of Schottky barrier heights for
metals on semiconductors, for example Kurtin, M c G i l l , and Mead (1969)
and Ihm, Louie, and Cohen (1978).
It should be emphasized that the concept of electronegativity is one
that apphes to individual atoms, and takes no account of surface dipoles.
Electronegativity cannot readily be measured. Furthermore, the linear
dependence of electronegativity on work function suggested by Gordy
47
 Table 1.3 Values of work functions and electro­
negativities for some common metals (eV)

Metal Work function" Electronegativity''

Pt 5.65 2.2
Ni 5.15 1.8
Pd 5.12 2.2
Au 5.1 2.4
Co 5.0 1.8
Cu 4.65 1.9
Mo 4.6 1.8
W 4.55 1.7
Fe 4.5 1.8
Cr 4.5 1.6
Sn 4.42 1.8
Ti 4.33 1.5
Al 4.28 1.5
Ag 4.26 1.9
Ta 4.25 1.5
Ga 4.2 1.6
In 4.12 1.7
Mg 3.66 1.2
Ca 2.87 1.0
Ba 2.7 0.9
Cs 2.14 07

'Values for polycrystalline specimens, from

Michaelson(1977).
"From Pauling (1960). Values refer to the common
oxidation states of the elements.

and Thomas is clearly inadequate since it takes no account of the

contribution of the surface dipole to the work function of a solid (this
inadequacy is clear by inspection of Table 1.3). One might expect a
simple relationship between Xm and <t>y, the volume contribution to the
work function, but as mentioned earlier the latter quantity cannot be
measured.
It is therefore not clear that much is to be gained by using the concept
of electronegativity, which ignores the existence of surface dipoles
altogether, rather than work function, which assumes (incorrectly) that
the interface dipole is the algebraic sum of the surface dipoles for the
free surfaces. For this reason it is the latter quantity that is used most
extensively in the remainder of this book.

48
2 Experimental studies of metals on
semiconductors

2.1 Metals on silicon and germanium

There have been extensive investigations of electrical contacts to silicon,

motivated by two main reasons. The first is the need to gain a fundamental
understanding of metal-semiconductor interfaces in general, and it might
be expected that this would be best achieved on a simple elemental semi­
conductor where high-quahty single crystal material is plentiful. The
second reason is that silicon is by far the most important semiconductor
in microelectronics, and the technology of contact fabrication to sihcon is
of the utmost importance in the manufacture of microcircuits. Indeed, the
interplay between fundamental studies and technological applications has
always been particularly strong in this area.
Although extensive studies of metal-sihcon interfaces have been
reported, relatively Uttle work has been pubhshed for contacts to
germanium. Most of this section, therefore, will be concerned with metals
on silicon. The scientific hterature contains an abundance of accounts of
metal layer growth mechanisms, chemical reactions at interfaces, and the
determination of electrical properties of interfaces and Schottky barrier
heights. For more detailed discussions the reader is referred to reviews by
Sze (1981), Calandra, Bisi, and Ottaviani (1984), and L e Lay, Derrien,
and Sebenne (1986). Many metals form silicides when the contact is
annealed, and indeed some are formed at room temperature. These are of
great importance in technology and are dealt with more comprehensively
in Chapter 5.

2.1.1 Clean silicon

Most studies of the growth of metals on clean silicon indicate that the
interface is often non-abrupt, even when deposition is onto surfaces
maintained at room temperature. Contacts deposited on clean silicon are
free of contamination layers such as oxides at the interface, and it may be
expected that the interpretation of the electrical behaviour of such
interfaces would be simpler than that of comparable ones involving
deposition onto etched surfaces. However, this is not so, largely because
of chemical reactions and interdiffusion processes which are common at
intimate interfaces. Indeed, it appears that non-abrupt interfaces are

49
obtained for gold and copper contacts even if the silicon surface is
maintained at liquid nitrogen temperature while the metals are deposited
(Abbati and Grioni, 1981).
Many investigations have involved the deposition of metals on crystals
cleaved in vacuum. Such surfaces may not be physically perfect since the
cleaving process is known to cause some damage which may take time to
anneal out. The first extensive apphcation of this method to sihcon was
made by Archer and Atalla (1963) and, although the vacuum used by
them was only 10~* Torr, the metal was evaporated so rapidly that there
should have been negligible contamination of the interface. They
determined the barrier heights from CIV measurements. Their
experiments were repeated by Turner and Rhoderick (1968) using a
rather better vacuum (~ 10"^ Torr). The latter also deduced the barrier
heights from CIV data and found very good agreement with the results of
Archer and Atalla. Similar measurements on cleaved n-silicon using low
work-function metals (Mg, Ca, K and Na) were carried out by Crowell,
Shore, and L a Bate (1965) and by Szydlo and Poirier (1973). The barriers
obtained with these metals were so low that they could only be deduced
from IIV characteristics. The collected data from all four sets of experi-
ments are shown in Fig. 2.1. Wherever possible, values of work functions

1.0
VFe

0.9 Au
Ag

0.8 -
Al{8 Cu Au
oPb Dpe
0.7 - A1{Ç Ni
TAg 15

o. 0.6 - AlTPb Cu
Co T N i

0.5 - Mg
+ K D Van Otterloo et ai.
+ Na T Thanailakis & Rasul (I/V)
0.4 - ACa
V Thanailakis & Rasul \ciV)
X Turner & Rhoderick
0.3 - + Szydlo & Poirier
o Archer & Atalla
A Crowell, Shore & La Bate
0.2
2.0 3.0 4.0 5.0 6.0
(eV)

Fig. 2.1 Barrier heights on cleaved sihcon (111) surfaces. Wherever possible, the work
function values are those obtained under U H V conditions by Thanailakis and Rasul
(1976). Other values are as given by Rivière (1969).

50
obtained in ultra-high-vacuum systems have been used. The barrier
heights almost all fall in the range between 0.4 and 0.85 eV and appear to
form two groups, one in the region of 0.7 eV, and the second around
0. 45 eV.
Some very careful measurements on cleaved silicon surfaces have been
carried out by Thanailakis (1974, 1975) and Thanailakis and Rasul
(1976). These authors cleaved sihcon crystals in a vacuum of 5 x 10~"
i b r r and measured their work functions by means of the Kelvin vibrating-
capacitor method, and then evaported a metal film. The work function of
the metal and the height of the Schottky barrier were then measured
without breaking the vacuum. The barrier heights were measured by I/V,
CIV, and photoelectric methods, and could be estabhshed to within
± 0.01 eV. Measurements of barrier heights for several metals deposited
in ultra-high vacuum on cleaved sihcon crystals were also carried out by
van Otterloo and de Groot (1976) and by van Otterloo and Gerritsen
(1978). These are all included in Table 2.1 and Fig. 2.1. The following
important points emerge from the data:

1. In general there is excellent agreement between the values of ^

obtained from the IIV and photoelectric methods, but they are both
substantially smaller than the value obtained from CIV characteristics
by an amount which is much greater than the image-force lowering. It
will be recalled from Chapter 1 that the IIV and photoelectric methods
measure the zero-bias barrier height ^bo> while the CIV method
measures the flat-band barrier height
2. There is generally good agreement between the values of barrier height
obtained from the C / F data by the various groups.
3. Broadly speaking, metals with large work functions yield large values of
^b whereas low-work-function metals give much lower values. This is
shown most clearly in Fig. 2.1. However, there is a great deal of scatter
in the experimental data for a given metal.

It is instructive to examine the development of the Schottky barrier

height as ultra-thin metal layers are deposited on a clean surface. Photo-
emission studies have been reported by Purtell et al. (1983) for palladium
on silicon, and by Clabes, Rubloff, and Tan (1984) for vanadium
deposited on vacuum-cleaved and heat-cleaned silicon (111) surfaces.
The data of Clabes et al. are shown in Fig. 2.2. The Fermi levels for the
clean cleaved ( 2 x 1 ) surface and the heat cleaned (7 x 7) surface are
positioned at energies of ~ 0.8 eV and ~ 0.6 eV, respectively, below the
conduction band. Upon deposition of 2 A of vanadium, the Fermi level
positions at the surfaces of both crystals become identical, demonstrating
that the different surface structures do not lead to different values of ^b-
The Fermi level does not reach its final position until more than 2 A of

51
Table 2.1 Schottky barrier heights (in eV) for a range of metals deposited on clean cleaved silicon
Metal Work Thanailakis and Rasul (1976) Turner and Archer and Other
function Rhoderick(1968) Alalia (1963)
Photo av OV C/V W Photo C/V

Pt 5.30 0.71 0.71 0.82

Au 5.10 0.73 0.73 0.82 0.82 0.81 0.68= 0.82»
Ni 5.10 0.59 0.59 0.74 0.70 0.68
Co 4.97 0.61 0.61 0.81
Cu 4.55 0.62 0.62 0.75 0.79 0.77
Fe 4.58 0.63 0.63 0.98 0.7» 0.71-0.77"
Ag 4.41 0.68 0.68 0.79 0.79 0.76 0.71-0.83' 0.70-0.81'
Pb 4.25 0.61 0.61 0.72 0.79
Al 4.17 0.61 0.61 0.70 0.76 0.77 0.70" 0.72"
Mg 0.56" 0.55-0.58"

'Varma e/a/. (1977).

"Van Otterloo and Gerritsen (1978).
' Van Otterloo and de Groot (1976).
The work function values are those determined under U H V conditions by Thanailakis and Rasul (1976).
 0.85

V/Si(lll)
(2X1)
0.80 - hv=m ev
A

•o

0.60
Deposit Anneal

0.55
2 4 200 350 500
Clean
Thickness (A) Temperature (°C)
Fig. 2.2 Schottky barrier height (n-type barrier) for the as-deposited and reacted
V/Si(lll) interface as determined from synchrotron radiation photoemission studies of
core level band-bending shifts. (From Clabes, Rubloff, and Tan, 1984. Copyright American
Institute of Physics.)

vanadium has been deposited, and anneahng leads to a further change in

the barrier height. A n identical result was obtained for palladium on
silicon by Purtell et al. Detailed studies of chemical reactions at the
interface demonstrate that a vanadium sihcide is formed upon annealing,
whereas the interface formed at room temperature is reported to be
abrupt. Palladium, on the other hand, forms a silicide at room tempera­
ture and the equilibrium barrier height is reached after the deposition of a
few monolayers of the metal and without annealing. This equilibrium
value is nearly identical to that determined by I/V and C/V methods for
thick palladium contacts on sihcon. The data demonstrate that the
detailed nature of the interface chemistry does influence the final barrier
height. The above constitute examples where the formation of Schottky
barriers on silicon has been studied for extremely thin overlayers. It may
be concluded that a few monolayers of the overlayer are sufficient to
generate barriers with heights close to those associated with thick
overlayers.

2.1.2 Etched and oxidized silicon surfaces

The effect of exposing cleaved sihcon surfaces to oxygen before

depositing the metal contact has been investigated by several workers.

53
Archer and Alalia (1963) studied the effect of cleaving in oxygen at a
pressure of 10"'' Torr and exposing the surface to this ambient for a few
seconds before depositing the metal. They found a negUgible effect on the
barrier height for platinum, palladium, nickel, and gold, but a reduction of
around 0.1 eV for copper, silver, and aluminium. It is noteworthy that
Turner and Rhoderick (1968) found no difference in barrier heights on
cleaved and etched surfaces for gold and nickel (they did not investigate
platinum and palladium) but a significant decrease on etched surfaces for
copper, silver, and aluminium. Varma et al. (1977) studied gold and silver
contacts on clean and oxidized (111) silicon surfaces and reported that
the interfacial contamination reduced the mechanical strength of the
adhesion between metal and semiconductor. They also reported that the
interface contamination led to a reduction in the differences between the
values of barrier heights obtained by the I/V and C/V methods. Varma
and his co-workers also reported that the barrier heights for gold and
suver increased with time, ultimately yielding values closer to those
measured for intimate interfaces. Time-dependent, or 'ageing', effects are
particularly important for metals on chemically etched surfaces, as
pointed out by Turner and Rhoderick (1968), who observed variations in
the measured Schottky barriers of around 0.2 eV over a period of 1000
hours for A u contacts. The ageing of gold and copper contacts on
oxidized silicon has also been studied by Mottram et al. (1979). For
copper contacts, the ageing was such as to increase the barriers,
ultimately leading to values closer to those measured for copper on clean
surfaces.
The ageing of metal-silicon contacts may be accelerated by heating,
and shows a time-scale consistent with the migration of charged ions, as
often observed in M O S transistors. A second possibility is that the metal
undergoes a chemical reaction with the oxide layer. Detailed studies of
the nature of the silicon oxide layer by photoelectron spectroscopy
(Hollinger and Himpsel, 1983) show the presence of sihcon, not just in
Si02 but in a sub-oxide as well. This sub-oxide is denoted SiO,,, where
x<2. It seems that many metals, including aluminium and silver, can
lead to a reduction of the sub-oxides even at room temperature, and in
some cases the SiOj can also be reduced.
Measurements of barrier heights on chemically etched silicon surfaces
have also been made by several other workers, notably Kahng (1963),
Jäger and Kosak (1969), and Hirose, Altaf, and Arizumi (1970). Their
results are generally similar to those of Turner and Rhoderick, although
they did not study ageing effects. In all cases the donor density was such
that the theoretical value of the image-force lowering was less than the
experimental error, and there was no tendency for barrier heights
obtained by C/V characteristics to be consistently higher than those
obtained by I/V or photoelectric data. Figure 2.3 shows data from the

54
 0.9

0.8

0.7 oCu
AI
fAg "W 9Ni
F7
WT xCu
0.6
TaT x + "Mo
OMg "ivio X Hirose, Altaf, and Arizumi A Card
O Turner and Rhoderick • Basterfield, Shannon and Gill
0.5 V Jäger and Kosak • Gulknechl and Slnitt
+ Salticb • Landkammer
D Cowley and Zeltler • Crowell, Sarace and Sze
oPb n Cowley e Wilkinson
0.4

0.3
3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0
*n, (eV)

Fig. 2.3 Barrier heights on chemically etched silicon surfaces. The values of ^„ are those
given by Michaelson (1977).

above sources, together with a few other results on individual metals,

plotted against metal work functions.
It is apparent from Fig. 2.3 that a wide variation exists in the values of
reported by different groups for the same metal (e.g. aluminium).
Some of the scatter is undoubtedly due to variation in the surface
treatments used by different authors and can be partly attributed to the
lack of information on ageing effects. Furthermore, some workers heated
the silicon to drive away moisture whereas others did not. A l l that can be
said with confidence is that metals with a high work function (e.g. gold
and platinum) tend to give large barriers, while metals with low work
function (e.g. magnesium and calcium) tend to give small barrier heights.
The problems associated with chemical reactions between metals and the
oxide layer, coupled with uncertainties in the appropriate values of metal
work functions, also add to the difficulties of applying the Bardeen
theory to deduce detailed information about interface states from
Fig. 2.3.
The results of Turner and Rhoderick (1968) taken in isolation show a
more pronounced correlation between barrier height and work function
than do all the data taken collectively. This is probably because they used
a standard method of preparation and allowed for ageing effects. Their

55
data relating to can be reasonably approximated by a straight line
of slope about 0.5 and were analysed in terms of the Bardeen model.
They yielded a value of » 1.2 x 10" eV~' m~^, which is typical of
experimental values obtained by other methods for etched silicon
surfaces. The results were consistent with a value of of 0.31 eV above
the top of the valence band. It does not seem profitable to indulge in any
more elaborate attempt at comparing theory and experiment; what is
significant is that the results of one particularly extensive set of
experiments can be qualitatively explained in terms of the Bardeen
model.
A n interesting series of measurements has been made by Arizumi,
Hirose, and Altaf (1968, 1969) who prepared Schottky barriers by
evaporating binary alloys of pairs of the metals gold, silver, and copper
on to sihcon. They found that for any pair the barrier height varied
linearly with alloy composition between the values corresponding to the
pure metals. One might expect the work function of the alloy to vary in a
linear manner with composition of the alloy provided the band structures
of the constituent metals are similar, but a review of the experimental
data by Rivière (1969) shows this not to be the case.
The dependence of barrier height on donor density has been studied
by several workers (Kahng, 1963; Saltich, 1969; Archer and Yep, 1970)
and in no case was there any clear evidence of a variation of (j)^ with
A/j.The temperature dependence of the barrier height has been studied
by several workers with conflicting results. Crowell, Sze, and Spitzer
(1964) measured the temperature dependence of gold-silicon barriers
using the photoelectric method and found that ^t, was a decreasing
function of temperature with a functional dependence almost identical
with the temperature dependence of the energy gap. Arizumi and Hirose
(1969) also found the height of the gold-sihcon barriers to decrease
more or less linearly with increasing temperature with a coefficient of
about - 3 X 10"" eV K " ^ On the other hand, Cowley (1970) observed
that the temperature dependence of titanium-silicon barriers was much
less than that of the energy gap, and Cowley and Zettler (1968) stated
that the barrier heights of their chromium-silicon diodes were the same,
within the experimental error, at 196 K and 298 K. O n theoretical
grounds one would expect a temperature variation of the energy gap to
result in a temperature variation of because of the change in - ^Q.
Barriers on p-type silicon have been thoroughly investigated by Smith
and Rhoderick (1971), who made measurements on six different metals.
The barriers were generally lower than on n-silicon and are shown in
Table 2.2. The results can be combined with the data of Turner and
Rhoderick on n-silicon to give the following results for ^ ^bp- The sum
is equal, within the limits of experimental error, to the energy gap of
silicon (1.1 eV) in all cases except that of lead, which is slightly low.

56
Table 2.2 Barriers on n- and p-type silicon
Barrier Metal
(eV)
Ag Al Au Cu Ni Pb
0.56 0.50 0.81 0.69 0.67 0.41
0.54 0.58 0.34 0.46 0.51 0.55
I.IO 1.08 1.15 1.15 1.18 0.96

2.1.3 Germanium

In view of its great importance in the early days of semiconductor

technology, it is surprising how little information is available about
Schottky barriers on germanium. There was a thorough study of pomt
contacts in connection with early microwave-mixer development, but
these contacts were always 'formed' in such a way that the junction was
not a simple metal-semiconductor contact but possibly a p - n junction
(see Shockley, 1950). A n extensive series of measurements were made by
Jäger and Kosak (1969) who prepared Schottky barriers by evaporating
a series of metals onto silicon and germanium heated to 300°C. The
barrier heights were in all cases lower on germanium than on sihcon by
between 30 and 50 per cent, but there appeared to be no simple relation­
ship between the two. No attempt to explain their result in terms of the
Bardeen model was made.
Schottky barriers on etched germanium surfaces have also been
investigated by Thanailakis and Northrop (1973). Their results do not
agree very well with those of Jäger and Kosak, possibly because they did
not heat the germanium prior to evaporation. Thanailakis and Northrop
analysed their data in terms of the Bardeen model in a manner similar to
that used by Turner and Rhoderick to interpret their results on silicon.
They found that their results were consistent with a density of surface
states of 2 X 10" eV~' m"^ and a position of the neutral level 0.13 eV
above the top of the valence band. In the case of aluminium contacts they
found pronounced ageing effects and departures from ideal behaviour
which they attributed to the fact that aluminium oxidizes very easily and
tends to reduce the GeOj film on the etched surface of the germanium.
There is very little information available about contacts to cleaved
germanium surfaces, apart from some rather sparse results by Mead and
Spitzer (1964). They report barrier heights of 0.45 eV and 0.48 eV for
gold and aluminium, respectively, but the number of samples studied was
extremely small and few details are given of the experimental procedures.

2.1.4 Discussion

2.1.4.1 Lack of reproducibility A general point that emerges very clearly

57
 from the preceding considerations is that the experimental measurements
of barrier heights by different methods and by different workers are not
very consistent and show considerable scatter. Bearing this in mind, it
does not seem profitable to attempt to fit the best straight line to barrier-
height versus work-function or electronegativity data by the method of
least squares, as some authors have done. It is important to question
whether the lack of reproducibihty of barrier height for a given metal is
an experimental artefact, or due to something more basic such as varia­
tions in the detailed structure of the interface. It is clear that, for etched
surfaces, most of the variation is due to differences in the method of
surface preparation and/or varying degrees of ageing, as is clearly shown
by the work of Turner and Rhoderick. But in some cases, especially for
intimate contacts, there may be a more fundamental cause arising from
the microscopic structure of the interface. The photoemission data of
Clabes et al. (1984), illustrated in Fig. 2.2 for vanadium on sihcon,
suggest that the latter may often be the case.
Much hght has recently been shed on the importance of the micro-
structure of the interface by studies of nickel disilicide on sihcon. Like
the majority of the transition metals, nickel can form a series of silicides
(NijSi, NiSi, or NiSi2) which are metalhc in nature and which are capable
of forming Schottky barriers in the same way as ordinary elemental
.metals. These silicides are usually formed by heat treatment of the metal/
silicon couple, and are discussed at some length in Section 5.3. A
surprising feature of the process is that in many cases a thin layer a few
units cells thick having the sihcide structure and having bonds which are
characteristic of the silicide may form even at room temperature
(Rubloff, 1983), and the nickel-sihcon interface provides an example of
this (Cheung et al, 1980). Tung (1984) showed that two types of NiSij
contacts are possible on heat-cleaned S i ( l l l ) surfaces, and that these
types can be formed by using slightly different preparation procedures.
These are caUed type A and type B , respectively, and the structures
associated with them as determined by transmission electron microscopy
are illustrated in Fig. 2.4. It may be seen that in the type B structure the
NiSiz lattice is rotated by 180° with respect to that in type A , so that
there is a slight difference in the geometry of the bonds at the NiSij-Si
interface. Detailed / / F a n d C / F studies indicate a barrier height of 0.78
eV for the type A interface and 0.65 eV for type B. These results aroused
considerable interest because, if correct, they present the first clear
demonstration of the effect of interfacial structure on barrier height. The
experiments were repeated by Liehr et al. (1985) who claimed that the
lower barriers were due to structural imperfections at the interface, and
that contacts of types A and B both yield barrier heights of 0.78 eV,
provided they are structurally perfect. More recently, however, Tung et
al (1986) have carried out a more extensive study and have reaffirmed

58
 (a) Type A NiSiz/Si (111) (b) Type B NiSiz/Si (111)

Fig. 2.4 The structure of Type A and Type B NiSiz on Si(lll). (From Tung, 1984.
Copyright American Institute of Physics.) The diagram shows a projection of the lattice
viewed in a (110) direction.

their original conclusion that types A and B lead to different values of

the Schottky barrier height. These data have been confirmed by
Hauenstein etal. (1985).
It is apparent, therefore, that variations in of 0.13 eV or more are
observed due to different microscopic configurations at the interface for
the most thoroughly studied and most weU controlled contact to sihcon.
Nickel contacts deposited on clean silicon often consist of mixtures of
phases even in the absence of heat treatment. Slightly different prepara­
tion procedures may unavoidably give contacts in which different phases
predominate, and this may be one of the reasons for the large variations
observed in the reported barrier heights for nickel. Furthermore, it is
quite likely that such variations will be observed for other metals; it
would be most surprising if nickel were unique in this respect. Many of
the interfaces formed for transition metals on sihcon are highly perfect
and there is no evidence that defects at the interface play a dominant role
in determining the Schottky barrier heights. However, the studies of
Liehr et al. (1985) demonstrate that structural imperfections at the
interface do influence the barrier heights, by an amount of the order of
0.1 eV, and certainly contribute to the poor reproducibihty of data by
different workers. It is therefore unhkely to be profitable to search for
good linear dependences of barrier heights on work functions or electro­
negativities until the reproducibihty problem has been tackled for a
wider range of contacts, in the manner carried out by Tung et al. for
nickel disilicide on silicon.

59
 Despite these comments it is still possible to make certain generaliza­
tions about the data presented in the preceding sections. For metals on
atomically clean and on oxidized surfaces there is a clear dependence of
the barrier height on the metal forming the contact, though this
dependence is much less than the Schottky-Mott theory predicts, since
the observed barrier heights lie within a range of about 0.4 eV for a total
variation of nearly 3 eV in the work function. It is obvious, therefore,
that there must be some pinning mechanism at work, and this imphes the
existence of interface states of some sort or other. Furthermore, there is a
very roughly Unear dependence of the barrier height on the metal work
function. Tove (1983) has considered this dependence in more detail for
a range of metals which form silicides. He examined the dependence of
barrier heights on both work functions and electronegativities of the
metal contact and found that the former gives at least as good a linear fit
as the latter.

2.1.4.2 Etched surfaces From a theoretical point of view, one would

expect contacts on etched surfaces to be easier to analyse because the
oxide film to a certain extent 'decouples' the semiconductor from the
metal. It may be expected that one can then talk of interface states which
are a property of the semiconductor-oxide combination and that the
Bardeen model should be a reasonably good approximation. The barrier
height should then show a linear dependence on metal work function as
long as the surface state density is a constant and the surface dipole
contribution to the work function is unchanged when the metal is
deposited. A s we have seen there is a good deal of scatter in the
experimental data, but if the results from a single laboratory using one
particular method of surface preparation are considered in isolation, a
straight line can often be roughly fitted to the data. For etched sihcon
surfaces, the results of Turner and Rhoderick can be understood in terms
of the Bardeen model by postulating a surface state density of about 10"
eV~' m~^ and an oxide layer thickness of about 20 A. The surface state
density is comparable with that obtained for M O S transistors and is
substantially lower than the density of surface states on a clean surface.
(For the free cleaved surface the total number of surface states is about
one per surface atom, which for sihcon is around 5 x 10^^ m~^.) It is
difficult to know what thickness to assume for the oxide layer in the
studies of Turner and Rhoderick, but even if this is reduced to 5 A the
surface state density is stiU an order of magnitude lower than that
expected for a clean surface. Some metals chemically react with the
oxide, thus making a precise determination of its thickness difficult, and
possibly leading to a metal dependence of the thickness.
Although the Bardeen model does form a useful basis for the
comparison of theory and practice with etched surfaces, there are certain

60
fundamental limitations to its validity. The most questionable assumption
is that the surface contributions to the metal work function and semi­
conductor electron affinity remain unchanged upon contact (or at any
rate their difference remains unchanged). The model also neglects any
effect of image forces on the interface states. Another serious criticism is
that, as Archer and Yep (1970) have pointed out, the Bardeen model
predicts a much greater dependence of barrier height on dopant
concentration than that observed in practice.
The data illustrated in Fig. 2.3 can be analysed in a different way. Kar
(1975) made direct measurements of the quantity <l>'^ - x'^, where (f)'^ and
X\ are respectively the work function of the metal and the electron
affinity of the silicon measured relative to the bottom of the conduction
band of SiOj. It is then possible to plot as a function of ^m~"Zs> thus
avoiding having to make any assumptions about the surface dipole
contributions to ^„ and Xs if one assumes that the interfacial layer has the
composition Si02. However, there is no better correlation than when the
metal work functions are used. A n alternative way is to plot (j>^, as a
function of the electronegativity of the metal rather than its work
function, but again there is no evidence that this is any more satisfactory.

2.1.4.3 Intimate contacts Turning now to intimate contacts, the origin

and form of surface states on cleaved (111) silicon surfaces are quite well
understood, as considered in Chapter 1. The density of states is of the
order of 5 X 10^^ eV~^ m~2 (Many, Goldstein, and Grover, 1965) and is
consistent with the fact that the Fermi level at the free surface is
positioned about 0.3 eV above the valence band edge (Allen and Gobeh,
1962). Upon contact with a metal the surface structure and the
distribution of surface states is changed. Louie, Chelikowsky, and Cohen
(1977) have estimated theoretically the Schottky barrier height at
intimate aluminium-sihcon interfaces assuming that it is controlled by
MIGS, and arrive at a value of 0.6 ± 0.1 eV, in good agreement with
experiment. In a crude fashion one can extend the Bardeen theory and
adapt eqn (1.5) to the case of intimate metal-semiconductor contacts by
noting that the quantity qdD^ in the expression for y [eqn (1.6)]
represents the dipole moment due to the interface states, and may be
taken as the total charge in the tails of the metal-induced gap-states
multiplied by the attenuation length. Louie, Chelikowsky, and Cohen
(1977) and Yndurain (1971) estimated a value of y of the order of 0.2 for
metals on silicon. This is consistent with a roughly linear fit to the data
presented in Fig. 2.1. It could be argued that Fig. 2.1 is more consistent
with two pirming energies, one at around 0.45 eV and the other around
0.75 eV below the conduction band edge, as discussed in Section 1.4.2,
particularly if the barriers determined from C / F data are considered
alone and those metals which readily react with sihcon to form sihcides

61
are ignored. However, the spread in the data is too large for any
conclusive statements along these lines to be made.
The experimental data of Tung for types A and B NiSij on sihcon
(111) show that the barrier height depends on the microscopic structure
of the interface. The most likely reason for the difference of 0.13 in the
barrier heights is that the precise form of the electron wave-functions,
and possibly the bond lengths at the interface, are very slightly different
in type A and type B structures, yielding slightly different interface
dipoles in the two situations. Close examination of Fig. 2.4 reveals that
the essential difference between the type A and type B configurations is
that, after allowing for the 180° rotation of the NiSiz lattice, the silicon
atoms in the positions labelled X have different positions relative to the
nearest nickel atom in the two cases. It would be reasonable to assume
that this produces a difference in the bonding electron wave-functions,
giving rise to a different charge distribution around the sihcon atoms in
position X , which are at a distance of about 5 A from the nearest nickel
atoms. If one supposes that the different charge distribution gives rise to
a difference a ^ in the effective charge at X , one finds that a need only
have a value of approximately 0.02 to cause a change of 0.1 V in the
electrostatic potential difference across the interface.
It is still not fully established whether the significant differences
between the barrier heights determined on the one hand from I/V and
photoelectric measurements and on the other hand from C/V data as
reported, for example, by Thanailakis and Rasul (1976), are intrinsic
properties of an ideal metal-semiconductor contact or artefacts arising
from the method of preparation. The fact that they could be reduced by
exposing the metal to oxygen suggests the former. Thanailakis and Rasul
have attempted an explanation in terms of M I G S , though they found
difficulty in explaining the magnitude of the effect, especially in the case
of iron. The fact that Missous, Rhoderick, and Singer (1986c) were able
to obtain very good agreement between I/V and C/V measurements of
in A l - G a A s contacts prepared by M B E suggests that the effect is not an
intrinsic property of ideal contacts, unless it is pecuhar to silicon.
A completely different approach to explaining the variation of barrier
heights for transition metals on silicon has been adopted by Andrews
and PhiUips (1975). These authors use a chemical bond approach and
relate the barrier height to the heat of formation of the silicide which is
formed when the metal interacts with the silicon. Ottaviani, Tu, and
Mayer (1980) have questioned the interpretation of barrier heights of
sihcides in terms of heats of reaction, and have instead attempted to
relate the barriers in a linear fashion to the eutectic temperature of the
appropriate transition-metal-silicon compound. The reasoning behind
these two approaches seems to be that both the heat of formation and the
eutectic temperature reflect the strength of the chemical bond between

62
the metal and silicon, and this in turn is related to the charge transfer at
the interface and hence to the strength of the interfacial dipole. However,
there is no indication that either of these approaches gives any better
linear fit to the experimental data than a simple plot of barrier height
against work function, and neither is capable of explaining the
differences observed for type A and type B NiSij contacts to sihcon.
At the present time, therefore, no fully conclusive statements can be
made about the fundamental physics of barrier formation for metals on
silicon and germanium. Though the nature of the electronic states on
clean cleaved surfaces is well understood, it is not clear how these relate
to states at interfaces with metals, except in the case of the specific
calculations of Louie, Chelikowsky, and Cohen (1976) for aluminium on
sihcon. What is clear is that there must be some pinning mechanism
which makes the barrier height less dependent on the metal than the
simple Schottky-Mott theory would indicate, and that in some cases the
magnitude of the interfacial dipole can have a significant effect. It is
highly hkely that the Schottky barrier heights are determined by a
combination of mechanisms, with the relative importance of these
depending on the detailed preparation conditions. Unfortunately it is
extremely difficult at present to carry out detailed calculations of real
interfaces which fully incorporate all the relevant mechanisms, partly
because of the difficulty of the calculations themselves, and partly
because the precise physical and chemical nature of the interfaces is not
usually well established.

2.2 Metals on III-V semiconductors

2.2.1 General

A wide range of surface-science techniques has been appUed to study the

microscopic interactions and Schottky barrier formation for metals
deposited on clean and oxide-covered III-V semiconductors. The
majority of these investigations have been applied to ultra-thin metal
layers deposited on semiconductor surfaces prepared by heating, by
cleavage in vacuum, by sputtering and annealing, or prepared by
molecular beam epitaxy (MBE). Extensive studies of Schottky barrier
heights for thick metal layers on III-V semiconductors have also been
carried out using I/V, C/V, and photoresponse methods. In some cases the
metals have also been deposited by M B E . In this section, the data for
ultra-thin metal layers are presented first (Section 2.2.2) followed by
those for thick layers (Section 2.2.3). Chemical and metallurgical
interactions at interfaces between metals and III-V semiconductors will

63
not be considered in detail; full reviews have been presented elsewhere,
for example by Smha and Poate (1978), BriUson (1982b), Ludeke
(1983), Wilhams (1983, 1985) and Spicer et al. (1986). However, a
number of important general points which emerge will be briefly outlined
here.
Interfaces formed by deposition of metals onto clean (110) surfaces of
III-V semiconductors are rarely perfectly ordered and atomically abrupt.
In many cases interdiffusion occurs, and in others strong chemical
reactions lead to new chemical phases at the interface. The detailed form
of the interface is generally influenced by the quality of the substrate
surface, by the temperature of the substrate during deposition of the
metal, and by post-deposition aimealing cycles. Consider, for example,
the deposition of silver onto clean (110) InP surfaces at room
temperature. In this case the interface is abrupt, but considerable out-
diffusion of indium into the silver is observed if the InP surface is first
hghtly sputtered with argon ions (McKinley et al., 1982). Heating of the
abrupt Ag-InP junction leads to the diffusion of A g into the semi­
conductor, accompanied by out-diffusion of In into the metal contact.
Broadly speaking, chemical reactions at metal/III-V semiconductor
interfaces can be well accounted for by using thermodynamic data
associated with bulk compounds. These aspects have been considered in
detail by Brillson (1982a), McGilp (1984), and by Pugh and Wilhams
(1986). It is important to consider the formation of binary and ternary
compounds as weU as alloy formation between the deposited metal and
the group III element. Zunger (1981) and Ihm and Joannopoulos (1981,
1982) suggested that for ultra-thin overlayers, chemical reactions may be
activated by the energy released when the metal atoms form nuclei, or
small clusters, on the surface. Ludeke and Landgren (1981)
demonstrated that aluminium atoms are highly mobile, even at room
temperature, on GaAs(lOO) and (110) surfaces and that chemical
reactions do indeed appear to accompany the formation of nuclei. The
surface diffusion of A l , and consequently the formation of nuclei, was
found to be dependent on the surface stoichiometry and on the form of
the reconstruction. Clearly, therefore, chemical reactions at interfaces
may depend in a highly complex way on the nature of the semiconductor
surface.
Metal films deposited onto room-temperature (110) III-V semi­
conductor surfaces are generally polycrystaUine. However, single crystal
epitaxial layers of aluminium may be grown on (100) surfaces of gallium
arsenide prepared by M B E (Cho and Dernier, 1978). Such contacts are
considerably more stable at high temperatures than polycrystaUine
contacts due to the reduction in grain boundary diffusion (Missous,
Rhoderick, and Singer, 1986b), and yield abrupt interfaces when
deposited at room temperature.

64
2.2.2 Ultra-thin metal layers on gallium arsenide and indium phosphide

The progressive development of Schottky barriers when ultra-thin metal

layers are deposited on clean surfaces has been more thoroughly
investigated for the III-V semiconductors than for any other group of
solids. The approaches most often used have been ultraviolet photo-
emission, soft X-ray photoelectron spectroscopy, and contact potential
studies.

2.2.2.1 Gallium arsenide Clean cleaved (110) surfaces of GaAs are free
of surface states in the band gap, provided the surface has a low step and
defect density, as was pointed out in Chapter 1. Since there are no gap
states, the energy bands are flat up to the surface, whatever the bulk
doping, and the Fermi levels at the surfaces of n- and p-type samples are
located close to the conduction and valence bands, respectively.
Deposition of thin metal layers leads to band bending with the Fermi
level shifting towards the centre of the gap, for both n- and p-type
crystals.
Spicer et al. (1980a,b) have reported that the Fermi level reaches its
final position very rapidly when A l , Ga, and In are deposited on
GaAs(llO) cleaved surfaces. The position of the Fermi level as a
function of coverage is shown in Fig. 2.5; it may be seen that the final
position is reached for coverages well below one monolayer (~ 8 x 10^"
atoms cm~'^). A t these very small coverages, detailed analysis of the
photoemission spectra for aluminium indicates that the atoms are
laterally dispersed on the surface of the GaAs and are not sufficiently

1.0 1.5
Metal coverage (10"/cm^)

Fig. 2.5 Position of the Fermi level at the GaAs(llO) cleaved surface as a function of
metal coverage, for several metals. (From Spicer ec al., 1980b. Copyright American
Institute of Physics.)

65
close together to form a continuous metal. It is noteworthy that the final
Fermi level positions for n- and p-type crystals in Fig. 2.5 differ by
around 0.2 eV. Furthermore, similar pinning energies were reported for
caesium and for oxygen on the GaAs surface. These observations,
together with the fact that most interfaces are non-abrupt, led Spicer et
al. (1980a,b) to propose the 'unified defect model' where the Fermi level
at the semiconductor-metal interface is pinned by defects in the semi­
conductor (see Section 1.6.3). These defects may be generated when the
metal atoms condense on the semiconductor surface. Since the barriers
appear to be fully formed before the adsorbed layers develop metallic
properties it was argued that an interpretation in terms of M I G S was
unlikely. For metals on GaAs(llO) cleaved surfaces it was proposed that
two defect levels are involved, a donor state at 0.9 eV below the
conduction band, and an acceptor at an energy of 0.65 eV. The donor
level is presumed to be responsible for Fermi level pinning in p-type and
the acceptor in n-type crystals.
Tang and Freeouf (1984) and Ludeke (1986) have pointed out that the
data of Fig. 2.5 were not corrected for the effects of a non-uniform
surface potential, and that these should be unportant in heavily doped
samples. Briefly, the argument is that, for low metal coverages, the band
bending due to the charged defects is confined to the immediate vicinity
of the metal ad-atoms, so that there are fluctuations in the positions of
the band edges in directions parallel to the interface. For sub-monolayer
coverages and high doping levels there may be virtually no band bending
at the points between the metal atoms (or clusters). Ludeke, Chiang, and
Miller (1983) have studied this possibihty further for silver deposited on
n- and p-type GaAs(llO) surfaces. Silver is a relatively unreactive metal
and the position of the Fermi level as a function of the metal thickness
was investigated. The experimental points for n- and p-type samples did
not meet but remained separated by around 0.2 eV for coverages in
excess of 1 A . The data were then corrected to account for non-
uniformities in the surface potential, i.e. for some bending of the energy
bands parallel to the surface between the A g clusters. When this was
done a single Fermi level stabilization energy could be derived for n- and
p-type crystals. Ludeke et al. found that the saturation position of the
Fermi level, i.e. the pinning poshion, was not reached until several
Angstroms of silver are deposited; this slow approach to stabilization was
attributed to the fact that A g forms clusters, rather than a uniform layer,
on the surface.
More recently Newman et al. (1986) have presented the results of a
more complete study for a wider range of metals on GaAs(llO) surfaces.
The Fermi level pinning positions for ultra-thin layers of metals on n-
and p-type crystals are shown in Fig. 2.6(b); also shown are the measured
pinning energies for the bare heat-cleaned (100) surface and for the case

66
 CBM
1.4
1.3
GaAs (110)
1.2
X thick n-type
1.1
ti thin n-type
1.0
0.9
0.8
0.7 ^ A-
0.6
0.5
0.4
0.3
0.2 (a)
0.1
VBM
=• O, Ge Gd Ti Ga In Ni Ag Au
,S Si ;Mn; A \ . C r ; Sn ; Cu : Pd ,
i-j M Electronegativity

CBM
1.4
1.3
1.2 GaAs (110)
A thin n-type
1.1
o thin p-type
1.0
0.9
0.8
1« 0.7 -
0.6
0.5 Po
0.4 o
0.3
0.2 (b)
0.1
VBM
g" O2 Ge Gd Ti Ga In Ni Ag Au
.^S Si , I Al ,Cr , ; Cu , P d ,
Electronegativity
3 Ut Lft OS ON ^ OOV£,vON>.t^

Fig. 2.6 Fermi-level pinning positions for (a) ultra-thin coverages of adatoms and thick
metal coverages on clean cleaved n-GaAs(llO) surfaces; (b) ultra-thin coverages of
adatoms on n- and p-type GaAs(llO) surfaces. (From Newman et at., 1986. Copyright
American Institute of Physics.)

of oxygen adsorption on the (110) clean surface. Here it may be seen that
the same pinning energies are observed for some metals on n- and p-type
crystals; A u , Ag, and C u are in this category. The measured barriers for
most metals lie within a narrow range of about 0.3 eV, being largest for
the electronegative metals on n-type crystals.
Few measurements of Fermi level pinning for thin metal layers on

67
(100) surfaces of GaAs have been reported. One such study, for A g on
(100) surfaces of GaAs grown by M B E , has been reported by Ludeke,
Chiang, and Eastman (1982). The interfaces were reported to be abrupt,
and silver layers deposited on different reconstructed faces were found
to yield different values of barrier height. The measured barrier for A g
on the c(2 X 8) surface was 0.83 eV whereas a value of 0.97 eV was
measured for the (4 x 6) surface.

2.2.2.2 Indium phosphide Schottky barrier heights for metals on n-InP

are generaUy substantially smaller than those for the same metals on
GaAs, although the band gaps of the two semiconductors differ by only
around 0.1 eV. Chemical interactions for a wide range of metals on clean
In? have been investigated by a number of groups (Brillson et al, 1982;
McKinley et al, 1982; Hughes, McKinley, and Wilhams, 1983;
Kendelewicz et al, 1985; Houzay, Bensoussan, and Barthe, 1986). Most
metals appear to react to form phosphides, leading to the release of
phase-segregated indium. Some metals, for example gold, tend to form
alloys with the group III element. Comprehensive reviews of these inter­
actions have been given by Brihson (1982b) and WiUiams (1985).
McKinley et al (1980) investigated the Fermi level shifts of clean
cleaved n-InP(llO) surfaces when silver was deposited, and showed that
the Schottky barrier height of around 0.5 eV was largely developed for
an overlayer thickness of a few Angstroms. Detailed studies of Fermi-
level stabilization for a range of metals have been reported by BriUson et
al (1982) and their data for n-InP are presented in Fig. 2.7. The Fermi

6 8 10 12 14 16 18
Metal layer thickness (A)
Fig. 2.7 Position of the Fermi level below the conduction band edge as a function
of metal coverage for Ti, Al, Ni, Au, Pd, Ag, and Cu on InP(llO) cleaved surfaces. The
Ef position is derived from absolute In-Ad and P-2p binding energies. (From Brillson
et at., 1982. Copyright American Institute of Physics.)

68
level at the clean surface is located close to the conduction band and
moves away from this position with increasing metal coverages. In many
cases the Fermi level position has not reached its final position even for
thicknesses of 10 A, and the shifts cover a broad range up to 0.7 eV
below the conduction band. The shifts show no simple dependence on
work function but there is evidence that those metals with the largest
work functions (or electronegativities) yield the highest barriers.
Wilhams, Varma, and Montgomery (1979) and Spicer et al. (1980a)
concluded that most metal-InP interfaces are non-abrupt and dis­
ordered, with a high density of defects which would be capable of
pinning the Fermi level at the interface. This will be considered further in
Section 2.2.4.

2.2.3 Thick metal contacts on gallium arsenide and indium phosphide

There have been extensive studies of electrical contacts with thick metal
layers on III-V semiconductors; these have been reviewed by Robinson
(1985). In this section, attention will be confined largely to those studies
where metals were deposited on characterized and well-controlled semi­
conductor surfaces; contacts to chemically treated surfaces will be
mentioned but not considered in detail. Again, the most complete studies
have been carried out for GaAs.

2.2.3.1 Clean gallium arsenide The first results for metals on clean
cleaved GaAs surfaces were those of Spitzer and Mead (1963) who used
the photoelectric and capacitance methods to measure the Schottky
barrier heights of a series of metals deposited on n- and p-type crystals.
Similar measurements were made by Smith (1969a), and more recently
by Newman et al. (1985) and Ismail, Palau, and Lassabatere (1984) for
n-type crystals. The data are summarized in Table 2.3. Again it may be
seen that the barrier heights obtained by the C/V method are generally
slightly larger than those determined from the 1/V technique. For those
situations where barrier heights have been measured on both n- and
p-type crystals it is found that the sum + is generally close to the
value of the band gap, i.e. the Fermi level position is the same for a given
metal on both types of crystal. Little information is available regarding
the crystallographic nature of the metal contacts whose barrier heights
are hsted in Table 2.3, but it is likely that most, if not all of them, are
polycrystalhne except for those made by M B E . It is of interest to
compare the values of ^ obtained here with those for ultra-thin metal
layers considered in the previous section. Figure 2.6(a) shows such a
comparison (after Newman et al., 1986). It may be seen that, in general,
barrier heights derived for thick and thin layers of a given metal are in
good agreement, the worst case being for A l and In. Notice that no

69
Table 2.3 Schottky barrier heights for various metals on clean n-GaAs(llO) and (100) surfaces

(110)

Newman era/. (1985) Mead and Spitzer (1964) (100): Waldrop (1984) Other

Metal I/V C/V C/V PR 1/V C/V I/V C/V

Cu 0.87 0.94-1.08 0.83-0.90 0.82 0.96 0.96 0.85' 0.85'

Pd 0.85 0.88 0.91 0.93
Ag 0.85-0.90 0.95-0.99 0.90-0.95 0.88 0.90 0.89 0.82" 1.03"
0.90' 0.88'
Au 0.92 0.99-1.05 0.93-0.98 0.90 0.89 0.87 0.88" 0.98"
0.94' 0.95'
Al 0.80-0.85 0.84-0.93 0.78-0.92 0.80 0.85 0.84 0.85" 0.87"
0.78" 0.78"
Ti 0.83 0.83
Mn 0.72 0.75 0.81 0.89
Pb 0.80 0.91
Bi 0.77 0.79
Ni 0.77 0.82 0.77 0.91
Cr 0.67 0.72 0.77 0.81
Fe 0.72 0.75
Mg 0.62 0.66
Pt 0.90-0.98 0.86
Be 0.82 0.81
Sn 0.77 0.82 0.68-0.74 0.79' 0.80'
Ba 0.94 1.05'
In 0.76' 0.77'
Co 0.76 0.86
Sb 0.83" 0.74"

" (100): Svensson and Anderson (1985).

"(100): Missous, Rhoderick, and Singer (1986a).
'(110): Smith (1969a).
"(110): Ismail, Palau, and Lassabatere (1984).
barrier heights for thick G a contacts on GaAs are included in Fig. 2.6. In
fact, gallium seems to display anomalous behaviour since it has been
reported to form ohmic contacts to both n-type (Woodall, Lanza, and
Freeouf, 1978), and p-type GaAs (Bachrach and Bianconi, 1978).
A n extensive study of Schottky barrier heights for a range of metals on
heat-cleaned (100) surfaces of n- and p-type GaAs crystals has been
made by Waldrop (1984). No information was provided to indicate
which reconstructed form was displayed by the clean semiconductor
surface, or whether or not the metal contacts were single crystal or
polycrystalline. The barrier heights estabhshed by / / F a n d C / F for n-type
crystals are presented in Table 2.3. Figure 2.8 shows barrier heights for
n- and p-crystals; again it may be seen that the largest barriers are for
electronegative metals and are very similar to the barriers measured for
metals on cleaved surfaces. In almost all cases + ^bp is nearly equal to
the band gap though some, particularly M g , seem to deviate from this
behaviour. The barrier heights measured by Waldrop show no simple
dependence on the metal work function. This has been further confirmed
by Barret and Massies (1983) who succeeded in growing epitaxial single
crystal aluminium layers with two different orientations on the
GaAs(lOO) surface. Though the work function of the two aluminium
layers differed by 0.35 eV, both yielded precisely the same value of
Schottky barrier height.
It has been reported that barrier heights for aluminium deposited on
(100) faces of GaAs depend on the particular reconstruction adopted by

£ . =0
Cu Pd Ag Au Al Ti Mn Pb Bi Ni Cr Co Fe Mg
Fig. 2.8 Schottky-barrier height of ideal metal contacts on heat-cleaned GaAs(lOO) as
measured by the UV method: • n-type contacts (right-hand scale); O p-type contacts (left-
hand scale). (From Waldrop, 1984. Copyright American Institute of Physics.)

71
the surface before the metal is deposited (Cho and Dernier, 1978; Wang,
1983). This result is not supported by the detailed studies of Missous,
Rhoderick, and Singer (1986c) who deposited epitaxial aluminium on
(100) surfaces of n- and p-type layers of GaAs grown by M B E . The
aluminium films were almost perfect single crystals, with the (100) A l
plane parallel to the (100) GaAs plane but the two lattices rotated by 45°
relative to each other. Provided the epitaxial aluminium films were grown
under the most stringent vacuum conditions it was found that the diodes
formed were almost perfectly ideal, with CIV and IIV yielding identical
values of the barrier height, within the experimental error of ± 0.01 eV.
These rigorous conditions included switching off all ion-gauges and using
a cryopump to maintain an extremely low background pressure of C O
and CO2 during growth. Deviations of the growth conditions from the
most stringent led to non-ideal diodes and differences in the barrier
heights measured by the CIV and IIV techniques. The barrier heights did
not show any dependence on the reconstruction of the GaAs(lOO)
surface. It was also demonstrated that, to within the experimental error,
^bn + ^bp was equal to the band gap of GaAs. The same group (Missous,
Rhoderick, and Singer, 1986a) also studied the Schottky barrier heights
for antimony layers deposited on n- and p-type crystals of GaAs grown
by M B E . The barrier heights reported are presented in Table 2.3 and
were found to be very close to those reported for aluminium. The
detailed nature of the interfaces for these epitaxial A l - G a A s interfaces
has been investigated by Eaglesham et al. (1987) using transmission
electron microscopy. They were found to be abrupt, with almost perfect
registration between the A l and GaAs lattices, apart from the inevitable
misfit dislocations which were necessary to accommodate the 1% differ­
ence between the A l and the GaAs lattice parameters (after allowing for
the 45° rotation).
Missous, Rhoderick, and Singer (1986b) also measured Schottky
barrier heights for polycrystaUine aluminium layers on n-GaAs crystals
grown by M B E . The barrier heights prior to annealing were identical to
those reported for the epitaxially grown A l contact, though the IIV
characteristics were not so nearly ideal.

2.2.3.2 Oxidized gallium arsenide There have been a large number of

measurements of Schottky barrier heights for a wide range of metals
deposited on chemically etched and air-cleaved n- and p-type GaAs.
These have been reviewed by Robinson (1985), and almost all lie in the
range between 0.65 eV and 1.0 eV for n-type crystals. Ismail, Palau, and
Lassabatere (1984) and Kendelewicz et al. (1985) report little difference
in the values of measured by the IIV method for Ag, A u , and Pd on
air-cleaved GaAs compared with data for the same metals on surfaces
cleaved in ultra-high vacuum. The influence of interfacial oxide layers

72
generated by exposure of the GaAs cleaved surface to water vapour was
studied by Montgomery and Williams (1982) for the case of gold
contacts, and again little influence of the oxide on ^ was observed.
Perhaps the most detailed study of barriers formed by metals on
chemically etched n-GaAs crystals was carried out by Smith (1969a) and
the data obtained are presented in Table 2.4. It may be observed that
similar barriers are produced for a given metal on the (TTT) and (110)
faces, though the barriers for gold on the (100) and (111) surfaces are
somewhat lower than on the other surfaces. A dependence of barrier
height on orientation was also reported by Kahng (1964) for A u - G a A s .
Comparison of the barrier heights with corresponding ones obtained by
Smith for metals on clean cleaved (110) surfaces in Table 2.3 show that
the barrier heights for In and Sn are somewhat lower on the chemically
etched surfaces. Though oxygen at the interface appears to have httle
effect on the A u - G a A s barrier heights, sulphur and selenium atoms at
the interface do appear to have a significant effect (Massies et ai, 1980;
Waldrop, 1984).
Table 2.4 Experimental barrier height data on etched n-type
GaAs (from Smith, 1969a)

Metal Surface Barrier height (eV)

From From Photoresponse

C/V I/V

Au (TTl) 0.95 0.91 0.90

Cu (111) 0.84 0.82 0.82
In (111) 0.64 0.62
Ni (111) 0.81 0.77 0.78
Pb (111) 0J7 0.76
Sn (111) 0.67 0.67 0.68
Au (110) 0.93 0.89 0.90
Cu (110) 0.86 0.84
In (110) 0.64 0.64
Sn (110) 0.68 0.67
Au (100) 0.83 0.79 0.81
Au (TIT) 0.89 0.87

It could be argued that the existence of an insulating oxide layer

separating the metal and semiconductor should reduce considerably the
density of any M I G S in the semiconductor. Likewise, it could be argued
that the generation of defects in the semiconductor by the adsorbed
metal would be reduced, since the adsorbed atoms do not come into
contact with the free semiconductor surface. However, it is not certain
that the thin oxide does completely separate the metal and semi­
conductor, since diffusion of metal atoms through the oxide and chemical

73
reactions between the metal and the oxide can occur. Strong chemical
reactions between metals and oxide layers on GaAs have been observed
and studied by Kowalczyk, Waldrop, and Grant (1981). Prior to the
deposition of metals, the GaAs(lOO) surfaces were covered by oxide
layers around 10 A thick, with a composition of either AS2O3 plus
G a 2 0 3 , or G a 2 0 3 alone, depending on preparation procedure. It was
observed that Cu, Ag, and A u did not lead to a reduction of the oxides,
whereas Mg, A l , T i , and Cr led to complete reduction of the oxides and
the formation of new compounds such as AI2O3 and MgO. The experi­
ments also demonstrated that strong band bending exists at the free
surface of n-type GaAs covered by the mixed A s 2 0 3 - G a 2 0 3 oxide, and
that this band-bending does not change substantially when metals are
deposited. The band-bending at the free surface covered with just G a 2 0 3
is smaU, and increases when metals are deposited, eventually yielding the
same Fermi level position at the surface as that observed for surfaces
covered by the mixed oxide.

2.2.3.3 Indium phosphide A s was mentioned in Section 2.2.2.2, metal

contacts on n-InP almost invariably yield relatively small Schottky
barrier heights. Furthermore, the reported barriers appear to be more
dependent on the detailed chfemistry of the interface than for metals on
GaAs. Barrier heights on atomically clean cleaved (110) surfaces have
been investigated by a number of groups. Williams, Montgomery, and
Varma (1978) reported barriers in the range 0.4-0.5 eV for A u , Ag, and
C u on n-type crystals and substantially lower barriers for the more
chemically reactive metals A l , Fe, and N i . Ismail, Palau, and Lassabatere
(1984) reported ohmic contacts for A l on clean cleaved n-InP and
substantially larger barriers for A g and A u . They also studied the same
metal on n- and p-type crystals and confirmed that + ^bp is roughly
equal to the band gap. Newman et al. (1985) have also measured barrier
heights for a range of metals on clean cleaved (110) surfaces, and by
utilizing small area diodes, were able to measure barrier heights as small
as 0.3 eV by the I/V method. The data of various groups for metals on
clean cleaved n-InP surfaces are presented in Table 2.5. Newman, van
Schilfgaarde, and Spicer (1987) have investigated a large number of thick
metal contacts on cleaved p-type InP (110) surfaces and have found that
the sum + ^bp is i n all cases equal to the band-gap within the experi­
mental error, indicating the same Fermi level position on n- and p-type
crystals.
It is important to emphasize that barrier heights reported by different
workers for a particular metal show considerable scatter, by as much as
0.1-0.2 eV in some cases. Photoemission studies, described in Section
2.2.2, clearly show that chemical reactions nearly always occur at the
interfaces between metal contacts and InP, leading to mixed phases at the

74
Table 2.5 Measured Schottky barrier heights for metals on clean cleaved n-InP

Metal Newman et al. Williams et al Ismail et al. Others

(1985) (1977,1978,1986) (1984)

IfV I/V CfV 1/V

Ag 0.54 0.42 0.47 0.515

Cr 0.45
Cu 0.42 0.49
Au 0.42 0.43 0.50 0.43
Pd 0.41
Mn 0.35
Sn 0.35 ohmic-0.4"
Al 0.325 ohmic ohmic 0.22"
Ni 0.32 ohmic
Ga 0.6
In 0.35

'Humphreys et a/. (1985).

"Slowick, Brillson, and Richter (1986).

interface. It is likely that the rather large scatter in the barrier height
measurements for a given metal is associated with the degree to which
different phases form and this, in turn, is dependent on precisely how the
contact is fabricated.
Many metals, for example aluminium, react with the surface in such a
way as to release phase-segregated indium, and Ismail et al. (1986) have
demonstrated that excess indium at the interface can significantly
influence the measured barrier heights. Williams, Montgomery, and
Varma (1978) and BriUson et al. (1982) suggested that the measured
Schottky barrier heights were related to the interface reactions, being
lower on n-type crystals for those metals which tend to react to form
phosphides. However, more recent studies by Maani et al (1984),
Newman et al (1985), and Williams et al (1986b) show that this view is
an oversimplification. Gallium contacts interact with the clean InP
surface in a manner similar to aluminium, yet ^ for galhum contacts to
n-InP is - 0 . 6 e V compared to ~0.2 eV for aluminium (Maani et al,
1984) . Metals such as gold and silver yield ohmic or low barrier contacts
when deposited on surfaces subjected to a light ion bombardment and
annealing cycle (Farrow et al, 1978; Houzay, Moison, and Bensoussan,
1985) . According to Farrow and his co-workers such surfaces generaUy
have small droplets of metallic indium on them, and it is probably the
presence of these that leads to the low barriers, since indium is known to
yield low barrier heights on clean InP.
Electrical contacts to oxidized InP have been extensively investigated,
but there is poor agreement between the results of different groups. In
some cases it has been reported that when metals are deposited on air-
exposed faces identical barrier heights are obtained to those for clean
surfaces (Newman et al, 1985). In other cases exposure of the clean

75
cleaved surfaces to air prior to the deposition of gold and silver contacts
appeared to reduce the barrier heights (Williams, Varma, and McKinley,
1977). Similarly, contradictory reports appear in the literature regarding
barrier heights for metals on chemically etched surfaces. Several groups
report barriers for A u on chemically etched n-IilP in the range 0.43-0.55
eV (Kim et al, 1976; Roberts and Pande, 1977; Morgan, Howes, and
Devlin, 1978; Hokelek and Robinson, 1983; Williams, Varma, and
McKinley, 1977; Ismail, Palau, and Lassabatere, 1984), while Kamimura,
Suzuki, and Kunioka (1980) obtained surprisingly high barrier heights of
around 0.83 eV. Recently, Sa and Meiners (1986) have reported barrier
heights of 0.92 eV for mercury contacts on chemically etched n-InP.
To summarize, Schottky barrier heights measured for metals on both
clean and oxidized InP show considerable scatter and are more severely
affected by chemical processes at the interface than for corresponding
metals on GaAs. The barriers appear to be influenced by phase-
segregated indium at the interface and by the formation of compounds
such as Ga^In]_^P which complicate the interpretation of the experi­
mental data. The oxide layer on InP is also a complex one (see Wilmsen,
1985). The most thermodynamically stable oxide is InP04, which is a
wide band gap insulator. However, semiconducting oxides such as In203
may also be formed at the InP surface under certain circumstances. A s
pointed out by Williams, Varma, and McKinley (1977), the variation in
barrier height for a given metal on chemically etched InP is almost
certainly due to variations in the composition of the interfacial oxide
layer. It is reported that metals such as aluminium reduce the oxide,
generating aluminium oxide and indium and again producing a mixed-
phase interface (Wilhams et al, 1986b). Silver has been reported to be
unreactive with the clean cleaved InP surface (McKinley, Parke, and
Williams, 1980) but appears to react strongly with the oxide layer on InP,
again giving rise to phase-segregated indium and a mixed-phase inter­
face. Clearly the oxide layer on InP does not act as a barrier to chemical
reactions. The complicated behaviour of metals on InP is unfortunate
from an applications point of view, since it makes it difficult to form good
rectifying contacts to n-type crystals.

2.2.4 Discussion—metals on gallium arsenide and indium phosphide

Schottky barrier heights measured by different techniques for thick and

thin metal layers on gallium arsenide are in broad, though not perfect,
agreement. Barriers measured by different groups for metals on indium
phosphide show a greater spread of values and appear more dependent
on how the contact is made. For most metals on GaAs the Fermi level at
the interface lies in the range 0.7-0.9 eV below the conduction band.

76
whereas for metals on InP this range is 0.2-0.6 eV. The most electro­
negative metals tend to yield the largest barriers on n-type crystals.
Furthermore, for the most thoroughly studied case, namely A l on GaAs,
the barrier height is not sensitive to the metal work function (as
influenced by the crystallographic orientation of A l ) or to the nature of
the epitaxy associated with it, and does not appear to depend on the
surface reconstruction of the semiconductor. This imphes that the Fermi
level is strongly pinned by interface states and the influence of any
interface dipole layer is small by comparison.
Clearly there is strong Fermi level pinning by interface states at metal¬
GaAs interfaces, and it is natural to enquire as to the origin of these
states. A s pointed out in Chapter 1, there are no surface states in the
forbidden gap for clean cleaved GaAs and InP(llO) surfaces, and conse­
quently £p is not pinned at the free cleaved surface of these semi­
conductors. However, when a metal is adsorbed the structure of the
surface layer and the matching condition for the electron wave-functions
will inevitably change, and it may be possible for interface states to be
induced in the band gap. If this were the dominating mechanism one
might expect the interface state distribution, and consequently the
barrier height, to be dependent on the semiconductor surface recon­
struction, on whether the metal is deposited on (100) or (110) surfaces,
and on the metal work function. This is evidently not the case for A l on
GaAs.
Woodall and Freeouf (1982) pointed out that interdiffusion and
chemical reactions between metals and GaAs or InP often lead to inter­
faces which are anion rich. They suggested that the barrier heights are
determined by anion-rich clusters at the metal-semiconductor interface
in accordance with the simple Schottky-Mott theory, and estimated the
magnitudes of these barriers for a range of semiconductors. It seems
hkely that if large regions rich in arsenic exist at the A l - G a A s interface
they would influence barrier heights; however, for the most perfect
epitaxial interfaces prepared by M B E there is no evidence to indicate the
existence of such clusters. Nevertheless, at this stage one cannot discount
this 'effective work function' model, and it is possible that mixed phases
at the metal-InP interface may account for the lack of reproducibility of
barrier heights measured for this semiconductor.
The 'unified defect' model was put forward by Spicer et al. (1980a,b)
following a series of experiments which included the observation that
thin layers of A l , In, and G a on GaAs appear to lead to different Fermi
level pinning energies for n- and p-type crystals. They interpreted this
result in terms of discrete donor and acceptor levels at 0.9 and 0.65 eV,
respectively, below the conduction band. This result is at variance with
those for thick metals on GaAs where + ^bp is nearly equal to the
band gap, indicating that the same interface states control the position of

77
Ep for n- and p-type crystals. It is therefore necessary to ask further
whether the photoemission technique does in fact yield the correct
barrier height. As pointed out earlier, non-uniform surface potentials
associated with cluster formation can lead to complications in the
interpretation of photoemission data, but this is not believed to entirely
account for the 0.2 eV observed difference in the saturation values of the
barriers measured for n- and p-GaAs (see Newman et al, 1986).
Zur, M c G i l l , and Smith (1983) explained the difference between the
results for thick and thin metal films in the following way. If one assumes
that in the case of submonolayer coverage there is a neghgible charge on
the adatoms, the charge on the defect states must balance that in the
depletion layer, which means that only acceptor-like states can be
operative in the n-type case and only donor-hke states in the p-type case.
The Fermi level cannot therefore lie below the acceptor level in n-type or
above the donor level in p-type material, resulting in two pinning
positions. But with a thick metal film, there can be a comparatively large
charge of either sign on the surface of the metal, and since it is now the
charge on the defect states plus that on the metal that balances the charge
in the depletion layer, either the donor or the acceptor level may lead to
pinning as described in Section 1.4.2. In this case one should observe the
same pinning position in both n-type and p-type material.
However, it is not obvious that the assumption that the metal adatoms
cannot hold a significant charge is a valid one. They will have donor and
acceptor states associated with them, and these may lie within the
forbidden gap. Furthermore, according to the unified defect model the
number of adatoms exceeds that of the defects which they generate. If the
possibility of charge residing on the metal atoms is taken into account,
the donor (or acceptor) states should be able to pin the Fermi level in
both n-type and p-type samples, as in the thick-film case. In principle,
one should be able to explain the pinning behaviour of the donor and
acceptor levels in terms of the two-level extension of the Bardeen model
developed in Section 1.4.2, together with the field dependence of the
barrier height discussed in Section 1.4.3, the latter being necessary to
explain the dependence on the doping of the semiconductor. However,
attempts in this direction have met with only limited success (see, for
example, Spicer et al, 1984), and it must be admitted that the
appearance of different Fermi-level positions when ultra-thin metal films
are deposited on n- and p-type specimens of a particular semiconductor
has not yet been satisfactorily explained.
Several groups have considered the origin of defect levels which may
lead to Fermi level pinning in GaAs and InP. Theoretical estimates of
pinning levels based on anion and cation vacancies (Daw and Smith,
1980, 1981) and antisite defects (Allen and Dow, 1982; Ahen, Sankey,
and Dow, 1986) are broadly consistent with the experimental data. They

78
predict pinning energies close to mid-gap for GaAs and in the upper part
of the band gap for InP. However, such calculations are not particularly
accurate, and it would be unwise to place too much rehance on them. It is
also important to emphasize that there is no absolutely conclusive
evidence to date that defects are responsible for the Fermi level pirming
behaviour at interfaces of metals with GaAs and InP.
The contribution of M I G S to the barrier formation in GaAs and InP is
also difficult to assess. The positions of the neutral levels calculated by
Tersoff (1984, 1985) (see Section 1.5.2) are 0.78 and 0.58 eV below the
conduction band edges in GaAs and InP, respectively, with a
comparatively small dependence on the metal. Zang, Cohen, and Louie
(1985, 1986) have considered in more detail the way that Fermi level
pinning by MIGS may be influenced by the atomic arrangement at the
interface and conclude that, for A l on GaAs (HO), the crystaUographic
structure has only a small effect, of the order of 0.1 eV. It may therefore
be concluded that the Schottky barrier heights established for most
metals on clean GaAs and InP are at least not inconsistent with the
M I G S model. Silver is one of the least chemically reactive metals on InP
and the barrier height of 0.55 eV reported for silver on clean (110)
surfaces is close to the value of the neutral level predicted by Tersoff. It
could then be argued that the lower barriers observed for other metals
are associated with lack of abruptness of the interface due to inter-
diffusion and chemical reactions between the metal and the semi­
conductor. This would be consistent with the observation that the
effective Schottky barriers of silver contacts on n-InP crystals are
reduced upon armealing, and at the same time interdiffusion occurs
across the interface (Williams, Varma, and McKinley, 1977; McKinley,
Parke, and Williams, 1980). However, at the present time there exists no
unambiguous evidence in support of M I G S and their importance is
therefore difficult to evaluate.

22.5 Metal contacts to other III- Vsemiconductors

2.2.5.1 Binary compounds Extensive studies of Schottky barrier

formation on single crystals of gallium phosphide were reported by
Cowley (1966). The measured barrier heights were in the energy range
1.0-1.4 eV for n-type crystals and the measured values showed a
substantial dependence on surface preparation. Smith (1969b) also
noticed that consistent results for A u on GaP(IlT) surfaces could only
be obtained if the substrates were heated to 120°C during evaporation of
the metal. He measured barriers of 1.33 ± 0.02 eV and 0.88 ± 0.03 eV
for gold on n- and p-type crystals, respectively. The sum of the two
barriers is in good agreement with the band gap of gallium phosphide.
Brief studies of microscopic interactions of several metals with clean

79
gallium phosphide have been carried out by Hiraki et al (1979); the
interfaces, in general, appear less complex than for indium phosphide.
Studies of the microscopic chemical reactions and surface Fermi level
positions have been reported by Lindau et al. (1978) for thin layers of a
few metals on clean cleaved (110) gallium antimonide surfaces (GaSb has
a band gap of 0.7 eV). Gold gives rise to strong reactions, leading to the
formation of a A u - G a alloy together with antimony. For all the ad-layers
studied, namely Cs, Ga, Sb, and A u , as well as for oxygen, the surface
Fermi level was found to be located within 0.2 eV of the valence band
edge, i.e. in the lower half of the band gap. Poole et al. (1987) have made
Al-n-GaSb contacts by M B E in which the (thick) aluminium films were
deposited as single crystals. They found good rectifying characteristics
with a barrier height of 0.55 ± 0.02 eV as determined by both the 7/Fand
C / F methods. It is clear that there must be some pinning mechanism at
work which makes the barrier height on GaSb essentially independent of
the metal.
A study of surface Fermi level shifts for thin metal layers on heat-
cleaned InAs (100) surfaces has been carried out by Brillson et al.
(1986) for increasing metal coverages. The Fermi level stabilization
energy covers a range of 0.6 eV, even though the band gap is only 0.36
eV. Aluminium and gold contacts led to stabilized Fermi level positions
in the conduction and valence bands, respectively. In contrast, Walpole
and Nill (1971) and Mead and Spitzer (1964) observed gold contacts to
be ohmic on etched (100) surfaces of n-type InAs, corresponding to a
Fermi level in the conduction band, with C/V measurements showing a
substantial barrier on p-type material. Brillson et al. found that A u on
air-exposed InAs (100) surfaces also yields a Fermi level position in the
conduction band at the surface, at least for gold coverages up to a few
monolayers, so that with InAs the nature of the surface causes a very
drastic change in pinning behaviour.
Indium antimonide has been studied by Mead and Spitzer (1964) who
found barrier heights of 0.17 and 0.18 eV for gold and silver,
respectively, on n-type material. Korwin-Pawlowski and HeaseU (1975)
also investigated contacts to n-InSb and found the barrier properties to
be very sensitive to the etch which was used.

2.2.5.2 Ternary alloys Studies of metal contacts to aUoys of III-V

compound semiconductors have been increasing due to their importance
in opto-electronics and because the alloys can now be prepared fairly
routinely. Consider first the case of aluminium contacts to AljGa,_^As,
which have been investigated by several groups. The alloy consists of two
different cations but has a common anion, and the variation of the
barrier height with x is of interest in view of the so-called 'common anion
rule'. This 'rule' was proposed by McCaldin, M c G i l l , and Mead in 1976

80
and states that the Schottky barrier height for p-type compound
semiconductors is determined by the anion species (i.e. the group V
element) so that should be independent of a: in a system such as
AljjGai_^As, where only the cation changes with the compositional
parameter jc. Figure 2.9 shows experimental data from several groups for
contacts deposited on (100) surfaces of n-type Al^Gai_^As. The figure
shows Eg-^bn plotted as a function of x. According to the data of
Okamoto, Wood, and Eastman (1981) for aluminium, in which the layers
were prepared by M B E , it appears that Eg — jibn is constant for a: < 0.4 as
predicted by the common anion rule, but for jc>0.4 it increases. A t the
point, X" 0.4, the alloy changes from a direct to an indirect gap semi­
conductor. Similar studies by Missous (1987), however, indicate that £ g -
^bn increases across the whole compositional range, in disagreement with
the common anion rule. Contradictory data has also been reported for
gold on chemically etched Al^Gaj.^As surfaces. Gol'dberg et al. (1972)
report that E^-^^,„ is constant for < 0.4, but similar investigations by
Best (1979) show a contmuous variation across the whole compositional
range; this is illustrated in Fig. 2.9. The heights of barriers for gold on
GajtIn,_^P were investigated by Kuech and McCaldin (1980), and were

2.5

Ol 1 I I I 1 i 1 I I

0 0.2 0.4 0.6 0.8 1.0

GaAs X AlAs
Fig. 2.9 Comparison of tlieory and experiment for Schottky barriers on GaAs and AlAs
and their alloys: x Goldberg et al. (1972); O Best (1979); • Okamoto, Wood, and
Eastman (1981); theory based on anion vacancies (Daw and Smith, 1981);
theory based on pinning by antisite defects (Allen and Dow, 1981);
according to common anion rule. (From Robinson, 1985. Copyright Plenum Press.)

81
reported to yield a constant value of Eg-^bn^ in accord with the common
anion rule.
Brillson et al. (1986) have carried out a detailed photoemission study
of A l , In, Ge, and A u on heat-cleaned Ga^In]_^As (100) surfaces; their
data are presented in Fig. 2.10. It may be seen that there is no single
Fermi level pinning energy and the data are in complete disagreement
with the common anion rule. Their results are in contradiction with those
of Kajiyama, Mizushima, and Sakata (1973) on the same alloy; the latter
authors found that Eg-^bn remains approximately constant with x.

The compositional dependence of ^b for A u on n-Al^Ga,_^Sb has been

studied by Chin, Milano, and Law (1980), who found £g-^bn to increase
with increasing x. Studies on Ga^IUj.^Sb have been reported by Keeler,
Roth, and Fortin (1980), and the system Au-GaAsi-^^Pj. has been studied
by Rideout (1974) who found Eg-^bn^O.SS eV, independent of x, in
apparent contradiction to the common anion rule, since in this case the
anion element is changing. Escher et al. (1976) have reported measure­
ments of barriers for gold on p-GayIni_^Asi_^P^ as x was varied. The
aUoy was prepared on an InP substrate and the parameter y was varied to
maintain lattice matching. It was found that ^bp increased monotonically
from x= Oto x= I with the ratio ^bpZ-'^g remaining roughly constant.

2.3 Metals on other semiconductors

Metal contacts to semiconductors outside those in groups IV and III-V

82
have received relatively little attention. Goodman (1964) made a
thorough study of contacts on etched single crystals of cadmium
sulphide, and Spitzer and Mead (1963) reported barrier height measure­
ments for a number of metals on n-CdS cleaved in a vacuum of 10~^ Torr
in a stream of the appropriate metal. Barrier heights ranging from ohmic
for aluminium to 0.88 eV for platinum were obtained by the photo-
response and capacitance methods. Kusaka, Matsui, and Okazaki (1974)
found that the A u - C d S barrier height for etched surfaces depended on
the surface polarity, being 0.83 eV on the (0001) and 0.73 eV on the
(0001) surfaces, respectively. Since the etching behaviour is different for
these two surfaces, it is not known whether the difference in barrier
heights is an intrinsic property of the semiconductor surface or arises
from a difference in the properties of the interfacial layer. Brillson
(1982a) studied interfaces formed by the deposition of several metals on
CdS, CdSe, ZnS, ZnSe, and ZnTe, and showed that most of the inter­
faces are not abrupt. Brucker and Brillson (1981) have confirmed that
A u forms large barriers on clean n-CdS and n-CdSe and that A l forms
ohmic contacts on both materials. The formation of metal contacts on
n-CdTe has been extensively studied by Patterson and Wilhams (1982)
and Dharmadasa, Herrenden-Harker, and Wilhams (1986). Again A u
and A l form high and low barriers, respectively, when deposited onto
clean cleaved (110) n-CdTe surfaces. More generally, the barrier heights
on CdTe were found to be highly dependent on the metal, but there
appeared to be no simple dependence on parameters such as work
function or electronegativity. The barriers were also found to be
dramatically influenced by the presence of oxide layers on the surface.
Soft X-ray photoelectron spectroscopy demonstrated that most
interfaces were non-abrupt. Furthermore, several metals were found to
react strongly with the native oxide layer generated by chemical etching
or by exposure to air, and the measured barriers in several cases
appeared dependent on the degree of reaction with the oxide. It appears
that the measured barriers for metals on CdTe may be significantly
influenced by the presence of tellurium at the interface, which accounts
for the wide variation of Schottky barrier heights reported in the
hterature for chemically etched CdTe crystals.
In spite of their importance as photodetectors, not much is known
about the I V - V I compounds (lead sulphide, etc). Nill et al. (1970) have
studied contacts to lead telluride and reported that surface states are
unimportant, so that the barrier heights conform approximately to the
Schottky-Mott limit. Since the electron affinity of lead telluride is 4.6 eV,
which is greater than the work function of either lead or tin, these metals
should give ohmic contacts on n-type material, and rectifying contacts on
p-type material with barrier heights greater than the band gap. This
prediction is in accordance with the C/Fcharacteristics.

83
 Soft X-ray photoemission has been apphed by Hughes, McKinley, and
Wilhams (1982) to investigate the interaction of metals with the semi­
conductor gallium selenide. This material crystallizes with a layered
structure, somewhat like graphite, where each layer consists of two
planes of galhum atoms sandwiched between plans of selenium atoms.
Because of its layered structure the material can be cleaved to yield
highly perfect surfaces. Hughes and his co-workers showed that metals
such as aluminium chemically react with the GaSe surface liberating
phase-segregated gallium, whereas metals such as gold, silver, and tin are
far less reactive. The unreactive metals lead to large variations in
Schottky barrier height roughly consistent with the Schottky-Mott limit.
The reactive metals, however, did not show such a variation. This was
taken as evidence that the defects generated by the strong interface
reactions lead to strong pinning of the Fermi level for the reactive metals.
Apart from silicon and germanium, there is only scanty information
about group I V semiconductors. Gold contacts to p-type diamond have
been examined by Mead and Spitzer (1964) and by Glover (1973), who
found barrier heights of 1.35 and 1.73 eV, respectively. The former
authors have also looked at gold and aluminium contacts on n-type
silicon carbide and found barriers of 1.95 and 2.0 eV, respectively.
In view of their immense practical importance in the pre-silicon and
pre-germanium era, it is surprising how httle is known about contacts to
cuprous oxide and selenium. Early estimates of the height of the C u -
CujO (p-type) barrier, prepared by oxidizing copper or by reducing
cuprous oxide, gave barrier heights of about 0,4 eV (Henisch, 1957),
while measurements by Assimos and Trivich (1973) gave a value of 0.75
eV. Lanyon and Richardson (1971) have found the barrier between
p-type selenium and a series of metals to lie in the range 0.30-0.55 eV.

2.4 Summary and conclusions

The application of surface and interface science techniques, during

recent years, has shown clearly that interfaces formed between metals
and semiconductors are complex regions whose physical properties are
highly dependent on the preparation procedure. In many instances the
metal contact is deposited onto surfaces covered by unknown
contaminants which may give rise to interface states and influence both
the mechanical and electrical properties of the contact. Very often, the
metal is deposited onto insulating oxide layers on the semiconductor, for
example Si02 on silicon. In some cases the oxide layer has the effect of
'decouphng' the metal and semiconductor so that the interface states at
the oxide-semiconductor interface are not influenced by the metal. In
such cases a qualitative interpretation of the experimental results relating

84
to Schottky barrier heights at the interface has usually been possible in
terms of the Bardeen model. In some situations the metal contact itself
can react chemically with the oxide layer, giving rise to a complicated
mixture of phases at the interface; in other cases the oxide layer is
perhaps only thick enough to partially decouple the metal and serai-
conductor. In view of these complexities it would be most surprising if
any one model, for example the Bardeen model, were able fully to
account for the electrical barriers formed at metal-semiconductor
interfaces.
For understanding the fundamental physics of contacts to semi­
conductors the most important studies are those where metals have been
deposited onto atomically clean and ordered surfaces. In a few cases, for
example NiSij on sihcon and aluminium on gallium arsenide, it has been
possible to prepare epitaxial single crystal contacts with highly perfect
interfaces to the semiconductor. These studies, particularly those of Tung
for NiSiz on silicon, have demonstrated that the detailed form of the
interface crystallography does influence the Schottky barrier heights in
certain cases. Most contacts are polycrystaUine and detailed investiga­
tions have shown clearly that chemical reactions and interdiffusion are
common at intimate metal semiconductor interfaces, even when the
substrate is held at room temperature or below as the metal is deposited.
The products formed in the reactions are complex and are governed by
kinetic as well as thermodynamic considerations. They are highly
dependent on the details of the contact preparation procedure, and
therefore may be difficult to reproduce accurately. Clearly, where strong
chemical reactions do occur, it might be expected that parameters such
as the work function of the metal would be less relevant than those
associated with the reaction products.
In order to compare theoretical predictions with experimentally
measured Schottky barrier heights it is desirable to consider only those
measurements made on unreactive interfaces where interdiffusion across
the interface is minimal. Although interfaces between gold and III-V
semiconductors do not fully satisfy these criteria, several groups have
nevertheless adopted the barrier heights measured for gold contacts as a
basis for comparison with their theoretically derived barriers. A n
example is presented in Fig. 2.11. The points represent experimentally
determined barrier heights for gold on a range of III-V semiconductors
and their alloys, whereas the full line represents the predicted barriers
assuming that the Fermi level is pinned at the energies of anti-site
defects, in this case cations on anion sites (after Sankey et al, 1985). The
agreement in the trends between experiment and theory is remarkably
good. However, one should be cautious not to assume on this basis that
anti-site defects determine the heights of Schottky barriers in general.
Similar trends are observed if one assumes pinning by vacancies (Daw

85
 2.0

-0.51 I I \ I I
AlAs GaAs GaP InP InAs GaAs

Fig. 2.11 Fermi-level pinning energies for gold on a range of III-V semiconductors: •
experimental values; calculations based on pinning by antisite defects, i.e. Group
III atoms occupying Group V sites; O calculations based on pinning by metal-induced gap
states, according to Tersoff (1985). (Based on Sankey et ai, 1985. Copyright American
Institute of Physics.)

and Smith, 1980, 1981), though the exact pinning energies in this case
are not well accounted for. Furthermore, the experimental data
presented in Fig. 2.11 are also well described by MIGS. Tersoff has
estimated the energy of the neutral level for a range of III-V and II-VI
semiconductors; the open circles in Fig. 2.11 correspond to these values
for the appropriate semiconductors. Assuming the Fermi level to be
pinned close to ^Q, it may be seen that the M I G S model is just as satis­
factory as the anti-site defect model.
The apparent agreement between what appear to be two quite different
models is, at first sight, rather surprising. However, as recently pointed
out by Flores and Tejedor (1987), the energy associated with a III-V
anti-site defect will follow closely the energy of the mid-point between
the cation and anion dangling bond energies. This mid-point energy
corresponds to the neutral level in the M I G S model. It is therefore
perhaps not surprising that the two models considered above lead to the
same trends for contacts to III-V materials, so that it is difficult to
distinguish between them simply by comparing theoretically predicted
barrier heights with experiment.
A general point to emerge from the wide range of studies reported in
the literature is that barrier heights are much more dependent on the
metal for contacts to larger band-gap ionic semiconductors than to low
gap covalent ones. Kurtin, McGiU, and Mead (1969) analysed a large
body of data using the relationship

where C is a constant and Xm the electronegativity of the metal. The

86
quantity 5 was found for a range of semiconductors by plotting an
approximately linear relationship between ^j, and Xm- The analysis
yielded small values of S for the covalent materials, but a value close to
unity for the large gap ionic materials such as Z n O and ZnS. This
transition was referred to as the covalent-ionic transition and much
effort has been made to undertstand its origin. Although Schlüter (1978)
has reanalysed the original data and shown that the covalent-ionic
transition is not well defined, it is clear that barrier heights for metals on
ionic materials are more dependent on the metal work function than for
the same metals on covalent semiconductors. It is important to stress that
this can only be taken as a rough guide; the data need considerable
updating and contain results on surfaces prepared by several different
techniques.
A possible explanation for the larger dependence of the barrier height
on the metal in ionic semiconductors has been put forward by Louie,
Chelikowsky, and Cohen (1977) in terms of MIGS as discussed in
Section 1.5.2. These authors considered aluminium contacts on silicon,
gallium arsenide, and zinc sulphide crystals. It was shown that, because
of the large band gap in ZnS, the M I G S penetrate a much shorter
distance into the semiconductor compared with those in the smaller gap
materials such as silicon (see Fig. 1.16). Furthermore, the calculated
interface state density in ZnS is about a factor of three smaller than in
GaAs. By incorporating these values in eqn (1.6) it may be seen that y
(or S) is likely to be significantly larger in ZnS than in Si or GaAs.
Phillips (1974) and Tersoff (1986) have also pointed out that Fermi level
pinning by M I G S will be less effective in the more ionic high band-
gap semiconductors, and have attempted to describe this in terms of
models involving the dielectric constant. Following the argument of
Tersoff, the parameter 5 is a measure of how effectively the interfacial
dipole is screened by the MIGS, and the screening depends on the
optical dielectric constant in such a way that S is proportional to E~^.
This means that S will be greatest for those solids such as CdS and ZnS
which have relatively small values of £<„. The model has been further
refined by Monch (1986).
At the present time one has to admit that the experimental data
available do not enable any single model of Schottky barrier formation to
be favoured above all others. In most cases it seems likely that the barrier
heights are determined by a combination of mechanisms, with the
relative importance of each depending on the method of preparation of
the contact, which determines the crystallographic and metallurgical
structure of the interface. Our general understanding of the mechanisms
governing barrier formation is unlikely to improve until more experi-
ments are carried out on systems which are as perfect and as well charac-
terized as NiSij on silicon and epitaxial aluminium on gallium arsenide.

87
The best advice we can give readers who wish to know what barrier
height will be obtained by depositing a particular metal on a particular
semiconductor is that they should consult one of the several data bases
which exist t to find out whether the combination has been investigated
already, and if so that they should then repeat the measurement under
their own experimental conditions.

tOn-line data bases which cover metal-semiconductor contacts are provided by Inspec
(Institution of Electrical Engineers, Station House, Nightingale Road, Hitchin, Herts SG5
I r J , UK). Reviews in book form containing data on contacts to silicon and gallium
arsenide are available from the EMIS Group of Inspec [EMIS Datareviews Series No. 2
(gaUium arsenide, 1986) and No. 4 (silicon, to appear in 1988)].

88
3 Current-transport mechanisms

3.1 Introduction

This chapter discusses the transport mechanisms which determine the

conduction properties of Schottky barriers. It assumes that a barrier has
been estabhshed as described in Chapter 1 and says nothing about the
factors which determine the height of this barrier except where they may
affect the //Frelationship.
The various ways in which electrons can be transported across a
metal-semiconductor junction under forward bias are shown schemati­
cally for an n-type semiconductor in Fig. 3.1. The inverse processes
occur under reverse bias. The mechanisms are:
(a) emission of electrons from the semiconductor over the top of the
barrier into the metal;
(b) quantum-mechanical tunnelling through the barrier;
(c) recombination in the space-charge region;
(d) recombination in the neutral region ('hole injection').
a

Fig. 3.1 Transport processes in a forward-biased Schottky barrier.

89
It is possible to make practical Schottky-barrier diodes in wiiich (a) is the
most important and such diodes are generally referred to as 'nearly ideal'.
Processes (b), (c), and (d) cause departures from this ideal behaviour.

3.2 Emission over the barrier

3.2.1 The two basic mechanisms

Before they can be emitted over the barrier into the metal, electrons must
first be transported from the interior of the semiconductor to the
interface. In traversing the depletion region of the semiconductor, their
motion is governed by the usual mechanisms of diffusion and drift in the
electric field of the barrier. When they arrive at the interface, their
emission into the metal is determined by the rate of transfer of electrons
across the boundary between the metal and the semiconductor. These
two processes are effectively in series, and the current is determined
predominantly by whichever causes the larger impediment to the flow of
electrons. According to the diffusion theory of Wagner (1931) and
Schottky and Spenke (1939), the first of these processes is the limiting
factor, whereas according to the thermionic-emission theory of Bethe
(1942) the second is the more important.
The difference between the two theories is well brought out by the
behaviour of the quasi-Fermi level for electrons in the conduction band
of the semiconductor.! According to the diffusion theory, the concentra­
tion of conduction electrons in the semiconductor immediately adjacent
to the interface is not altered by the applied bias. This is equivalent to
assuming that at the interface the quasi-Fermi level in the semiconductor
coincides with the Fermi level in the metal. In this case, the quasi-Fermi
level droops down through the depletion region as shown in Fig. 3.2.
This behaviour is in sharp contrast with the situation in a p - n junction
under bias, where the quasi-Fermi levels for both types of carrier are
generally assumed to be flat throughout the depletion region (Shockley,
1950).
The electrons emitted from the semiconductor into the metal are not in
thermal equilibrium with the conduction electrons in the metal but have
an energy which exceeds the metal Fermi energy by the barrier height
(~ 1 eV). They can be described loosely as 'hot' electrons, and the energy

tThe quasi-Fermi level ^ is a hypothetical energy level which is introduced to describe

the behaviour of electrons under non-equilibrium conditions. It has the property that it
predicts correctly the concentration of electrons in the conduction band if the electrons are
assumed to be in thermal equilibrium at the lattice temperature and the quasi-Fermi level is
used in the Feriiii-Dirac distribution function in place of the equilibrium Fermi level. It
also has the property that the electron current in the x direction is given by qnfi{d^/dx),
where n is the concentration of electrons and their mobility (see Shockley, 1950).

90
 1 IV 1
d

/
kTlq \ V )i
t

1 V

Fig. 3.2 Electron quasi-Fermi level in a forward-biased Schottky barrier:

according to diffusion theory, according to thermionic-emission theory. The
broken circle shows the energy distribution of electrons which make their last collision at a
distance / from the interface.

distribution of electrons on the metal side of the interface is as shown in

Fig. 3.3. Gossick (1963) has suggested that the hot electrons in the metal
can be thought of as a species of electron different from the ordinary
conduction electrons, and that they can be described by their own
quasi-Fermi level. A s the hot electrons penetrate into the metal, they
lose energy by colhsions with conduction electrons and the lattice and
finally come into equilibrium with the conduction electrons in the metal.
Their quasi-Fermi level falls until it ultimately coincides with the metal
Fermi level.t This process is rather hke recombination in a semi­
conductor. This viewpoint implies that the electron quasi-Fermi level at
the interface does not have to coincide with the metal Fermi level, and it
is now possible to envisage that the quasi-Fermi level remains flat
through the depletion region as in a p - n junction (see Fig. 3.2). This is
the assumption made in the thermionic-emission theory of Bethe (1942).
Bearing in mind that the gradient of the quasi-Fermi level provides the
driving force for electrons, one can see that, according to the diffusion
theory, the main obstacle to current flow is provided by the combined
tGossick's concept of a separate quasi-Fermi level for the hot electrons in the metal is
strictly valid only as long as they have an almost MaxwelHan velocity distribution. This
implies that the distribution must also be nearly isotropic. In practice this will not be so
because the electrons are emitted into the metal within a narrow cone (see p. 98) and do
not acquire an almost isotropic distribution until they have made several collisions within
the metal. Moreover, as we shall see in Section 3.2.3, the assumption of an isotropic
velocity distribution is also not valid in the semiconductor close to the interface, so it is
really more accurate to picture the quasi-Fermi level as undefined near the interface with
an almost discontinuous drop from its position in the semiconductor to within the
metal.

91
 «(£)

Fig. 3.3 Energy distribution of electrons emitted into the metal.

effects of drift and diffusion in the depletion region whereas, according to

the thermionic-emission theory, the bottleneck lies in the process of
emission of electrons into the metal. In practice, the true behaviour will
lie somewhere between the two extremes of the diffusion theory and the
thermionic-emission theory.

3.2.2 The diffusion theory

To derive the current/vohage characteristic according to the diffusion

theory, we write the current density in the depletion region in the usual
way as
dn
J=qniuS' + qD, (3.1)
dx
where n is the concentration of electrons in the n-type semiconductor, fi
their mobility, Z)„ their diffusion constant, ^ the electric field in the
barrier, and — g the charge on the electron.
It is only possible to write the expression for the current in the simple
form of eqn (3.1) if it is legitimate to use the concepts of a mobility and a
diffusion constant which are independent of the electric field. This is not
the case near the top of the barrier where S' has its maximum value.
Moreover, if the electron distribution function changes appreciably
within the electron mean free path, as is the case near the top of the
barrier, it is not even justifiable to split up the current into drift and
diffusion components which are independent of each other. However, if
such a simplification is not made, the analysis becomes extremely
unwieldy; we shall therefore assume the validity of eqn (3.1), but it must
always be borne in mind that the accuracy of the analysis ultimately rests
on the truth of this assumption. This point is considered further in
Section 3.2.3.
We now introduce the quasi-Fermi level for electrons C„ defined by

92
 n = N^txp[-q(E^-Q/kT} (3.2)
where is the effective density of states in the conduction band, is
the energy of the bottom of the conduction band, and we have used the
Boltzmann approximation to the Fermi-Dirac function. Making use of
Einstein's relationship ju/D^ = q/kT, it is possible to write eqn (3.1) in the
form

J=qMn^ (3.3)

which shows that the gradient of ^„ supplies the 'driving force' for
electrons. Combining eqns (3.2) and (3.3) gives

J= q^Kexp{-q{E,-^„ykT}
dx

= kTfiN^cxpi - qE^kT) £ {exp(qUkT)}. (3.4)

Integrating eqn (3.4) between x = 0 and x=w gives

tv
cxp{qEJkT)dx= [txp{qUkT)Z
kTfiNj
= cxp{ql;^{wykT\ - expiqUOykT}. (3.5)
Hence the current/voltage relationship is completely determined if E^ is
known as a function of x and if the values of ^„(0) and ^„(H') can be
specified for a particular value of applied bias. It is convenient to take
the Fermi level in the metal as the zero energy level so that ^„(H')= V,
since the applied voltage F is equal to the difference between the Fermi
levels at the terminals of the diode expressed in electron volts. (We
ignore any voltage drop in the neutral region of the semiconductor
between x = w and the ohmic contact.) The assumption made in the
diffusion theory is that the concentration of electrons on the semi­
conductor side of the interface is unaltered by the application of bias, i.e.
that ^„(0) = 0. Since the driving force for electrons is the gradient of
the assumption that ^„(0) = 0 is equivalent to assuming that the impedi­
ment to current flow is provided entirely by the processes of drift and
diffusion in the depletion region.
To evaluate the integral on the left-hand side of eqn (3.5), we shall use
the depletion approximation (see Appendix A ) with a constant donor
density Ng. Equation (A.5) then gives

E,(x) = ^ + ^(x'-2wx) (3.6)

93
where is the barrier height and (= e^r^o) is the permittivity of the
semiconductor. The integral may now be written as

exp(qE,/kT)dx = a'^exp(q^/kT)exp{ - a^w^) exp(y^)dy

) Jo
= a~'exp(q(^/kT)F{aw) (3.7)

where a = {ci^Ng/le^kTy^. F{aw) is known as Dawson's integral and has

been tabulated numerically (for example, MiUer and Gordon, 1931). If
aw > 2, F(aw) is approximately equal to (ZaH')"^ [This approximation is
equivalent to neglecting the x'^ term in eqn (3.6), or assuming that is
constant and equal to its maximum value é'^^^ throughout the depletion
region.] The condition aw > 2 is equivalent to > 4fc7; and is generally
satisfied except for very large values of forward bias. Making this
approximation, we can write eqn (3.5) as
J=2kT/xNyw{exp{qV/kT) - l]/exp{q(^^,/kT)
= ç7V,//^„„exp(- q<^,/kT)[expiqV/kT) - 1} (3.8)
since, by Gauss's theorem, the maximum field strength is given by
^max = ^^d^^/^s =2kTa'^w/q. Equation (3.8) gives the dependence of current
on voltage as predicted by the diffusion theory; it is almost, but not quite,
of the form of the ideal rectifier characteristic J= jQ{exp(qV/kT) —1].
The difference arises because S'^^^ is not independent of bias voltage
but is proportional to ( F ^ Q - Vy^^. For large values of reverse bias, the
current does not saturate but increases roughly as | F | ^''^

3.2.3 The thermionic-emission theory

In Bethe's thermionic-emission theory, the assumption is made that the

current-limiting process is the actual transfer of electrons across the
interface between the semiconductor and the metal. The inverse process
under reverse bias is analogous to thermionic emission from a metal into
a vacuum but with the barrier height replacing the metal work
function The effects of drift and diffusion in the depletion region are
assumed to be negligible, which is equivalent to assuming an infinite
mobility. It foUows from eqn (3.3) that d^„/dx is negligibly small, so that
the quasi-Fermi level for electrons remains flat throughout the depletion
region and coincides with the Fermi level in the bulk semiconductor as in
a p-n junction. This in turn implies that the concentration of electrons
on the semiconductor side of the interface is increased by a factor
exp(qV/kT) when a bias voltage V is applied. The situation can be
visualized by imagining the existence of a thin insulating layer at the
interface through which the electrons can tunnel with probability p. If

94
p<l, most of the electrons incident from the semiconductor will be
reflected back into the semiconductor and will remain in thermal
equilibrium with those in the bulk. Inspection of Fig. 3.2 shows that the
electron concentration on the semiconductor side of the boundary is now
given by
n = iV,exp{ - q{^^, - V)/kT]. (3.8a)
For a semiconductor with spherical constant-energy surfaces, these
electrons will have an isotropic Maxwellian distribution of velocities, and
the number incident per second on unit area of the insulating layer is
given by elementary kinetic theory (see, for example, Loeb, 1961) as
ni>/4, where V is the average thermal velocity of electrons in the semi­
conductor. The current density due to electrons passing from the
semiconductor into the metal is therefore

Jsm- ^ ^ e x p { - q(<^- VykT], (3.8b)

since only a fraction p of the electrons can tunnel into the metal. There is
also a flow of electrons from the metal into the semiconductor which is
unaffected by the application of bias because the barrier seen from
the metal remains unchanged.! For zero bias, the current from semi­
conductor to metal just balances this current so that

J^.-^exp{-q^/kT).
Hence

= ^^exp(-q^,/kT){exp(qV/kT)-l]. (3.9)

For a Maxwellian distribution of velocities v = (SkT/nm*y^''-, where m* is

the effective mass of electrons in the semiconductor.
One of the implicit assumptions of the thermionic emission theory is
that, if the imaginary insulating layer becomes so thin that p approaches
unity, the number of electrons incident on the interface per second is
unchanged, so that the current can be calculated by simply putting p=l
in (3.9). This is a very drastic assumption, since if p = 1 no electrons are
reflected back into the semiconductor and the velocity distribution of
electrons at the top of the barrier becomes unidirectional and the
concept of a quasi-Fermi level very dubious. Baccarani and Mazzone
(1976) have used the Monte Carlo method to calculate the trajectories of
electrons crossing the barrier, assuming that they have an isotropic

tWe are neglecting for the time being any possible bias dependence of ^f,.

95
Maxwellian distribution at the edge of the depletion region. They find
that the velocity distribution of electrons crossing the barrier is very
close to a unidirectional Maxwelhan distribution, which they call a 'hemi-
Maxwellian' distribution, and that the concentration of these electrons is
almost exactly one half of the concentration predicted by (3.8a). The
mean velocity of such a distribution in the direction normal to the inter­
face is v/2, so the current density due to electrons flowing from the semi­
conductor to the metal is very close to that predicted by (3.8b) with
p = 1. Such a conclusion was later confirmed by Baccarani (1976) using
an analytical method, and more recently by Berz (1985). It seems, there­
fore, that the rather drastic assumption that we may take p = l without
changing the flux of electrons incident on the interface is justifiable.
Substitution of p = l and N, = 2(2nm*kT/hY^ in (3.9) gives the
current/voltage characteristic according to the thermionic-emission
theory as
J = A*T^exp{-q<^^/kT){exp(qV/kT)-l) (3.10)
where
A * = 47rm*#V/i3. (3.11)
^ * is the same as the Richardson constant for thermionic emission
except for the substitution of the semiconductor effective mass m* for
the free-electron mass m; it has the value
A* = 1.2 X 10«(m*/w) A m - 2 K"^ (3.11a)
provided the conduction band of the semiconductor has spherical
constant-energy surfaces. This is the case with galhum arsenide, for
which (m*/m) = 0.072.
Bethe's derivation of the current/voltage relationship proceeds rather
differently from the derivation outlined above. Using largely intuitive
arguments, he writes the expression for the current from the semi­
conductor to the metal as

^s,„=9^-|iVcexp{-?(^-F)/fcr}

where 'v/2 is the average velocity of an electron in the Maxwell-

Boltzmann distribution in one cartesian direction, and the (second)
factor 1/2 arises because only electrons moving towards the potential
maximum can get across it'. Thus it would appear that Bethe arrived at
essentially the right answer by using purely intuitive reasoning.
There is an alternative approach to be found in the books by Henisch
(1957) and Spenke (1958), according to which the electrons are assumed
to have a Maxwellian distribution at the edge of the depletion region
(x = w). The number of electrons which can surmount the barrier per

96
second is then calculated on the assumption that the mean free path / is
large compared with w. The expression for the current density calculated
in this way is identical to eqn (3.10). It can easily be shown that this
approach yields a result which is independent of the position of the
particular plane at which the electrons are assumed to have a Maxwellian
distribution; in other words, if the electrons were assumed to have a
Maxwelhan distribution at x = x' (x' < w) which is in equilibrium with the
bulk of the semiconductor, the current would be the same as if one took
x' = w. If one now allows x' to approach zero (i.e. one assumes that the
electrons are in thermal equihbrium with the bulk of the semiconductor
right up to the metal, with a flat quasi-Fermi level), the two approaches
are seen to be equivalent. It foUows that it is not necessary to assume that
/ is large compared with w, as is often done; on the contrary, it was argued
by Bethe that the condition for the validity of the thermionic emission
theory is the much weaker one that / must exceed the distance d in which
thebarrier falls by an amount kT/q from its maximum value.
The question of the correct value of m* for a semiconductor with
eUipsoidal constant-energy surfaces is comphcated and has been
thoroughly discussed by Crowell (1965). H e shows that the value of m*
associated with a single energy minimum is

where /, m, and n are the direction cosines of the normal to the interface
relative to the principal axes of the elhpsoid, and m^, m^, and are the
components of the effective-mass tensor. For semiconductors with
spherical constant-energy surfaces such as gallium arsenide. A* is
independent of direction but, for semiconductors with anisotropic
constant-energy surfaces such as silicon and germanium, A* depends on
orientation even if the crystal has cubic symmetry. For electrons in
silicon, the appropriate values of m* obtained by summation over the six
valleys are:

m* = 2m, -I- 4(m,/n,)'^^ = 2.05m for (100) directions

and
1/2
nil + 2mim,
OT*=6 = 2.15m for (111) directions

where m, and m, are the longitudinal and transverse effective masses,

respectively, and m is the mass of a free electron. For electrons in
germanium, the values are:

m* = 4 mf + 2 m , m , y ^ ^ ^ ^ ^ ^ ^ for (100) directions

97
and
m*=m, + {m^ + 8m,m,)i''2 = 1.07m for (111) directions.
The value of vl* for silicon is much larger than that for gallium arsenide,
partly because of the larger effective-mass components and partly
because of the six valleys in silicon.
A n interesting and sometimes important feature of the thermionic-
emission process is that the electrons injected into the metal under
forward bias are confined to a narrow cone of directions. A s we have
seen, in the semiconductor the electrons moving towards the metal have
a hemi-Maxwellian distribution of velocities with a mean velocity
~ {dikT/my^. When the electrons cross the interface, the component of
velocity (strictly crystal momentum) parallel to the boundary is con­
served, but the component perpendicular to the boundary is increased by
an amount corresponding to the energy difference between the top of the
barrier and the bottom of the conduction band in the metal; this is
shghtly greater than the Fermi velocity Vp The electron trajectories
in the metal are therefore confined to a cone or jet of half-angle
tan""^{(3fcr/m*)^''Vvp)== 5°. Conversely, an electron may only pass from
the metal into the semiconductor if its trajectory lies within this cone. For
this effect to be present, one would expect the interface to have to be flat
on a scale comparable with the wavelength of the electron wave-function
(~ 100 A ) , since this is a necessary condition for conservation of the
component of crystal momentum parallel to the boundary. This condi­
tion might be met in extremely high-quality interfaces (for example
cleaved surfaces) but is unlikely to be satisfied by ordinary mechanicaUy
polished surfaces, so it is not clear how important the 'jet' effect is hkely
to be in practical diodes.

3.2.4 The effect of the image force on the current/vohage relationship

The current/voltage relationship predicted by the thermionic-emission

theory [eqn (3.10)] is of the form of the ideal rectifier characteristic
/ = 7o{exp(gF/A:r)-l}, with JQ = A*T'^exp{-q^^/kT), provided the
barrier height is independent of bias. However, as we saw in Chapter 1,
there are several reasons why the barrier height may depend on the
electric field in the depletion region and hence on the applied bias. In
particular, even in a perfect contact with no interfacial layer, the barrier
height is reduced as a result of the image-force by an amount A ^ ^ i which
depends on the bias voltage. The effective barrier that electrons must
surmount before they can reach the metal can therefore be written as
= ^b ~ A^bi- Moreover, as we saw in Section 1.4.5, in the presence of an

98
interfacial layer depends on the bias voltage so that may be bias
dependent for two reasons. Such a bias dependence of will modify the
current/voltage characteristic.
Let us suppose that d^JdV happens to be constant so that we may
write <j)^ = ~ (A^bi)o + /SK where ^,,0 and (A^|,i)o refer to zero bias. The
coefficient /3 is positive because always increases with increasing
forward bias. The current density now becomes
J = A*T'exp[- q[<l,,Q - (A^bi)o + PV]/kT][exp{qV/kT) - 1)
= /oexp(- PqV/kT){exp{qV/kT) - 1} (3.12)
where
jQ=A*T^exp[-q[<l>,, - {A<l>^doVkT].
We can write eqn (3.12) in the form
/ = jQexp{qV/nkT)[l - e x p ( - qV/kT)] (3.13)
where

~ = l - p ^ l - (d(^,/dV). (3.14)

n is often called the 'ideality factor'. If d^JdV is contant, n is also

constant. For values of F greater than 3kT/q, eqn (3.13) can be written as
J = JQexp(q V/nkT). (3.13 a)
As we shall see, there are other mechanisms besides a bias dependence
of (notably thermionic-field emission and recombination in the
depletion region) which give a current/voltage characteristic of the form
of eqn (3.13), and there may be some general reason why all transport
mechanisms should yield a relationship of this form. Equation (3.13) is
often written in the literature in the form

fqV^
exp [nkTJ —1 (3.13b)

This form is incorrect because the barrier lowering affects the flow of
electrons from metal to semiconductor as well as the flow from semi­
conductor to metal, so that the second term in the curly bracket must
contain n. The difference between eqns (3.13) and (3.13b) is negligible
for V> 3kT/q, but the correct form of eqn (3.13) has the advantage that n
can be found experimentally by plotting ln[//{l - e x p ( - qV/kT)}] against
V This graph should be a straight line of slope q/nkT if n is constant,
even for V<3kT/q and for reverse bias (Missous and Rhoderick, 1986).
More usually d<j>JdV is not constant and the plot of
ln[//{l - exp(- qV/kT)]] against F is not Unear. The ideality factor

99
defined by eqn (3.14) is now a function of bias, but it is still a useful
concept which can be obtained from the experimental J/V characteristic
through the relationship
1 kT d
\n\J/{l- exp{ - qV/kT)]] (3.14a)
n q dV

or, for V> 3kT/q,

kTdjln J)
(3.14b)
q dV
The parameter n is generally a function of V, and can only be specified
for a particular operating point on the characteristic. Equation (3.14b) is
more commonly found in the hterature, but eqn (3.14a) has the
advantage that it may be used for V< 3kT/q and for reverse bias.
For the particular case where there is no bias dependence of ^(,> we
have (d^^/dV) = - (дA^^„/дV) and A^^i is given by eqn (1.26a) as
1/4
q'N, kT

SO that, from eqn (3.14),

1 1
4
f q'Ng ^
1/4
kT
\ -3/4

If V is restricted to values less than ^^/4, n is roughly constant. Taking

= £s = 10"'" F m~', appropriate to silicon or gallium arsenide, and
^t,-^" 0-5 eV, n has a value of about 1.02 for A^,, = 1 0 " m"^. The effect
of image-force lowering on the forward characteristic is therefore
neghgible for Schottky barriers with A^<j < 10^^ m~^, although it may be
more important under reverse bias because of the larger electric field in
the depletion region.

3.2.5 The combined thermionic-emission/dijfusion theory

Several authors (Schultz, 1954; Gossick, 1963; Crowell and Sze, 1966)
have combined the thermionic-emission and diffusion theories by con­
sidering the two mechanisms to be in series and effectively finding the
position of the quasi-Fermi level at the interface which equahzes the
current flowing through each of them. The most fully developed theory is
that of Crowell and Sze, who introduce the concept of a 'recombination
velocity' Vr at the top of the barrier; is defined by equating the net
electron current into the metal to v^{n — n^f), where n is the electron

100
density at the top of the barrier and « 0 its value at zero bias.t The terms
v,n and V^HQ represent the electron flux from semiconductor to metal and
the flux in the reverse direction, respectively. The concept of a recom­
bination velocity is quite general and can be applied, for example, to a
p - n junction. In terms of the thermionic-emission theory outlined in
Section 3.2.3, which assumes that no electrons which pass over the
maximum of the potential barrier are scattered back into the semi­
conductor, v^ = v/4 where is the mean thermal velocity of electrons in
the semiconductor, since CroweU and Sze took the electron distribution
to be fuUy MaxweUian.
If we regard the position of the quasi-Fermi level ?„(0) at the interface
as an adjustable parameter, the reasoning which leads up to eqn (3.9) can
be adapted to give the thermionic-emission current as
4 = qKv,cxp(- q^,/kT)[exp{q^„{OykT) - 1]. (3.15)
Equation (3.5) can be used to give the current hmited by drift and
diffusion in the depletion region as
kTjuK{exp{qV/kT) - expiqUOym
Jd = ^ (3.5a)
exp{qE,/kT)dx
0
where we have used ^^(w) = F If the effect of the image force is taken
into account, it is the quasi-Fermi level Sn(-^m) at the maximum of the
barrier which should be used rather than Cn(0). and the lower limit of
integration should be x^. Since the thermionic-emission current must
equal the current determined by drift and diffusion, ^„(0) can be
eliminated between eqns (3.15) and (3.5a) and, by writing 7,^ = Jg = J, we
finally obtain

qN,v,cxp(-q<^/kT){exp{qV/kT) -1}
= T;^ (3-16)
1 -I- — exp{-q^kT) exp{qE,/kT)dx
MkT Jo
which, following CroweU and Sze, may be written in the form

1 + [V,/Vg)
where
q
exp(-q^,/kT) exp(qEykT)dx (3.18)
fxkT 0
tif the barrier height is bias dependent, n„ is also bias dependent and is equal to the
hypothetical electron density which would exist at the top of the barrier if the barrier
height at zero bias were equal to the barrier height at bias V.

101
The significance of v^, is that Vg[exp(qV/kT) -1} would be the mean
velocity due to drift and diffusion of electrons at the top of the barrier if
the diffusion theory were to apply. The integral in the expression for
can be expressed in terms of Dawson's integral, as in eqn (3.7) and, if one
adopts the same approximation as was used in deriving eqn (3.8), namely
that the electric field is constant and equal to its maximum value S'^^^,
eqn (3.18) simphfies to Vj = n^'^^^.
If Vg > v„ eqn (3.17) reduces to eqn (3.9) with in place of v/4 and p
taken as 1; the bottleneck to current flow is the actual emission of
electrons into the metal, and the thermionic-emission theory applies. If
Vg < v„ eqn (3.17) reduces to eqn (3.8) when is taken as equal to juS'^^/,
the current is controUed by drift and diffusion in the depletion region and
the diffusion theory apphes. The condition for the vahdity of the
thermionic-emission theory (Vj > v,) is equivalent to

A^max > I- (3.19)

Equation (3.19) can be obtained more directly by noting that the condi­
tion that the main impediment to current flow lies in the thermionic
emission process rather than in drift and diffusion is that the coefficient
qN^juS"^^^ which occurs in (3.8) must greatly exceed the coefficient
qN^v/4 in (3.9). The condition (3.19) can be written in the equivalent
form
q^^,,>m*mt,,
since ju = qtjm*, or
lqS>^,,>kT, (3.20)
where is the mean time between colhsions of electrons in the semi­
conductor and /(= VT^) their mean free path. In deriving eqn (3.20) we
have used v= (8kT/jim*y^^ and have ignored a factor of 2/jt. Equation
(3.20) is equivalent to Bethe's criterion that the mean free path must
exceed the distance d(= kT/qS'^^^) in which the barrier faUs through an
amount kT/q.
The physical significance of this condition can be understood by
reference to Fig. 3.2. Electrons which pass over the barrier into the metal
will, on the average, have made their last collisions at a distance / from
the maximum.! Immediately after making these collisions, the electrons
will have a Boltzmann distribution of energies with an average energy of
ikT above the bottom of the conduction band due to motion normal to
the interface. Those electrons moving towards the metal will only be able

tMore strictly, at a distance 1/2 from the maximum, since / is the mean free path in three
dimensions.

102
to surmount the barrier if their kinetic energy due to motion normal to
the metal exceeds the amount by which the conduction band rises within
the distance /. Hence, if the conduction band rises by much more than kT
in the distance /, only a small fraction of the electrons will be able to
enter the metal. Most of them will be reflected back into the semi­
conductor, and the concentration and velocity distribution of the
electrons for X < / will not differ significantly from those of a Maxwellian
distribution in equilibrium with the bulk of the semiconductor. This is the
condition for the current to be limited by thermionic emission, and it will
clearly hold ii l> d, where d is the distance within which the conduction
band fahs by an amount kT/q from the maximum (see Fig. 3.2). But if
Kd, nearly all the electrons will be able to surmount the barrier, and the
flow of electrons into the metal will be limited by drift and diffusion
through the space-charge region in accordance with the diffusion theory.
The barrier region between x = 0 and x = d plays a role equivalent to
that of the thin insulating region postulated in Section 3.2.3.
Figure 3.4 shows the dependence of d on and F^ for a semi­
conductor of permittivity 10""'° F m~', each curve corresponding to a
given value of d. The left-hand scale shows the mean free path / for
silicon as a function of N^. If we assume, with Berz, that the Bethe
condition for thermionic emission is satisfied if / > 5 c?, it can be seen that

Fig. 3.4 Graph showing conditions under which the Bethe criterion is satisfied for n-type
silicon Schottky diodes. The curves correspond to constant values of the width d of the kT
layer'. Also displayed vertically is a scale showing the electron mean free path / cor­
responding to the various donor concentrations. The condition for the current to be deter­
mined by thermionic emission for a particular combination of band-bending and donor
concentration is that / should greatly exceed d. (From Berz 1985. Copyright Pergamon
Journals Ltd.)

103
for A/j = 5 X 10^2 the Bethe condition is satisfied if Vg > 0.2 V, but
for Ng = 10^' it is not satisfied for any reahstic value of V^.
CroweU and Sze's analysis is not strictly valid because the electron
distribution is far from Maxwellian near the top of the barrier. Because
of this, one should not use the standard transport equations and the
concept of a quasi-Fermi level ceases to have any meaning. However, as
Berz (1985) has shown, the distribution at x = is almost Maxwellian.
One can therefore use the concept of a quasi-Fermi level for x> d and
can justifiably use it to calculate the electron concentration at x = d, so
that n{d) = N^exp{q[^^(d) - E^{d)]/kT]. According to Berz's analysis,
the actual electron concentration at x = 0 is given by e~'n(rf)/2, and
since the recombination velocity for this hemi-MaxweUian distribution is
v/2 we can write

4 e-'cxp{qmd)-E,(d)]/kT} -exp{-q^/kT
2 2
I
qvN, kT
exp{q Ud)-E,{d)- — kT} - exp(-q<J,,/kT)

(3".15a)
where e is the base of natural logarithms.
Since E,(d) + (kT/q) = ^^, (3.15a) is equivalent to (3.15) with ?„(rf) in
place of ?n(0) and = v/4. One should also write ^„(rf) in place of ^„(0)
in (3.5a) and take the lower limit of the integral as d rather than 0, with
the consequence that the approximate form of (3.18) becomes
Vj = fiS'{d). Hence the modification of the Crowell-Sze theory which is
necessary to take into account the non-Maxwellian distribution merely
involves replacing the quasi-Fermi level at x = 0, where it has no
meaning, by the quasi-Fermi level at x = d, where it does have a meaning,
together with a trivial change in Vj.

3.2.6 Hot-electron effects

At the top of the barrier the electric field may be quite large (typically
~ 10^ V m"'), and hot-electron effects may impair the vahdity of the
anlyses in Sections 3.2.2 and 3.2.5. We should not expect to be able to
use values of mobility and diffusion constant which are independent of
the electric field (Ryder and Shockley, 1951). However, one cannot take
the electric field into account simply by using a field-dependent mobility;
with zero bias, the electrons are in thermal equihbrium with the lattice in
spite of the buih-in electric field and, as Stratton (1962) has shown, the
effect of the field is to cool the electrons under forward bias when they
are moving against the field, and to heat them under reverse bias when
they are moving with the field. Unfortunately, Stratton's analysis uses the

104
boundary condition ^„(0) = 0, which hmits it to the case of the diffusion
theory. The influence of hot-electron effects on the thermionic-emission
diffusion theory has been discussed by Stokoe and Parrot ( 1 9 7 4 ) , who
conclude that in a typical case the forward current is reduced by the
order of 30%. However, Stokoe and Parrott's analysis is open to doubt
because of their use of questionable boundary conditions. It appears that
the main cool-electron effect on the thermionic-emission theory will be a
reduction in 7te due to a reduction in v, but the magnitude of this effect is
likely to be small. Baccarani and Mazzone ( 1 9 7 6 ) did not observe any
hot- or cool-electron effects resulting from their Monte Carlo calcula­
tions.

3.2.7 Refinements of the thermionic-emission theory

Crowell and Sze ( 1 9 6 6 ) incorporated several additional refinements into

the theory, the most important of which concern the value of the
Richardson constant A*. A* was first introduced in eqn ( 3 . 1 0 ) by using
v/4 instead of Crowell and Sze's recombination velocity (or, according
to the more exact theory, a recombination velocity of v / 2 together with
one-half the electron concentration used by CroweU and Sze). This
procedure assumes that all the electrons incident on the interface at the
potential maximum cross into the metal and do not return. However,
according to quantum mechanics, an electron may be reflected by a
potential barrier even if it has sufficient energy to negotiate this barrier;
furthermore, even after an electron has passed over the barrier, it may be
scattered through a large angle with the absorption or emission of a
phonon so that it returns whence it came. The latter process is most
likely to take place between the metal and the position of the potential-
energy maximum as modified by the image force, i.e. within the semi­
conductor. In this region, the bottom of the conduction band slopes very
steeply, and for most of its path within the semiconductor the electron
will have enough energy to emit an optical phonon. (The energy of an
optical phonon is about 0.06 eV in silicon and 0 . 0 4 eV in gallium
arsenide.) For electrons with energies just above the threshold, the
emission of an optical phonon is more likely than any other form of
scattering. Crowell and Sze have calculated the probability f^ of an
electron reaching the metal without being scattered back into the semi­
conductor; the effect of this on the thermionic-emission process is to
multiply the recombination velocity by a factor fp, which is, of course,
less than unity. Their paper gives graphs of ^ as a function of the
maximum electric field <^^a„ in the depletion region.f fp approaches unity
asymptoticahy as increases because the position of the potential
fThe maximum electric field is the field due to the donors alone and neglects the field
due to the image.

105
maximum gets nearer to the metal, so that the distance within which
phonon scattering is effective is reduced; it decreases with increasing
temperature because of the increasing phonon population. For silicon, ^
is about 0.95 at room temperature for a field of 4 x 10^ V m~', which
corresponds to a diffusion potential of 0.5 V and a donor density of 10^^
m~^. For gallium arsenide, the value is somewhat lower, about 0.85,
because of the larger optical phonon interaction in polar compounds.
The effect of quantum-mechanical reflection has also been investigated
by CroweU and Sze, who included in their work the additional effect of
tunnelling of electrons through the top of the barrier. These two effects
combine to multiply the thermionic-emission current by a factor which
depends both on temperature, through the temperature dependence of
the electron energy, and on the maximum electric field S'^^^ because this
determines the height and shape of the potential barrier. There is some
uncertainty in the calculations because an assumption has to be made
about the exact form of the image potential close to the surface of the
metal. For silicon and gallium arsenide, ^ rises above unity for fields in
excess of about 4 x 10^ V m^' because of the effect of tunnelling. Below
this value, falls below unity because of reflection and approaches a
value of about 0.6 for gallium arsenide and 0.5 for silicon.
Although Crowell and Sze's ideas about phonon scattering and
quantum-mechanical reflection are undoubtedly qualitatively correct, it
is not clear how reliable they may be quantitatively. The neglect of
acoustic phonon scattering must result in an overestimate of j ^ , and the
use of the effective-mass approximation in the calculation of is of
doubtful validity when the bottom of the conduction band slopes very
steeply, as it does between the potential maximum and the surface of the
metal. This latter point has been considered by James (1949). For the
case in which the thermionic-emission theory is vahd, the effect of fp and
/q is to replace A* in eqn (3.10) by A**=fpf^A*. [In the general case the
effect is to replace in eqn (3.17) by ^j^v,.] Probably the most that can
be said with confidence is that A** may be less than A* by as much as
50%. In practice, because a variation in A** by a factor of two has the
same effect on the current as a change in of less than kT/q, the differ­
ence between A* and A** is not very important. Because fp and f^ depend
on the applied bias (through S'^^^), A** is not strictly a constant, but this
effect tends to be masked in practice by image-force lowering of the
barrier and other effects. Crowell and Sze compute the effective values of
A** for electrons in gaUium arsenide and for sihcon ( H I ) surfaces at
300 K to be 4.4 X 10" A m"^ K"^ and 96 x 10" A m"^ K"^, respectively;
the large difference between these values is the result of the small effec­
tive mass in gallium arsenide and the six equivalent valleys in the conduc­
tion band of silicon. Andrews and Lepselter (1970) have re-examined
Crowell and Sze's computed values of fp and for silicon and find the

106
best average value of ^ * * to be slightly higher, nameley 112 x lO" A
K"^. They also give the value of 32 x 10" A for holes in p-type
silicon.

3.2.8 Comparison with experiment

Crowell and Beguwala (1971) have calculated the position of the quasi-
Fermi level at the interface by eliminating J from eqns (3.15) and (3.5a)
(see p. 101), and have concluded that, for semiconductors with a fairly
high mobility such as germanium, silicon, and gallium arsenide, the
droop in the quasi-Fermi level through the depletion region
{^„(H') - S„(0)) is negligibly small, typically less than kT/q for a silicon
diode with moderate forward bias. This justifies the fundamental
assumption of the thermionic-emission theory. Their conclusion has been
confirmed by Rhoderick (1972), who computed ^„(0) for several
practical diodes by inserting the experimentally determined J/V charac­
teristic into eqn (3.5a). There are therefore both theoretical and practical
reasons for believing that Schottky diodes made from fairly high-mobility
semiconductors should conform to the thermionic-emission theory
rather than the diffusion theory.
Among the most nearly ideal diodes yet reported are those of Arizumi
and Hirose (1969), who made measurements on Schottky diodes made
by evaporating A u , A g , and C u onto chemically cleaned sihcon surfaces.
Applying Rhoderick's analysis to the highest current densities reported
in their paper, one finds that KJw) - ^„(0) « 10"^ eV, so that the diodes
should conform to the thermionic-emission theory. The forward charac­
teristics showed a linear dependence of In 7 on V over more than four
decades, and the n values were in the range 1.01-1.02, independent of
temperature between 100 and 350 K and explicable in terms of image-
force lowering of the barrier. The barrier heights obtained by assuming
the thermionic-emission equation (3.10), with A* replaced by ^4** and
given the value 96 x 10" A m~^ K~^, agreed to within 2% with those
obtained from C / F measurements and from the temperature dependence
of the J/V characteristics. These data agree very closely with the pre­
dictions of the thermionic-emission theory. However, for large forward
bias when the band bending is very small, one would expect deviations
from the pure thermionic-emission theory if becomes comparable with
because of the reduction in fiS"^^^. Such behaviour is analogous to
high-level injection in p - n junctions and has been observed by Wilkinson
(1974) and Wilkmson, Wilcock, and Brinson (1977) in titanium-sihcon
Schottky diodes.
Next to silicon, the semiconductor which has been most frequently
used for Schottky-barrier studies is gallium arsenide. The most ideal
diodes reported seem to be those made by molecular beam epitaxy
107
( M B E ) by Missous, Rhoderick, and Singer (1986b,c). These consisted of
an epitaxial film of aluminium deposited on the (100) face of GaAs in
such a way that the (100) A l plane was paraUel to the (100) GaAs plane.
The A l film was an almost perfect single crystal, with an abrupt interface
as revealed by transmission electron microscopy. Figure 3.5 shows a
logarithmic plot of 7/(1 - exp(- qV/kT)] against V, as proposed in
Section 3.2.4. The graph is hnear over the whole range of V from + 0.5
V to - 1 . 0 V, showing that a characteristic of the form (3.13) accurately
represents the data. The deviation from hnearity above + 0.5 V is due to
series resistance. The value of n is about 1.01, though it is difficult to give
a precise figure because the temperature was not known to better that
1%. This value of n can be explained by image-force lowering alone for a
barrier height of 0.8 eV and donor density of 10^^ m~^. A calculation of
the droop in the quasi-Fermi level between the semiconductor bulk and
the interface (Rhoderick, 1972) gives about 10~^ eV, so one may be
confident that thermionic emission is the controlling mechanism. The
agreement between the values of the barrier height as deduced from the
J/V and C/V characteristics depends on the value adopted for A**. The
agreement is only moderately good (within 0.03 eV) if one takes the

-1.0 -0.5 0 0.5 1.0

Bias voltage (V)
Fig. 3.5 Logarithmic plot of //(1 - exp(- qV/kT)] versus V for an epitaxial Al-GaAs
diode (Missous and Rhoderick, 1986).
108
commonly accepted value of 8.6 x 10" A m~^ (Gol'dberg, Posse, and
Tsarenkov, 1975), but is excellent if one takes a lower value of 3 x 10" A
-pjiere is recent experimental evidence for this lower value
(Srivastava, Arora, and Guha, 1981), which corresponds to the
theoretical value given by (3.11a) with m*/m = 0.072 and f^f ~ 0.35, and
is only slightly lower than the theoretical value of 4.4 x 10" A m"^ K"^
calculated by Crowell and Sze (1966).
We conclude that in Schottky diodes made from fairly high-mobility
semiconductors the forward current is limited by thermionic emission
provided the forward bias is not too large. Even comparatively low-
mobihty semiconductors like GaP and CdS follow the thermionic-
emission theory quite closely (Rhoderick, 1972), and eqn (3.19) seems to
be unduly severe as a condition for the validity of this theory. There is a
dearth of rehable data on low-mobihty semiconductors which are likely
to obey the diffusion theory, and the only results known to the authors
which appear to conform to the diffusion theory are some early data
relating to C u O (see Henisch, 1957) and some more recent measure­
ments on amorphous silicon (Wronski, Carlson, and Daniel, 1976).

3.3 Tunnelling through the barrier

3.3.1 Field and thermionic-field emission

Under certain circumstances it may be possible for electrons with

energies below the top of the barrier to penetrate the barrier by
quantum-mechanical tunnelhng. This may modify the ordinary ther­
mionic process in one of two ways which may be understood by
reference to Fig. 3.6. In the case of a very heavily doped (degenerate)
semiconductor at low temperature, the current in the forward direction

Fig. 3.6 Field and thermionic-field emission under forward bias. The diagram refers to a
degenerately doped semiconductor for which ^ is negative. (From Padovani and Stratton,
1966. Copyright Pergamon Journals Ltd.)
109
arises from the tunnelhng of electrons with energies close to the Fermi
energy in the semiconductor. This is known as 'field' emission. If the
temperature is raised, electrons are excited to higher energies and the
tunnelhng probability increases very rapidly because the electrons 'see' a
thinner and lower barrier. O n the other hand, the number of excited
electrons decreases very rapidly with increasing energy, and there will be
a maximum contribution to the current from electrons which have an
energy above the bottom of the conduction band. This is known as
'thermionic-field' emission. If the temperature is raised still further, a
point is eventually reached in which virtually all of the electrons have
enough energy to go over the top of the barrier; the effect of tunnelling is
neghgible, and we have pure thermionic emission.
Tunnelhng can be most easily understood if one recognizes the
existence of solutions of Schrödinger's equation which correspond to
energies within the forbidden gap of the semiconductor and which have
the nature of exponentially damped waves rather than travelling waves.
These can be most simply visualized by supposing that the relationship
q{E - £•<.) = fi^k^/lm*, which relates the kinetic energy of an electron to
its wave vector k and effective mass m *, remains valid if the energy E of
the electron is less than the energy E^ of the bottom of the conduction
band. In this case the wave vector becomes imaginary and we may write
k = ik', where k' is real. Since the electron wave-function is proportional
to txp(ikx) = exp(- k'x), the wave-function decays exponentially with
distance and the probability of finding the electron at the position x,
which is proportional to the square of the wave-function, is proportional
to exp(- 2k'x) where

\i E^- E is not constant (for example when the bands are not flat), k' is a
function of position and the probabihty of finding the electron at the
position X is given by P = exp(-2//:tiJc). This is known as the W K B
approximation. If this result is used to calculate the probabihty of a tri-
angular barrier being penetrated by an electron with energy A £ less than
the height of the barrier, it is found that

P = exp -^(2qm * f \ A E f ^ / n ^ \ (3.21)

where ^ is the electric field in the barrier. We can apply this to a

Schottky barrier with a diffusion potential F^, if A i i is sufficiently small
for the top of the barrier to be thought of as triangular. Making use of the
result that, according to the depletion approximation, the maximum field
in the barrier is given by ^^^^ = (2qNg Fj/eJ'^^, we find that
110
 P = exp - | ( A £ f V £ o o l ^ r (3.22)

where £ 0 0 (using the nomenclature of Padovani and Stratton, 1966) is a

parameter which plays an important role in tunnelling theory. It has the
dimensions of energy divided by charge (or energy in eV) and is given by
1/2
•'OO (3.23)

1/2
= 18.5 X 10 -15 eV (3.23a)

where m * {= m,m) is the effective mass of electrons in the semi­

conductor, £5 ( = £„£o) its permittivity, and the donor concentration is
expressed in m~^. Values of JEQO for semiconductors with various values
of Ng, m^, and are shown in Fig. 3.7.
For moderately doped semiconductors, the onset of thermionic-field
emission can be considered to be roughly equivalent to a reduction in the
barrier height by an amount A ^ equal to ( A E ) , , where ( A E ) ,

102> 10» 10^3 1024 ,025

Fig. 3.7 Values for E^^ as a function of for various semiconductors.

111
corresponds to a transmission probability of e The barrier lowering for
which this occurs is
I3]
2/3 2/3 1/3
J (3.24)
\2I \ 1 \ 1
This very crude estimate works remarkably well up to EQO ~ ^V,
corresponding to Ng < lO^" m~^ for sihcon. If we take as an example
galhum arsenide with A^^ = 10^^ m~^ and EQQ = 2 x 10~^ eV and assume
Vg = 0.8 V, we find that Aji) « 0.02 eV, which is barely observable. If,
however, Ng is increased to lO^-' m~^, A ^ « 0.04 eV, which is beginning
to be more significant.
The extension of the theory of tunnelling through Schottky barriers
beyond the simple analysis given above is complicated and beyond the
scope of this book. The theory has been developed by Padovani and
Stratton (1966) and by Crowell and Rideout (1969); both these analyses
are very mathematical, but the essential features are as follows:
1. In the forward direction, field emission occurs only in degenerate
semiconductors and, because of the very small effective mass, shows
up at lower concentrations in gallium arsenide than in most other
semiconductors.
2. The physical significance of EQQ is that it is the diffusion potential of a
Schottky barrier such that the transmission probability for an electron
whose energy coincides with the bottom of the conduction band at the
edge of the depletion region is equal to e~'. Therefore the ratio kT/qE^Q
is a measure of the relative importance of thermionic emission
and tunnelling. A s a rough guide, we should expect field emission if
kT<qEQQ, thermionic-field emission if kT-qE^Q, and thermionic
emission if kT> qE^^. A more exact analysis shows that the tempera­
ture below which field emission occurs is given by
kT< IqEUH- 4^/^) + ( - 2£oo/^)"')"' (3.25a)
where I is the distance of the Fermi level below the bottom of the
conduction band in the bulk semiconductor and is negative in a
degenerate semiconductor. The temperature above which the process
may be described as thermionic-field emission is given by
kT> 2^£oo{ln(- 4^1,/^))-'. (3.25b)
The ranges of temperature and concentration over which gallium
arsenide Schottky barriers exhibit field and thermionic-field emission
are shown in Fig. 3.8.
3. The upper temperature limit for thermionic-field emission is given by
cosh^(qEoo/kTysinh\qEoo/kT) < 2 Fd/3£oo (3.26)
112
 Afd(cm-')
Fig. 3.8 Ranges of temperature and concentration over which GaAs Schottky barriers
exhibit field and thermionic-field emission. (Padovani, 1971. Copyright Academic Press.)

where Vg (= ~ ^ ~ V) is the diffusion potential. The transition

between thermionic-field and pure thermionic emission therefore
depends on the applied bias and is shown in Fig. 3.18 as a plot of
diffusion potential against temperature.
4. The current/voltage relationship is of the form
7 = /,exp(F/£;o){l - e x p ( - qV/kT)} (3.27)
where
Eo = EooCOth(qEoQ/kT).
Note that (3.27) is similar in form to (3.13), which was derived for
the case of a voltage-dependent barrier height. A t low temperatures
(kT/qEQQ <1), E Q ' ^ EQQ so that the slope of the graph of
ln[7/{l - exp(- qV/kT)]] against V is independent of temperature.
This is the case of field emission. A t high temperatures (kT/qE^Q > 1),
Eg is slightly greater than kT/q, and the slope can be written as q/nkT,
where n = qEQ/kT= (^qEQQ/kT)coth(qEQQ/kT). There is therefore a
smooth transition from field emission, through thermionic-field
emission, to pure thermionic emission. Figure 3.9 shows values of E Q ,
obtained experimentally from the slope of the I/V curve for a gold-
gallium arsenide diode with a donor density of 5 x 10^^ m~^, plotted
as a function of kT/q (Padovani and Stratton, 1966). The solid curve
represents the theoretical expression for EQ with EQQ taken as 17 meV.
The agreement between theory and experiment is excellent.
113
 20 30
kVq (meV)
Fig. 3.9 Experimental values of £0 as a function of temperature for a Au-GaAs diode.
(From Padovani and Stratton, 1966. Copyright Pergamon Journals Ltd.)

5. The pre-exponential term is weakly dependent on bias voltage. It is

a comphcated function of temperature, barrier height, and semi­
conductor parameters for which the reader is referred to the original
papers by Padovani and Stratton and by Crowell and Rideout. For
thermionic-field emission, Padovani and Stratton give

J ^ J^q{7iEU<k-V-^)] 1/2
•exp{-(^,-^yE,} (3.27a)
kTcosh{qEoo/kT)

where 7„ [= yl*T^exp(- q^/kT)] is the 'flat-band current density'.

Chang and Sze (1970) have published graphs showing the depen­
dence of /5 on donor density for silicon Schottky diodes.
6. The maximum of the energy distribution of the emitted electrons
occurs at an energy E^ = Vg(cosh(qEQo/kT)]~'^ above the bottom of
the conduction band in the bulk semiconductor.
Figure 3.10 shows the variation of n and E^/Vg as a function of
kT/qEoQ. If we somewhat arbitrarily take n > 1.02 and E^/V^ < 0.95 as
criteria for the transition from pure thermionic emission to thermionic-
field emission, we see that it occurs when kT/qE^^ < 4. In galhum
arsenide this corresponds to A^^ > 10^^ m~^ and 10^^ m~^ at 300 and 77
K , respectively, while in silicon the corresponding values of Ng are about
a factor of four greater.
The analyses of Padovani and Stratton and of Crowell and Rideout
both neglect the reduction of the barrier by the image force and the
quantum-mechanical reflection of electrons with energies exceeding the
height of the barrier. In a later paper, Rideout and Crowell (1970) have

114
Fig. 3.10 Ideality factor (n) and position of maximum of energy distribution of emitted
electrons (£„) as a function of кТ/дЕ^о.

taken both these effects into account for thermionic-field emission. They
find a significant increase in and a small increase in n, primarily when
the conduction process is close to pure thermionic emission. Figure 3.11
shows their computed forward current/voltage characteristics presented
in normahzed form as a function of the band bending (= - V-
They also present the same data in tabular form.
Chang and Sze (1970) have taken both image-force lowering and
quantum-mechanical reflection into account, and have also used Fermi-
Dirac statistics so that their results are apphcable to degenerate semi­
conductors. However, they present their results in the form of curves
computed specifically for sihcon, and their analysis cannot be readily
adapted to other semiconductors.
The general situation with regard to field and thermionic-field
emission is that the analyses of Padovani and Stratton and of Crowell
and Rideout seem to be adequate to explain the experimental data taking
into account the uncertainties arising from imperfect knowledge of the
barrier parameters (see, for example, Padovani, 1971). This is rather
surprising in view of a rather serious shortcoming of the theoretical
model, namely that it ignores the fact that the space charge associated
with the donors is not continuously distributed but consists of discrete
charges. For example if one takes ~ 0.5 V and A''^,« 10^^ m~^, there
will only be about four layers of donor atoms (assuming them to be
arranged in a regular cubic lattice) within the depletion region. There are

115
 qVJkT
Fig. 3.11 Forward J/V characteristics for thermionic and thermionic-field emission,
allowing for image-force lowering and quantum-mechanical reflection. The ordinate is the
ratio of the current density to the 'flat-band' current density J^ = A*T^ exp( - q^/kT); the
abscissa is the band-bending (=(<b~ V~ in units of kT/q. £ „ is a measure of the
modification of the barrier by the image force and is equal to (q^/2c,)N\". £,^ = 0
corresponds to no image-force lowering. (Rideout and Crowell, 1970. Copyright Pergamon
Journals Ltd.)

two effects. First, because the barrier consists of a number of discrete

potential wells centred around each donor, its shape will no longer be
parabolic so the tunnelling probabihty will be altered; however, this
effect will be partially reduced by the Debye screening due to the
conduction electrons. Secondly, because the number of donors within the
depletion layer is smaU and they are not arranged in a regular lattice,
there will be spatial fluctuations in the precise shape of the barrier and an
average has to be taken. The second point was considered by Chang and
Sze who found that it could lead to errors in the calculation of / of the
order of 10% for Ng = 102" m-3 and 100% for A^j = 1 0 " m"^.

3.3.2 Ohmic contacts

Field emission is of considerable importance in coimection with ohmic

contacts to semiconductors, which usually consist of Schottky barriers

116
on very highly doped material. What matters in this case is the specific
differential resistance around zero bias [i.e. (dV/dJ)y^Q]. The full
theoretical expression for R^ for all values of the ratio qEoo/kT is
complicated and has been discussed by Y u (1970). He shows that the
dependence of /?(. on doping level is determined predominantly by the
factors:
exp({ib/^oo) for field emission (qEQ^/kXP 1)
exp{^b/^oo coth{qEQQ/kT)) for thermionic-field emission
(qEoo/kT- 1)
exp(<3'^b/^^) for thermionic emission {qE^Q/kT-^ 1).
Thus for low doping (the thermionic-emission regime) R^ is independent
of Ng, while for highly doped material the conduction process is field
emission and In R^ is proportional N^^''^. The thermionic-field regime
bridges the two. Figure 3.12 shows some experimental determinations
made by Hooper, Cunningham, and Harper (1965) of R^ plotted as a
function of Ng for n-type sihcon at room temperature. The theoretical
curves are those computed by Vilms and Wandinger (1969) by using
either a simple parabolic barrier, the height of which was adjusted to
allow for the image force (curve A ) , or a truncated parabolic barrier

Fig. 3.12 Contact resistance versus doping density for contacts to n-type silicon.
Experimental data by Hooper, Cunningham, and Harper (1965). Curves A and B refer to
two different theoretical models (see text). (From Vilms and Wandinger, 1969. Copyright
Electrochemical Society.)

117
(curve B). The agreement between theory and experiment is good for the
higher donor densities.
Theoretical calculations of have been published by Chang, Fang,
and Sze (1971) for contacts to silicon and gallium arsenide. Their
computed values for silicon are in reasonable agreement with those of
\^lms and Wandinger, and they find that the values of R^ for gallium
arsenide are significantly lower than those for silicon because of the
much smaller effective mass.

3.4 Recombination in the depletion region

The importance of recombination in the space-charge region has been

convincingly demonstrated in a classic paper by Y u and Snow (1968).
The recombination normally takes place via locahzed states, and
according to the theory (Shockley and Read, 1952; Hall, 1952), the most
effective centres are those with energies lying near the centre of the
forbidden gap. The theory of the current due to such recombination
centres in Schottky diodes is similar to that for p - n junctions (Sah,
Noyce, and Shockley, 1957), and the current density for low forward bias
is given approximately byf
J, = J,oexp{qV/2kT)[l - exp(- qV/kT)} (3.28)
where J^Q = qriiW/lr^. Here «1 is the intrinsic electron concentration,
proportional to exp(- qE^/2kT), w is the thickness of the depletion
region, and is the lifetime within the depletion region. This simple
result embodies several rather drastic assumptions, namely that the
energy levels of the centres coincide with the intrinsic level, that the
capture cross-sections for electrons and holes are equal, and that the
centres are distributed in a spatially uniform manner. None of these
assumptions is likely to be true in practice, especially the equality of the
electron and hole capture cross-sections, which may differ by as much as
three orders of magnitude. Sah, Noyce, and Shockley have shown that,
depending on the ratio of the capture cross-sections, the n value for
recombination current may be between 1 and 2. The simple expression
(3.28) for 7f is therefore a gross simplification, though it is usually
accepted without question. If one accepts (3.28), the total current density
is given by

7 = 4 + 7, = /,o{exp(?F/A:T; - 1) + loexp(qV/2kT)[l - exp(- qV/kT)]

= {J^Qexp(qV/kT)+J^oexp(qV/2kT)}{l - exp(- qV/kT)]

tThe actual expression derived by Sah, Noyce, and Shockley shows that J, is propor­
tional to sinh(qV/2kT), which can be written in the form (3.28). This expression is similar
in form to (3.13) and (3.27).

118
where, assuming the thermionic-emission theory, /,o = A * * r ^
exp{ - q(l>JkT). The ratio of the thermionic to the recombination current
is proportional to

This ratio increases with r^, V, and E^, and decreases with Also, since
£g -H F - is usually negative for n-type semiconductors and small
values of K the ratio increases with T. Thus the recombination com­
ponent is likely to be relatively more important in high barriers, in
material of low lifetime, at low temperatures, and at low forward-bias
voltage. It is much more important in gallium arsenide than in silicon.
When recombination current is important, the temperature variation of
the forward current shows two activation energies (Fig. 3.13).t A t high
temperatures the activation energy tends to the value ~ K charac­
teristic of the thermionic-emission component; at low temperatures it
approaches the value {E^ - F ) / 2 , characteristic of the recombination
component.
Recombination current is a common cause of departure from ideal
behaviour in Schottky diodes. Figure 3.14 shows a hypothetical diode
characteristic which has been synthesized by adding a recombination
term
7, = 3 X 10-^ exp(A:/2){l - exp(- x)]A

3 4

Fig. 3.13 Forward current of platinum-silicon Schottky diode as a function of tempera­

ture. (From Yu and Snow, 1968. Copyright American Institute of Physics.)

t Any process which shows a temperature dependence of the form exp(- EJkT) is said
to have an activation energy E^.

119
 Forward bias (V)

10-^ •
•
•
•
10-' •
•
<
S 10-^

* 10-'

(b)
I 10^

10-'

•/
10-^ • A/
*//
1

() 0.1 0.2 0.3 0.4 0.5 0.6

Forward bias (V)
Fig. 3.14 Plots of the function /= 10-«(exp(A:)-1) +3 x 10"' exp(ji:/2)(l - exp(-;c)),
where X- q{y- 20T)/kT. In (a) the dots show a conventional logarithmic plot of / versus V
and the straight line represents the function / = 1.6 x 10"' exp(^K/1.05A:r). In (b) the dots
show a logarithmic plot of //{1 - exp(-^F/feT)) versus V and the lines represent the
functions /= 10-' exp(9mr) and / = 3 X 10"' txp[qV/2kT).

120
to the ideal thermionic-emission term
4 = 10-'x(exp(x)-l}A
Here X = q(V- lOiykT, corresponding to a series resistance of 20 Q. A
conventional logarithmic plot of / against V (Fig. 3.14a) is well fitted
over four decades by
/ = 1.6 X 10-''{exp(9F/1.05A:r))A
Recombination current may therefore cause apparent deviations of n
from unity and of the pre-exponential term from the ideal value
A ** T^exp( - q<l)^/kT). However, a logarithmic plot of //{1 - exp( - q V/kT)]
against F (Fig. 3.14b) shows the two components quite clearly, and
illustrates very vividly the advantage of this method of plotting the I/V
characteristics. The importance of recombination current in causing
small departures from ideal behaviour has been frequently overlooked in
the literature. Such departures become more pronounced at low
temperatures.

3.5 Hole injection

If the height of a Schottky barrier on n-type material is greater than half

the band gap, as is often the case, the region of the semiconductor
adjacent to the metal becomes p-type and contains a high density of
holes. One might expect some of these holes to diffuse into the neutral
region of the semiconductor under forward bias, thus giving rise to the
injection of holes. Hole injection at metal contacts was extensively
studied in the early days of semiconductors, and the early work has been
summarized by Henisch (1957). It is convenient to distinguish between
the behaviour of plane contacts formed by evaporation and that of point
contacts.

3.5.1 Hole injection in plane contacts

If one assumes that the hole quasi-Fermi level coincides with the Fermi
level in the metal and remains flat through the depletion region, the hole
current density can be written, using ordinary p - n junction theory, as

J, = ^^{cxp{qV/kT)-l} (3.29)

where Po (= nf/Ng) is the equilibrium concentration of holes at the edge

of the depletion region, and and L are the diffusion constant of holes
in the bulk semiconductor and the thickness of the neutral region,

121
respectively. ( L is assumed to be less than the hole diffusion length.) The
assumption of a flat quasi-Fermi level for holes is justified in Appendix
D. Since the electron current is given by eqn ( 3 . 1 0 ) if one assumes the
thermionic-emission theory, the hole injection-ratio is given by

(3.30)
7h + 7, 7e '^exp( - q<l>JkT)'
N,LA**T'>

a result which seems to have been first derived by Henisch. The injection
ratio increases with because of the reduction in J^, and decreases with
increasing Ng because of the decrease in and consequent reduction in
/j,. If one substitutes typical figures of ~ 0-8 eV, ~ 10^^ m~^, and
L « 5 X 10"^ m for sihcon diodes, one finds y ~ lO"", so that hole injec­
tion is generally negligible in practically useful diodes. It is for this reason
that Schottky diodes are often described as 'majority-carrier' devices.
Equation ( 3 . 2 9 ) takes into account only the diffusion component of
the hole current density. Scharfetter ( 1 9 6 5 ) has pointed out that, for
large forward bias, the electric field in the neutral region of the semi­
conductor gives rise to a significant drift component, and for very large
currents the injection ratio rises linearly with /. The critical current
density at which this increase in begins to take place is given by
/(. = qD^NJL, where is typically of the order of 10^ A m"^ in practical
diodes. Scharfetter's theory has been extended by Green and Shewchun
( 1 9 7 3 ) who have found that there is a hmit to the growth of y^, with 7.
This limit arises partly because of high-level-injection effects when the
minority-carrier density in the neutral region becomes comparable with
the majority-carrier density, and partly because the supply of holes at the
contact is limited by thermionic emission. The result of these two effects
is that the injection ratio may pass through a maximum and decrease at
high current densities.
Scharfetter's theory has been partially confirmed by some experi­
mental results of Y u and Snow ( 1 9 6 9 ) on gold-sihcon diodes. They
found that at low current densities the experimental value of the injection
ratio agreed well with eqn ( 3 . 3 0 ) and that y,, began to increase when /
approached the value 7^ given above, as shown in Fig. 3.15. However, the
measurements were not continued to sufficiently high current densities to
establish whether y^ increases linearly with 7 in agreement with
Scharfetter's theory, or passes through a maximum as predicted by Green
and Schewchun. More recently. Green and Shewchun's theory has been
confirmed by Hargrove and Anderson ( 1 9 8 6 ) for platinum silicide-
silicon diodes. They were able to observe the predicted maximum in y^.
If hole injection becomes appreciable at high current densities, one
might expect that the additional electrons which must enter the neutral
region to maintain charge neutrality would increase the conductivity and
reduce the series resistance. Conductivity modulation arising from this
 1(A)

Fig. 3.15 Injection ratio y{ = J^/J) as a function of current for a Au-Si Schottky diode.
(From Yu and Snow, 1969. Copyright American Institute of Physics.)

cause lias been reported by Jäger and Kosak (1973) and by Stolt et al.
(1983) as arising in high-power Schottky rectifiers; it has also been
observed by Andersson, Hyder, and Berg (1973) in sihcon particle
detectors. The effect is noticeable only with large barrier heights and in
weakly doped material. According to Green and Shewchun, conductivity
modulation can cause an inductive component of impedance to appear
under large forward bias in weakly doped semiconductors. Hole
injection is also important when Schottky diodes are used under 'punch-
through' conditions as in B A R R I T T diodes (Sze, Coleman, and Loya,
1971). A further discussion of hole injection is given in Appendix C.
Because of the occurrence of «f in eqn (3.30), the injection ratio can
be large in small band-gap materials (physically this arises because />o is
proportional to n?). Hole injection is therefore much more significant in
germanium than in silicon or gallium arsenide, a fact which explains its
importance in the early days of semiconductor physics. There is a wide
range of injection phenomena in low band-gap semiconductors, which
has been comprehensively described by Henisch (1984). Henisch also
discusses the complications that arise in the so-called 'relaxation' semi­
conductors, in which the dielectric relaxation time e/o exceeds the
minority carrier lifetime. This book is concerned only with 'lifetime'
semiconductors, in which the minority carrier lifetime exceeds the
dielectric relaxation time; all semiconductors of technological import­
ance fall into this category.

3.5.2 Hole injection in point contacts

The minority-carrier injection ratio in a point-contact rectifier may be

quite large (approaching unity), and the operation of point-contact
123
transistors depended on this fact. The reason for this difference in
behaviour is not clear. A l l the work on point-contact transistors was
carried out using germanium, and there is almost no reliable information
available about the relevant barrier heights; it is therefore impossible to
say whether eqn (3.30) would predict a large or a small injection ratio.
Equation (3.30) is in fact quite sensitive to <j>^, N^, and (the latter
through nf), and as we have already seen, the low band-gap of
germanium compared with silicon and gallium arsenide would undoub­
tedly cause a considerable increase in y^- However, the conditions of
operation of point-contact rectifiers are vastly different from those
pertaining to plane contacts, especially with regard to the current density.
In their original paper on the point-contact transistor, Bardeen and
Brattain (1949) quote currents of ~ 10""-' A flowing in tips of radius 5
fim; this corresponds to a current density of about lO'' A m"^. With such
a high current density the change in potential at the edge of the depletion
region due to the forward bias must be quite large and the bands almost
flat, so that eqn (3.30) does not apply. Under such conditions the hole
current is determined more by the transport processes in the neutral
region than by the contact itself, and values of y^ approaching unity are
possible, as was shown by Braun and Heiusch (1966). According to
Clarke, Green, and Shewchun (1974), the hole current is accentuated by
the spherical symmetry, which causes the hole concentration at a
distance r to decrease as and so increases the diffusion rate.

3.6 Reverse characteristics

According to the thermionic-emission theory, the reverse current density

of an ideal Schottky diode should saturate at the value JQ =
A**T^e,xp{- q^^/kT). There are several causes of departure from this
ideal behaviour, and these are outlined in the foUowing sections.

3.6.1 Field dependence of the barrier height

If for any reason is dependent on the electric-field strength in the

barrier, the reverse characteristics will not show saturation. There are
several possible mechanisms, all of which predict that should be a
decreasing function of S'^^y., the maximum field strength in the barrier.
Since S'^^^ increases with reverse bias V^, it follows that decreases with
increasing V„ and the current does not saturate but increases propor­
tionally to exp{qA(j>JkT), where Aj^^ is the lowering of the barrier due to
the field.
The simplest form of barrier lowering is that due to the image force.
The corresponding value of A^j, is given by eqn (1.26a) with V replaced

124
by - V^. Since A ^ b i is proportional to Vy* for large values of reverse
bias, a plot of In J against Vy* should give a straight line, the intercept of
which on the In / axis gives JQ. Arizumi and Hirose (1969) have
described sihcon Schottky diodes with a donor density of 10^' in
which the reverse characteristics can be completely explained in terms of
image-force lowering up to a reverse bias of 10 V, and Missous,
Rhoderick, and Singer (1986c) have found similar behaviour in epitaxial
aluminium-gallium arsenide diodes made by M B E . These diodes appear
to be the most nearly ideal that have yet been made.
More usuaUy, the barrier lowering necessary to explain the lack of
saturation is considerably greater than that due to the image force. One
of the commonest causes of a field-dependent barrier height is the
presence of an interfacial layer, and this wiU be discussed in Section 3.8.
Andrews and Lepselter (1970) observed a lack of saturation in the
reverse characteristics of sihcon-sihcide Schottky diodes which could be
explained in terms of a field-dependent barrier height. These diodes were
made by depositing R h , Zr, or Pt on sihcon and heating the combination
to such a temperature that the metal reacted with the silicon to form a
silicide. A l l these sihcides exhibit metalhc conductivity, and a Schottky
barrier is formed between the sihcon and the metal silicide which is
thought to be free from any interfacial layer. Andrews and Lepselter
were able to explain their reverse characteristics by assuming an
empirical field dependence of the form A^^, = a ' ^ „ „ with values of a' in
the range 15-35 A. A n example of how successfully this empirical
analysis fits their data is given in Fig. 3.16. The origin of the term a' was
presumed not to be the same as that of the parameter a introduced in
Section 1.4.3 because of the assumed absence of an interfacial layer.
A possible explanation of the a'<^n,ax term has been postulated by
Andrews and Lepselter and others (Parker et al, 1968; Broom 1971) as
follows. The silicide is supposed to form a perfect metal-semiconductor
junction with the sihcon and, according to the Heine model (Section
1.5.2), the wave-functions of the conduction electrons in the metal
penetrate into the forbidden gap of the semiconductor in the form of
exponentially damped evanescent waves. These exponential tails con­
stitute an electric dipole which distorts the barrier shape in such a way
that its height is reduced, and to a first approximation the reduction is
proportional to S"^^^. But although this explanation may be qualitatively
correct, there are difficulties in applying it in a quantitative manner.
According to both Parker et al. and Broom, the position of the maximum
of the potential barrier required to explain their data is about 50 A from
the interface. It is a very poor approximation to assume that the tails of
the conduction-electron wave-functions can be represented by a simple
exponential over this whole distance when the bands are bent as
drastically as they are near the potential maximum. Furthermore, both

125
 .001 .01 0.1 1.0 10 100
K,(V)

Fig. 3.16 Reverse characteristics of a ZrSij-Si Schottky diode. The theoretical curves are
based on a field-dependent barrier height of the form "'^max> with a'=15 A.
(Andrews and Lepselter, 1970. Copyright Pergamon Journals Ltd.)

Andrews and Lepselter and Parker et al. assume that the contributions
due to the image force and to wave-function penetration are additive.
This is only true if the effect of the image force is negligibly small; in
general, the two effects interfere with each other and are not additive.
Moreover, such an explanation is very questionable when one remembers
that Arizumi and Hirose and Missous et al. observed reverse charac­
teristics that could be explained in terms of image-force lowering alone.
It should be appreciated that the existence of an 'intimate' (i.e. chemi­
cally abrupt or oxide-free) interface does not necessarily rule out an
explanation in terms of an interfacial layer. In all theories of interface
states, the electrical centre of the interface state must be separated from
the metal, otherwise the states would have no observable effect. This
separation may come about because the states are spread put, as in the
Heine model, or because they are localized at some distance from the
metal, as in the 'unified defect' model of Spicer and his co-workers (see
Section 1.5.3). In either case, the region of semiconductor between the
surface of the metal and the electrical centre of the interface states
behaves like an insulating layer, as in the Bardeen model.

3.6.2 The effect of tunnelling

Tunnelling through the barrier becomes significant at lower doping levels

126
Ill lllC ICVCISC UllCL-llUll m a i l lll LIH.^ iwivvcina uix^v/nv/n ijwwt*uo\^ ^^±^л^J
voltages involved are usually much greater; furthermore, the apphcation
of only a moderately large reverse bias can cause the potential barrier to
become thin enough for significant tunnelling of electrons from the metal
to the semiconductor to take place even at low doping levels. The effect
of tunnelling can again be described as either field or thermionic-field
emission as shown in Fig. 3.17.
Figure 3.18 shows the boundaries between the thermionic and
thermionic-field regimes on a plot of band bending F^ against tempera­
ture, with £ 0 0 as a parameter (Padovani and Stratton, 1966). The curves
are based on eqn (3.26), and apply to both forward and reverse bias. It
can be seen from the figure that, at room temperature for reverse biases
around 3 V, departures from pure thermionic emission begin at donor
concentrations greater than 10^^ m"^ for silicon (corresponding to
EQQ= з meV), compared with 5 x lO^^ m"^ for forward bias. In the
thermionic-field regime the current/voltage relationship has been
derived by Padovani and Stratton as
7= - 7 , e x p ( F y E ' ) (3.31)
for values of reverse voltage exceeding 3 kT/q, where
E' = Eoo/{(qE,o/kT) - іапЩЕ,,/кТ)]
-1
q
кт £0/
since EQ = EQQCOth{qEaQ/kT) (see Section 3.3.1).
Thermionic-field emission

Fig. 3.17 Field and thermionic-field emission under reverse bias. (After Padovani and
Stratton, 1966.)

127
 о , 10 20 30 40
kTlq (mV)

Fig. 3.18 Boundaries between thermionic and thermionic-field regimes on a plot of

band-bending (V^) against temperature with E^^ as a parameter. The curves are derived
from Padovani and Stratton's (1966) eqn (51) modified as follows: (1) - £ is replaced by
-E + ^2 (their notation, see Padovani, 1971, p. Ill); (2) E^ is replaced by - E^. The latter
modification seems to be necessary because the condition for thermionic-field emission
must depend on the band-bending V^, which is equal to Ев + ^2 ~ ^ in their notation, since
Kj determines the shape of the potential barrier. If these modifications are made, their eqn
(51) is identical with their eqn (33) and their Figure 13 becomes identical with their Figure
3 when expressed in terms of V^. The two figures must coincide for small values of bias in
either direction. The curves drawn here therefore apply to both forward and reverse biases,
values of less than ф^-^ (our notation) denoting forward bias (i.e. negative V,).

It is instructive to compare this result with eqn (3.27) which, though

introduced in the context of forward bias, applies equally to reverse bias.
For reverse bias voltages exceeding 3 A:T/g, (3.27) becomes

1
/ = - J^exp
" ' F r - -
which is identical with (3.31) as far as the exponential term is concerned.
For reverse bias, is a slowly varying function of applied bias given by
1/2
Яфь
7.= exp(-^ib/Eo). (3.32)
cosYi {qEao/кТ)

For small values of reverse bias and for values of doping density such
that kT> qEgo (the condition for thermionic-field emission), this expres­
sion for 73 coincides with that for forward bias [eqn (3.27a)]. Thus the

128
complete expression for reverse bias agrees with that for forward bias
near the origin as it should, since the characteristic must be continuous.
Equation (3.31) can be written as
7= -J,exp(qV,/n'kT),
where n'~' = 1 — Here n (= qEg/kT) is the ideahty factor for
forward bias shown in Fig. 3.10. Hence the concept of an ideality factor
can be used under reverse bias, but it is not the same as the ideality
factor for forward bias.
At higher doping concentrations field emission may occur. Unhke the
case of forward bias, field emission can take place in non-degenerate
semiconductors because it involves tunnelling of electrons from the
metal into the semiconductor, and the metal is always degenerate. The
analysis of Padovani and Stratton predicts that the transition from
thermionic-field to field emission occurs when (^l^^kT/qEg^vy^)1
where Fj = + ^ ~ I- The boundaries between the two regimes are
shown in Fig. 3.19, which is a plot of against kT with EQQ as a
parameter. The barrier height has been taken as 1 eV. For silicon,
donor concentrations in excess of about 5 x 10^" m~^ are necessary if
field emission is to occur at room temperature. The expression given by

kVq (mV)
Fig. 3.19 Boundaries between field and thermionic-field emission on a plot of band-
bending (Fj) against temperature, with £oo as a parameter. The curves involve approxi­
mations which are vaUd only if V,> ^f,, i.e. if is greater than about 2 V. (Derived from
Padovani and Stratton, 1966, Fig. 12.)

129
Padovani and Stratton for the J / V characteristic under field-emission
conditions is complicated, but in the very low temperature limit it
simplifies to

00 2^b
J = A* exp (3.33)
3£oo(^b+J^r-^) 1/2

Figure 3.20 shows the forward and reverse characteristics at 4.2 K of a

gallium arsenide diode with a donor concentration of 2 x 10^" m~^
(Padovani, 1971), the reverse current being due to field emission. A t low
bias voltages the reverse current exceeds the forward current as was
predicted originally by Wilson (1932).
TunneUing is one of the most common causes of 'soft' reverse charac­
teristics. It is particularly important near the edge of the metal contact,
because the crowding of the field lines causes an increase in the field
strength which decreases the barrier width and also exaggerates the
image-force lowering. The effect is accentuated if the surface of the semi­
conductor adjacent to the metal is accumulated (i.e. the electron
concentration is increased) due to the presence of positive surface

10-'^

0.0 0.1 0.2 0.3 0.4 0.5 0.6

V(V)
Fig. 3.20 Forward and reverse characteristics at 4.2 K of a GaAs Schottky diode having
Na = 2x 10^" m - l (From Padovani, 1971. Copyright Academic Press.)

130
charges, as this makes the barrier at the edge even thinner (Yu and Snow,
1968).
Edge effects may be minimized by using a method of surface pre­
paration which does not cause accumulation (Smith, 1968) and almost
totaUy eliminated by using a guard-ring. The simplest type of guard-ring
is a concentric metal ring which is maintained at the same potential as the
main contact (Padovani, 1968; Tove, Hyder, and Susila, 1973), but this is
effective only if the separation between the guard-ring and main contact
is comparable with the depletion width. This restricts its use to high-
resistivity material. A more effective method is to use a diffused p-type
ring under the edge of the contact (Lepseher and Sze, 1968) as shown in
Fig. 3.21. The presence of the p-type material eliminates the high field at
the edge and places a reverse-biased p - n junction in parallel with the
Schottky barrier. The p - n junction can be made to have good reverse
characteristics, and the resulting diode is then very nearly ideal with a
reverse characteristic determined by image-force lowering. Hole injec­
tion from the p - n junction is normally greater than from the Schottky
diode alone, and if it is important to minimize hole injection, two back-
to-back p - n junctions may be used (Saltich and Clark, 1970). Edge
effects can also be minimized by using a mesa structure (Gutknecht and
Strutt, 1974). A review of methods avoiding edge effects has been given
by Tove (1982).
Al

Fig. 3.21 Schottky diode with p-type guard ring (Lepselter and Sze, 1968. Copyright
A.T.&T.)

if edge effects are avoided, a Schottky diode will eventuaUy exhibit

breakdown which, depending on the donor density, is due either to
thermionic-field emission or to impact ionization causing an avalanche of
electron-hole pairs. The dependence of the avalanche breakdown
voltage on donor density is the same as for a planar p - n junction (Sze
and Gibbons, 1966; Okuto and Crowell, 1974), and the transition from
avalanche breakdown to thermionic-field emission takes place at donor
densities between 5 x 10^^ and lO^" m~^ for sihcon at room temperature
(Chang and Sze, 1970). Сое (1971) has reported the fabrication from 25
Q-cm silicon of aluminium-silicon diodes with multiple guard-rings
which exhibit reverse breakdown voltages in excess of 1 kV. Nomograms
showing the breakdown vohage of ideal Schottky diodes made from
sihcon and gallium arsenide with a range of dopant concentrations have

131
been published by Miller, Lang, and Kimerling ( l y / /) ana are repro­
duced in Appendix A .

3.6.3 Generation in the depletion region

If the effects of tunnelling and image-force lowering are reduced to

negligible proportions by the use of a low donor density, there may be an
appreciable reverse current due to the generation of electron-hole pairs
in the depletion region. This is the inverse of process (c) (Section 3.1)
and gives rise to a current-density component 7g ~ qn^w/2r^, where w is
the width of the depletion region and the lifetime within the depletion
region, /g increases with reverse bias because w is proportional to
(Vgo+V^y^'^. Like the inverse process, generation current is most
important in high barriers and in low-lifetime semiconductors, and is
more pronounced at low temperatures than at high temperatures because
it has a lower activation energy than the thermionic-emission component.
It is a frequent cause of the lack of saturation of the reverse current in
gallium arsenide Schottky diodes.

3.7 Transient effects

It is pertinent to enquire at this stage whether there are any transient

effects in Schottky diodes which arise from a mechanism analogous to
minority-carrier storage in p - n junctions. In a p - n junction, the
application of forward bias results in the injection of minority carriers
(predominantly electrons into the p-type side if the donor density greatly
exceeds the acceptor density). If the bias is suddenly reversed, these
injected minority carriers must be cleared away before the diode assumes
a high-resistance state, and a large current momentarily flows in the
reverse direction. The charge associated with the injected minority
carriers is equal to Jr^^ per unit area, where 7 is the forward current
density and r^^ is the recombination time of electrons in the p-type region
(see, for example, Sze, 1981).
The analogue of this process in a Schottky diode is not the hole
injection discussed in Section 3.5, which refers to the injection of those
carriers (holes) which contribute very little to the current, but rather the
injection of electrons into the metal. A s we saw in Section 3.2.1 the
electrons injected into the metal are 'hot' in the sense that they possess
much more energy than corresponds to thermal equilibrium, and under
the application of reverse bias they can diffuse back into the semi­
conductor until such time as they have lost so much energy that they can
no longer surmount the barrier. The time taken for this to happen is
effectively equal to the mean free time for electron-electron colhsions in

132
tne meiai, since m sucn a collision tne not electron loses on average
about half its excess energy (Loveluck and Rhoderick, 1967); this time
plays a role equivalent to the electron-recombination time in a p-n
junction. The mean free path for electron-electron colhsions when the
hot electron has 1 eV excess energy above the Fermi level is about 500 A
(Ritchie and Ashley, 1965) and the mean free time is about 10"'" s. This
is neghgibly small, and minority-carrier storage effects arising from this
mechanism have not been observed in Schottky diodes.
It is natural to ask whether there may be an observable recovery time
in a Schottky diode associated with the charge due to the injected holes
because, although the hole current is very much smaller than the electron
current, the hole recombination time in the semiconductor ( ~ 10~* s
in a typical silicon diode) is many orders of magnitude greater than the
electron-electron mean free time in the metal. The injected-hole charge
per unit area is given by J^r^y,, which is equal to yhJr.y, where 7 is the total
forward current density and the hole-injection ratio given by eqn
(3.30). The charge storage arising from hole injection can be described in
terms of an effective recovery time which is equal to the injected
charge per unit area divided by the total current density 7 and is given by
y^rrh. Although may be comparatively long, y^ is so smaU that the
recovery time due to injected holes is usually not more than about 10"" s
and is negligible for most purposes. A t high current densities rises
because of the increase in y^ and depends in a complicated way on the
boundary conditions at the back (ohmic) contact as discussed by
Scharfetter (1965). In practice, the recovery time of a Schottky diode is
determined by circuit considerations rather than by any intrinsic
electronic process associated with the conduction mechanism.

3.8 The effect of an interfacial layer

Unless they are manufactured by cleaving the semiconductor in an ultra­

high vacuum or by M B E , Schottky diodes nearly always have a thin
oxide layer between metal and semiconductor. This interfacial layer may
be considered to be an insulator, even though it may be so thin (of the
order of 10 A) that it does not possess the band-structure characteristic
of a thick oxide. A n idealized band diagram for such an imperfect
Schottky barrier is shown in Fig. 3.22.
The insulating layer has three effects:
1. Because of the potential drop in the layer, the zero-bias barrier
height! is lower than it would be in an ideal diode, provided there is
no charge contained in the layer.
tBy 'the barrier height' we mean i-^- maximum height of the bottom of the
conduction band in the semiconductor relative to the metal Fermi level. This is distinct
from the barrier presented by the insulating layer.

133
Fig. 3.22 Schottky barrier with interfacial layer: zero bias; forward
bias.

2. The electrons have to tunnel through the barrier presented by the

insulator so that the current for a given bias is reduced in a manner
equivalent to a reduction in A**.
3. When a bias is apphed, part of the bias voltage is dropped across the
insulating layer so that the barrier height ^t, is a function of the bias
voltage (see Section 1.4.5). The effect of this bias dependence of the
barrier height is to change the shape of the current/voltage charac­
teristic in a manner which can be described in terms of an ideality
factor n defined by eqn (3.14).
Figure 3.23(a) shows the forward I/V characteristics of some gold-
silicon Schottky diodes prepared by Card (Card and Rhoderick, 1971a)
in which a thin film of SiOj was intentionally grown on the sihcon before
the metal was deposited. It is apparent that the current decreases with
increasing film thickness due to the effects mentioned above. Effect (1)
can be eliminated by measuring V^Q independently and normalizing the
curves to a common zero-bias diffusion potential. Figure 3.23(b) shows
the curves so normalized; these now exhibit the combined effects of (2)
and (3). The reduction of current due to tunnelling through the oxide
layer is clearly visible, but there is also a departure from ideal behaviour
which may be described by an «-value which increases with increasing
oxide thickness. The dependence of the n-value on the oxide thickness
may be explained in terms of the analysis of Section 3.2.4. According to
eqn (3.14) we have

dV'

134
 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(a) V(V) (b) K(V)

Fig. 3.23 (a) Forward I/V characteristics of Schottky diodes with oxide interfacial layers
of various thicknesses, (b) Same data as (a) normalized to a common diffusion potential
(Card and Rhoderick, 1971a).

We shall ignore image-force lowering so that the bias dependence of is

contained entirely in and hence d^JdV= dfy,/dV. From eqn (1.25),
after some manipulation and making use of eqn (1.17), we find that

dV E,^^, + aqNg
so that
1 ^ Cs^max

or
aqN,
n =l +

a
= 1 +
w

= 1+ ^ (3.34)
w{6qD, + E;) ^ ^
where w is the width of the depletion region. We have made use of the
result (from the depletion approximation) that S'^^^ = qN^w/E^. A similar
result, but omitting the effect of surface states, has been obtained by
Strikha (1964). Card and Rhoderick (1971a) have extended this analysis
to the case where the interface states are in equilibrium with the bulk of
the semiconductor rather than with the metal. They suppose that the
interface states can be divided into two groups, such that all the states in
one group (density 7)53) are in equilibrium with the metal while those in
the other group (density D^y,) are in equilibrium with the semiconductor.
The general expression for n becomes

135
 {d/e;){{ejw) + qPj
n = 1+ — . (3.35)

The dependence on D^^ and D^^, can be understood physicaUy because

states m equilibrium with the metal tend to hold constant (i.e. to
reduce n), while states in equilibrium with the semiconductor tend to
hold Vg constant (i.e. to increase n). Equation (3.35) has been used by
Walker (1974) as the basis of a method of measuring surface state
densities from I/V characteristics. Generally speaking, if 6 < 30 A most
of the states are in equilibrium with the metal, while if 5 > 30 A most of
the states are in equilibrium with the semiconductor. Films of thickness
of the order of 20 A usually lead to values of n in the range 1.3-1.5. The
transition coefficients of the films used in Card's experiments were
always greater than was expected on the basis of the known band
structure of SiOj, being typically of the order of 10~^ for a 20 A film.
Gray (1973) has suggested that this may be due to the presence of
pinholes. A n alternative explanation is that the effective barrier pre­
sented by the insulator may be decreased by the image force, since within
the insulator the image gives rise to an attractive force both as the
electron approaches the metal and as it leaves the semiconductor,
resulting in a potential in the shape of an inverted U . However, as
Kleefstra and Herman (1980) and Tugulea and Dascalu (1984) have
shown, if the permittivity of the semiconductor exceeds that of the
insulator (as is usually the case) there is also a positive potential spike
within the semiconductor through which the electron has to tunnel before
it can enter the insulator (Fig. 3.24), and this has the opposite effect.
There is a further possible mechanism, not hsted in Section 3.1 or
shown in Fig. 3.1, which may contribute towards the current if there is an
interfacial layer. This consists of the capture of electrons from the
conduction band by interface states, followed by tunnelling from the
interface states into the metal. But some thought is necessary before one
can be sure that this mechanism really contributes an additional current,
and one must consider what would happen to an electron if it were not
captured by an interface state. If the answer is that an electron not
captured by an interface state would simply tunnel direct from the semi­
conductor conduction band into the metal, then the current via interface
states occurs at the expense of the ordinary thermionic current, and does
not constitute an additional component. But if the electron would other­
wise have been reflected back into the semiconductor, the current via
interface states can provide an additional contribution. This may occur if
the probability of tuimelling through the interfacial layer is small, or if
either of the processes represented by /p and in Section 3.2.7, namely
phonon backscattering and quantum-mechanical reflection, is significant.
The current contribution from interface states in this case has been

136
 Distance ——•

Fig. 3.24 The broken line shows the effect of the image force on the band-diagram for an
interfacial layer with < e,. (After Kleefstra and Herman, 1980.)

discussed theoretically by Strikha and Kil'chitskaya (1968) and by Card

(1975b). The magnitude of the interface state current is, of course,
criticaUy dependent on the parameters one chooses to represent the
capture cross-section and density of interface states, and also on the
probability that an electron in an interface state can tunnel into the metal.
For the particular set of parameters chosen by Card, the current due to
tunnelhng from interface states turns out to be of the same order as the
thermionic emission current for a sihcon Schottky diode if there is an
interfacial layer of S i O i of thickness 10 A between the silicon and the
metal, and a density of interface states of 10'^ m"^ eV~'. For nearly ideal
Schottky diodes with a thinner interfacial layer or a smaller density of
interface states, the contribution due to tunnelling from the interface
states seems to be unimportant, for the reason already mentioned.
Another effect of an interfacial layer is to increase the minority-carrier
injection ratio under forward bias. For very thin layers, the effect arises
predominantly because the electron current is limited by thermionic
emission and is therefore proportional to the probability of electrons
tunnelling through the oxide layer, whereas the hole current is controlled
by diffusion in the neutral region of the semiconductor and is relatively
unaffected by the presence of the interfacial layer. For thicker layers

137
there is a major reahgnment of the bands in the semiconductor with
respect to the Fermi level in the metal, with the result that far more holes
are able to tunnel from the metal into the semiconductor. The effect has
been investigated by Card and Rhoderick (1973) in the context of gold-
silicon diodes, and exploited by Livingstone, Turvey, and Allen (1973)
and Haeri and Rhoderick (1974) to improve the injection efficiency of
electroluminescent diodes. Values of у of about 0.1 have been obtained
in this way.
In the reverse direction, the presence of an interfacial layer causes the
effective barrier height to decrease with increasing bias, so that the
reverse current does not saturate. The effect is formally equivalent to a
reduction in barrier height of the form A ^ ^ ^ ct^max (see Sections 1.4.3
and 3.6.1). Card and Rhoderick (1971b) have shown that, because of the
reduction in barrier height, the reverse current of a diode with a fairly
thick interfacial layer may actually be greater than that of a diode with a
very thin layer (see Fig. 3.25). Although the electrons have to tunnel
through the barrier presented by the insulator, this barrier is very thin
and is overcompensated by the reduction in the Schottky barrier. This
effect is a common cause of the 'soft' reverse characteristics of diodes
prepared under poor vacuum^ conditions or with a contaminated semi­
conductor surface. Thanailakis and Northrop (1971) have found that in
aluminium-germanium diodes with an observable interfacial layer, the
lowering of the barrier under reverse bias can exhibit a long time
constant of the order of tens of minutes or even hours; this can be
explained in terms of a model involving slow interface states.
3.9 The To'effect

If the departure of n from unity arises from image-force lowering or from

interface effects, n, should be independent of temperature, but if it is due
to thermionic-field emission or to the effect of recombination in the
depletion region, n will be temperature dependent. The majority of
Schottky diodes exhibit n values which depend on temperature.
Padovani and Sumner (1965) and Padovani (1967) have shown that for
some Schottky diodes made from sihcon or gallium arsenide the J/V
characteristic can be well represented by the equation
J = AT^exp{-<l,^,/k{T+ To)}[exp{gF/fc(T+ To)} - 1]
where TQ is a parameter which is independent of temperature and voltage
over a wide range of temperatures. This is equivalent to writing
n = l + (TQ/T), SO the experimental results imply a very special sort of
temperature dependence of n. Padovani (1971) states that, for a total of
25 diodes made on the same slice of gallium arsenide, the values of TQ
varied from 10 K up to 100 K . Moreover, other diodes have been
reported (for example Arizumi and Hirose, 1969; Gol'dberg, Posse, and
Tsarenkov, 1975) which exhibit almost ideal J/V characteristics with
To 0. The effect is evidently not an intrinsic property of ideal Schottky
barriers but an artefact introduced by the fabrication process. Various
attempts have been made to explain such a temperature dependence in
terms of tunnelhng (Crowell and Rideout, 1969), particular distributions
of interface states (Saxena, 1969; Levine, 1971; Rhoderick, 1975;
CroweU, 1977), and a non-uniformly doped surface layer (Padovani,
1971; Crowell, 1977). It is quite possible that more than one of these
mechanisms may operate simultaneously.
It is curious that for these diodes the temperature dependence of the
exponential term involving j^^o should apparently have the same form as
that of the term involving V. Although all the mechanisms which have
been invoked to explain the form of the current/vohage characteristic
(i.e. n values greater than unity) also affect the zero-bias barrier height,
they do not do so in a simple manner. There is also a temperature
dependence of which arises from the variation of the band gap with
temperature, but this makes no contribution to the temperature depen­
dence of the term involving V.

3.10 Numerical analysis of current flow

Over the last decade or so there has been steady progress in the
numerical modelling of current-flow phenomena in semiconductor
devices (see, for example, Engl, Manck, and Wieder, 1976; Selberherr,
1984), and it is natural to ask whether such procedures lend themselves
to the analysis of current transport in Schottky diodes.
As we have seen, for ideal single-carrier diodes (i.e. neglecting hole
injection and recombination in the depletion region) the current is
usually limited by thermionic or thermionic-field emission if the (n-type)
semiconductor has a high electron mobility. In such a situation, the
current transport mechanism in the depletion region is unimportant and
the bulk properties of the semiconductor only appear through the resist­
ance of the neutral region between the edge of the depletion region and
the ohmic contact. Deviations from this ideal model are usually due to
departures from a perfectly abrupt metal-semiconductor interface, such
as the existence of an interfacial layer or of complex metallurgical phases
adjacent to the metal. In such cases, the precise nature of the barrier is
usually highly speculative, and it is this largely unknown factor which
hmits our abihty to analyse the current/voltage characteristic. If the exact
shape of the barrier close to the interface is known, the current flow can
usuciUy be adequately explained in terms of thermionic or thermionic-
field emission, modified where necessary by the field dependence of the
barrier which results from ац interfacial layer, and there is no need to
resort to numerical modeUing. In extreme cases where the Bethe
criterion is not satisfied due to the existence of a low barrier and/or a
large forward bias so that diffusion effects become important, the
analysis of Crowell and Beguwala (1971) is generally adequate to explain
the experimental data.
However, in the case of two-carrier systems (for example when hole
injection or recombination within the depletion region is important)
numerical computation may have an important role to play. Here, the
current flow may be dependent on such parameters as the spatial
distribution of recombination centres and their capture cross-sections,
the diffusion coefficient of minority carriers in the neutral region of the
semiconductor, and the precise geometry of the device, which cannot
easily be dealt with analytically but may be handled by numerical
computation. A detailed discussion of computer modelling techniques is
beyond the scope of this book, and the reader is referred to papers by
Green and Shewchun (1973) and by Tove, Elfsten, and the Uppsala
group (Masszi et al, 1981; Elfsten and Tove, 1985), or to the book by
Henisch (1984).

140
4 The capacitance of a Schottky barrier

4.1 The capacitance of an ideal diode under reverse bias

The depletion region of a Schottky barrier behaves in some respects like a

parallel-plate capacitor. It is important to know what factors determine its
capacitance, not only because reverse-biased diodes are used in practice
as variable capacitors (varactors), but also because measurements of the
capacitance under reverse bias can be used to give information about the
barrier parameters.

4.1.1 The general case

Consider the ideal Schottky diode (without any interfacial layer) with a
band diagram as shown in Fig. 4.1(a). The semiconductor is assumed to
be n-type. The solid lines refer to a reverse bias V, and the broken lines to
a larger reverse bias V, + A F , . If the reverse bias is increased from F^ to
Fr + A Fr, electrons in the conduction band of the semiconductor recede
further from the metal and increase the width of the depletion region from
wto w+ Aiv. A t the same time, if there is a significant concentration of
holes in the semiconductor immediately adjacent to the metal (as will
happen if the surface of the semiconductor is invertedf), the hole
concentration will decrease because the hole quasi-Fermi level coincides
with the Fermi level in the metal (see Appendix D). The change in the
charge in the depletion region gives rise to capacitance. In addition, there
is a parallel conductance due to the reverse current of the diode, so the
equivalent circuit consists of a capacitor in parallel with a resistor.
There are three sources of charge in the barrier region: the positive
charge 2 d 'lue to the uncompensated donors in the depletion region; the
positive charge due to the holes in the valence band; and the negative
charge (2m due to electrons on the surface of the metal. It is not
immediately apparent which of these charges have to be taken into
account in calculating the capacitance, but the question can be resolved
by considering the various contributions to the current through the
depletion region. When the bias changes, the total current through the
depletion region consists of a conduction current plus a displacement
current. The displacement current density 4 (= E^dS'/dt) arises because
fThe surface is said to be inverted if the barrier height is so large that the region of the
semiconductor adjacent to the metal is p-type.

141
Fig. 4.1 (a) Schottky barrier under reverse bias: bias voltage V/, bias
voltage V, + AF,. (b) Charge density due to mobile carriers (charge on surface of metal not
shown.)

the electric field in the depletion region increases with time as the bias
increases. The conduction current consists of two parts: a current density
/^1 due to the drift and diffusion of electrons injected over the barrier
from the metal, and a current density J^2 which arises because there is a
flow of electrons and holes out of the depletion region as the negative bias
increases. exists even if the bias is constant in time, and constitutes the
reverse current of the diode; it is independent of position throughout the
depletion region. The component 7^2 exists only if the bias is changing.

142
О'лА ЯПЧе"^ бее aide -Кете i^«,3.f- 4^а.^1^'^^'^^"Ь-
regions of the semiconductor where the space-charge density changes
with time. If the bias increment AF^ varies sinusoidally with time, the
component 4 i is in phase with AF^ and gives rise to a paraUel
conductance, while the components J^2 '^^^ h are in phase-quadrature
with AFr and give rise to a capacitance. Both J^^ and ^ vary with x, but
satisfy the continuity equation in such a way that the sum /c2 + 4 is
constant.
The charge density due to the mobile carriers (electrons and holes) is
shown in Fig. 4.1(b). In addition, there is the fixed positive charge (not
shown in the diagram) due to the ionized donors, which exactly cancels
the charge due to the electrons in the bulk of the semiconductor. To the
right of the plane x = x^, where E'^ - Ey(x^ > ^ + (3kT/q), the hole density
will be less than a few per cent of (assuming A^^ ~ Щ; and to the left of
the plane X == Xj, where E^(X2) - E^(°o)> 3kT/q, the electron density will
be less than a few per cent of Ng. Therefore, within the range Xj to X2 the
charge density is essentially equal to the charge on the donors qNg and is
independent of time. A s the bias varies, electrons enter or leave the
depletion region from the bulk of the semiconductor and holes enter or
leave from the metal, so that in the region x, to X2 the component /^2 of
the conduction current density must be zero, and the only quadrature
component will be J^. Hence between Xj and X2 the capacitive current will
be equal to the displacement current Jg = e^dS'/dt. From the usual circuit
definition of capacitance, the displacement current density can also be
written as Jg = C(dAF/df),t where С is the differential capacitance per
unit area, so that
^dAF, Z?-^
С—:— = e.
dt " dt
N o w ^ is an explicit function of V, so we may write
df^^ df^ dAF,
dt dV, ^ dt

and therefore
dS'
C = . . - . (4.1)

The electric field strength S' must be evaluated within the region x, to X2,
where the total charge density is essentially equal to the donor charge
density and is independent of bias.
The field strength^ can easily be found by applying Gauss's theorem to
a closed surface bounded by two planes parallel to the interface, one
t is assumed independent of time, and the time dependence is contained in A V,.
143
being situated between and X2 and the otner rar 1^1 stiit ijiWifiCl ^^..^1-
semiconductor where the bands are flat and the charge on the electrons'
compensates the charge on the donors. If the bias varies, the charge
enclosed by this surface varies because of the change in the width of the
depletion region, and the variation in this charge is equal to the change in
the total charge per unit area Qg due to the uncompensated donors.
Hence
EM = AQa
or
dS> ^ dQg

and from (4.1)

C = ^ . (4.2,

This result shows that the capacitance can be calculated by considering

only the charge due to the uncompensated donors and ignoring the charge
due to the holes. Alternatively, since the entire barrier region must be
electrically neutral because there is no electric field in the metal or the
interior of the semiconductor,! Qn, + Qh + Q d 0 and we may write

_ ^((2m + Q h )

This shows that the holes are to be regarded as 'belonging' to the metal
because in Fig. 4.1(a) they enter from the left. The capacitance defined in
this way is a differential capacitance per unit area; i.e. it represents the
response of the depletion region to a change A F , in the reverse bias. It can
be measured experimentally by superposing a small a.c. voltage on the
steady d.c. bias F^ and measuring the capacitance in the normal way by
means of a bridge which responds only to the a.c. component of the
current.

4.1.2 The case in which the minority carriers are negligible

The differential capacitance can easily be expressed in terms of the

diffusion voltage and donor density if the effect of the holes can be
neglected. This will be the case if the barrier height is such that the top of
the valence band at the interface is below the metal Fermi level by at
least ^ + {?>kT/q); this is equivalent to taking x^ = 0. In this case the
electric field at the interface will be due only to the uncompensated
tSee footnote on p. 19.
144
''uuiiuiS aiiu can uc xuuiiij iiuiif cqii \p.\.|'oi /vppciiujpt D wiui //^ pui equal
to zero SO that

o2
' max Nd + ^cxp{-qV,/kT)
[ 9 /
where Fj is the diffusion vohage corresponding to the reverse bias V,. If
qVa> 3 A:rthe last term in the bracket is neghgible and

»2
' max 2 « K , - * ?

From Gauss's theorem the charge due to the uncompensated donors is

given by
kT 1/2

so that

dV,
1/2 kT -1/2
q^M
FH- (4.3a)

Since Vg = Fdo + F„ where Fjo is the diffusion voltage at zero bias, eqn
(4.3a) may be written
1/2 1-1/2
(qeM^
C = ^dO + K - (4.3b)
\ 2 1
Equation (4.3b) shows that a graph of C " ^ as a function of F , should be a
straight hne with a slope of l/qe^N^ and a negative intercept - 1^ on the
V, axis equal to — V^Q + (kT/q) as shown in Fig. 4.2, so that V^Q = F , +
(kT/q). The distance ^ of the Fermi energy below the conduction band,
and hence the barrier height = ^do + ^. can be obtained by finding Ng
from the slope of the hne, smce ^ = (kT/q) \n(NJNa).
Equation (B.l) is obtained with the aid of Boltzmann statistics, and
also assumes that all the donors are ionized. Goodman and Perkins
(1964) have considered the effects of degeneracy of the conduction
electrons and of incomplete ionization of the donors. They have shown
that the combined effect is to multiply the term kT/q in the radical of
(4.3a) or (4.3b) by the factor

2a
K =— - In a,
3
145
Fig. 4.2 Variation of C"^ with V, for Schottky barrier.

where a ( = A^d/^d) is the fraction of the donors which is ionized in the

bulk of the semiconductor, and Fj{-^) is the Fermi-Dirac integral of
order (see, for example, Blakemore, 1962). In the case of strong
degeneracy, ^ is negative. If the donors are incompletely ionized but
there is no degeneracy (as would be the case for deep donors with
lying between the donor level and the bottom of the conduction band so
that ^ is positive), F3/2(- ^)/Fi/2(~ ^) has the value 3/2, and K = a-lna,
a result which was first obtained by Dewald (1960). The correction is not
significant unless a is as low as about 0.1, in which case a-In a" 2.4. If
a exceeds 0.5, the correction, amounts to less than 20% of kT/q. The
slope of the graph of versus V, is still given by 2/qe^Ng, where Ng is
the total donor density. If the conduction electrons are strongly
degenerate (^ < 0) but the donors are completely ionized, which would
be the case if the donor ionization energy is negligibly smaU as for very
heavy doping, « = 1 and 7^3/2(~ l ) / ^ i / 2 ( ~ ^) tends towards the value
3q\^\/5kT, so that K approaches the value 2q\$\/5kT, and the term kT/q
in the radical of (4.3) is replaced by 2| ^ | / 5 .
If the term kT/q is omitted from eqns (4.3), the result is equivalent to
using the depletion approximation (see Appendix A ) , which assumes that
at the edge of the depletion region the electron density faUs abruptly
from the value Ng to a value negligible compared with N^. In this case the
relationship between C and Vg can be found very simply from first
principles, because according to the depletion approximation the
diffusion voltage is given by

qN,w^ Ql
F> =
2e, 2qE,N,
and the capacitance per unit area by
C-(qe,N,/2Vgy^^=e,/w (4.4)
where w is the width of the depletion region. Equation (4.4) shows that

146
according to the depletion approximation the capacitance is the same as
that of a parallel-plate capacitor with a dielectric of permittivity and
thickness w. This result is easy to remember but is somewhat fortuitous.
The depletion region differs from a parallel-plate capacitor in that w is a
function of the bias voltage and the charge is a space charge rather than a
surface charge on the electrodes. The correction term - kT/q which
appears in the more exact equations (4.3) expresses the effect of the
penetration of the electrons into the depletion region (the so-called
'Debye tail'). Nomograms giving the capacitance and depletion-width of
Schottky barriers on silicon and gallium arsenide according to the
depletion approximation have been published by Miller, Lang, and
Kimerling (1977) for a range of dopant concentrations and diffusion
voltages. They are reproduced in Appendix A .

4.1.3 The effect of minority carriers

If the barrier height exceeds E^-^, the hole density adjacent to the
metal will exceed the donor density (assuming ~ N^), and the electric
field at the interface given by eqn (B.l) will be due partly to the holes and
partly to the uncompensated donors. The charge due to the holes must
be subtracted from the total positive charge to find Qg. The integral
involved caimot be solved analytically, but it can be evaluated
numerically using the G functions of semiconductor surface theory (see,
for example, Frankl, 1967) or by numerical computation. Schwarz and
Walsh (1953) have made detailed calculations for germanium and have
shown that, if the capacitance is expressed in the form C<x(V^+ V^)~^^^,
Vi is not equal to F^o - (kT/q) when the holes are taken into account but
depends on the bias voltage V^. If the barrier height exceeds £ g ~ I , the
effect of the holes is appreciable and results in Fi being less than
Fjo - (kT/q). A plot of against. is therefore not linear, and if the
low-bias part of the curve is extrapolated to cut the F^ axis, the intercept
will underestimate V^Q. The problem has also been discussed by Green
(1976).
For the general case, a quahtative treatment is possible which gives a
simple picture of the effect of the holes on the barrier shape and on the
capacitance. It is obvious that the holes will not have much effect on the
shape of the barrier unless their concentration is comparable to N^.
Remembering that the hole quasi-Fermi level is coincident with the
metal Fermi level (see Appendix D), it is clear that the condition that
p> Ngis that the top of the valence band must he below by less than
an amount or that the bottom of the conduction band must lie above
^p by more than E^ - | . Suppose that this condition is satisfied to the left
of the plane x= Xj (Fig. 4.3). Then for x> x^ the space charge is due
almost entirely to the uncompensated donors and the barrier shape is

147
Fig. 4.3 Shape of Schottky barrier showing effect of minority carriers (holes).

parabolic. For x< the space charge becomes predominantly that of

the holes, and because of the exponential energy dependence of p the
barrier rises very steeply. The contribution of the uncompensated donors
to the barrier potential is as shown by the broken line in Fig. 4.3, which is
simply an extrapolation of the parabolic curve. Because the barrier rises
very steeply for X < X j , the distance x^ is very small (< 100 A) and the
parabolic extrapolation makes an intercept at A; = 0 (point A ) which is
not very different from the height E^ - ^ of the barrier at x = Xy Hence
the total charge due to the uncompensated donors is as if the barrier
height were reduced to E^ - | , and the diffusion voltage to Eg - 2^ + F^.
If the position of the point A were independent of bias, the graph of
versus would be Unear and the intercept would correspond
approximately to a zero-bias diffusion voltage of E^ - 2$, or to a barrier
height of £g - ^. But because the position of A changes sUghtly with bias,
the graph is not quite Unear, though the deviation from linearity is
generally negUgible except for low band-gap semiconductors. The main
effect is that the barrier height inferred from plots of versus will
be less than the true value.
Minority carriers only have a significant effect on the capacitance if the
barrier height is nearly equal to the band gap. In the case of gallium
arsenide, for example, the largest known barrier height on n-type
material is about 1.0 eV, and even in this case the hole concentration at
the interface is negligibly small. However, with germanium (E^ = 0.7 eV)
it is possible to obtain barrier heights as large as 0.6 eV, and, as Schwarz
and Walsh have shown, minority carrier effects (lack of linearity of
versus plots and underestimation of F^o) can be quite appreciable.
A n extreme case in which the effect of minority carriers on the
capacitance is very large has been analysed by Walpole and Nill (1971).
These authors discussed what happens if the barrier height exceeds the
semiconductor band gap, as is apparently the case for lead barriers on
p-type lead telluride and gold barriers on p-type indium arsenide. The
band diagram for such a situation is shown in Fig. 4.4, where the semi­
conductor is assumed to be degenerate. A s in the case analysed by
Schwarz and Walsh, the graph of against F is found not to be linear
148
for small bias voltages although it becomes linear for large values of V. If
the barrier height is regarded as a variable, the intercept F , on the voltage
axis can be shown theoretically to depend on the diffusion voltage V^Q as
shown in Fig. 4.5. It is apparent that there is a maximum value of F ,
slightly larger than the band gap which is reached when F^Q is

Metal

Fig. 4.4 Shape of Schottky barrier on p-type material i „> (From Walpole and Nill,
1971. Copyright American Institute of Physics.)

200 400 600 800 1000

Built-in potential Vdo (mV)
Fig. 4.5 Dependence of V, on V^o for p-InAs at 4.2 K. (From Walpole and Nill, 1971.
Copyright American Institute of Physics.)
149
approximately equal to E^. A n explanation of this can be given in terms
of the argument of the previous paragraph. For a degenerate semi­
conductor, I is negative, so the bottom of the conduction band must lie
below the metal Fermi level if the density of minority carriers (electrons
in the p-type case) is to exceed the acceptor density. The value of the
capacitance is therefore as if the barrier height were approximately equal
to JEg — ^ ( = £ g + I ), corresponding to a diffusion voltage of E^—2^,
and an increase in true barier height beyond this value does not result in
any increase in Fj. If the voltage intercept is found experimentally to be
equal to this maximum value, one can only set a lower limit to the barrier
height. It is as if the inversion region in Fig. 4.4 were part of the metal.

4.2 The effect of an insulating layer with interface states

If a Schottky diode has an insulating interfacial layer, as occurs when

there is an oxide film on the surface of the semiconductor, its capacitance
is altered because the layer modifies the dependence of the charge in the
depletion region on the bias voltage. The capacitance of the interfacial
layer is effectively in series with the capacitance of the depletion region,
but because the latter is non-linear the overall capacitance is a rather
compUcated function of the parameters involved.

4.2.1 Very thin interfacial layers

The effect of an interfacial layer on the capacitance has been analysed in

some detail by Cowley (1966) and by Crowell and Roberts (1969) in
terms of the Bardeen model discussed in Section 1.4. The analysis is
easiest if the layer is sufficiently thin for the occupation of any interface
states which may exist at the insulator-semiconductor interface to be
determined by the Fermi level in the metal; this will be the case if the
layer is not more than about 20 A thick. The interface states are then
emptied or filled by the tunnelling of electrons from the metal and need
not be taken into account explicitly in calculating the capacitance, for the
same reason that the minority carriers make no direct contribution in
the situation discussed in Section 4.1.3. The capacitance therefore is still
given by dQJdV^. The interface states may, however, affect the
capacitance indirectly because they may modify the relationship between
the charge due to the uncompensated donors and the bias voltage. This
has been analysed by Cowley, who showed that the graph of against
V, remains linear with a slope of 2/qe^N^, as in the ideal case discussed
in Section 4.1.2 where there is no interfacial layer. However, the intercept
V\ on the axis is modified and is given by

150
 1/2/ kT
= Kio + a (4.5)
<? 2E,

where

a
e, + qdD,

In deriving this result it was assumed that a remains constant as

changes; i.e. that the density of states is constant for those interface
states whose occupation is changed by the appUcation of bias, and also
that the effect of the holes can be neglected. We can write eqn (4.5) in the
form
kT kT 1/2
Vi = ^^o-^ + </> "AO +h
4
where is the zero-bias barrier height, and = la^qN^/e^. If we were
to deduce the barrier height from the C / F data in the usual way by
adding I + kT/q to V[, we should obtain the value
kT 1/2
1/2 + ^ =
4
from eqn (1.24a). Since in all practical cases is less than the flat-band
barrier height t'Y about two orders of magnitude, the C / F method
gives essentially ^g. This result was confirmed by Crowell and Roberts. If
we take the case considered at the end of Section 1.4.5 with d = 20 A
and = 10^^ m~^, the value of a is about 60 A if does not exceed
about 10''' e V " ' m~^. In this case the difference between ^^o K
would be approximately 0.02 eV, which is about equal to the
experimental error of a careful measurement. If 6 is increased to 30 A
and to 10^^ m~^, the difference would be about 0.1 eV, which is quite
significant. The error would be reduced by a high density of surface
states, since these reduce a and tend to lock the height of the barrier
relative to the Fermi level in the metal.
It is possible to show that, for the particular case where the occupation
of the interface states is determined by E'p, Cowley's expression for the
capacitance can be written in the form

- = — + ^ (4.6)

where
1/2 / i^rp \ -1/2
, = ^ , and q = qD,.
o
Q and are respectively the capacitance of an ideal Schottky barrier
151
having a diffusion voUage and the capacitance of a parallel plate
capacitor of thickness 6 and permittivity e,. We shall refer to them as the
depletion layer capacitance and the oxide capacitance. The quantity
Cs = qD^ can be regarded as an additional capacitance due to the effect
of the interface states, and is in parallel with the oxide capacitance.
Physically, this arises because a change AP^ in the potential across the
interfacial layer causes a change qD^AVl in the charge residing in the
interface states if the density of states is constant over the energy range
qAVi and the occupation of the states is determined by the Fermi level in
the metal. The equivalent circuit is therefore as shown in Fig. 4.6(a),
where for simplicity we have omitted the parallel conductance of the
diode arising from the thermionic emission current.

(b)

Cd

Q RsKH

(c) (d)
Fig. 4.6 Equivalent circuit for Schottky barrier with insulating interfacial layer (parallel
conductance not shown): (a) thin interfacial layer, low frequency; (b) thin interfacial layer,
high frequency; (c) thick interfacial layer, low frequency; (d) thick interfacial layer, high
frequency.

If the interface states are not continuously distributed in energy but

form a discrete level, as some models suggest, they can make no
contribution to the capacitance unless their occupation can be changed
by a change in V^. Suppose the occupation probability of the interface
states is ^ . Then the change in interface charge resulting from a change
A l ^ is qN,(df,/dV;)AVi= qN,(df,/dE,){dEydV,)AVi, where N, is the
concentration of the discrete interface states per unit area and E^ their
energy. Assuming a Fermi-Dirac distribution for f^, df/dE^ =
- qf^(l -f^ykT, and if E^ is expressed in electron-volts dEJd I^ = - 1, so
that

qK^AV, = ^Ul-fOAV,
kT
152
and

As might be expected, Q = 0 i f / 5 = 0 or 1, and has its maximum value

when coincides with Ef.
Cowley's analysis which leads to eqns (4.5) and (4.6) assumes that the
population of the interface states can vary sufficiently rapidly for them to
be instantaneous equilibrium with the metal. This is equivalent to
assuming that their time constants are much shorter than the inverse of
the measuring frequency. Such an assumption is consistent with the
assumption that the interface states are in equiUbrium with the metal
since, if the interfacial layer is thin enough to enable good communica­
tion between the interface states and the metal, their time constants will
be very short and they should be able to follow a measuring frequency of
around 100 kHz. If the thickness of the interfacial layer exceeds about 10
A, or if the frequency of the a.c. voltage is increased above 100 kHz, the
time constant for establishing equilibrium between the interface states
and the metal may be so long that the population of the interface states is
not able to follow the measuring signal. The equivalent circuit now
becomes as in Fig. 4.6(b), where represents the turmelling resistance
and the product is equal to T„.
Cowley compared his theoretical analysis with some experimental
results on gallium phosphide Schottky diodes made by evaporating the
metal either in a conventional vacuum system fitted with an oil-diffusion
pump or in a system fitted with an ion pump capable of obtaining a
vacuum of 10~* Torr. In the latter case he found good agreement
between the values of the barrier heights deduced from the C/Fdata and
those obtained by the photoelectric method, whereas in the former the
value deducted from the C / F characteristics was consistently higher than
that obtained photoelectrically by as much as 0.5 eV. Cowley explained
the discrepancy by supposing that in the oil-pumped system oil vapour
condensed on the semiconductor before the metal was deposited, so that
an interfacial layer was formed, and that the better vacuum conditions
obtainable in the ion-pumped system resulted in negUgible contamina­
tion. But although the results could be described qualitatively in terms of
the above model, the magnitude of the discrepancy was much greater
than could be explained by assuming the interface states to be in
equilibrium with the metal.

4.2.2 Thick interfacial layers

We now consider the case where the insulating layer is so thick that not
only is there no tunnelling of electrons between the interface states and

153
the metal, but there is also so little tunnelling of holes from the metal to
the valence band that the hole quasi-Fermi level now corresponds with
the semiconductor Fermi level. This situation is similar to that which
exists in a M O S capacitor, and the population of the interface states is
determined by electron exchange with the semiconductor conduction
and valence bands, as in the familiar Shockley-Read-Hall recombination
mechanism. The steady-state occupation probabihty for the interface
states is now determined by the semiconductor Fermi level. It can be
shown that the interface states again give rise to a capacitance qD^ (for a
continuous distribution) or q^NJ^{l - fy/kT(ioT a discrete level), but in
this case the interface-state capacitance is in parallel with the depletion-
layer capacitance. The equivalent circuit is now as in Fig. 4.6(c),
provided the population of the interface states can follow the a.c. voltage
instantaneously. If the time constant arising from the Shockley-Read-
Hall process is such that the interface states cannot follow the a.c.
measuring signal, the equivalent circuit is as shown in Fig. 4.6(d). The
case of a thick insulating layer has been extensively discussed by
NicoUian and Goetzberger (1967) in the context of M O S capacitors. The
variation of with V is generally not linear, and it is not easy to
deduce the value of V^Q.

4.2.3 The general case

The intermediate case, in which the insulating layer is neither so thin that
the interface states communicate exclusively with the metal, nor so thick
that they only communicate with the semiconductor, gives rise to an
equivalent circuit which is a synthesis of Figs 4.6(b) and 4.6(d) and is
shown in Fig. 4.7. If either or RSRH is infinitely large, we recover
4.6(d) and 4.6(b), respectively. The series combination of R„ and R^^^
constitutes a d.c. path which is responsible for the d.c. current via
interface states referred to in Section 3.8.
The situation represented by Fig. 4.7 often arises in almost-ideal
Schottky diodes, but as yet we have no completely satisfactory analysis
which enables us to interpret the capacitance-voltage-frequency

Co, Ca
1
I
o 0

—^AA/— —^/^A—
^SRH

Fig. 4.7 Equivalent circuit for Schottky barrier with insulating interfacial layer (general
case).

154
 relationships in a tractable maimer. Freeman and Dahlke (1970)
discussed the theory of tunnelling through thin insulating layers down to
15 A in thickness. However, they considered only the case of a non-
inverted surface, with the interface states exchanging electrons
predominantly with the majority band; their analysis is therefore not
directly applicable to a Schottky barrier in which, for barrier heights
greater than one-half the band gap, the surface of the semiconductor is
inverted (i.e. for an n-type semiconductor, the hole concentration
exceeds the electron concentration). Moreover, they did not take into
account any changes in the band-bending resulting from the variation of
charge in the surface states, so that their treatment of the electrostatics is
incomplete. Card and Rhoderick (1971a) assumed that the interface
states can be divided into two categories, one of which communicates
exclusively with the metal and the other only with the semiconductor,
and that these two categories can be characterized by different densities
of states. This model has been extensively discussed by Fonash (1983).
But although the model has the merit of simplicity and is relatively easy
to analyse, it lacks reaUsm since it does not allow any of the states to
communicate with both the metal and the semiconductor.
It is possible to make some general observations on the circumstances
under which the capacitance arising from the interface states might be
experimentally observable. It can be seen from Fig. 4.6(b) that if the
interface states exchange electrons predominantly with the metal, the
capacitance Q is effectively in parallel with the oxide capacitance Q^,
whereas if their population is determined by the semiconductor Fermi
level, as in Fig. 4.6(d), Q is effectively in parallel with the depletion
capacitance Q . Now in a typical case, with ô « 10 A and a depletion
width of approximately 1000 A, Q „ « 100 Q , and it is clear that the
effect of is only likely to be observable if the second situation prevails.
It should be noted that if the tunnelling probability between the interface
states and the metal is very low, it does not immediately follow that the
interface state population is governed by E^, since if they exchange
electrons predominantly with the valence band (as may happen if the
semiconductor surface is inverted due to a high barrier) and the hole
quasi-Fermi level coincides with Ef, the interface state population will
also be determined by Ef. For Fig. 4.6(d) to hold, either the dominant
electron exchange must be with the conduction band (i.e. a low barrier or
large forward bias) or ô must be so large that the hole quasi-Fermi level
no longer coincides with Ef.

4.2.4 Interface state spectroscopy

Analyses of the effect of interface states on the capacitance of nearly-

ideal Schottky diodes have been pubUshed by Barret and Vapaille (1979)

155
and by Muret and Deneuville (1982). They show that if the interface
states exchange electrons predominantly with the conduction band, so
that their occupation is determined by (or £|r), there is an additional
capacitance which has a maximum value when coincides with the
energy of the interface states (assuming them to have a discrete energy
distribution). This situation only arises under moderately large forward
bias and for interface states lying above Ef, assuming an n-type
semiconductor. This is because the condition for the interface states to
exchange electrons predominantly with the conduction band is that
c„n> CpP (see Section 4.5.1), and if the barrier height exceeds one-half
the band-gap this will only be the case under foward bias. Since is
effectively in parallel with C^, the depletion capacitance under forward
bias has to be subtracted by deducing from an extrapolation of the
linear plot of C~^^ against F which is obtained under reverse bias. If the
states lie below Ef they will remain occupied for all values of forward
bias and will make no contribution to the capacitance. (Interface states
lying below Ef can, however, be observed in p-type semiconductors.) If
the interface state occupation is not determined by the capacitance
maximum does not occur when E^ coincides with the interface state
energy, and the trap parameters can only be deduced in a complicated
and indirect way from analyses of the variation of with bias,
frequency, and temperature. The situation is even more complex if, as is
usually the case, the interface states form a continuous band, with
capture cross-sections for electrons and holes which vary throughout the
band. These analyses have been used to interpret experimental data on
silver, gold, and nickel contacts to silicon by Muret (1982), and on C r - S i
and A l - G a A s interfaces by Barret, Chekir, and Vapaille (1983).
Both the analyses of Barret and Vapaille and of Muret and Deneuville
seem to be deficient in one respect, namely that they do not take into
account the potential drop across the interfacial layer which separates
the interface states from the metal. Such a separation must occur if the
interface states are to have an observable effect on the barrier height and
on the capacitance. The potential drop across the interfacial layer varies
with bias because of the change in charge in the interface states, and has
the effect of altering the energies of the interface states and of the band
edges relative to the Fermi level in the metal. (It is precisely this effect
which is responsible for the piiming action of interface states.) This has
three consequences: first, it modifies the charge in the interface states
and hence their contribution to the capacitance; second, it alters the
diffusion potential and therefore the depletion capacitance C^; and third,
the change in diffusion potential modulates the thermionic emission
current through the diode at the a.c. frequency, and since this modulation
is not in phase with the a.c. component of the voltage, it gives rise to an
additional capacitive component. The last effect has been discussed by

156
Werner, Ploog, and Queisser (1986). Although it is not difficult to obtain
a self-consistent expression for the response of the interface charge to a
change in apphed bias when the interfacial layer is taken into account, it
is much more difficult to find a tractable expression for the capacitance
which can provide a straightforward interpretation of the experimental
data.
Another omission which is common to all existing treatments of
interfacial phenomena is that they do not consider the effect of the image
potential on the energies of the interface states. If the concentration of
charged interface states is large (say about 10^* m"^, so that their average
separation from each other is about 10~* m or 10 A), then the lateral
separation between the interface states is about the same as their
separation from the metal. Their images in the metal approximate to a
continuous sheet of charge, and the potential produced by this image
charge is precisely the contribution of the interface charge to the
potential drop across the interfacial layer which we have referred to in
the previous paragraph. However, if the concentration of charged
interface states is small (say about 10'^ m~^ or less, either because is
low or because the probability of the states being charged is small) then
the average separation between the charged interface states is much
greater than their separation from the metal, and the electric field
produced by such a charge distribution is Uke that of discrete charges.
Each charged state then experiences an image potential as discussed in
Section 1.6, which may amount to about 0.03 eV or more depending on
the value of £j, for d~ 10 A. Since this decrease in potential exceeds kT/q
and depends on whether or not the state is charged, the image potential
is bound to modify the occupation probability of the interface states.
It appears, then, that we do not yet have a completely satisfactory
theory that enables us to interpret the contribution made by interface
states to the capacitance of forward-biased Schottky diodes in a simple
and rehable manner, and estimates of interface-state parameters which
have been deduced from such measurements should be treated with
caution. Furthermore, existing analyses do not discuss the dependence of
the capacitance on the separation between the interface states and the
metal, so they do not differentiate between true interface states and bulk
states which happen to lie close to the interface. One could, of course,
take the view that there is no essential difference between interface states
and bulk states which happen to lie near the metal; such a viewpoint
would be particularly appropriate to the unified defect model proposed
by Spicer and his co-workers. In this model, the barrier height is assumed
to be determined by defects which are created close to the surface of the
semiconductor when the metal is deposited, and the 'interfacial layer'
consists of the thin layer of semiconductor which separates the defect
states from the metal.

157
4.3 Non-uniform donor distribution

We saw in Section 4.1 that, for an ideal Schottky diode with a uniform
concentration of donors, the plot of against V is linear with a slope
equal to and in Section 4.2 it was seen that this remains true in
the presence of a thin interfacial layer. This result is frequently made use
of as a means of measuring the donor density. If the semiconductor is
compensated, then must be replaced by - N^. We shall now show
that, if the donor density is non-uniform, the plot of against F i s not
linear but its slope at any particular bias voltage is inversely proportional
to the donor density at the edge of the depletion region.
Suppose that the donor distribution is non-uniform and that varies
with the distance x into the semiconductor as shown in Fig. 4.8(a). Let us
assume that we can use the depletion approximation, according to which
the electron density is negligible compared with N^ix) throughout the
depletion region and rises abruptly to a value equal to Ni{w) at the edge
of the depletion region of thickness w at a reverse bias V,. The charge
density will then be equal to qN^(x) between x=0 and x= w, and zero
beyond x= w. The electric field strength S' will be zero beyond x= w,
and from Gauss's theorem the field strength at the position x' will be

(a)

X •

Fig. 4.8 (a) Non-uniform donor distribution, (b) Resulting electric field distribution in
depletion region.

158
equal to the total charge between the planes x = x' and x = w multiplied
by £7^; i.e. it will be proportional to the area under the A'd(.*) curve
between x! and w. The variation of S' with x will be as shown by the solid
line in Fig. 4.8(b), and the total drop in potential across the depletion
region will be equal to the area under this curve. Suppose the reverse bias
is increased to + A V„ resulting in an increase in the width of the
depletion region from w to w + A w. A n additional amount of charge
qNi{w)IS.-w will be 'uncovered' at the edge of the depletion region, and
the field strength at any point within the depletion region will be
increased by an amount qN^(w)S.wk^ as shown by the broken line in Fig.
4.8(b). The drop in potential across the depletion region, which is now
equal to the area under the broken Hne, has clearly been increased by an
amount w\qNi<^\v)IS.wl£^ and this must equal the increase in bias voltage
AF„ so that Aw/AF, = EjyvqN^{w). The increase in the charge due to the
uncompensated donors (A<2d) is q^J(W)S.w, and the capacitance can be
found by allowing A F , to become infinitesimally small so that

AF, w'
In other words, as with a uniform distribution of donors, the capacitance
is equal to that of a parallel-plate capacitor of thickness w and
permittivity Cj. If is plotted as a function of F„ the slope of the curve
will be
aw

This is the same result as for a uniform distribution of donors, except

that N^(w) is the donor density at the edge of the depletion region.
Hence, if w is varied by changing the reverse bias, the slope of the graph
of C"^ against F^ will give N^at x= w. The value of w corresponding to a
particular value of F^ can be obtained from the capacitance through the
relationship C = e/w, assuming the depletion approximation to be vaUd.

4.4 C/V methods of measuring dopant distributions

The analysis of the previous section forms the basis of a technique which
is extensively used for determining the doping profile of semiconductors,
and a number of instruments for the automatic determination of impurity
profiles from C / F measurements have been developed (see, for example,
Baxandall, CoUiver, and Fray, 1971). The principle of the latter is to use
a signal of frequency about 100 k H z to measure C, while the reverse
bias is modulated at a lower frequency around 1 k H z so that dC/dV, can
be obtained directly. From C and dC/dV, it is possible to determine

159
dC~'^/dV, and w automatically. A more complicated method due to
Copeland (1969) is discussed in Section 4.5.2. In practice, the Schottky
barrier is formed either by evaporating a metal such as gold onto a
suitably etched surface or by using a mercury contact contained in a
capillary tube (see, for example, Hammer 1970). In semiconductors like
indium phosphide which form low barriers, the high parallel
conductance may cause measurement difficulties. This may be overcome
by deliberately creating a thin interfacial layer to reduce the reverse
current; for example, by oxidizing the surface of the semiconductor.
Provided the layer is thin enough for any surface states which may be
present to be in equilibrium with the metal, dC''^/dV^ will not be affected
(see Section 4.2.1). There are many possible sources of error which must
be taken into account. Goodman (1963) has discussed general causes of
error in C/V measurements, and Smith and Rhoderick (1969) have
considered the effect of methods of surface preparation. A critical
analysis of the errors inherent in C / F profihng has been given by Amron
(1967), and a comprehensive review by Blood and Orton (1978).
Generally speaking, if the impurity profiles obtained from C / F data are to
be reliable, the I/V characteristics of the Schottky diode should be not
very far from ideal, with n values less than about 1.1.
A more fundamental difficulty in applying the capacitance technique to
the measurement of impurity profiles arises if the gradient of the
impurity density is very large and the mean carrier density is low; one
cannot then be sure of the existence of charge neutrality even in the bulk
of the semiconductor. We saw in Section 4.1 that the capacitance is given
by dQ^/dV^, where is the charge due to the uncompensated donors. If
Fj changes, changes owing to the movement of tiie mobile carriers
(electrons if the semiconductor is n-type) in and out of the depletion
region, and not because the donors move. One might expect, therefore,
that the value of dC~'^/dV, given by eqn (4.7) should contain n{w) in the
denominator rather than NJw), where n{\v) is the electron density at the
edge of the depletion region. If the donor density is uniform, the bulk of
the semiconductor is electrically neutral so that n= (assuming
complete ionization of the donors) and eqn (4.7) is correct. A s an
example of a case in which eqn (4.7) would not be correct, suppose that
the donor density suddenly changes from a value N^i for x< XQ to a
value N^2 for X > XQ, where N^^ > N^2- The electron density does not fall
abruptly from N^^ to N^2 at x= XQ but decreases smoothly, being less
than iVji for some distance inside the highly doped part and greater than
N^2 for some distance inside the weakly doped part as shown in Fig. 4.9.
The region on each side of XQ is therefore not electrically neutral, but has
a net positive charge density for x< Xg and a net negative charge density
for X > XQ. This non-neutral region constitutes a dipole layer which
causes band bending as in a p - n junction. The distance over which the

160
electron density differs significantly from the donor density on each side
of the junction is approximately equal to the appropriate Debye length
(= E^kT/q^N^y^^. The space-charge region therefore extends further
into the more weakly doped side. If N^2" 10^^ m " ^ L152 0.1 jum, and an
attempt to obtain an impurity profile for such a distribution from C/V
data would show the transition spread over approximately 0.1 fim
instead of being abrupt. Impurity distributions in which the density
changes appreciably in distances of less than 0.1 jum cannot therefore be
investigated by this technique if the minimum donor density is less than
about 10^' m~^. This limitation was first pointed out by Kennedy, Murley,
and Kleinfelder (1968).

Fig. 4.9 Electron concentration (n) resulting from an abrupt change in donor density

At the present time, the only way of dealing with this problem is to
make some reasonable assumption about the Ukely form of the donor
distribution and then to use a computer-modelling technique to calculate
what the versus Fgraph should look like. If this does not coincide
with the experimentally observed curve, one revises the originally
assumed form of the distribution and repeats the process until
reasonable agreement is reached. This was first done by Johnson and
Panousis (1971) for the case of high-low homojunctions, and extended
to heterojunctions by Kroemer et al. (1980). Both these groups of
authors found that the apparent donor density obtained from eqn (4.7) is
approximately, but not exactly, equal to the free electron density that
would exist in the system if there were no Schottky barrier, and that the
shape of the apparent donor distribution depends on which side of the
system contains the Schottky barrier (i.e. in the case of the high-low
junction, whether it is in the highly doped or the weakly doped side).
Procedures for computer modelling of heterojunctions have recently
been pubUshed by Whiteaway (1983) and by Missous and Rhoderick
(1985). The problem is particulary acute with heterojunctions because of
the abrupt discontinuities in the valence and conduction bands. The
general problem of interpreting C/V characteristics when there are
161
abruptly varying dopant distributions has been discussed by Kroemer
and Chien (1981) and by Rhoderick (1984).
A n alternative approach to the determination of dopant distributions
using C/V measurements is the electrochemical technique invented by
Ambridge and Faktor (1974, 1975). In this technique, the metal contact
is replaced by an electrolyte such as 'Tiron' or N a O H . The method relies
on the fact that an electrolyte is a conductor and behaves in some
respects hke a metal, with the 'redox' potential replacing the Fermi level.
A potential barrier whose height depends on the redox potential is
produced within the semiconductor, accompanied by a depletion region
which exhibits capacitance in the same way as an ordinary Schottky
barrier. A reverse bias can be applied between the electrolyte and the
metal, and the differential capacitance measured using a suitable
electrode. The derivative of the versus V relationship yields the
donor density at the edge of the depletion region, as in eqn (4.7). The
edge of the depletion region is made to move into the semiconductor, not
by increasing the reverse bias as in conventional C / F profiling but by
etching the semiconductor electrolytically. The amount of semiconductor
removed (and hence the position of the edge of the depletion region) can
be calculated by integrating the etching current to obtain the total charge
that has passed and using Faraday's law of electrolysis, assuming the
charge carried by each semiconductor ion is known. The advantage of
this technique is that it is unnecessary to evaporate a metal contact onto
the semiconductor, so that it is quicker and more convenient than the
conventional method. The electrochemical technique has been reviewed
by Blood (1986) and commercial instruments are now available.

4.5 The effect of deep traps

By a 'deep trap' we mean a localized electron state in the bulk of the

semiconductor which is so far removed from the conduction or valence
bands that it is not ionized at room temperature. In this respect it differs
from shallow donors and acceptors which are usually completely ionized.
If the energy level of such a trap varies with respect to the Fermi level as,
for example, in the depletion region of a Schottky barrier where the
bands are bent, the occupation of the level (and therefore its state of
charge) will also vary. Because the degree of band-bending depends on
the application of a bias voltage, the state of charge of the trap will
depend on the bias and as a result the capacitance will be affected. The
effect of traps on the capacitance of a Schottky barrier affords a very
convenient method of detecting and characterizing an extremely low
concentration of defects.

162
4.5.1 The population of deep traps under reverse bias

Consider a Schottky barrier in an n-type semiconductor which contains a

single type of deep trap of energy E^ as shown in Fig. 4.10. The traps are
assumed to be distributed uniformly throughout the semiconductor. In
the absence of bias, the traps will be occupied by electrons if they lie
below E-p (i.e. to the right oi x = am Fig. 4.10) and will be empty above
E-p, assuming the zero-temperature approximation to the Fermi-Dirac
function. If a reverse bias is appUed (Fig. 4.11), the Fermi level is spht
into quasi-Fermi levels for electrons and holes. A s was seen in Chapter 3,
it is a very good approximation to assume that under forward bias the
electron quasi-Fermi level is horizontal through the depletion region
and coincides with the Fermi level in the semiconductor, while the hole
quasi-Fermi level is also horizontal and coincides with the Fermi level
in the metal. This is also a fairly good approximation under reverse bias
except that, as shown by Crowell and Beguwala (1971), the electron
quasi-Fermi level rises appreciably as it approaches the metal. We shall
ignore this complication, since in the region where the electron quasi-
Fermi level rises, the electron concentration is so small that it is
unimportant, and assume both quasi-Fermi levels to remain flat. The
question now to be answered is what determines the occupation of the
traps.

Fig. 4.10 Charge state of deep traps (zero bias). The traps are assumed to be donor-hlce,
i.e. neutral when occupied, positively charged when empty.

In the general case, the occupation of the traps is determined by

Shockley-Read-HaU statistics (see, for example, Blakemore, 1962)
according to which the probability of a trap being occupied by an
electron is given by
o^nv + ep
/ = (4.8)
CT„nv + e„ + o-ppv + gp

163
Here andCTpare the capture cross-sections of the trap for electrons
and holes, respectively, and v is the mean thermal velocity (assumed to
be the same for electrons and holes). For convenience we shall write
CT„v= c„ and OpV = Cp, so that cjc^ = o^/a^. The probabiUties per unit
time of the trap emitting an electron to the conduction band or a hole to
the valence band are and e^, respectively, and n and p are the
concentrations of electrons and holes. Detailed balancing arguments
show that and are related to c„ and Cp by
e„= c^n, e x p { g ( £ , - E>^lkT\ = c „ n e x p { 9 ( £ , - e„)/fcr} (4.9a)
and
ep = Cp/Ji exp{9(Ei - E^/kT] = Cpp exp{?(ep - E:)/kT\ (4.9b)
where is the intrinsic carrier concentration and E-^ the intrinsic Fermi
level.
It is convenient to introduce an energy Ey defined by

kT
= £ i + — I n
Iq
is a parameter associated with a particular type of trap, and has the
significance that if the trap energy lies above E^ electron emission is
more likely than hole emission, while if the trap energy hes below E^ the
converse applies. From eqns (4.9a) and (4.9b) and the definition of the
quasi-Fermi levels, it can easily be shown that the following inequalities
hold:

if£,>£, (4.10a)
(4.10b)

>s i f £ , > 2 £ , - ? „ = £,p (4.10c)

(4.10d)
i f £ , < 2 £ i - e p = £<,„ (4.10e)
if^„ + ^p>2E,. (4.10f)
Because of the exponential energy dependences of the emission rates and
carrier densities, an inequaUty of more than 2kT/q in one of the energy
relationships results in a difference of an order of magnitude or more
between the corresponding emission or capture rates. The energies E^^
and defined by (4.10c) and (4.10e) are known in photoconductivity
theory as the demarcation levels for holes and electrons, respectively. It
follows from their definitions that E^^ - = ^dp ~ Cp; also, since
^p> 2 £ i if c^n>CpP, then Ean and E^^ lie below and gp respec­
tively if c„n > CpP, and above and ^p respectively if CpP > c„n.

164
 Consider now the band diagram under reverse bias, shown in Fig. 4.11.
The line representing intersects ^^at x = a and t„ at jc = c. The plane
x = b is where £ i is midway between C„ and ^p, so that, at x = b,
+ 5p = 2£i. Hence for x<b, > c^n, while for x> b, c„n > c^p. It is
important to reaUze that Ei is not a constant of the band structure, but
depends on the capture cross-sections of the particular type of trap
under consideration; it will vary from trap to trap, unless all the traps
happen to have the same value of the ratio o^/o^. This ratio can vary
enormously, and can have values from more than 10^ to less than 10~^,
depending on whether the traps are donor-like or acceptor-like. But even
if the ratio is as large as 10^ or as small as 10""^, will not differ from
by more than a few times kT/q, so that is always close to the centre of
the energy gap. We shall assume a barrier height exceeding E^/2, so that
£ , Ues above Ef at the interface. Figure 4.11 also shows the behaviour of
the demarcation levels. Note that if + ^p = 2E^, E^^ = ^p, and if
t„ = £ i , = E^,so that £<jp passes through the intersections of with
x = b and of with x = c. Similarly E^^ passes through the intersections
of with x = a and of with x=^ b.

Now consider the situation for 0 < jc < a. Throughout this region
CpP> c„n. It follows from the relationships (4.10) that for E^<E^<E^
the only important processes are hole capture and hole emission, so that
the trap only interacts with the valence band. Hence

c,p + e, 1 +exp[q{E,-i,)/kT}

165
from (4.9b), i.e. / has the form of a Fermi-Dirac distribution with a
Fermi level equal to ^p. In the zero-temperature approximation, the trap
will be occupied if it lies below ^p and empty if it lies above it. If the trap
energy lies above E^, then e„ > e^, and f~ e^/^CpP + e„), but the effect of
this is simply to make / even smaller. The increase of e„ above E^^, which
follows from (4.10e), is likewise of no consequence, so that we may take
the distribution as a Fermi-Dirac distribution controlled by over the
whole of the energy gap.
Between x = a and x=b,vie still have c^p > c„n. The demarcation level
Ea„ now Ues below ^p, and there is a range of trap energies between E^„
and ^p where, as a result of the relationships (4.10), both of the capture
processes are negligible compared with either of the emission processes,
so that

^ ~ e„ + gp ~ 1 + exp{2(£, -E,)/kT} ^"^-^^^

from (4.9). This function resembles a Fermi-Dirac distribution and

changes abruptly from nearly 1 to almost zero as E^ increases through E^.
The increase in c^p relative to gp above ^p, in accordance with (4.10d),
merely brings / closer to zero, and the reduction in e„ below E^^
according to (4.10e) simply makes / approach 1 more closely, so / is
effectively given by (4.11) throughout the entire range of
Between x = b and x = c, c„n > c^p, but there is still a range of trap
energies between and E^^ for which both of the capture processes are
negligible compared with either of the emission processes as a result of
the relationships (4.10), so that / is given by (4.11). The increase in c„n
relative to gp above E^^ is unimportant since they are both negUgible
compared with e„, so that/remains almost zero. Likewise, the increase in
Cn« relative to for trap energies below ^„ simply brings/closer to 1, so
that/is effectively given by (4.11) throughout the forbidden gap.
To the right oi x= c, c„n> c^p, and E^p falls below Above E^p we
have c„« > gp, so that

/« = ^- (4.12)

from (4.9a), i.e. / has the form of a Fermi-Dirac distribution with a Fermi
level equal to C„- Below E^p, Cp increases relative to c„n, but this simply
brings / closer to 1, so that / is essentially given by (4.12) throughout the
forbidden gap.
We can summarize the preceding arguments by saying that between
X = 0 and x = a the trap occupation probabiUty has the form of a Fermi-
Dirac distribution with a Fermi energy equal to the hole quasi-Fermi
level Cp, while to the right of jc = c it has the form of a Fermi-Dirac

166
distribution with a Fermi energy equal to the electron quasi-Fermi level
Between X = fl and x = c the occupation probability is equal to e^/
(e„ + Cp), so that traps lying above are almost completely empty, while
traps lying below El are almost completely full. The assumption is often
made that the occupation of the traps is determined by Ç„ throughout the
depletion region, but it is clear from the above discussion that this is only
the case to the right of the plane x = c, where c„« > c^p and E^^ lies
below Ç„. Although this holds for all values of E„ the trap occupancy will
only change in response to a change in Ç^, resulting from a change in bias
voltage, if the Une representing the trap energy £ , (not shown in Fig.
4.11) intersects Ç„ to the right of x = c. This can only happen if lies
above Ey, and since E^- E^, the trap must lie in the upper half of the
band-gap. In the same way, all traps in the region 0 < x < a are controlled
by the hole quasi-Fermi level Çp, but their occupation will only change in
response to a change in band-bending (brought about by a change in bias
voltage) if the line representing E^ intersects Çp to the left of x = a. This
can only happen if E^<Ej^, i.e. if the trap lies in the lower half of the gap.

4.5.2 The contribution of traps to the capacitance under reverse bias

Before discussing the effect of deep traps on the capacitance of a

Schottky barrier, we must consider carefully how the capacitance is
measured in practice. Nearly all capacitance measurements involve the
use of an a.c. bridge operating with a small-amplitude test signal of
frequency To measure the capacitance as a function of reverse-bias
voltage, the bias is changed at a rate which is slow compared with the test
frequency 0)^. In some applications the bias consists of a d.c. component
modulated by a small a.c. component which varies sinusoidally at a low
frequency (t)^. Because the traps do not fill or empty instantaneously
when the bias changes, their contribution to the capacitance will depend
on how the two frequencies o)^ and compare with the inverse of the
time constant associated with the filling of the traps. The various cases
have been reviewed by Kimerling (1974). We shall consider the
particular situation in which the traps lie in the upper half of the band-
gap and are controlled by the electron quasi-Fermi level, which is
assumed to be flat throughout the depletion region. It can easily be
shown that in this case the trap time constant is given by T, = (e„ + c„n)"^
Since the only traps whose occupation changes with bias are those close
to the quasi-Fermi level, and for these e„ « c„n, it follows that the
mean time constant of the traps in question is equal to (2e„y\

CASE 1. 2e„> (i)^> (o^,

In this case the traps are able to follow instantaneously both the test
signal and the bias variation. Suppose that the traps are donor-like (i.e.

167
neutral when occupied, positively charged when empty), so that they are
positively charged to the left of the plane x = y [Fig. 4.12(a)] and neutral
for x> y. The donors are uncompensated within the depletion region of
width w, so the variation of charge density with x is as shown by the solid
line in Fig. 4.12(b). If the reverse bias is increased from to + AFp
the charge distribution will become as shown by the broken line in Fig.
4.12(b), and the positive charge will increase by an amount ^A/jAw due
to the uncompensated donorsf and by an amount qN^Ay due to the
change in the charge state of the traps. The contribution qN^Ay is
included because the traps exchange electrons with the conduction band
and hence with the bulk of the semiconductor; there is no component /^j
of the current density to the left of the plane x = y (see Section 4.1.1), so
the capacitance is given by C= q{N^/S.w-^ A'tA_y)/AFf. The increment
AFr includes both the test signal of frequency (o^ and the bias variation of
frequencyft>b-The electric-field intensity will be as shown in Fig. 4.12(c),
and the increase in the potential drop across the depletion region (AF^)
must be equal to q{NiWA.w+ NjAyys^. The differential capacitance is
therefore
^ ^ A Q ^ 8,{N,Aw + NAy)
AF, N^wAw + N.yAy' ^ ' ^
Although Fig. 4.12(b) shows uniform donor and trap densities, it can
easily be shown, using an argument similar to that in Section 4.3, that eqn
(4.13) is still correct for the case of non-uniform donor and trap distribu­
tions provided is taken to be the donor density at x = w and A^, the
trap density at x = y. If the donor density is uniform, w - j is given
according to the depletion approximation by

1/2
w- y = = A (4.14)
qN,
so that A w = Ay. In this case

C = ^ ^ i ^ ^ , (4.15)

and the capacitance is no longer given by W ; it is as if the effective

depletion width were an appropriately weighted average of w and y. It is
easily shown that in this case
dC~^ _ 2

tit is assumed throughout this chapter that there are no shallow acceptors present. If
there is a concentration of shallow acceptors, Af, must everwhere be replaced by

168
169
It is important to realize that the term Ni in the denominator is the trap
density at x = y and not at x= w. Since the effective depletion width
given by eqn (4.15) Ues between y and w, the apparent trap distribution
will be incorrect if it is deduced from eqn (4.16) with the assumption
that y= EJC. If the traps are acceptor-like (neutral when empty), the
expression for the capacitance is more complicated because the
denominator of eqn (4.14) now contains A^<, - A^„ and A is not constant
unless both A/(j and A/, are constant, so that in general A w is not equal to
^y

CASE 2. (o^> 2e„> (Oy,

In this case the traps cannot follow the test signal, and the capacitance is
determined entirely by the movement of electrons at the edge of the
depletion region so that

C = - . (4.17)
w
However, the mean occupation of the traps (averaged over a time that is
long compared with co~^) is able to follow the variation of the bias
voltage at the frequency (Wb and affects the relationship between w and
Vj. For donor-like traps, the change in w and y due to a change in the bias
voltage A is still given by

A F, = - (A/d vvA w + N^yAy).

(AF, here refers only to the slow change in bias voltage at frequency (Wb-
It does not include the component at frequency co^ due to the test signal.)
Consequently

dV, el dV, qeM + N,{y/w)} ^ • ^

if one assumes a uniform donor density so that Aw = Ay. Hence, for large
bias voltages when y/w tends towards unity, the apparent donor density
tends towards + N^. If A^^ is constant, eqn (4.18) can be used to infer
A',, but it must be remembered that A/^, is the trap density atx = y and not
at the position x = w given by eqn (4.17); since w - >' ( = A) is constant
for a uniform donor distribution the apparent trap distribution inferred
from eqns (4.17) and (4.18) will be shifted by a distance X. Equation
(4.18) can be rewritten as

(4.18a)
ac"' 2
where A = w- y, and if and are constant, a plot of dV,/dC ^
170
against C gives a straight Hne with an intercept at C = 0 that gives the
value of A/j + Nf. This was first pointed out by Senechal and Basinski
(1968a). The case of acceptor-like traps is again very comphcated
because 2 is not constant, so that Ay is no longer equal to A w.
If the test frequency o)^ is increased from a low value so that there is a
gradual change from Case 1 to Case 2, there will be a smooth change
from the low-frequency capacitance Qf [given by eqn (4.15) for the case
of a uniform donor density] to the high-frequency value given by
(4.17). Hence
^ _ ^ _ £sNt{w-y)
w{NiW + N,y)
It can be shown (Zohta, 1973) that at an intermediate frequency the
capacitance is given by

1+ CO'T*'

where

r* = (2e„)-

The equivalent circuit of the system is as shown in Fig. 4.13(a), where the
effect of the traps is represented by a lossy capacitor of time-constant
Cti?,= r*, and Q = £s/w. Cj is the capacitance of an ideal Schottky
barrier having a depletion width w and no traps. Clearly, C,, = Q + C,
and Chf= Q , so that Ct= Q - Qf. Figure 4.13(a) is in turn equivalent to
the simple parallel circuit shown in Fig. 4.13(b), which is the form of
circuit assumed by most measuring instruments. Comparing Figs. 4.13(a)
and 4.13(b) one can easily show that

1+0) 2_*2
S
and

R- l+ftj^T*'

I—WW-"
c, «1 r'=vg'

(a) (b)
Fig. 4.13 Equivalent circuit for Schottky barrier containing deep traps, (a) Actual
equivalent circuit, (b) Form of circuit assumed by most LCR bridges.

171
These equations have also been obtained by Pautrat et al. (1980). The
charging and discharging of the traps therefore not only give rise to a
frequency-dependence of the capacitance, but also cause a parallel
conductance to appear which is simply related to the frequency-
dependent part of the capacitance.
The study of the capacitance and conductance of Schottky barriers
containing deep levels has led to a new technique known as admittance
spectroscopy (Losee, 1975; Vmcent, Bois, and Pinard, 1975; Pautrat et
al., 1980). The method involves the measurement of the capacitance and
conductance as a function of frequency at a fixed temperature, or as a
function of temperature at a fixed frequency. The quantity G'/m^ is
proportional to (o^T*/(l + CDIT*^), and if either o)^ or r* is varied, G7a)^
goes through a maximum when CD^T* = 1. Since r* is proportional to e;;"'
and e„ = c^n^ e x p { g ( £ t - E-^/kT] from (4.9a), varying the temperature
provides an effective means of changing T * . In this way pairs of values of
(Wj and T c a n be obtained, where (o-r is the frequency corresponding to
the peak value of G'/o)^ at temperature T.
The relationship between r* and e„ can be written as

where a = C,f/C|,f and P = N^/(Nf+ N^), so that the condition <WJT*= 1

is equivalent to

= ^ = ^ exp{9(£, - E.ykT} [from (4.9a)]

= ^exp{-g(£,-£.)/^r}.

The capture coefficient c„ ( = o„v) is proportional to T^^^ if o„ is

assumed independent of temperature, and the effective density of
states is proportional to T^^^, so that (Oj is proportional to
r^exp{ - q(E^ - E^)/kT]. Hence the slope of a plot of ]n((Oj/T^) against
1/T gives E^- E„ and the intercept on the vertical axis gives if the
effective mass (which determines v and N^) is known. The quantity a is
obtained from the experimental data, while /? can be found using Cope-
land's method as described in the next section (Case 3).
The foregoing analysis assumes that the trap occupation is determined
by the electron quasi-Fermi level tn> and, as we saw in Section 4.5.1, such
traps only respond to a change in bias voltage if they he in the upper half
of the band-gap. In p-type material, a similar analysis shows that the
traps which can be studied by admittance spectroscopy are those whose
occupation is controlled by the hole quasi-Fermi level tp, and which lie
in the lower half of the band-gap. Measurements on both types of

172
material are therefore necessary if the whole range of trap energies is to
be investigated.
C A S E 3. a)^> a)i,> 2e„

In this case the traps cannot follow either the test signal or the
modulation of the bias voltage. The capacitance is determined entirely by
the movement of electrons at the edge of the depletion region and is
given by eqn (4.17); the traps do not affect the variation of depletion
width with bias voltage so that

(4.19a)

if the traps are donor-like, and

(4.19b)

if the traps are acceptor-like. In eqn (4.19b), unUke Cases 1 and 2, TV, is
the trap concentration atx= w.
Case 3 includes as a special case the method of determining donor
distributions invented by Copeland (1969). In Copeland's technique a
current containing an a.c. component at frequency is driven through
the diode, and the a.c. voltage components across the diode at
frequencies u>s and 2(o^ are measured using tuned amplifiers. The
component at frequency co^ is proportional to C " ' and gives the
depletion width w, while the component at frequency 2a)^ can be shown
to be proportional to ATj' where is the donor density at д; = ж This
result is true whether traps are present or not. It was pointed out by
Zohta (1970) that if Copeland's method is used to find Л',, the result can
be combined with eqns (4.16) or (4.18a) to obtain Л^, for Cases 1 and 2,
respectively. If and Л/, are both constant, the slope of a plot of
dV/dC'^ against С gives Я from eqn (4.18a), and the trap energy E^ can
then be obtained from eqn (4.14).
It is clear from the above considerations that if traps are present the
capacitance under reverse bias will be a complicated function of
frequency.! Conversely, if the capacitance is found experimentally to be
independent of the test frequency, this can be regarded as evidence for
the absence of traps. If traps are present, great care must be taken in
using the C/V technique to measure impurity profiles, and the only
reliable method is to ensure that the test and bias modulation frequencies
satisfy the conditions of Case 3. A comprehensive discussion of the
problems of impurity profiling in the presence of traps has been given by
Kimerling(1974).
tUnless the conditions are such that the traps are always full. This will be the case if
E, < £ , and also E , < at the interface (assuming an n-type semiconductor).

173
4.5.3 Transient measurements

There is an important class of measurements in which the reverse bias on

a Schottky barrier is changed instantaneously from one value to another,
and the resulting change in capacitance is observed as a transient. The
measurement can be made by using a capacitance meter which gives a
voltage output proportional to the capacitance under test; the capaci­
tance can then be observed as a function of time by displaying the meter
output on an oscilloscope or chart recorder. The method was first used
by WiUiams (1966) and subsequently refined by Senechal and Basinski
(1968a, 1968b) and by Sah and Neugroschel (1976).
Consider a Schottky barrier containing a uniform concentration of
traps A^, under a reverse bias voltage V^. If the traps are donor-like and
he in energy above E j , so that their population is governed by the
electron quasi-Fermi level, the band diagram will be as shown in Fig.
4.14(a). Those traps which are at more than a distance from the metal

"'2(0)
^ \ W2(<») ^

A
\ ^
\
. //=0

(b)
Fig. 4.14 Shape of Schottlcy barrier containing deep traps: (a) with reverse bias V,
(steady state) and (b) after increase of reverse bias to Kj, showing shape of barrier at / = 0
and f = 00.

174
will be neutral, and those that are within of the metal will be positively
charged. If the differential capacitance Cy (per unit area) is measured
with a test frequency to which the traps cannot respond, C , will be given
by C ] = £ S / H ' I . Suppose now that at time / = 0 the reverse bias is
instantaneously increased to V2. Because the traps cannot change their
charge state instantaneously, they wUl remain positively charged up to a
distance y, from the metal at t = O"*". Since the reverse bias is increased,
the depletion width must increase instantaneously to ^2(0), and the
capacitance €2(0) measured at O"*" will be equal to £/^2(0). As time
proceeds, those traps which lie above the electron quasi-Fermi level at
distances between y^ and y2(t) from the metal will lose their electrons
and become positively charged, since the electron-emission rate e„ is now
much greater than the electron-capture rate c„n. The positive charge
density is thereby increased, and the depletion width will decrease.
Finally, as f - 00, all the traps above the electron quasi-Fermi level will
become positively charged, the depletion width will decrease to a value
W2(°°), and the capacitance will become C2(°o) = £S/H'2(°°)> S O that
C, > C2(°°) > C2(0). The variation of C with time is shown schematically
in Fig. 4.15. The rate of increase of capacitance from €2(0) to C2(°°) is
determined by the time constant of the traps, and it has been shown by
Seneschal and Basinski (1968a) that the time variation obeys the equation

-^2 = Y [A^d + N,{1 - exp(- t/T)] - ATV.CHJI - e x p ( - t/r)]/e,]

(4.20)
where
AK=F2-F,
Ac-2 = ( C 2 ( o r - c r 2
and Chn, i ^ the harmonic mean 2CjC2{t)/[Ci + C2(t)]. Here r is equal to
e~\ and /V(j and A^, are assumed constant.

(=0

Fig. 4.15 Time variation of capacitance when reverse bias is increased at / = 0

(schematic).
175
 Equation (4.20) can be understood in the following way. If we let the
voltage increment AV, become infinitesimally small so that AV/AC~'^
becomes dV^/dC~^ and become C, then eqn (4.20) becomes identical
to eqn (4.19a) for t= 0*. This is because the traps cannot instantaneously
follow the change in bias, so that Case 3 of Section 4.5.2 applies. If
f — 00, eqn (4.20) becomes identical with eqn (4.18a) because the traps
are able to respond to the change in bias after an infinitely long time and
Case 2 applies. For arbitrary values of t, we may regard eqn (4.20) as
derived from eqn (4.18a) if we suppose that a fraction (1 - exp(- t/r)) of
the traps have been able to follow the change in bias.
If the measurements made by observing €2(1) as a function of t for
various values of V2 are analysed by plotting AV,/AC~^ as a function of
for particular values of t, the relationship should be hnear as
confirmed by Fig. 4.16, which shows Senechal and Basinski's experi­
mental results for A u - G a A s Schottky barriers. It can be seen that the

T 1 1 r

• r=306 K
X r=24I K

Fig. 4.16 Plots of - A F / A C ~ ^ against the harmonic mean capacitance (C|,„) for a A u -
GaAs Schottky barrier. (From Senechal and Basinski, 1968b. Copyright American
Institute of Physics.)
176
straight lines corresponding to different values of / all intersect at a
common point; this must occur when the right-hand side of eqn (4.20) is
independent of t, i.e. when
XC^Je, = 1. (4.21)
The value of A P ^ / A C ~ ^ at which the intersection occurs, namely
qe^NJl, is the same as would be obtained if t/x were vanishingly small.
The latter condition can be obtained by coohng the specimen so that the
electron-emission rate becomes very small and the trap-time constant
very long. It is seen from Fig. 4.16 that the data for T= 241 K do in fact
give a constant value of A I ^ / A C ~ ^ which passes through the intersection
point within experimental error.
By extrapolating the straight lines in Fig. 4.16 to cut the vertical axis,
one obtains A(j+ 7V,{1 - exp(-f/r)) as a function of /, and from these
data A/j, N^, and x can be found. The value of obtained in this way by
Senechal and Basinski agreed with the value obtained from the inter­
section point. Knowing the value of at the intersection point, one
may use eqn (4.21) to find X and hence obtain the energy of the trap with
respect to the Fermi level in the interior of the semiconductor from eqn
(4.14). A n alternative way of obtaining an approximate value of the trap
energy is to measure x (and hence e„) as a function of temperature.
Equation (4.9a) can then be used to find £ , - E-^ if the temperature
variation of o^, v, and can be ignored.
Lang (1974) has discussed a variant of the transient technique in which
the reverse bias is reduced for a short time (or is even changed to a
forward bias) and is then returned to its original value. The capacitance
decreases as a result of the application of the pulse, because more traps
become neutral, and when the pulse has ended it returns to its original
value as described by Senechal and Basinski with a time constant which
is measured directly by means of a special dual-gated detector. Traps
with energy below £ i have their population determined by the hole
quasi-Fermi level and give a capacitance transient of the opposite sign.
This technique is known as deep-level transient spectroscopy (DLTS)
and is widely used for studying deep levels in semiconductors. The
absolute sensitivity depends on the dopant concentration, but under
favourable circumstances a deep level concentration as low as 10'^ m~^
can be detected. The technique has been reviewed by Miller, Lang, and
Kimerling (1977) and by Johnson (1986), and commercial instruments
are available.

4.5.4 The effect of light

Numerous investigations of traps have been reported which depend on

the change in the charge state of the traps brought about by exposure to
light and the subsequent effect of this change in charge state on the
177
capacitance of a Schottky barrier. There are many variants of the
method, all of which are generally described by the term photocapaci-
tance.
Consider the case of a Schottky barrier in an n-type semiconductor
containing donor-like traps which are all occupied by electrons and
therefore electrically neutral. (This may come about because the traps all
lie below the zero-bias Fermi level, or because they have been filled by
an application of forward bias as will be discussed in Section 4.6.2.) Now
suppose that a reverse bias is applied. If the trap population is deter­
mined by the electron quasi-Fermi level, all the traps within a distance y
of the metal [Fig. 4.12(a)] will gradually lose their electrons by thermal
emission. However, at sufficiently low temperatures (say ~10 K ) , the
thermal emission rate will be so low that the traps will retain their
electrons for a very long time. The capacitance measured at a moderately
high frequency will be determined by the total positive space charge in
the depletion region, which is the charge density qN^ due to the shallow
donors. If the barrier is now uniformly illuminated with light whose
quantum energy exceeds the difference in energy between the trap and
the bottom of the conduction band, the traps will lose their electrons by
photoexcitation and become positively charged. Retrapping of the photo-
excited electrons can be neglected because they are removed as soon as
they enter the conduction band by the electric field in the barrier, so the
density of positively charged traps will tend towards A^, with a time
constant TQ - (^CTQ)"^ where ^ is the flux of incident photons per unit
area per second and OQ the optical cross-section of the traps for electron
emission. The square of the capacitance will increase by a factor
(N^ + N^yN^ with the same time constant TQ, SO that if ^ is known a
measurement of TQ allows the optical cross-section to be determined.
The energy of the traps can be determined fro the photothreshold.
This technique was first used by Sah, Rosier, and Forbes (1969) using
p-n junctions. It is capable of great sensitivity because of the com­
parative ease with which small changes in capacitance can be measured.
(A change of one part in 10^ is easily detectable if the initial capacitance
is around 10~'° F.) Sah and his co-workers were able to detect a concen­
tration of 10^'' m""^ sulphur atoms in siUcon, which corresponded to a
total number in the depletion region of only 10^, using this method. It has
been adapted to Schottky barriers by many workers, especially by Henry
et al. (1973) and by Hamilton (1974). A review has been given by
Grimmeiss (1974). A similar technique in which the traps are emptied by
raising the temperature (thermally stimulated capacitance) instead of by
photoexcitation has been used by Sah et al. (1972).
A comprehensive analysis of almost all conceivable transient capaci­
tance effects in p-n junctions subjected to all possible external stimuli
has been given by Sah et al. (1970). Most of their analysis is equally
178
applicable to Schottky barriers. Comparisons of the various methods
have been given by Lang (1974), by Grimmeiss and Ovren (1981), and
by Bourgoin and Lannoo (1983).

4,6 The capacitance under forward bias

4.6.1 The diffusion capacitance

A p - n junction under forward bias exhibits a capacitance due to the

injection of minority carriers into the neutral regions on each side of the
junction, in addition to the space-charge capacitance of the depletion
region. This diffusion capacitance increases exponentially with the
forward-bias voltage and, for moderate values of forward bias, it
dominates the space-charge capacitance. The diffusion capacitance is
closely linked to the phenomenon of minority-carrier storage, in which a
p-n junction suddenly switched from forward to reverse bias exhibits a
comparatively high reverse current until the injected minority carriers
are cleared away. One can think of this transient current as discharging
the diffusion capacitance.
It is natural to ask whether a Schottky diode may also exhibit diffusion
capacitance when forward biased. We saw in Section 3.7 that there is
virtually no minority-carrier storage in a Schottky diode because the
electrons injected into the metal rapidly lose energy through electron-
electron collisions; they cannot then surmount the barrier if the bias is
suddenly reversed. A corollary of this is that a Schottky diode exhibits no
diffusion capacitance, and it is for this reason that when C/V measure­
ments are used to find the height of a Schottky barrier they are frequently
extended into forward bias as the total capacitance is still determined by
the capacitance of the depletion region. It is often difficult to make
measurements in forward bias, however, because of the large shunt
conductance of the diode. It should also be appreciated that the capaci­
tance does not continue to satisfy the relationships (4.3) for very high
forward bias, since they are only valid 'iiqV^> kT.

4.6.2 The effect of traps

We now consider what determines the electron population of traps under

forward bias. We shall assume that the current is determined by the
thermionic-emission theory, so that the electron quasi-Fermi level ^„ is
flat through the depletion region and coincides with £p, while
coincides with Ef. The quasi-Fermi levels and demarcation levels are
shown in Fig. 4.17, where we have assumed that the barrier height and
bias voltage are such that c^p > c„n at the interface. A n important differ­
ence between Figs 4.11 and 4.17 is that in the former E^^ lies above E^^,

179
Fig. 4.17 Schottky barrier witii deep traps under forward bias, showing demarcation
levels of electrons (£d„) and holes (£dp)-

while in the latter the reverse is true. A s a result, there is nowhere in Fig.
4.17 corresponding to the region a<x<c in Fig. 4.11 in which both of
the capture processes are negUgible compared with either of the emission
processes.
To the left of the plane x = a in Fig. 4.17 (where + &p = 2 £ i ) ,
CpP > c„n. The dominant processes are hole capture and, for £ , < tp,
hole emission. Hence in this region f'^e^/^c^p + e^), which gives a
Fermi-Dirac distribution determined by ^p. The increase in e„ relative to
CpP for £ , > E^^ merely brings / even closer to zero. To the right of jc = a,
CpP<c„n, and the dominant processes are electron capture and, for
£•( > electron emission, so that / « e^/(c„n + e„), which gives a Fermi-
Dirac distribution controlled by The increase of relative to c„«
below E^p merely brings / even closer to 1. Hence to the left oix=a the
trap population is determined by ^p, while to the right of x = a it is
determined by A s the forward bias is increased, the electron
concentration increases and the plane x = a moves to the left until,
finally, if c„n> CpP at the interface, the trap population will be controlled
by ^„ throughout the depletion region.
The assumption that is usually made is that the occupation of the traps
is everywhere determined by the electron quasi-Fermi level as we
have seen, this is only true as long as c„n > c^p, which is not necessarily
the case near to the metal. It is more likely to be vahd in large-band-gap
semiconductors like gallium arsenide than in smaller-band-gap materials
like silicon, because, for a given barrier height, E^ lies closer to at the
interface in the former case, and it does not require such a large forward
bias to obtain the condition c„n > c^p throughout the depletion region.
If the trap occupation is determined by the electron quasi-Fermi level
throughout the depletion region, the application of a forward bias such
that ^„ lies above the trap level at the interface is often used to fill all the

180
traps and so provide a convenient set of initial conditions for transient
capacitance or photocapacitance studies. It has also been used by
Roberts and Crowell (1970, 1973) as the basis of a technique of
measuring the energy level of the traps. Their method relies on the fact
that, if the capacitance of a Schottky barrier containing traps is measured
with a test frequency u)^ low enough for the traps to respond (Case 1,
Section 4.5.2), the traps will make a contribution to the capacitance as
long as ^„ intersects the trap level If a forward bias is applied that is
sufficiently large to make 5„ lie above at the interface, the traps will
remain full throughout the period of the test frequency and will make no
contribution to the capacitance. This condition reveals itself as a point of
inflection in the plots of against V.
The situation discussed by Roberts and Crowell is very similar to that
in which there are interface states separated from the metal by a thin
insulating layer. This was discussed in Section 4.2, where we saw that
interface states lying above Ef (assuming an n-type semiconductor) can
make a contribution to the capacitance under forward bias. Since
Roberts and Crowell's technique only detects deep states which are near
the metal, there is no clear distinction between them and interface states.
As we have already pointed out, the theory of the contribution of
interface states to the capacitance of Schottky barriers is still in a rather
unsatisfactory state; these remarks apply equally to the capacitance due
to deep states which lie close to the metal.

181
5 Practical contacts

In previous chapters, apart from admitting the possibihty of an insulating

interfacial layer, we have assumed that all contacts are metallurgically
ideal in the sense that they involve an abrupt transition from a pure metal
to a perfectly homogeneous semiconductor. In practice the methods of
preparation are such that this ideal is rarely attained, and the purpose of
this chapter is to discuss the various methods of manufacture and the
resulting deviations from ideal behaviour. Reviews of the various
methods of depositing metals onto semiconductors have been given by
Fraser (1983) and by Sze (1985).

5.1 Methods of manufacture

5.1.1 Point contacts

Although point contacts were extensively used as detectors in the early

days of radio and later as microwave mixers and detectors, they have
now completely given way to extended-area contacts because of the
vastly superior reproducibility and reliabihty of the latter. The electrical
properties of point contacts are very far from ideal and depend very
much on the way that contacts are 'formed'. They have been extensively
discussed in the books by Torrey and Whitmer (1948) and by Henisch
(1957), and have recently been revived as a research tool by Mcllroy and
Pepper (1985).

5.1.2 Evaporated contacts

The overwhelming majority of practical contacts are made by either

evaporation or sputtering. Most evaporated contacts are made in a
conventional vacuum system pumped by a diffusion pump giving a
vacuum around 10"^ Torr, usually with a Uquid-nitrogen trap. This
method of depositing metal films has been extensively discussed by
Holland (1956). The lower-melting-point metals such as aluminium and
gold can usually be evaporated quite simply by resistive heating from a
boat or filament, while the refractory metals like molybdenum and
titanium are generally evaporated by electron-beam heating. Most
frequently the semiconductor surface is prepared by chemical etching,
and this invariably produces a thin oxide layer of thickness about 10-20 A ;

182
the precise nature and thickness depend on the exact method of
preparing the surface. The effect of surface preparation on the
characteristics of evaporated silicon Schottky barriers has been
discussed by Turner and Rhoderick (1968). Interfacial layers can also be
caused by water or other vapour adsorbed onto the surface of the
semiconductor before insertion into the vacuum system. Such adsorbed
layers can usually be removed by heating the substrate to between 100°C
and 200°C prior to evaporation (Smith, 1969b).
Even if the semiconductor surface were devoid of an oxide layer to
start with, as would be the case if it were prepared by cleaving a crystal
so as to expose a fresh surface, an oxide layer would form by exposure to
air in the time taken to transfer the semiconductor to a vacuum chamber
and pump the system down to its final pressure. For example, in a
vacuum of 10"^ Torr the gas molecules which strike the surface would, if
they all adhered, build up a monolayer in about 10~^ s. If the sticking
coefficient is as low as 10~^, as it is believed to be for oxygen on silicon
(Joyce and Neave 1971), it would take much longer ( ~ 100 s) for a
monolayer to build up. However, the semiconductor surface is exposed
to much higher pressures for several tens of minutes during the pump-
down period, so there is plenty of time for an oxide to form.
The pump-down period can be avoided if the surface is prepared by
cleavage or ion etching within the vacuum system after the final pressure
has been reached, but an oxide layer can still be formed during the
evaporation unless the vacuum is maintained at 10~' Torr or better while
the evaporation is in progress. To maintain pressures of this order
requires a very good vacuum system, because the heat produced during
the evaporation accelerates outgassing. A n interfacial layer may also
form as a result of the adsorption of vapour from a diffusion pump, and
Cowley (1966) has shown that this source of contamination can be
avoided by the use of an ion pump. For this reason it is now common to
use ion-pumped systems capable of achieving a vacuum of 10~'-10~^
Torr.
The only way of ensuring that there is no interfacial layer is to
evaporate a film onto a semiconductor immediately after it has been
cleaved in an ultra-high vacuum at a pressure around 10~'° Torr
(Thanailakis, 1974; van Otterloo and Gerritsen, 1978; Newman et ai,
1985b). The time taken for a monolayer of adsorbed gas atoms to form
at this pressure is so long ( ~ 10^ s) that there is virtually no possibility of
an interfacial layer being established or of the metal film being
contaminated. But although contacts prepared in this way may be
chemically ideal, they are still not physically ideal because of the
mechanical damage to the surface during the cleaving process, and they
frequently show current/voltage characteristics with ideality factors n
considerably greater than unity. This method of preparing contacts is far
183
too slow and cumbersome to be used in industrial processes, and there is
no potential advantage in doing so.

5.1.3 Sputtered contacts

Many authors have described contacts in which the metal has been
deposited by sputtering. For example, Gutknecht and Strutt (1971) have
reported aluminium-silicon diodes which were prepared by r.f.
sputtering, though they give no details of either the sputtering gas or the
pressure. They used 'sputter-etching' (i.e. sputtering in which the silicon
was shielded from the aluminium) to etch the surface of the silicon prior
to deposition of the metal. The diodes were almost ideal, with ideaUty
factors of between 1.01 and 1.02, and barrier heights of 0.78 V which
coincides with the value obtained on cleaved surfaces (see Section 2.1.1).
Presumably the sputter-etching produces an almost ideally clean surface
which has approximately the same surface-state density as a cleaved
surface. Sinha and Poate (1973) have reported the fabrication of nearly
ideal Schottky diodes of tungsten on gallium arsenide by r.f. sputtering.
Mullins and Brunnschweiler (1976) made molybdenum-silicon
Schottky diodes by d.c. sputtering in an argon atmosphere at a pressure
of 0.1 Torr. Again they used sputter-etching to clean the surface of the
silicon. They found that diodes prepared using a sputtering voltage of 1
k V were almost ideal, but that for greater sputtering voltages the I/V
characteristics became non-ideal to an extent which depended on the
voltage and on the sputtering time. They were able to explain their results
by supposing that the sputtering produced defects near the surface of the
silicon within a depth which increased with the d.c. voltage, the density of
these defects being a linear function of the sputtering time. The defects
apparently behaved like positively charged donors, and the additional
space charge produced by these donors caused a narrowing of the barrier
which in turn led to turmelUng of the electrons. They were able to give a
good explanation of the observed I/V characteristics by assuming an
exponential distribution of defects below the surface with a characteristic
length in the range 10-100 A. Similar results have been obtained by
Fonash, Ashok, and Singh (1981) for M o - S i contacts, and by Finetti et
al. (1984) for T i N - S i contacts. The same sort of behaviour has also been
observed with sputtered A u - M o contacts (Devlin, 1980) and T i - W - A u
contacts (Weinman, Jamison, and HeHx, 1981) on gallium arsenide. The
almost ideal characteristics observed by Gutknecht and Strutt were
presumably due either to the different sputtering conditions used by
these workers or to the fact that they annealed their diodes after the
sputtering had taken place.
Sputtering is frequently used to make contacts to practical devices
because it produces metal films with good mechanical adhesion. It is
184
 beginning to supplant evaporation in semiconductor technology because ^
it gives better control of deposition rate and thickness, eliminates
contamination from the boat, and can be easily applied to alloys; it is also
used for depositing metal films prior to siUcide formation (see Section
5.3).

5.1.4 Chemical deposition

Comparatively little attention has been given to the possibility of

depositing the metal chemically, perhaps because of the comparative
ease with which the lower-melting-point metals can be evaporated. But
because of its simplicity and cheapness, chemical deposition is becoming
more popular, especially for refractory metals (Fraser, 1983). Crowell,
Sarace, and Sze (1965) deposited tungsten on germanium, silicon, and
gallium arsenide by reacting tungsten hexafluoride with the appropriate
semiconductor. The resulting Schottky diodes were close to ideal with n
values not exceeding 1.04. This technique is now becoming widely used
in integrated-circuit technology (Shaw and Amick, 1970; Brors et al,
1984). Kano et al. (1966) prepared molybdenum-silicon diodes by
reducing molybdenum pentachloride with hydrogen and found them to
have nearly ideal rectifying characteristics. Furakawa and Ishibashi
(1967) report having made both ohmic and rectifying contacts to n-type
and p-type gallium arsenide by reducing stannous chloride with
hydrogen.
Deposition from solution is occasionally used. GoI'dberg, Posse and
Tsarenkov (1971) reported the manufacture of almost ideal Schottky
diodes from gallium phosphide by depositing gold or nickel by an
electroless process. For gold they used a mixture of HAUCI4 and HF, and
for nickel a mixture of N i C l j , КаНгРОг, H3C6H5O7, and NH4CI. The
diodes prepared in this way had n values of 1.02-1.03. The same authors
have prepared nearly ideal Schottky diodes on gallium arsenide 'by
chemical precipitation of gold' but give no details of the process
(GoI'dberg, Posse, and Tsarenkov, 1975). A list of solutions suitable for
plating contacts onto gallium arsenide has been given by GoI'dberg,
Nasledov, and Tsarenkov (1971). Dorbeck (1966) has reported the
fabrication of rectifying contacts on gallium arsenide by electroplating
gold from an aqueous solution of gold and potassium cyanide or copper
from a solution of potassium copper cyanide, and Datta et al (1980a,b)
used an electroless process for making palladium and copper contacts to
siUcon. A review of methods of plating metals onto semiconductors has
been given by Hillegas and Schnäble (1963).
Sullivan (1976) has described a novel method of depositing strongly
adherent films of silver, gold, palladium, and platinum from commercial
plating solutions onto gallium arsenide, gallium phosphide, and silicon.

185
The area of the (n-type) semiconductor on which deposition is required
is first damaged by bombardment with charged particles either from an
accelerator or from a glow discharge. The semiconductor is then
immersed in the plating solution and illuminated by light with sufficient
quantum energy to produce electron-hole pairs. Deposition of metal
occurs on the damaged area by an electrolytic process in which the
damaged area acts as cathode and any undamaged area as anode, no
external connections being necessary. The method has been more fully
investigated by England and Rhoderick (1981).
The information available in the literature about chemical-deposition
methods is extremely sparse. This is probably because from a techno­
logical standpoint the methods may not be compatible with other
processes involved in fabricating a semiconductor device, and also
because those that are compatible tend to be shrouded in commercial
secrecy. They seem to be used only in very special circumstances.

5.2 The effects of heat treatment

Most contacts used in semiconductor devices are subjected to heat

treatment. This may be deliberate, to promote adhesion of the metal to
the semiconductor, or unavoidable, because high temperatures are
needed for other processing stages which occur after the metal is
deposited. Deliberate heating is often rather loosely referred to as
'sintering' or 'firing'. It is important to avoid the melting of rectifying
contacts because the interface may become markedly non-planar with
sharp metaUic spikes projecting into the semiconductor. When this
occurs, tunnelling through the high-field region at the tip of the spike
may severely degrade the electrical characteristics (Andrews, 1974).
Unless alloying of the contact is desired (for example in the formation of
ohmic contacts), it is necessary to keep temperatures to which contacts
are subjected below the eutectic temperature of the metal-
semiconductor system. A s examples, the eutectic temperatures of alloys
of siUcon with the three common contact metals gold, aluminium, and
silver are 370, 577, and 840°C, respectively.

5.2.1 Contacts to silicon

Even at temperatures substantially below the eutectic temperature,

migration of the semiconductor through the metal may occur. These
metallurgical changes have been extensively studied recently by using the
newly developed techniques of Rutherford backscattering, Auger
electron spectroscopy, and secondary-ion mass-spectrometry (SIMS).
(These techniques are described in several excellent review articles, for

186
 example, Mayer and Turos, 1973; Sigurd, 1974; Morgan, 1975; Mayer
and Poate, 1978; Robinson, 1980.) For example Hiraki et al. (1971) have
shown by using Rutherford backscattering that at temperatures as
low as 200°C there can be substantial migration of siUcon through a gold
film. The extent of the migration is very sensitive to the condition of the
surface of the silicon before evaporation of the gold, and can be almost
completely suppressed by the presence of a thin oxide film at the
interface. It cannot be explained simply in terms of homogeneous
diffusion but probably involves grain-boundary diffusion as well. The
effect of the migration is to make the interface depart very considerably
from a perfect silicon-metal junction, and the electrical characteristics
become very non-ideal. In a later paper Hiraki et al. (1977) have
reported similar behaviour for a wide range of semiconductors at
temperatures as low as room temperature. Generally speaking, the
diffusion constant of sihcon in a metal is many orders of magnitude
greater than the diffusion constant of the same metal in siUcon, so that
diffusion on the silicon side of the interface can usually be ignored
(McCaldin, 1974).
In device appUcations thermal degradation may come about as a result
of heating caused by the passage of current through the diode. Gerzon et
al (1975) have examined the 'burn-out' mechanism in metal-silicon
microwave mixers subjected to pulses of microwave power. They found a
close correlation between the power necessary to cause burn-out and the
temperature at which burn-out can be simulated by heating alone. This
temperature (and therefore the amount of power necessary to cause
burn-out) was found to depend on the choice of metal.
It is usually difficult to relate the degradation of the / / F characteristics
to the observed metallurgical changes. The change in the I/V
characteristics cannot usually be explained simply in terms of a change in
barrier height, but as a rule the whole shape of the characteristic alters to
such an extent that it is clear that we no longer have a simple Schottky
barrier. Occasionally the characteristic can be explained by supposing
that atoms which behave as donors or acceptors diffuse into the
semiconductor, or that electrically active defects are created, so that the
effective density of dopant in the semiconductor is changed. If the
dopant density increases, the barrier gets thinner and thermionic-field
emission (or even field emission) may occur. If the diffusing atoms or
defects introduced into the semiconductor have the opposite polarity to
that of the original dopant, the effective dopant density decreases and it
occasionally happens that a p - n junction may form. A good example of
this is the case of aluminium-siUcon contacts, which have been
extensively studied because of their technological importance. Chino
(1973) first reported that aluminium-silicon contacts, heated above
450°C showed a significant change in I/V characteristics which, in the

187
case of n-type silicon, could be described in terms of increases in the
barrier height and ideality factor. This change in characteristics was
accompanied by pronounced pitting of the sihcon surface. Basterfield,
Shannon, and Gill (1975) have given a convincing explanation of these
observations in the following terms. A t temperatures around 500°C
silicon is taken up into solid solution by the aluminium to an amount
determined by the solubility Hmit at the particular temperature. On
cooling, a layer of silicon is formed immediately adjacent to the
aluminium which contains aluminium in solid solution. This layer of
silicon is doped p-type, because aluminium is an acceptor. The solid
solubihty of aluminium in sihcon at 500''C is about 5 x 10^'' m~^, so that
the net space-charge density is negative near the interface, and the bands
become bent downwards as shown in Fig. 5.1. If the distance from the
maximum of the barrier to the interface is less than the electron mean
free path, the structure behaves like a Schottky barrier of height

Metal Semiconductor

Fig. 5.1 Band diagram for Al-Si contact after heat treatment. (From Basterfield,
Shannon, and Gill, 1975. Copyright Pergamon Journals Ltd.)

The aluminium concentration quoted above is of the right order of

magnitude to explain the observed barrier height increase of about 0.1
eV. This model has been extended by Card (1974), and Card and Singer
(1975) have shown by using Auger electron spectroscopy that if
aluminium-sihcon contacts are heated to 500°C for 20 min the
concentration profiles of the various elements are changed as shown in
Fig. 5.2. In the heat-treated sample the aluminium signal extends 100¬
200 A further into the silicon than in the untreated sample. The
aluminium concentration in this tail is of the order of 10^^ m~^; this is

188
 80|- Alxl-,^ Sixl^^

70
3 60 • • Before
heat treatment
50|- • 4 aluminium
silicon
5) 4o[- carbon
o oxygen

30

< 10 V O X 1 °'*"'jr\

0 2000 4000 6000 8000 10000

Etching time (s)

Alxl- Si X I
70
60 After 550°C
heat treatment
50 • aluminium
' silicon
40 o oxygen
» carbon
30 . Al in
unheated
20 '*, sample
oc«^"./Cxl0
OXIOn
10
.\'^Oxl CxlO-

0 2000 4000 6000 8000 10 000

Etching time (s)
Fig. 5.2 Auger depth profile of Al-Si contact before and after heat treatment at 550°C.
The horizontal axis gives the time for which the contact has been subjected to etching by
ion bombardment and can be regarded as a rough measure of the depth. The vertical axis
gives the amplitude of the Auger signal and is proportional to the concentration of the
particular element. (From Card and Singer, 1975. Copyright Elsevier Sequoia.)

adequate to explain the change in the electrical characteristics. If the

silicon is p-type, the aluminium tail increases the negative space-charge
density and causes the depletion region to become narrower, so that the
effective barrier height is reduced due to tunnelhng (Card, 1975a).

5.2.2 Contacts to compound semiconductors

There is considerable interest in the effect of heat treatment on metal

contacts to compound semiconductors, especially the III-V (A,„Bv)
compounds such as gallium arsenide. The behaviour of these contacts is
comphcated and varied, and depends on both the metal and the
semiconductor.

189
 Sinha and Poate (1978) have classified the various reactions in terms
of the electronegativities of the metals (see Section 1.8). The first group
consists of the noble metals (Cu, A g , Au) which are very electronegative.
For these metals, the dominant effect of heating involves the out-
diffusion of the electropositive component A,„ into the metal. The
second group consists of the transition metals next to the noble metals in
the periodic table (i.e. N i , R h , Pd, Os, Ir, and Pt), which are moderately
electronegative and show a strong tendency to attain closed electron
shells. O n heating, these metals form stable compounds both with the
metal Am and with the metalloid By. The third group consists of the
remaining transition metals (Ti, V , Cr, Zr, N b , M o , Hf, Ta, and W)
together with aluminium. They have fairly small electronegativities and
the interfaces of these metals with AmBy compounds are relatively
stable.
We shall illustrate this wide range of behaviour by taking one example
from each of the three groups. A s an example of the first group, consider
the case of gold on gallium arsenide. Figure 5.3 shows the I/V
characteristics of A u - n - G a A s diodes obtained by K i m , Sweeney, and
Heng (1974) after heat treatment at various temperatures, and Fig. 5.4
shows the corresponding gold and gallium distributions obtained by
SIMS. Above 350°C there is extensive migration of gallium into the gold

10" 1—1— ' ill

~RT
350
- 400 "C
500 °C-

10-« -
/250

1 TCQ n <l>h
10-' ill RT 1.01 0.85 "
250 1.01 0.82
350 1.01 0.86
III 400 1.07 0.87 -
500 2.95 0.59

/(= 5,067 xlO-'cm^

10-

III r 1 1 1 1 J [ 1
0.2 0.4 0.6 0.8 1

^'F(V)
(ii)
Fig. 5.3 (a) Forward and (b) reverse characteristics of a Au-GaAs contact after heat
treatment. (From Kim, Sweeney, and Heng, 1974. Copyright Institute of Physics.)

190
 Penetration depth
(fim)

Fig. 5.4 Secondary-ion mass spectra of a Au-GaAs contact after heat treatment. (From
Kim, Sweeney, and Heng, 1974. Copyright Institute of Physics.)

Up to a limiting concentration set by the solid solubility, followed by a

rapid increase in the gallium concentration at the A u - G a A s eutectic
temperature of 450°C. A t this temperature there is also substantial
migration of gold into the gallium arsenide. The arsenic (not shown)
remains relatively stationary up to 450°C but diffuses rapidly into the
gold as the temperature approaches 500°C. Similar results have been
obtained by Sinha and Poate (1973), by Todd et al. (1974), and by
Robinson (1976). The diffusion of gallium into the gold is accompanied
by the formation of gallium vacancies in the gallium arsenide. Madams,
Morgan, and Howes (1975) have suggested that these vacancies, or
possibly the gold atoms which have diffused into the GaAs, act as donors
and that the change in the I/V characteristics arises not because of a
change in barrier height but because the increased donor density causes
thermionic-field emission. Because the concentration of gallium in the
gold is determined by its solid solubility at temperatures below the
eutectic temperature, the total number of gallium atoms present in the
gold must be proportional to the thickness of the gold fihn, so that a thin
film will have a less serious effect on the characteristics of the diode than
a thick one.
Of the second group of transition metals, which form stable
compounds with both A,„ and By, we consider the case of platinum
contacts to gallium arsenide. This system has been studied by Sinha and
Poate (1973, 1978) using Rutherford backscattering, and by Kim,
Sweeney, and Heng (1974) using SIMS. Between 400°C and 500°C a
layered structure forms having the approximate composition
Pt|PtGa|PtAs2|GaAs. The reasons for this behaviour appear to be the
rapid diffusion of G a in Pt, coupled with the high stability of the
compound PtAs2 formed at the interface. The formation of PtGa is
favoured by the relatively large electronegativity difference between Ga

191
 and Pt. The presence of PtAsj at the GaAs interface seems to produce
stable Schottky barriers, so that the barrier height is insensitive to
annealing temperature.
A s an example of the third group, we consider first the case of
aluminium contacts to gallium arsenide. Here the issue is not clear cut.
Kim, Sweeney, and Heng (1974) reported that A l - G a A s contacts are
fairly stable, and found very little interdiffusion (as detected by SIMS),
with only minor changes in the electrical characteristics after heat
treatment up to 500°C. Johnson, Magee, and Peng (1976) found some
apparent interdiffusion at 500°C, but also observed erosion centres that
exposed the GaAs substrate and which may have given the illusion of
interdiffusion when studied by Auger depth profiling. By contrast,
Christou and Day (1976) reported considerable interdiffusion at 450°C,
with gallium appearing on the surface of the aluminium, though the
electrical characteristics remained relatively stable. They also found a
poorly defined interface in the as-prepared samples. This diverse
behaviour may be connected with the ways in which the aluminium films
were deposited. Christou and Day used an electron-beam evaporator
with the unusually high deposition rate of 100 A/s, which resulted in a
grain size of 500 A. The other groups did not specify their deposition
rates, but it is likely that they used lower rates, resulting in larger grain
sizes. One can then explain Christou and Day's results by supposing that
the variation in observed behaviour is a consequence of varying amounts
of grain-boundary diffusion.
This hypothesis is supported by the observation of Missous,
Rhoderick, and Singer (1986b) that A l - G a A s contacts formed by
depositing aluminium epitaxially on gallium arsenide by molecular beam
epitaxy (MBE) are extremely stable with respect to heat treatment up to
SOO'C, and exhibit abrupt interfaces both at room temperature and after
heating. These aluminium films were almost perfect single crystals, with
very few grain boundaries, and would be expected to be free from effects
due to grain-boundary diffusion. It seems reasonable, therefore, to
conclude that ideal A l - G a A s contacts are very stable, as stated by Sinha
and Poate (1978). This stability is probably due to the strong bond that
forms between aluminium and arsenic, which helps to stabihze the
interface. The small increase in the barrier height (by about 0.05 eV) that
has been observed by several workers and confirmed by Missous et al.
on annealing to 500°C can be explained by assuming an exchange
reaction between A l and Ga with the formation of a thin layer of
Ga^Al,_^As at the interface. (It is known that this latter compound
exhibits a higher barrier height than GaAs.) Srivastava and Arora (1981)
have found that this increase in barrier height is accompanied by a small
decrease in the Richardson constant A**, which they suggest may be due
to electrons tunnelUng through the Ga^^Ali.^As layer.

192
 The refractory metals in the last group show fairly straightforward
behaviour, with inert interfaces and stable barrier heights. A good
example is tungsten on gallium arsenide, which remains stable up to
500°C (Sinha and Poate, 1973). The behaviour of titanium on gallium
arsenide seems to be intermediate between that of tungsten and platinum
(Wada, Yanagisawa, and Takanashi, 1976; Buck, 1983). The T i - G a A s
interface remains stable on annealing up to 400°C, but above this
temperature a layered structure appears as in the case of Pt-GaAs.
However, there is little change in the barrier height or ideality factor even
up to SOCC. This has been attributed to the formation of TiAs2, which
forms a stable interface with GaAs. Wickenden et al. (1984) have
reported that amorphous films of N i - N b and Ta-Ir alloys form Schottky
barriers on n-GaAs which show little change after annealing to 350°C,
and are much more stable than polycrystaUine films. They attribute this
to the absence of grain-boundary diffusion in the amorphous films.

5.3 Silicides

The effect of heat on metal-sihcon contacts is particularly important if

the metal is capable of forming a siUcide which is a stoichiometric
compound. Most metals, including nearly all the transition metals, form
siUcides, and many of these silicides may form as a result of solid-state
reactions at comparatively low temperatures of about one-third to one-
half the melting point of the siUcide in degrees Kelvin (Mayer and Tu,
1974). Tu (1975) has suggested that silicide formation at such low
temperatures may occur as a result of metal diffusion into the silicon with
a consequent weakening of the Si-Si bonds. The metals which are known
to form siUcides by low-temperature thermal reaction are shown in Table
5.1.
Table 5.1 Silicide formation in the periodic system

Li Be

Ka [ugi Al
L_J ^ ^
K Ca Sc I Ti V Cr Mn Fe Co Nil Cu Zn Ga Ge
I , ^. I
Rb Sr Y I Zr Nb Mo I Tc Ru I Rh Pdl Ag Cd In Sn Sb
' I I '
Cs Ba La I Hf Ta W | Re Os | Ir Ft I Au Hg TI Pb Bi
I I I I

All metals form silicides except those enclosed in solid lines (McCaldin, 1974). The
metals enclosed in broken lines form silicides by solid-state thermal reaction at compara­
tively low temperatures (Rubloff, 1983). The rare earths Gd, Tb, Dy, Ho, and Er also form
silicides by low-temperature solid-state reaction.

193
 The vast majority of siUcides exhibit metallic conductivity (Neshpor
and Samsonov, 1960) so that, if a metallic siUcide is formed as a result of
heat treatment of a metal-silicon contact, the silicide-silicon junction
behaves like a metal-semiconductor contact and may exhibit rectifying
properties. Moreover, because the silicide-silicon interface is formed
some distance below the original surface of the silicon it is free from
contamination and very stable at room temperature; contacts formed in
this way generally show stable electrical characteristics which are very
close to ideal. They also exhibit very good mechanical adhesion. For
these reasons they are of considerable importance in semiconductor
technology. Silicide-siHcon Schottky-barrier diodes were first described
by Lepselter and Sze (1968) and by Andrews and Lepselter (1969,
1970). SiUcides are also used as interconnections in microcircuit
technology because of their metallic conductivity and good metallurgical
stability. Reviews of the use of siUcides in integrated circuit technology
have been given by Murarka (1980, 1983) and by Chen and Roth
(1984), while a comprehensive review of the formation, properties, and
characterization of transition-metal siUcides has been given by Nicolet
and Lau (1983).

5.3.1 Mechanism of formation

The kinetics of siUcide formation have been extensively studied using the
techniques of X-ray diffraction and Rutherford backscattering. Figure 5.5
shows the Rutherford backscattering of 1.75 M e V He"*" ions by a 1500-A
film of rhodium deposited on siUcon (Сое et al, 191 A). The figure shows

0.50 0.75 1.00 1.25 1.50

Energy of scattered ions (MeV)

Fig. 5.5 Rutiierford backscattering spectra of 1.75 MeV He+ ions scattered by a 1500-A
film of rhodium on silicon after heat treatment at various temperatures: a, 337°C; b, 483°C,
c,536°C (Coee/fl/., 1974).

194
the energy distribution of the He'"' ions scattered through an angle of
160° after heat treatment of the contact for 20 min at various
temperatures. The results corresponding to 377°C (curve a) are
indistinguishable from those obtained with no heat treatment. The group
of He""" ions with energies between 1.15 and 1.50 M e V are those scattered
by the nuclei of rhodium atoms in the metal film, the ions of higher
energy being scattered from the outer surface of the rhodium film and
the ions of lower energy from the surface of the film adjacent to the
silicon. The loss of energy of these ions is due partly to the loss of energy
of the incident ions through interaction with electrons in penetrating the
film, and partly to the recoil energy given up to the rhodium nuclei. The
He"*" ions with energies below 0.75 M e V are those scattered by siUcon
nuclei. These have a much lower energy than those scattered by the
rhodium nuclei. This is partly because the incident ions contributing to
this group have passed through the whole of the rhodium film, but mainly
because the recoil energy given up to a silicon nucleus is much greater
than that given up to a rhodium nucleus as a result of the much smaller
atomic weight of silicon.
At 483°C (curve b) conversion of rhodium to RhSi has begun. The
reaction begins at the rhodium-silicon interface and moves outwards
into the metal and inwards into the silicon. The ions with energies
between 1.10 and 1.30 M e V are those scattered by rhodium nuclei in the
RhSi phase. The yield (counts per unit energy range) of these ions has
decreased because the concentration or rhodium atoms in RhSi is less
than that in rhodium metal. In addition, the minimum energy of the
group of ions scattered by rhodium nuclei has been reduced because the
energy lost by the incident ions in penetrating both the unconverted
rhodium and the RhSi is greater than the energy lost in the original
rhodium film. A t the same time the maximum energy of the group of ions
scattered by silicon nuclei has increased to about 0.90 M e V because
silicon atoms have moved into the rhodium to form RhSi.
Finally, at 536°C (curve c) the conversion to RhSi is complete. The
yield of ions scattered by rhodium nuclei is constant as a result of the
uniform concentration of rhodium in RhSi, and the maximum energy of
the ions scattered by silicon has increased to a value (1.00 MeV)
corresponding to scattering by silicon nuclei located at the outer surface
of the RhSi. The thickness of the RhSi phase was found to increase
linearly with time; this indicates that the rate of growth is limited by the
reaction rate at one or both of the phase boundaries rather than by the
diffusion of siUcon or rhodium through the RhSi.
Similar studies of other metal-silicon systems have been reviewed by
Tu and Mayer (1978) and by Ottaviani (1979). Generally speaking,
several silicides with different chemical compositions may be formed,
depending on the growth temperature. In particular, the metal-rich

195
silicide (MjSi), the monosUicide (MSi) and the disihcide (MSij) have all
been found to occur when particular metal-sihcon couples are heated to
different temperatures. For cobalt and nickel, all three phases may
appear.
The nickel-silicon system has been extensively studied because of its
technological importance (Andrews and Koch, 1971; Сое, 1974; Tu et
al, 1974; Koos and Neumann, 1975; Сое and Rhoderick, 1976). The
three phases №281, NiSi, and NiSiz have all been identified at various
stages of the growth process. The first to form at a temperature of
approximately 300°C is NijSi. This is well established within a period of
a few hours. After about 10 h at the same temperature the NiSi phase
appears. Both phases appear first at the silicon surface and move pro­
gressively towards the outer surface of the nickel. The thickness of the
№281 phase is proportional to the square root of the reaction time,
whereas that of the NiSi phase is linearly proportional to the time. This
impHes that the rate of growth of the NijSi phase is limited by the
diffusion of one of the elements, whereas that of the NiSi phase is hmited
by the reaction rate at either or both of the phase boundaries. Above
750°C a third phase, NiSi2, appears. Unlike the two earlier reactions, the
conversion to NiSi2 is effected not by the movement of a planar interface
between adjacent phases, but by a gradual decrease in the average nickel
concentration throughout the NiSi over a temperature range of about
100°C. This suggests that there is nucleation of the disihcide at sites
throughout the NiSi phase rather than at the phase boundaries. Similar
behaviour has been reported by Ziegler et al. (1973) to occur in the
conversion of HfSi to HfSij. Such a process is likely to occur if grain
growth is important.
A n additional feature of the growth of NiSi2 is that is can be grown
epitaxially on the (111), (110), and (100) surfaces of silicon (Tu et al,
1974). NiSi2 has the cubic calcium fluoride structure with a lattice
constant of 5.43 A (compared with 5.28 A for siUcon), so there is a close
match in crystal structure and lattice spacing between the disihcide and
siUcon. In the case of the (111) surface it is possible to achieve two
different orientations of the NiSi2 depending on the thickness of a very
thin nickel film which is used as a 'template' to initiate silicide growth
(Tung, Gibson, and Poate, 1983a, 1983b). These two orientations differ
by a 180° rotation of the NiSi2 lattice relative to the silicon lattice, and
involve different atomic arrangements at the interface (see Fig. 2.4).

5.3.2 The sequence of silicide phase formation

One might expect that the composition of the first sihcide compound to
appear on heating a metal-sihcon couple could easily be predicted by
inspection of the metallurgical phase-diagram. However, we are not

196
dealing with a system in true thermodynamic equiUbrium, as is evident
from the fact that silicide formation takes place well below the lowest
eutectic temperature; moreover, in thin film structures the energy
associated with the interface plays a role which is quite different from
that which operates in bulk systems. It is therefore not an easy matter to
predict the sequence of compound silicide formation when the metal-
siUcon couple is heated.
Walser and Bene (1976, 1978) have proposed that the sihcide which
first nucleates has a composition corresponding to the most stable (i.e.
having the highest melting point) congruently meltingf compound
adjacent to the lowest-temperature eutectic point in the bulk phase
diagram. Consider, for example, the phase diagram for the Pt-Si system
(Fig. 5.6). The lowest-temperature eutectic point is at 830°C,
corresponding to 23 at% Si. The highest-melting compound that melts
congruently near to this composition occurs at the nearest maximum of

1600

1400

1200

1000

g
3 800
<u
a.
B

600

400

200
0 20 40 60 80 100
P' Atomic % Si Si
Fig. 5.6 Piiase diagram of the platinum-silicon system. (From Nicolet and Lau, 1983.
Copyright Academic Press.)
tA compound is said to melt congruently when it forms a single liquid phase having the
same composition as that of the solid.

197
the hquidus curve, and this has the composition PtjSi. It is this
cofnpound which is observed to form first when the Pt-Si system is
heated. The Walser-Bene rule appears to work well in nearly all cases,
though there are one or two exceptions. Thus for the M n - S i system the
rule predicts that the first phase to grow should be MusSij, whereas the
observed first phase appears to be MnSi (Tu and Mayer, 1978). It should
be stressed that we are here only talking about phases that are actually
observed to grow; not all the phases present in the phase diagram are
observed in low-temperature thin-film reactions. It is possible that some
phases may nucleate but not grow to a sufficient thickness to be observed
by currently available techniques.
The second phase to appear depends on whether the initial thickness
of the metal is much less than that of the siUcon or vice versa. If the metal
thickness is much less than that of the silicon, as is usually the case, the
growth of the first silicide phase will stop when all the metal is consumed,
and we are left with a silicide-silicon couple. But if, as occasionally
happens, the silicon is thinner than the metal, (for example if a thin film
of polycrystalline silicon is deposited on top of an existing metal film),
the first phase will stop growing when the silicon is consumed, leaving a
metal-silicide couple. The question that now arises is which compound
will appear at the next stage of the reaction. Tsaur et al. (1981) have
extended the Walser-Bene rule to predict the second phase. They
postulate that the next phase to be formed at the interface between the
first phase and the remaining silicon or metal is the congruently melting
compound which is nearest to the first phase in the phase diagram and
richer in the unreacted element. In the case of the Pt-Si system, for
example, if the silicon thickness is much greater than the metal thickness,
the first stage of the reaction leaves an interface between PtjSi and Si.
The rule of Tsaur et al. then predicts that the next phase to be formed is
PtSi, as is observed experimentally. The final phase (assuming the metal
thickness to be much less than the sihcon thickness) is always that
compound which is nearest to silicon in the phase diagram.

5.3.3 The abruptness of the interface

For some time it was thought that all silicide-silicon interfaces took the
form of a gradual transition from the silicide to the elemental silicon over
a distance of a few tens of angstroms. Such a hypothesis not only seemed
reasonable when one considered that the formation of the silicide
involves atomic diffusion, but was also apparently supported by several
experimental observations (for example Roth and Crowell, 1978;
Grunthaner, Grunthaner, and Mayer, 1980).
More recent work has shown that in most cases a graded transition
does not occur. Using high-resolution transmission electron microscopy

198
(ТЕМ), Schmid et al. (1981) were able to show that the epitaxial PdzSi-
S i ( l l l ) interface can be structurally abrupt on an atomic scale, though it
contains steps and misfit dislocations. A similar conclusion has been
reached by Tung, Gibson, and Poate (1983a) for the NiSi2-Si(100)
interface. It must be appreciated, however, that the Т Е М results only
show that there is an abrupt transition from the crystal lattice
characteristic of silicon to one characteristic of the sihcide. This does not
necessarily imply that there must be an abrupt transition from pure
siUcon to a compound having the composition of the silicide; there could
be a gradual transition in stoichiometry in which some atoms of Type A
occupy lattice sites that should be occupied by atoms of type B , or vice
versa. As Rubloff (1983) has pointed out, it is difficult to obtain
unambiguous evidence either way for the existence of compositional
gradients of this sort. Apparent gradients may be due to artefacts such as
non-planar interfaces, which may arise if the heat treatment gives rise to
pitting of the surface, or if there is an uneven growth of the silicide due to
the initial presence of an oxide film on the surface of the silicon.
The present position (1987) seems to be that siUcides which grow
epitaxially and form coherent interfaces (i.e. ones in which the silicide
and silicon lattices are in accurate registration with each other) can be
presumed to have an interface which is abrupt both structurally and in
composition. This may seem surprising when one considers that sihcide
growth involves diffusion, which is a stochastic process, but it should be
remembered that the equilibrium configuration is that which ^as the
lowest free energy, и - TS. If an abrupt interface has associated with it a
negative interfacial energy, the low internal energy U that goes with an
abrupt interface may outweigh the entropy term TS which would favour a
graded ii^terface.

5.3.4 Barrier heights

The Schottky diodes which result from sihcide formation usually show
nearly ideal characteristics with n-values of around 1.1 or less, and with
good agreement between the values of the barrier height determined
from the I/V and C/V characteristics. Table 5.2 shows barrier heights
associated with the rhodium-silicon and nickel-silicon systems (Сое,
1974; Сое and Rhoderick, 1976). The most obvious feature of these
results is the remarkable constancy of the barrier height, except for the
two highest temperatures in the case of rhodium. If one regards the
silicide-silicon junction as one to which the Bardeen model can be
applied, the constancy of the barrier height for a particular metal and its
silicides implies either that the work functions of the metal and of the
silicides are remarkably similar, or else that the barrier height is clamped
by surface states whose characteristics are independent of which

199
 Table 5.2 Barrier iieights of Ni-Si and Rh-Si systems (Сое, 1974)

Phases present Barrier height

(eV)

20 Ni-Si 0.70
230 Ni-(Ni2Si)-Si 0.71
324 Ni-Ni2Si-(NiSi)-Si 0.70
356 NijSi-NiSi-Si 0.70
377 Ni^Si-NiSi-Si 0.72
430 NiSi-Si 0.69
550 NiSi-Si 0.68
650 NiSi-(NiSi2)-Si 0.69
850 NiSi2-Si 0.70
20 Rh-Si 0.78
324 Rh-Si 0.78
377 Rh-(RhSi)-Si 0.79
430 Rh-RhSi-Si 0.79
536 RhSi-Si 0.74
600 RhSi-Si 0.70

particular silicide happens to be present. There is almost no information

available about work functions of silicides, nor is very much known
about interface states at a siUcide-silicon interface.
A very tempting hypothesis is that there may be a very thin layer of
one particular phase always present immediately adjacent to the silicon,
and that this layer is thin enough to escape detection by Rutherford
scattering or other techniques but thick enough to determine the barrier
height (Ottaviani, 1981). Such a hypothesis cannot apply in those cases
of epitaxial sihcide formation where an abrupt transition from the silicon
to the sihcide lattice structure is observed, unless the epitaxial silicide
hself constitutes the phase that determines the barrier height and there is
always a thin layer of this particular phase immediately adjacent to the
silicon. Experimental evidence for the case of the nickel silicides has
recently been interpreted in this way (Chang and Erskine, 1983). A
variant of this hypothesis is that there may be a region only a few
Angstroms thick at the interface where the electron bonding wave-
functions differ from those in the bulk silicon and are independent of
which particular silicide happens to be present. These modified wave-
functions are presumed to create a dipole layer (and possibly interface
states) which determines the barrier height (Schmid et al., 1983).
Experimental support for this assumption is provided by ultraviolet
photoemission spectroscopy (see, for example, Rubloff, 1983), which
shows that the chemical bonds at the interface between sihcon and the
near-noble transition elements (Pt, N i , Pd) resemble those in the bulk
silicide for thin ( ~ 5 A ) films of metal, even at room temperature. There
is also similar evidence for the existence of interface states near or within
the sihcon band-gap at very small metal coverages (Rubloff, 1982;

200
Braicovich, 1983). The picture that emerges is therefore one in which the
properties of the interface that determine the barrier height are
substantially independent of whether the 'metal' happens to be the
elemental metal or the metallic sihcide, and are also independent of the
composition of the silicide.
In the special case of the NiSi2-Si(lll) interface, the two possible
atomic arrangements referred to earlier give different barrier heights
(Tung, 1984), that of the type A interface being 0.65 eV and that of the
type B interface 0.79 eV. On the other hand, if no special precautions are
taken to produce either the type A or the type B arrangements, as in the
work whose results are shown in Table 5.2, the NiSi2-Si system exhibits a
barrier height of 0.70 eV. One can reconcile these results by supposing
that in the latter case there are regions having either orientation, and the
measured barrier height is heavily weighted towards the A value in an I/V
measurement because the saturation current density will be about 200
times larger for the lower barrier. In terms of the model of the previous
paragraph, one can suppose that the two atomic arrangements are
characterized by different bond wave-functions at the interface, giving
rise to different interfacial dipoles and possibly different interface states.
Table 5.3 gives the approximate Schottky barrier height of all the
siUcides which have been reported to date. Since the barrier height
appears to be related to the nature of the chemical bonds at the interface,
it is reasonable to ask whether it is correlated with any macroscopic
quantity which also depends on the silicon-metal bonds. Andrews and
Phillips (1975) have pointed out that there is a good correlation between
barrier height and the heat of formation of the silicide, which is a
measure of the strength of the chemical bonding between the silicon and
metal atoms. Likewise, Ottaviani, Tu, and Mayer (1980) have proposed

Table 5.3 Schottky barrier heights (^t) of various siHcides on

n-type silicon (after Murarka, 1983; Tu, Thompson, and Tsaur,
1981)

Disilicides h (eV) Other silicides h (eV)

TiSij 0.6 HfSi 0.53

VSij 0.65 MnSi 0.76
CrSij 0.57 CoSi 0.68
ZrSia 0.55 NiSi 0.7
MoSi2 0.55 NijSi 0.7
TaSiz 0.59 RhSi 0.74
WSi2 0.65 PdjSi 0.74
CoSi2 0.64 PtjSi 0.78
NiSi^ 0.7 PtSi 0.87
YSij 0.39 IrSi 0.93
DySi^ 0.37 0.85
ErSij 0.39 IrSi, 0.94
GdSi^ 0.37
HoSi2 0.37

201
that there is a correlation with the metal-silicon eutectic temperature,
since this also depends on the bond strength. Schmid (1985) has
suggested that one might expect a correlation with the electronegativity
of the metal atom, and the experimental data, although they show
considerable scatter, do support such a correlation both for those metals
which form silicides and for those which do not. A l l these attempts at
establishing a correlation between Schottky barrier heights and other
parameters tend to be quahtative in nature and the data show
considerable scatter; the justification for seeking a correlation with a
bulk property such as the heat of formation or an atomic parameter such
as the electronegativity is that these properties should be related to the
charge transfer which takes place across the interface when the metal
forms a silicide, and hence to the magnitude of the interfacial dipole
moment.
It will be seen from Table 5.3 that Schottky barrier heights between
silicides and n-type silicon cover the range from 0.93 eV for IrSi and
IrSij down to 0.37 eV for the rare-earth sihcides (Tu, Thompson, and
Tsaur, 1981). The latter value is of practical importance for the
manufacture of ohmic contacts and infra-red detectors, and also has the
interesting consequence that rare-earth silicide contacts to p-type siUcon
have rectifying characteristics.

5.4 Control of barrier heights

Practical applications often arise in which it is desirable to be able to

control the barrier height of a rectifying contact. One simple method
which suggests itself is the use of an alloy of two metals, as was reported
by Arizumi, Hirose, and Altaf (1968, 1969) who used alloys of the noble
metals to make contacts on silicon. They were able to obtain a linear
variation of barrier height with composition, but the practical application
of this principle would require extremely careful control of the
composition of the alloy.
Another method which is possible in principle is to control the
properties of the interfacial layer. Archer and Atalla (1963) and Saltich
and Terry (1970) have discussed the effect of exposure of the
semiconductor surface to oxygen before evaporation of the metal on the
barrier heights of silicon Schottky diodes, and Montgomery, Williams,
and Srivastava (1981) have reported that exposure of indium phosphide
to hydrogen sulphide causes appreciable lowering of the barrier,
ultimately producing ohmic contacts. They suggest that the modification
of the barrier height is not simply due to the creation of interface states
and/or a dipole layer at the interface, but that the HjS dissociates into
free sulphur which can dope the surface layer in the manner proposed by

202
Shannon and discussed below. A similar effect of exposure to HjS has
been observed by Massies et al. (1981) for gallium arsenide, but it has
not yet been found practicable to apply this method in a controlled
manner. Brucker and Brillson (1981) were able to obtain barrier heights
varying from 0.8 eV to ohmic behviour on CdS and CdSe by interposing
a very thin layer of aluminium (f < 2 A ) between the semiconductor and
a gold contact. The mechanism probably involved the formation of an
intermetallic compound at the interface, which either caused a narrowing
of the barrier to allow tunnelling to take place or produced new interface
states.
Several workers have used a thicker interfacial layer, obtained by
oxidizing the surface, to control barrier heights. The effective barrier
height that determines the current-voltage characteristic depends not
only on the diffusion potential, which is modified by the presence of the
interfacial layer as discussed by Card and Rhoderick (see Section 3.8),
but also on the tunnelling probability of the electrons through the layer.
This method has been apphed by Pruniaux and Adams (1972) to gaUium
arsenide and by Majerfeld and Wada (1978) to indium phosphide.
The only practicable method so far reported seems to be that of
Shannon (1974, 1976), who has shown that the effective height of a
Schottky barrier can be controlled over quite a wide range by
incorporating highly doped surface layers. These highly doped layers
were realized by using ion implantation, though more recently M B E has
also been used (Eglash et al, 1983). To reduce the effective barrier
height, the surface layer was doped with impurities of the same type as
those in the interior of the semiconductor (i.e. donors in the case of
n-type semiconductors). The increase in the doping level reduced the
thickness of the depletion region to such an extent that thermionic-field
emission took place, with a resulting increase in the saturation current
density which is equivalent to a decrease in barrier height. By careful
control of the donor concentration and distribution in the surface layer
so that the depletion region just filled the heavily doped layer under zero
bias, it was found possible to maintain good reverse characteristics. The
extra field produced by the application of reverse bias was generated
predominantly within the low-doped bulk of the semiconductor, so that
the surface field was not altered much and there was comparatively litde
increase in the thermionic-field-emission current. In this way Shannon
was able to reduce the height of nickel-silicon Schottky barriers by up to
0.2 eV by implanting 5-keV antimony ions with a mean range of about
50 A and a surface concentration of 10" m~^. The lower curve in Fig. 5.7
shows the effective barrier height for nickel on n-type sihcon as a func­
tion of the surface concentration of antimony ions.
To increase the effective barrier height. Shannon used surface layers
doped with impurities of the opposite type to those in the interior of the

203
 0.8 r

0.3-

0.21 , I , , I ,
0 2 4 6 8 10 12x10"
Antimony atoms implanted (cm^^)

Fig. 5.7 Effective barrier height of Ni contacts on n-type and p-type silicon after
implantation with various doses of 5-keV antimony ions. (From Shannon, 1974. Copyright
American Institute of Physics.)

semiconductor (i.e acceptors in the case of n-type semiconductors). The

effect of the implanted acceptors is similar to that of the aluminium
atoms in n-type silicon described in Section 5.2. If the concentration is
great enough, the bands will actually bend downwards as shown in Fig.
5.1 and, if the distance of the maximum of the barrier from the metal is
less than the electron mean free path, the system behaves like a Schottky
barrier of height j^b- Shannon used this principle to increase the height of
the Schottky barrier produced by nickel on p-type silicon by implanting
antimony ions. The upper curve in Fig. 5.7 shows the effective barrier
height as a function of the surface concentration of 5-keV antimony ions.
The same method has been used by Ponpon and Siffert (1975) to
increase the height of barriers on n-type silicon by implanting gallium
ions.

5.5 Ohmic contacts

As normally used in semiconductor technology, the term 'ohmic contact'

means one for which the //Fcharacteristic is determined by the resistivity
of the semiconductor specimen or by the behaviour of the device of
which the contact forms part, rather than by the characteristics of the
contact. It is not essential that the //Fcharacteristic of the contact itself is
linear, provided its resistance is very small compared with the resistance
of the specimen or device. In addition, the contact should not inject
minority carriers and should be stable both electrically and mechanically.
A n important property of an ohmic contact is its specific resistance

204
(resistance multiplied by area). A good ohmic contact should have a
specific resistance of less than about 10~' Q m^. The specific resistance
can be found by measuring the total resistance of a circular contact of
diameter d on a slice of semiconductor of thickness t and resistivity p.
The total resistance R^„^ is given (Cox and Strack, 1967) by

4i 4«c r.
= tan"'
nd d
where RQ is the resistance of the contact on the opposite face of the shce
(assumed to have infinite area). Other methods of measuring contact
resistance have been described by Berger (1972) and reviewed by Cohen
(1983).
The fabrication of ohmic contacts is still more of an art than a science,
and every laboratory tends to have its own favourite recipes which
involve particular metal or alloy systems, particular deposition methods,
and particular forms of heat treatment. A comprehensive Ust of recipes
for ohmic contacts to germanium, siUcon, III-V, and I I - V I compounds
has been given by Milnes and Feucht (1972), and to III-V compounds by
Rideout (1975) and Sharma (1981).
A l l the recipes appear to depend on one or other of the following three
principles:

1. If the semiconductor is one which conforms approximately to the

simple Schottky-Mott theory [eqn (1.2)], it should be possible in
principle to obtain a contact with a negative barrier height (which
should behave as an ohmic contact) by finding a metal with a work
function less than the work function of an n-type semiconductor or
greater than the work function of a p-type semiconductor, as
discussed in Section 1.3.1. Alternatively, if the barrier height is
determined by Fermi-level pinning, it occasionally happens that the
pinning position is within the conduction band or valence band, in
which case negative barrier heights should be obtainable on n-type or
p-type semiconductors, respectively. The ohmic contact formed by
gold on indium arsenide seems to operate in this way (Walpole and
Nill, 1971). Unfortunately there are very few metal-semiconductor
combinations with this property, though if the barrier height is
positive, but very small, the contact should have a low enough
resistance to be effectively ohmic.

2. The vast majority of ohmic contacts depend on the principle of having

a thin layer of very heavily doped semiconductor immediately
adjacent to the metal. The depletion region is then so thin that field
emission takes place and the contact has a very low resistance at zero
bias. This mechanism has already been discussed in Section 3.3.2.
205
3. If the surface of the semiconductor is damaged (for example by sand­
blasting), crystal defects may be formed near the surface which act as
efficient recombination centres. If the density of these is high enough,
recombination in the depletion region will become the dominant
conduction mechanism and wiU cause a significant decrease in the
contact resistance. This method is of mainly historical interest; it was
widely used during the early days of semiconductor research for
making contact to sihcon and germanium, the actual contacts being
obtained either by soldering or by the application of pressure.
The most widely used of these methods is the one described m (2). The
heavily doped layer may be formed separately from the deposition of the
metal, say by diffusion or ion implantation (Meyer, 1969), or by M B E
(Barnes and Cho, 1978). Most of the ohmic contacts to silicon used in
present-day integrated-circuit technology are made in this way, since the
area to which contact has to be made is often highly doped as a
consequence of the circuit design. A typical example of the present state
of the art is given by Cohen et al. (1982) who investigated platinum
siUcide contacts to ion-implanted n"*"- and p"'"-type silicon. Specific
contact resistances around 10~" Q m'^ were obtained for doping levels of
10^^ m~^ The same authors point out that one of the problems is the
elimination of the oxide film that normally exists on the silicon surface;
this problem can be minimized by the use of a contact metal that reduces
the oxide, such as aluminium (Cohen etal, 1984).
Alternatively, the heavily doped region may result from the deposition
and subsequent heat treatment of an alloy containing an element which
acts as a donor or acceptor in the semiconductor. The heat treatment
may be carried out in an ordinary furnace, or by irradiation with a laser,
infra-red lamp, or electron beam. O n heating, some of the semiconductor
dissolves in the metal, and on subsequent coohng it recrystalhzes with a
high concentration of the electrically active element in solid solution.
One of the most important properties of a suitable alloy for ohmic
contacts is that it should 'wet' the surface of the semiconductor when it
melts (Zettlemoyer, 1969). For silicon and germanium, the alloys most
commonly used are gold-antimohy for n-type and gold-indium for
p-type.
Because of its growing technological importance, much effort has been
devoted to the development of ohmic contacts for gallium arsenide, and
the subject has been reviewed by Yoder (1980) and by Braslau (1981,
1983). The most widely used alloys are those of gold or silver with
germanium and tin (for n-type material) and with germanium,
manganese, or zinc (for p-type material). Gold-germanium will work on
both n-type and p-type because germanium is an amphoteric impurity,
i.e. it may behave as either a donor or an acceptor depending on whether
it substitutes for a gallium or for an arsenic atom. For many years A u - G e
206
was the most widely used ohmic-contact alloy for n-type, GaAs, but it
suffers from the drawback that it forms a contact which 'balls-up' and
becomes laterally inhomogeneous. A detailed study by Iliadis and Singer
(1983) showed that this inhomogeneity appeared just above the eutectic
temperature (363°C), and that sintering to 450°C caused the appearance
of areas having three distinct compositions: an underlying matrix
composed of GaAs with a small trace of gold, on which lay agglomerated
areas of irregular shape consisting of gold with a trace of gallium, and
rectangular particles consisting of about 80% A u , 10% Ga, and 10% A s
with some incorporation of Ge. The germanium seems to play an
essential role in the 'balling-up' process, since A u - G a A s contacts with no
Ge remained homogeneous. Braslau (1981) claims that sintered A u - G e /
GaAs contacts show a contact resistance which is approximately
proportional to where A^^ is the background donor concentration in
the galUum arsenide, and a relationship of this form was also found by
lUadis and Singer, though Heiblum, Nathan, and Chang (1982) found a
much weaker dependence on A^^. Braslau explained his observed
dependence on A^^ by supposing that the effective ohmic contact was
confined to discrete Ge-rich islands (presumably the rectangular areas
observed by Iliadis and Singer) and that the contact resistance of these
areas was negligible. The measured contact resistance is then due to the
spreading resistance between the ohmic areas and the bulk of the GaAs
which is proportional to A/j'.
The 'balling-up' can be overcome by covering the A u - G e with a nickel
overlayer (Braslau, Gunn, and Staples, 1967). A t first it was thought that
the nickel acted simply as a blanket to hold down the underlying Au-Ge
film and keep it in a homogeneous state, but it was later realized that the
N i moves through the A u - G e to accumulate at the GaAs interface and
act as a wetting agent (Robinson, 1975; lUadis and Singer, 1983).
However, N i diffuses very rapidly in GaAs, and the thickness of the
overlayer must be kept to a minimum (Braslau, 1983). Robinson (1980)
has published a graph of specific contact resistance versus alloying
temperature for N i - A u - G e contacts to gallium arsenide, and a study of
the metallurgical changes involved has been made by Kuan et al. (1983).
Murakami et al. (1986) found that the deposition of a film of nickel 5 A
thick between the GaAs and the N i - A u - G e alloy lowered the annealing
temperature necessary to produce low-resistance contacts. The lowest
specific contact resistance that has been reported using N i - A u - G e alloys
is about 3 X 10~" Q m^ on GaAs having A^^ = 3 x 10^^ m~^ (Nathan and
Heiblum, 1982) or A^^ = 2 x 1 0 " m"^ (Inada et al, 1979). Similar values
of contact resistance have been obtained by Ghosh, Yenigalla, and
Atkins (1983) using N i - A u - T e ; tellurium has the advantage over
germanium that it is not an amphoteric impurity, so that compensation is
avoided.

207
 There are many cases of ohmic contact fabrication in which the simple
picture of a heavily doped barrier region does not seem adequate to
explain the observed behaviour. Sebestyen (1982) has proposed that
many such instances can be explained in terms of a model in which the
alloying process creates a region where the composition of the
semiconductor varies with distance from the metal in such a way as to
form a graded heterojunction. For example, indium is known to make an
ohmic contact to n-gallium arsenide despite the fact that it is not a donor
in GaAs. Sebestyen suggested that as a result of the heat treatment a thin
layer of InAs is formed which is separated from the GaAs by a transition
layer of In^Ga,_^As. Since there is a negative barrier between In and
n-InAs, and the transition layer would have a smooth variation in band-
gap without any discontinuities in the conduction band, the overall result
would be a low resistance contact to the gallium arsenide. Such a graded
structure has been synthesized using M B E by Woodall et al. (1981) and
found to have a contact resistance of about 10~'° Q m^. In some cases
the transition region may be disordered or even amorphous, and
evidence for such a disordered region in A u - G a A s contacts has been
obtained by Gyulai et al. (1971) from Rutherford backscattering data,
and by Magee and Peng (1975) using Т Е М .
Values of specific contact resistance below 10"'^ Q m~^ have been
achieved by Stall et al. (1979) using a novel structure in which a thin,
degenerately doped germanium film of thickness 250 A is grown by
M B E on an n"*" layer of gallium arsenide, and a gold contact made to the
germanium film. The discontinuity in the conduction band at the G e -
GaAs heterojunction is so small that it does not present a significant
barrier to current flow, and the low Schottky barrier height of the A u - G e
contact ( - 0 . 5 0 eV) combined with the high doping level in the
germanium results in a very low contact resistance.
A method of making ohmic contacts which has the merit of great
simplicity but is only apphcable- in a limited number of cases is that of
plating by deposition from a solution without the passage of an electric
current ('electroless' plating). To ensure that the metal should adhere well
it is often found to be necessary to roughen the surface of the semi­
conductor. Suhivan and Eigler (1957) made ohmic contacts to sihcon by
the electroless plating of nickel from a solution of nickel chloride, sodium
hypophosphite, ammonium citrate, and ammonium chloride. The process
was carried out at 100°C. On low-resistivity material (5 x 10"'* Q m) the
method gave low-resistance contacts to p-type material provided the
contact was not annealed, and a low resistance to n-type material after
annealing to 600°C. The p-type result is apparently due to the existence
of a low barrier, the n-type result to the penetration of phosphorus from
the hypophosphite into the sihcon at elevated temperatures. Kelly and
Wrixon (1978) reported having made ohmic contacts to n-GaAs by

208
electroplating a A u - S n N i - A u sandwich structure on to the GaAs,
followed by anneahng at 300''C. Specific contact resistances as low as
3 X 10"' Q were obtained.
In device apphcations the alloy which is used to form the ohmic (or
rectifying) contact is always covered with a layer of a good conductor
such as aluminium or gold, in order that a low-resistance coimection to
the external circuit may be made. When this is done, it is usually
necessary to interpose a diffusion barrier between the contact alloy and
the low-resistance layer, to inhibit diffusion from one to the other.
Diffusion barriers usually consist of refractory metals such as chromium,
vanadium, tungsten, or titanium. For example, the use of chromium as a
diffusion barrier between NiSi, PdzSi, or PtSi and aluminium has been
studied by Bartur and Nicolet (1984), and Eizenberg, Thompson, and Tu
(1982) have used vanadium as a barrier between GdSij and aluminium.
The resulting multi-layer structure is often quite complicated.

209
210
Appendix A: The depletion approximation

It is frequently necessary to know how the electrostatic potential and

electric-field strength in a Schottky barrier depend on the barrier height,
bias voltage, and impurity concentration, and for most purposes it is
sufficiently accurate to use an approximation known as the depletion
approximation. In this approximation the free-carrier density is assumed
to fall abruptly from a value equal to the density in the bulk of the semi­
conductor to a value which is negligible compared with the donor or
acceptor concentration, whereas in reaHty the transition occurs smoothly
over the distance in which the bands bend by about 3kT/q.
Let us consider the case of an n-type semiconductor. Outside the
depletion region the semiconductor is neutral, and within the depletion
region the charge density is equal to qN^, so that the variation of charge
density with distance is as shown in Fig. A . I . With no appHed bias there
can be no electric field within the neutral region of the semiconductor
otherwise a current would flow; even when a bias voltage is applied, the

I qN, •
P

X »- w
(a)

Fig. A . l Variation of (a) charge density, (b) electric-field strength, and (c) electrostatic
potential with distance according to the depletion approximation.

211
current is usually so small in relation to the conductivity of the semi­
conductor that the electric field in the neutral region is negligible. The
electric field is related to the charge density by Gauss's theorem so that
dS'/dx'= qN^/e^. The magnitude of ^ (which is negative since it is
directed in the negative x direction!) increases hnearly as the metal is
approached, rising from zero at x = w to a value of qN^w/e^ at the
interface so that S'= - qNa{w - X)/E^. If we take the electrostatic poten­
tial V to be zero within the neutral region so that ^(w) = 0, the potential
at X is given by

rp(x) =

qNi{w-x)
dx

2E

The magnitude of V rises quadratically as the metal is approached

and has the value qN^w^/2e^ at the interface. This result can be remem­
bered easily because the difference in potential across the depletion
region is simply equal to the average electric-field strength multiphed by
w, and if varies hnearly the average field strength is equal to
^l-^maxl = ^^^d'^/^s- The magnitude of xp at the interface is equal to the
diffusion potential Fj so that
2
Fd. = ^ . (A.1)

Making use of the relationships

qN,w (2d

where gd is the total charge per unit area due to the uncompensated
donors in the depletion region, we can write eqn ( A . l ) in the alternative
forms

K = ^^^^^ (A.2)
2qN,

tThis definition is necessary to enable us to write Gauss's theorem in the above form,
which implies that the positive direction of S" must be in the positive x direction. However,
in Chapter 2 it was convenient to use the opposite definition of the sign of ^ so that the
field would be positive in an n-type semiconductor.
212
and

(A.3)
2eM
According to eqn (A.3) the differential capacitance per unit area is given
by

1/2
C = (A.4)
2V.d / W

The energy of the bottom of the conduction band relative to the Fermi
level in the metal [eqn (3.6)] is given by

^c(^) = + {V(0) - rp(x)}

= h +— (x^ -2wx). (A.5)

Nomograms showing values of w and C corresponding to particular

values of dopant concentration and diffusion potential according to the
depletion approximation have been published for siHcon and gallium
arsenide by Miller, Lang, and Kimeriing (1977) and are reproduced in
Figs A . 2 and A . 3 .

213
Total reverse Depletion Capacitance Ionized Voltage
bias voltage depth W pF/cm' impurities/cm' Resistivity ii cm breakdown
(M) (C) (N) n-type (P„) p-type (pp) (BV)

-500
-5000
-100
5X10'^
•50

^50
F-io'*
10^ — 10' -10'
-100

500-f 50--
-50 r-10'
Example
5x10" ^10
VR=20V, N=10"/cni' 2'-
-500
W=5n
-10"--
1 0 ' - C=2,050 pF/cm' 10'-z
^10
p„=4.9 £2 cm
Pp=13 Q cm
50- 5^-
BV= 350 V -5
-5x10'-
-10'
: 5,000 -1

-10"
10' — 10^
-0.5 -50

5- 0.5--

-0.5
5x10"

F^IO" -0.1
1— 0.1-^-10' -10'

0.5- 0.05^-
-0.05

- 7 500,000 -5x10" — 0.1

-5
-10"
0.01 — —
-0.05
(A2)

Figs A . l and A.3 Nomograms showing depletion width, capacitance, and theoretical
breakdown voltage for Schottky barriers on siUcon (A.2) and gallium arsenide (A.3) at 300
K. (From Miller, Lang, and Kimeriing, 1 9 7 7 . ) The 'Total Reverse Bias Voltage (KR)' shows
the total band-bending ( = Ki„ -t- V,).

214
 Ionized Debye Breakdown
impurities/cm' length voltage
(AT) B.V.(volts)

Total reverse Depletion Capacitance Resistivity

bias voltage depth pF/cm^ p(n cm)
(C) n p

3x10"-
•:-30 -30 :
-3000 300--
-300 -1000
10"
-100 -10
rlOOO 100-
--0.3 -100
3x10"-•
:-300
-300 30-: -3
-300
-30
10'
-1000
rlOO 10-
—1
-0.1
-10
3xl0"--
:-3000 -100
-30 3--
-0.3

10" - 3
-10'
-10 1-
-0.1
- - 0.03 -1
3x10'° - - -30
--3x10'
-3 0.3-:
-0.03
-0.3
10"
-10'
-1 0.1-
_-0.01 rO.Ol -10
3x10' -0.1
-;-3xio'
-0.3 0.03- -
-0.003

10'
-0.03
rO.i 0.01- -3
—
•0.003 -0.001

(A3)

215
Appendix B: Exact analysis of the electric field in
a Schottky barrier

It is sometimes necessary to know more accurately than the depletion

approximation allows how the electric-field strength in a Schottky barrier
depends on the barrier height, bias voltage, and impurity concentration.
To this end, we must allow for the fact that the majority carrier con­
centration does not fall abruptly to zero but penetrates into the depletion
region, and must also take into account a possible contribution to the
field from minority carriers close to the metal. The analysis will be
developed for an n-type semiconductor.
Let the electrostatic potential within the semiconductor at a distance x
from the interface by ijj(x), and suppose that the zero of potential is
chosen so that V(°°) = 0- The energy of the bottom of the conduction
band at a point in the depletion region then exceeds that in the interior of
the semiconductor by an amount - qijj, and if the bands are bent
upwards as in Fig. B . l ip must be negative. Let the equilibrium density of
electrons in the interior of the semiconductor by «o and the equilibrium
density of holes immediately adjacent to the interface be ft. A t a distance
X from the interface the densities of electrons and holes depend on the
positions of their respective quasi-Fermi levels (see Section 3.2). For
electrons the thermionic-emission theory assumes that the quasi-Fermi
level remains horizontal throughout the depletion region and coincides
with the Fermi level in the bulk semiconductor and, even according to
the diffusion theory, the quasi-Fermi level is still flat almost up to the
interface. The assumption of a flat quasi-Fermi level should therefore be
a good one except possibly close to the metal, but since the electron
concentration is in any case negligible compared with A^^ in this region,
no significant error is caused by this assumption as long as qV^> kT.

Fig. B . l Band diagram of a Schottky barrier, showing definition of V and V^.

216
Assuming the semiconductor to be non-degenerate, the electron con­
centration in the depletion region is given by n = exp(qrp/kT), where
«0 is the electron concentration in the neutral region of the semi­
conductor (see Fig. B . l ) . For holes the usual assumption (which has been
justified by Rhoderick, 1 9 7 2 ) is that the hole quasi-Fermi level is
horizontal throughout the depletion region and coincides with the Fermi
level in the metal, so the hole concentration is given by
p=^p,cxp[-q(V, + xp)/kT]
where is the diffusion potential (or band bending) and p^ is the hole
concentration at jc = 0. is equal to - Vs where is the surface
potential.
The net charge density is given by
p^q(Na+p-n)
= q[N, + A exp{- q(V, + xl,)/kT] - «0 exp(^V/^r)].
The charge density and electrostatic potential are related by Poisson's
equation

^ = _ P

It is convenient to introduce new variables u = qip/kT and F= -du/dx

so that d^u/dx^ = F(dF/du). Hence
dF ^
F— = - -~{Ni +ftexp(M, - M) - «oexp(M)}
du SsKl
where Ms (= - qV^/kT) is the value of u at the surface. Integrating from
the interior of the semiconductor to the surface,

- ^ {^d"s - A [ l - exp(M,)] - « 0 [exp(M,) - 1]}

where F^ is the value of F at the surface. We have made use of the fact
that F and u both vanish in the interior of the semiconductor and is
assumed constant. The electric-field strength is given by
^ = - drp/dx = - {kT/q){du/dx) = kTF/q
so that
2kT
^Lx = {^d"s - A [ 1 - exp(M,)] - no[exp(M,) - 1]}.

Writing this equation in terms of (= - kTuJq) and making use of the

fact that for an n-type semiconductor «o is very closely equal to
(assuming complete ionization of the donors), we find

217
 E>2 2q D kT
' max K — + ^[\-exp{-qVJkT)]
[ 1 I

+ ^exp(-qV,/kT) (B.l)

an equation first derived by Atalla (1966).

Unless the barrier height is so large that the surface is strongly p-type,
the term involvmg is negUgible, and if qV^>kT we may neglect the
exponential term in the coefficient of so that, to a very good approxi­
mation,

o2
2qN, kT
' max V,- (B.2)

The first term on the right can be understood very simply because it is
the result obtained from the depletion approximation in which the charge
density is assumed to rise abruptly from zero to the value qN^ at the edge
of the depletion region. The second term is a correction due to the
transition region (sometimes referred to as the Debye tail or, incorrectly,
as the 'reserve' layerf) in which the electron density falls from KQ to a
value neghgible compared with A^^.
It is important to reahze that in this more exact theory the depletion
region does not have a precisely defined width, since the bottom of the
conduction band approaches its position in the bulk of the semi­
conductor asymptotically. However, if the finite conductivity of the
semiconductor is taken into account, the bands in the neutral region of
the semiconductor must slope in the opposite direction to those in the
depletion region for the case of forward bias, so that the conduction
band edge must pass through a minimum. We can take the position of
this mmimum as denoting the edge of the depletion region. A fuller
discussion of this point has been given by Moreau, Manifacier, and
Henisch(1982).

tThe 'reserve' regime of a semiconductor refers to the situation in which the donors are
not completely ionized. That is not the case here.

218
Appendix C: Comparison of Schottky diodes and
p-n junctions

Although the choice between Schottky diodes and p - n junctions for

device applications is usually dictated by purely technological considera­
tions, it sometimes happens that both of them are technologically
feasible, and the choice between them has then to be made by reference
to their electrical properties. A full discussion of this topic has been
given by Rhoderick (1976).

C . l Current-transport mechanisms

We shall compare a Schottky diode made from an n-type semiconductor

with a p - n junction having the same barrier height. [By the 'barrier
height' of a p - n junction we mean the quantity (j>^, in Fig. C.l(b).] The
Schottky diode has the property that the current is carried almost
entirely by electrons even though the semiconductor may be only
moderately doped, and to ensure that this is also true of the p - n junction
we must assume that the latter is made with the p side more lightly doped
than the n side.
Assuming that the current through the p - n junction is carried entirely
by electrons, the current/voltage relationship will be of the form
•/p„ = V { e x p ( g F / A : r ) - l ) (C.1)
where

/op„ = exp(- q(l,^/kT). (C.2)

(a) (b)
Fig. C . l Band diagram of (a) Schottky barrier and (b) p-n junction, showing definition of

219
Here is the effective density of states in the conduction band, and
and Lg are the diffusion constant and diffusion length, respectively, of
electrons in the p-type side. Equation (C.2) can be rewritten by making
use of the relationship (from kinetic theory) that = vlJ3, where 4 is
the mean free path of electrons in the p region and v their mean thermal
velocity. Moreover 4=VT:^e and Le = 7A^re> where is the mean time
between colhsions and T „ the lifetime of electrons in the p-type side; eqn
(C.2) then becomes
/op„ = {qKW{3r,r']cxpi- q,t>JkT) (C.3)
where

For a Schottky diode the current/voltage relationship is normally given

by the thermionic-emission theory, according to which
/s = /os{exp(?F/A:r)-l}. (C.4)
From eqn (3.9), takmg p = \, we may write
= (gA4v/4)exp(- q(l>^/kT). (C.5)
Comparison of eqns (C.3) and (C.5) gives

•^OS (^/"e) ' _ 1/2

« . (C.6)
•^Opn
For a silicon p - n junction, a typical value of x^^ is ~10~^ s while
T „ ~ 1 0 - " s, so {r^y^ > 10^ Hence for the same barrier height the
saturation-current density of a Schottky diode exceeds that of a silicon
p - n junction by a factor of 10^ or more. For a short-hfetime semi­
conductor like gallium arsenide the ratio is smaller but is still of the order
of 101
In practice, the barrier height of a Schottky diode is hkely to be
appreciably less than that of a p - n junction made from the same semi­
conductor. In the case of sihcon, for example, the lowest possible value
of ^Y,, which occurs when the p side of the p - n junction is very lightly
doped (say with « 10^° m"'), is about 0.8 eV, while it is easy to make
Schottky diodes with barrier heights in the range 0.5-0.6 eV. This fact,
taken in conjunction with the argument of the previous paragraph, has
the result that the saturation current density of a Schottky diode is likely
to exceed that of a p - n junction made from the same semiconductor by a
factor of 10^ or more. Expressed in a different way, for the same current
density the forward voltage drop across a p - n junction will exceed that
across a Schottky diode by at least 0.4 V. This makes a Schottky diode
particularly suitable for use as a low-voltage, high-current rectifier.
Conversely, since the reverse saturation current is also larger, a Schottky
220
aiOae IS nOl as SUliaUlC as a p—ll JUllCUUll lUl use a» a niKii-vunaj^s.-, iKjrt

"current rectifier.

C.2 Hole injection

In the case of a p - n junction, the term 'minority-carrier injection'

normally refers to the injection under forward bias of those carriers
which are predominantly responsible for current flow. For the case under
consideration this means the injection of electrons into the p-type side,
and it is this sort of minority-carrier injection which is utilized in bipolar
transistors and light-emitting diodes. The exact analogue of this process
for a Schottky diode made from an n-type semiconductor consists of the
injection of electrons into the metal. These injected electrons do not
influence the conductivity of the metal in any way, so there are no effects
attributable to the exact analogue of minority-carrier injection in a
Schottky diode.
However, the injection of holes from the metal into the semiconductor,
although usually very small, is not always completely neghgible. The
analogue of this process in the p - n junction under consideration would
be the injection of holes into the n-type side; this is generally assumed to
be negligible if the n side is much more heavily doped than the p side.
The transport of holes through the depletion region of a Schottky barrier
and their subsequent diffusion in the neutral region of the semiconductor
is identical to the mechanism of hole transport in a p - n junction, and we
may write
A = ^h{exp(gF/fcГ)-l],
where, by analogy with eqn (C.3),
= {qKv/(3r,r']cxp(- q<t>JkT). (C.7)
Here is the effective density of states in the valence band, is the
'barrier for holes' shown in Fig. C.l(a), and ^i, = Tri,/r^h- The hole
injection-ratio is therefore given by

or

n " r \ n exp{ - q{h - <l>y,ykT], (C.9)

since is the thermionic-emission current given by eqns (C.4) and (C.5)

and « A^,. Equation (C.8) is equivalent to eqn (3.30) if the concentra­
tion gradient of holes is determined by the hole diffusion length rather

221
than by the thickness of the quasi-neutral region L. A s eqn (C.9) shows,
increases with increasing because of the reduction in and
decreases with increasing because of the reduction in Jy,. For most
Schottky barriers > j^b> so the exponent is negative. Taking the case of
a gold-silicon Schottky barrier with ~ 0-8 eV and = 10^^ m"^ and
assuming {ry^^ ~ 10^, we find « 10"^; this is in reasonable agreement
with Y u and Snow's results shown in Fig. 3.15. This doping level
corresponds to « 0.9 eV. If we take = 10^° m"^, is reduced to 0.8
eV and y,, increases to ~ 10"^. The hole injection ratio is small in the
asymmetrical p - n junction considered earher and is even smaller for a
Schottky diode because, although the hole current should be the same
for a given forward bias voltage and a given donor denshy in both cases,
the electron current is much greater in the Schottky diode because of the
considerations of Section C . l . [From the point of view of potential
apphcations, it is the electron injection-ratio which is important in the
p - n junction under consideration (with > N^), and this is vastly
greater than the hole injection-ratio in the Schottky diode.] A s was
mentioned in Section 3.8, the hole injection-ratio of a Schottky diode can
be increased by the presence of a thin interfacial layer to an extent which
may have possible device apphcations.
One consequence of minority-carrier injection in a p - n junction is that
majority carriers are drawn in from the contact to maintain almost
perfect charge neutrahty ('quasi-neutrality'). Since minority-carrier
injection normally refers to the carriers which emanate from the more
strongly doped side, the effect is to increase the carrier concentration in
the more weakly doped side. This effect is known as conductivity
modulation and is of considerable importance in power rectifiers. In a
Schottky diode the injection of electrons into the metal has a neghgible
effect on the metal's conductivity. However, if the barrier height is
large and the semiconductor is of high resistivity (which gives a
comparatively small value of the injection of holes into the semi­
conductor may cause observable conductivity modulation in the semi­
conductor at high current densities where increases with as
discussed in Section 3.5 (J^ is effectively equal to the total current
density.) Conductivity modulation of this sort has been observed by
Jäger and Kosak (1973) in Schottky diodes made from high-resistivity
silicon.

C.3 Minority-carrier storage

Minority-carrier storage is a consequence of minority-carrier injection

and in a p - n junction refers to the injection of those carriers pre­
dominantly responsible for current flow, which for the case under

222
consideration are electrons. If the bias is suddenly changed from forward
to reverse, the electrons injected into the p side must be removed before
the diode attains a high-resistance state. The recovery time is approxi­
mately equal to r „ which in practice may be about 10~* s for a sUicon
rectifier. The use of p - n junctions as fast switching diodes or microwave
mixers is limited by this recovery time.
The exact analogue of minority-carrier storage for a Schottky diode
made from an n-type semiconductor is the storage of electrons after they
have been injected into the metal. A s we saw in Section 3.7, the time
taken for these injected electrons to lose so much energy that they are
unable to diffuse back into the semiconductor is about 10"^" s, which is
quite negligible. There is therefore no direct analogue of minority-carrier
storage in a Schottky diode.
There is, however, the possibility of storage effects associated with the
injection of holes into the semiconductor. The recombination time of the
holes (~ 10~* s) is quite long compared with the storage time associated
with the electrons injected into the metal, but the total charge stored
in this manner is approximately equal to Jy^T,^, = yhJe^rhJ this is still quite
small because of the smallness of the hole injection-ratio y^- The effective
storage time is given by = Qh/-^tot ~ yn^rh and is generally less than
10~" s. (We have assumed that the total current density /,<,, is effectively
equal to Je-) A t high current densities rises because of the increase in
y^ and depends in a compUcated way on the boundary conditions at the
ohmic contact as has been discussed in some detail by Scharfetter
(1965). Even Th is not observable experimentally, and in practice the
recovery tunes of Schottky diodes are limited by their RC product.
Minority-carrier storage in a p - n junction has associated with it a
capacitance—the diffusion capacitance—which is added to the ordinary
depletion-region capacitance when the junction is forward biased. One
can think of the charge stored in the injected carriers as constituting the
charge on a capacitor. Because there is virtually no minority-carrier
storage in Schottky diodes, there is no diffusion capacitance either, and
the capacitance of a forward-biased Schottky barrier is simply that due to
the space charge in the depletion region.

223
Appendix D: The hole quasi-Fermi level

Although hole injection in Schottky diodes made from n-type semi­

conductors is generally very small, it is not completely negligible, and
there are occasions when one needs to know the behaviour of the hole
quasi-Fermi level ?p (see Fig. D . l ) .
Within the depletion region the holes move by the processes of drift
and diffusion, but the actual process by which they enter the semi­
conductor from the metal is rather more subtle. We can think of the
empty states below the Fermi level in the metal as holes, and those
moving to the right are able to enter the semiconductor if the excess
energy (relative to a hole at the Fermi level) associated with their motion
normal to the interface exceeds <j>\,p{= — ^b)- The process is analogous
to the emission of electrons from the metal into the semiconductor when
a reverse bias is apphed, with the usual proviso that energy is measured
downwards for holes, and we can describe it in terms of the thermionic-
emission theory (Green and Shewchun, 1973). If there is a difference A^p
between Ef and the hole quasi-Fermi level in the semiconductor
immediately adjacent to the metal, the argument of Section 3.2.3 can be
adapted to obtain
A =ArT^ exp(- q<l>^p/kT)[l - exp(- qAykT)].
For the case of the A u - S i diode considered in Section 3.5.1, the hole
injection-ratio for an electron current density of about 10^ A m"^ is
about 10"'' (Yu and Snow, 1969), so that « 10"' A m"^ for a forward

Fig. D . l Schottlcy barrier under forward bias, showing the hole quasi-Fermi level
(schematic).

224
DiaS O l U . J U V. IS a u u u L ±\j j AI« ^» , — ^ ^^p .
(corresponding to = 0-8 eV) we find A^p « 10~''kT/q, so that the hole
quasi-Fermi level is effectively constant across the surface. This result is
independent of (or ^bp)> for a given total current, since the hole
current is proportional to the injection ratio, which is proportional to
exp(q^^,/kT) or exp(- q<^y,p/kT).
To find the rise in through the depletion region, we may proceed by
analogy with eqns (3.5) and (3.7) and write

exp(q<^^,/kT)exp(a'w^) exp(-/)d3;
kTfiMa
l-exp(-?^p(H')/fcr)
where fip is the hole mobihty, AT^ is the effective density of states in the
valence band, and the error integral appears instead of F{aw). Substi­
tuting the above value of and taking the appropriate values of ^p and
for silicon gives ^p(w) » 0.07kT/q "0.002 eV at room temperature,
where is defined as the zero of energy. Hence the hole quasi-Fermi
level is substantially flat through the depletion region, as in a p - n
junction. This result holds even more rigorously for gallium arsenide

225
Appendix E: Contacts to amorphous
semiconductors

E . l Introduction

The recent intense activity in the field of amorphous semiconductors has

brought with it an interest in contacts to îtmorphous materials. The most
characteristic property of amorphous semiconductors, as far as their
electronic properties are concerned, is the appearance of localized states
within the forbidden gap. These localized states are caused by the loss of
crystalline order, and can be pictured as unsatisfied (or 'danghng') bonds
which result from the imperfect coordination pattern around individual
atoms. There are consequently no well-defined band edges; instead, the
valence and conduction band states extend into thé gap, where they
become localized because of the lack of periodicity in the potential. The
energies at which the states become localized are known as 'mobility
edges', and the separation between the two mobility edges plays a role
equivalent to that of the forbidden gap in a crystalhne semiconductor.
The locahzed states between the mobility edges are characterized by a
neutral level, rather in the same way as surface states, such that if the
states are filled up to this level and empty above it the solid as a whole
will be electrically neutral. This level therefore determines the Fermi
level in a homogeneous bulk semiconductor.
The semiconductor which has been most studied in its amorphous
state is silicon (a-Si). If films of a-Si are produced in the normal way by
sputtering or evaporation, the density of locahzed states is so large that it
is impossible to change the position of the Fermi level to any significant
extent by doping with the usual substitutional impurities such as
phosphorus; moreover, the photoconductive properties of such material
are severely degraded by the high density of gap states, which act as
electron and hole traps. This problem has been largely overcome through
the production of a-Si by reactive sputtering in a hydrogen atmosphere
or by the decomposition of silane in a r.f. plasma. These methods result
in amorphous sihcon which contains a substantial amount of hydrogen,
denoted as a-Si:H. Many of the hydrogen atoms form bonds with sihcon
atoms in such a way that the number of unsatisfied bonds is substantially
reduced. In this way the density of localized states within the gap is
reduced to such an extent that the Fermi level can be changed by doping,
so as to create n-type or p-type conductivity, and the density of electron

226
traps is also reduced. The separation between the mobility edges in
a-Si:H, which corresponds roughly to the band-gap of crystalline siUcon,
is about 1.8 eV. A typical density-of-states distribution for a-Si is shown
in Fig. E . l .

e,-e (eV)
Fig. E . l Density-of-state distribution in amorphous silicon as a function of energy below
the conduction band mobility edge E^. The full line represents the experimentally deter­
mined total density of states. Curves A and B are the assumed division into acceptor-like
and donor-like states, respectively. £,0 represents the position of the Fermi level in undoped
material. (From Spear, Le Comber, and Snell, 1978. Copyright Taylor and Francis.)

E.2 Schottky barriers in a-Si:H

When a metal makes contact with an amorphous semiconductor, a

Schottky barrier is established as in the crystEiUine case. However, there
is an important difference between the two cases. Whereas in the crystal­
line case the charge density in the depletion region is generally deter­
mined by shallow donors or acceptors, so that the barrier is parabohc in
shape if the doping concentration is uniform, in amorphous semi­
conductors the charge density is also determined by the charges associ­
ated with the localized states in the gap. Since the electron population of
these states is determined by the position of the Fermi level in the gap,
the charge density due to the localized states is far from uniform and the
barrier shape is far from parabohc. For example, if the semiconductor is
undoped and the density of locahzed states is assumed to be constant
within a few tenths of an electron-volt of the neutral level (which is not a
bad approximation for a-Si:H), it is easy to show that the space-charge
density and the barrier potential both increase exponentially with
distance from the edge of the depletion region. The shapes of the

227
potential barrier corresponding to experimentally observed density-of-
states distributions for the locahzed states in both doped and undoped
samples of a-Si:H have been thoroughly analysed using numerical
computation by Spear, Le Comber, and SneU (1978) and by SneU et al,
(1979). Shapes of the barrier profiles for both undoped and doped speci­
mens, based on the locahzed state distribution shown in Fig. E . l , are
shown in Fig. E.2. Reviews of Schottky barriers in amorphous silicon
have been given by Nemanich (1984) and by Nemanich and Thompson
(1984).

Fig. E.2 Barrier profile calculated for n-type Schottky barriers in amorphous silicon for
various doping levels. E^,{x) denotes the amount of band-bending at a distance x from the
edge of the depletion region, - e, denotes the position of the Fermi level below the
mobility edge in the bulk. (From Spear, Le Comber, and SneU, 1978. Copyright Taylor and
Francis.)

E.3 Capacitance measurements

The capacitance properties of Schottky barriers in amorphous semi­

conductors show two important differences compared to crystalline
semiconductors, both of which stem from the large concentration of
localized states within the forbidden gap. Firstly, because the barrier is
generally not parabolic in shape, is not usually a hnear function of
the reverse voltage V,. In fact, C does not necessarily decrease with
increasing V^; for the simple case mentioned in the previous section m
which the density of localized states is assumed constant near to the
neutral level, the total charge in the depletion region is proportional to
the diffusion voltage, and the differential capacitance is constant and
independent of V,. If the charge density depends only on the potential
(i.e. the charge centres are spatially uniform), it is a straightforward
228
matter to show that dCMV^ is proportional to [p'(x) - (p(x)y/Q],
where p(x) is the charge density at position x, Q is the total charge in the
depletion region, and the expression is evaluated immediately adjacent to
the metal. Thus if p'(x) is very large next to the metal, as would be the
case if it is determined by the rapidly varying density of states near the
mobility edges, dC/dV^ may be positive and C can actually increase with
increasing F„ as was observed for a gold contact to undoped a-Si by
Spear, L e Comber, and Snell (1978). In doped amorphous semi­
conductors, however, the space charge may be determined by the dopant
atoms, and if these are uniformly distributed a linear dependence of
on V, will result as in the crystalhne case.
Secondly, because the charge density is largely associated with states
near the middle of the gap which can have quite long time constants, the
capacitance can be very frequency dependent. The problem is essentially
the same as that of the contribution of deep levels to the capacitance of a
crystalline barrier, as discussed in Section 4.5.2. The situation is very
complicated to analyse, and frequencies below 10"^ H z are sometimes
necessary if frequency effects are to be avoided. A s in the case of
admittance spectroscopy (p. 172), much information about the locahzed
states can be derived from the study of capacitance and conductance as a
function of frequency and temperature (Snell et al, 1979). The apphca-
tion of capacitance measurements (both transient and steady-state) to the
measurement of localized state densities in a-Si:H has been reviewed by
Cohen (1984).

E.4 Current/voltage characteristics

Schottky barriers in amorphous semiconductors display all the current-

transport mechanisms that operate in crystalhne semiconductors, though
the relative importance of the different mechanisms is considerably
modified by the bulk properties of the amorphous material. Good recti­
fying characteristics are obtainable with a-Si:H, provided the substrate
temperature during growth is kept between about 200°C and 300°C
(Deneuville and Brodsky, 1979). The parameters /Q and n are found to
depend quite critically on the growth conditions (silane pressure and
flow rate).
It would be natural to assume that in undoped amorphous semi­
conductors the dominant conduction mechanism should be described by
the diffusion theory, because of the very low electron mobility (// < 10~^
m^ V"^ s~' for a-Si:H), and the early experiments of Wronski, Carlson,
and Daniel (1976) were interpreted in this way. Later workers have
claimed that their results for Pd and Pt diodes on a-Si:H can be recon­
ciled with the thermionic-emission theory (Thompson et al, 1981);

229
however, their conclusion is based on the temperature dependence of JQ,
and is not decisive. According to the thermionic-emission theory,
JQ = A**T^ ехр(-дфь/кТ), so that a plot of ЩІ^/Т^) against 1/T
should give a straight hne. The prediction of the diffusion theory is that
/Q = qN^juS'^^,, exp(- qф^,/kT), and the temperature dependence of the
pre-exponential term is uncertain. For a parabohc dependence of
density-of-states on energy, as in the conduction band of a crystalhne
semiconductor, is proportional to Г^^^. If one assumes naively that /u
is determined by acoustic-mode lattice scattering (which is a poor
assumption even for crystalhne silicon), ju should be proportional to
T~^''^, so that the product should be independent of temperature,
and the conclusion of Thompson et al. seems to depend on the fact that a
plot of In Jo against \/T for their data gave a straight line. But one
cannot assume that is proportional to T^''^ for an amorphous semi­
conductor, and the temperature dependence of depends very much on
whether the important conduction process involves states above the
mobility edge or proceeds via a hopping mechanism between locahzed
states below the edge. In any case, the temperature dependence is
dominated by the exponential term, and, as Deneuville and Brodsky have
shown, an equally good straight line can be obtained whether one plots
^(/o/T'^) or In JQ against \/T. One would expect the Bethe criterion
(Section 3.2.5) in its simplest form fiS"^^ > v/4 not to be satisfied with
the very low values of that are found in amorphous semiconductors.
Values of deduced from the slope of a plot of In JQ [or \п{1^/Т^)\
against 1/T are somewhat larger than those which refer to the same
metal on crystalhne sihcon, for example 1.1 eV for platinum on a-Si:H,
compared with 0.81 eV for etched crystalline Si. This is not surprising in
view of the larger effective band-gap in the amorphous case. Barrier
heights determined in this way agree well with values obtained from
photoelectric measurements, and also with those obtained from the C/V
method (Wronski and Carlson, 1977).
Tunnelhng has been observed in doped a-Si:H with donor concentra­
tions above 10^^ m~^ (Snell et al, 1979), and this shows up most clearly
in the reverse characteristics. The current/voltage characteristics can be
explained reasonably well in terms of the thermionic-field emission
theory of Padovani and Stratton (Jackson et al, 1986) within the limits
set by the uncertainties in the doping density and in the appropriate
value of m* and by the assumption of a parabolic barrier. It is also
possible that locahzed states near the conduction-band mobihty edge
may significantly enhance the tunnelling current. It may seem at first
sight that tunnelling can only affect the current flow if the current is
determined by thermionic emission, since its effect is to increase the
probabihty of an electron passing from the semiconductor into the metal.

230
However, a moment's consideration shows that this is not so. TunneUing
effectively reduces the barrier height, and so reduces the height and
width of the depletion region through which the electrons are transported
by drift and diffusion; it should therefore enhance the current flow even
if this is best described by the diffusion theory.

231
232
References

Abbati, I. and Grioni, M . (1981). /. Vac. Sei. Technol. 19,631.

Allen, F. G. and Gobeli, G. W. (1962). Phys Rev. Ill, 150.
Allen, R. E. and Dow, J. D. (1981). /. Vac. Sei. Technol. 19, 383.
Allen, R. E. and Dow, J. D. (1982). Phys. Rev. B25,1423.
Allen, R. E., Sankey, O. F. and Dow, J. D. (1986). Surf. Set. 168, 376.
Ambridge, T. and Faktor, M . M . (1974). Gallium Arsenide and Related
Compounds, Inst. Phys. Conf. Series 24,320.
Ambridge, T. and Faktor, M . M . (1975). J. Appl. Electrochem. 5,319.
Amron, I. (1967). Eleetrochem. Technol. 5, 94.
Anderson, C. L., Crowell, C. R. and Kao, T. W. (1975). Solid-St. Electron. 18,
705.
Andersson, L. P., Hyder, A. and Berg, S. (1973). Nucl Instrum. Meth. 114,1.
Andrews, J. M . (1974). /. Vac. Sei. Technol. 11, 972.
Andrews, J. M . and Koch, F. B. (1971). Solid-St. Electron. 14,901.
Andrews, J. M . and Lepselter, M . P. (1969). Ohmie contacts to semiconductors,
p. 159. Electrochemical Society, New York.
Andrews, J. M . and Lepselter, M . P. (1970). Solid-St. Electron. 13,1011.
Andrews, J. M . and Phillips, J. C. (1975). Phys. Rev. Lett. 35, 56.
Andrews, J. M . and Alalia, M . M . (1963). Ann. N. Y. Acad. Set, 101,697.
Andrews, J. M . and Yep, T. O. (1970). /. Appl. Phys. 41, 303.
Arizumi, T. and Hirose, M . (1969). Jap. J. Appl. Phys. 8, 749.
Arizumi, T., Hirose, M . and Altaf, N. (1968). Jap. J. Appl. Phys. 7,870.
Arizumi, T., Hirose, M . and Altaf, N. (1969). Jap. J. Appl. Phys. 8,1310.
Assimos, J. A. and Trivich, D. (1973). J. Appl. Phys., 44,1687.
Alalia, M . M . (1966). Proc. Munich Symp. Microelectron., p. 123. Oldenberg,
Munich.
Baccarani, G. (1976)./. ^/7/?/. PÄy5. 47,4122.
Baccarani, G. and Mazzone, A. M . (1976). Electron. Lett 12,59.
Bachrach, R. Z. and Bianconi, A. (1978). J. Vac. Sei. Technol 15, 525.
Barnes, P. A. and Cho, A. Y. (1978). Appl. Phys. Lett. 33,651.
Bardeen, J. (1947). Phys. Rev. 71,717.
Bardeen, J. and Brattain, W. H. (1949). Phys. Rev. 75,1208.
Barret, C , Chekir, F. and Vapaille, A. (1983). J. Phys. C: Solid-St. Phys. 16,2421.
Barret,C.andMassies, J. (1983). J. Vac. Sei. TechnoLBl, 819.
Barret, C. and Vapaille, A. (1979). /. Appl Phys. 50,4217.
Bartur, M . and Nicolet, M.-A. (1984). 7. Eleetrochem. Soe. 131,1118.
Basterfield, J., Shannon, J. M . and Gill, A. (1975). Solid-St. Electron. 18,290.
Baxandall, P. J., CoUiver, D. J. and Fray, A. F. (1971). 7. Phys. E: Sclent. Instrum.
4,213.
Berger, H. H. (1972). Solid-St. Electron. 15,145.
Berz, F. (1985). Solid-St. Electron. 28,1007.

233
Best, J. S. (1979). Appl. Phys. Lett. 34, 522.
Bethe, H. A. (1942). MIT Radiation Lab. Rep. 43-12.
Blakemore, J. S. (1962). Semiconductor statistics. Pergamon Press, Oxford.
Blood, P. (1986). Semicond. Set. Technol. 1,7.
Blood, P. and Orten, J. W. (1978). Rep. Prog. Phys. 41,157.
Bourgoin, J. and Lannoo, M . (1983). Point defects in semiconductors. 11.
Experimental aspects. Springer-Verlag, Berlin.
Braicovich, L. (1983). Sutf. Sei. 132,315.
Braslau, N . (1981). J. Vac. Set. Technol. 19, 803.
Braslau, N. (1983). Thin Solid Films 104, 391.
Braslau, N., Gunn, J. B. and Staples, J. L. (1967). Solid-St. Electron. 10,381.
Braun, F. (1874). Pogg. Ann. 153, 556.
Braun, I. and Henisch, H. K. (1966). Solid-St. Electron. 9, 981.
BriUson, L. J. (1982a). Thin Solid Films 89,461.
Brillson,L. J. (1982b). Surf. Sei. Rep. 2,123.
Brillson, L. J., Brucker, C. F., Katnani, A . D., Stoffel, N . G., Daniels, R. and
Margaritondo, G. (1982). J. Vac. Sei. Technol 21, 564.
Brillson, L. J., Slade, M . L., Vittura, R. E., Kelly, M . K., Tache, W., Margaritondo,
G., Woodall, J. M., Kirchner, P. D., Pettit, G. D. and Wright, S. L. (1986). /. Vac.
Set. TechnolB4, 919.
Broom, R. F. (1971). Solid-St. Electron, 14,1087.
Brors, D. L., Monnig, K. A., Fair, J. A., Coney, W. and Saraswat, K. C. (1984).
Solid-St. Technol 27(4), 313.
Brucker, C. F. and Brillson, L. J. (1981). Appl Phys. Lett. 39,67.
Buck, C. M . (1983). Ph.D. Thesis, University of Manchester.
Calandra, C , Bisi, O. and Ottaviani, G. (1984). Surf ScL Rep. 4,271.
Card, H . C. (1974). Proc. Manchester Conf Metal-Semiconductor Contacts,
p. 129. Conference Series No. 22. Institute of Physics, London.
Card, H. C. (1975a). Solid State Commun. 16,87.
Card, H. C. (1975b). Solid-St. Electron 18,881.
Card, H. C. and Rhoderick, E. H. (1971a). /. Phys. D.Appl Phys. 4,1589.
Card, H. C. and BUioderick, E. H. (1971b). J. Phys. D:Appl Phys. 4,1602.
Card, H. C. and Rhoderick, E. H. (1973). Solid-St. Electron. 16, 365.
Card, H. C. and Singer, K. E. (1975). Thin Solid Films 28,265.
Chadi, D. J. (1979). Phys. Rev. B19,2074.
Chang, C. Y., Fang, Y. K. and Sze, S. M . (1971). Solid-St. Electron. 14, 541.
Chang, C. Y. and Sze, S. M . (1970). Solid-St. Electron. 13,727.
Chang, Y.-J. and Erskine, J. L. (1983). Phys. Rev. B28,5766.
Chelikowsky, J.R. and Cohen, M . L. (1974). Phys. Rev. BIO, 5095.
Chelikowsky, J. R. and Cohen, M . L. (1979). Phys. Rev. B20,4150.
Chen, J. Y. and Roth, L. B. (1984). Solid-St. Technol 27(8), 145.
Cheung, N. W., Culbertson, R. J., Feldman, L. C , Silverman, P. J., West, K. W. and
Mayer, J. W. (1980). Phys. Rev Lett. 52, 461.
Chin, R., Milano, R. A . and Law, H. D. (1980). Electron. Lett. 16, 626.
Chino.K. (1973). Solid-St. Electron. 16,119.
Cho, A . Y. and Dernier, R (1978). J. Appl Phys. 49, 3328.
Christou, A. and Day, H. M . (1976). J. Appl Phys. 47,4217.
Clabes, J. G., Rubloff, G. W. and Tan, T. Y. (1984). Phys. Rev. B29,1540.

234
Clarke, R. A., Green, M . A. and Shewchun, J. (1974). /. Äppl. Phys. 45,1442.
Сое, D. J. (1971). Paper 10.3 ESSDERC Conference, Munich.
Сое, D. J. (1974). Ph.D. Thesis, Manchester University.
Сое, D. J. and Rhoderick, E. H. (1976). /. Phys, D.Appl. Phys. 9,965.
Сое, D. J., Rhoderick, E. H., Gerzon, P. H . and Tinsley, A . W. (1974). Proc.
Manchester Conf. Metal-Semiconductor Contacts, p. 74. Institute of Physics,
London.
Cohen, J. D. (1984). In Semiconductors and semimetals (ed. R. K. Willardson and
A. C. Beer) Vol. 21, Part C, Chap. 2. Academic Press, London.
Cohen, M . L. (1980). Advances in Electronics and Electron Physics, Vol. 51, p. 1.
Academic Press, New York.
Cohen,M. L. (1982). Phys. ScriptaTX, 5.
Cohen, S. S. (1983). Thin Solid Films 104, 361.
Cohen, S. S., Piacente, P. A., Gildenblat, G. and Brown, D. M . (1982). J. Appl
Phys. 53, 8856.
Cohen, S. S., Kim, M . J., Gorowitz, В., Saia, R. and McNelly, F. F. (1984). Appl
Phys. Lett. 45, 414.
Copeland, J. A. (1969). IEEE Tram. Electron. Dmcei ED-16,445.
Cowley, A. M . (1966). J. Appl Phys. 37, 3024.
Cowley, A. M . (1970). Solid-St. Electron. 13,403.
Cowley, A. M . and Sze, S. M . (1965). J. Appl Phys. 36, 3212.
Cowley, A. M . and Zetder, R. A . (1968). IEEE Trans. Electron. Devices'EJ)-\S,
761.
Cox, R. H. and Strack, H. (1967). Solid-St. Electron. 10,1213.
Crowell, C. R. (1965). Solid-St. Electron. 8,395.
Crowell, C. R. (1977). Solid-St. Electron. 20,171.
Crowell, C. R. and Beguwala, M . (1971). Solid-St Electron. 14,1149.
Crowell, C. R. and Rideout, V. L. (1969). Solid-St. Electron. 12, 89.
Crowell, C. R. and Roberts, G. I. (1969). /. Appl Phys. 40, 3726.
Crowell, C. R., Sarace, J. C , and Sze, S. M . (1965). Trans. Metall Soc. AIME 233,
478.
Crowell, C. R., Shore, H. B. and La Bate, E. E. (1965). J. Appl. Phys. 36, 3843.
Crowell, C. R. and Sze, S. M . (1966). Solid-St Electron. 9,1035.
Crowell, C. R., Sze, S. M . and Spitzer, W. G. (1964). AppL Phys. Lett. 4,91.
Datta, A . K., Ghosh, K., Chowdhury, N . K. D. and Daw, A . N. (1980a). Solid-St.
Electron. 23, 905.
Datta, A. K., Ghosh, K., Mitra, R. N. and Daw, A. N. (1980b). Solid-St. Electron.
23,99.
Davydov, B. (1939). /. Phys. USSR, 1,167.
Davydov, B. (1941). J. Phys. USSR, 4, 335.
Daw, M . S. and Smith, D. L. (1980). J. Vac. Sei. Teehnol 17,1028.
Daw, M . S. and Smith, D. L. (1981). Solid-St. Commun. 37, 205.
Deneuville, A. and Brodsky, M . H. (1979). /. Appl Phys. 50,1414.
de Sousa Pires, J. (1978). Phys. Scripta 18,372.
de Sousa Pires, J., Donoval, D. and Tove, P. A . (1982). Solid-St Electron. 25, 989.
de Sousa Pires, J., Tove, P. A., and Donoval, D. (1984). Solid-St. Electron. 27,
1029.
Devlin, W. J. (1980). Electron. Lett. 16, 92.

235
Dewald, J. F. (1960). Bell. Syst. Technol. J. 39,615.
Dharmadasa, I. M., Herrenden-Harker, W. G. and Williams, R. H. (1986). Appl.
Phys. Lett. 48,1082.
Dharmadasa, I. M., McLean, A. В., Patterson, M . H., and WilHams, R. H. (1987).
J. Semicond. Sci. Technol. 2,404.
Dorbeck, F. H. (1966). Solid-St. Electron. 9,1135.
Duke, C. B. and Mailhiot, C. (1985). J. Vac. Sci. Technol B3,1170.
Eaglesham, D. J., Kiely, C. J., Cherns, D., and Missous, M . (1987). UnpubUshed
work at Bristol University.
Eaves, L., Grovenor, C. R. M., Schwartz, G. P. and Wieder, H . H . (1986). In
Properties of Gallium Arsenide. EMIS Datareviews Series No. 2 INSPEC.
London, Chap. 17.
Eglash, S. J., Pan, S., Mo, D., Spicer, W. E. and Collins, D. M . (1983). Jap. J. Appl
Phys. Suppl 22(1), 431.
Elmers, G. W. and Stevens, E. H. (1971). IEEE Trans. Electron. Devices ED-18,
1185.
Eizenberg, M., Thompson, R. D. and Tu, К. N. (1982). J. Appl Phys. 53,6891.
Elfsten, B. and Tove, P. A. (1985). Solid-St. Electron. 28,721.
Engl, W. L., Manck, O. and Wieder, A. W. (1976). Proc. Fifth European Solid¬
State Devices Research Conference, p. 1. Grenoble, 1975, Société Française de
Physique, Paris.
England, A. A . and Rhoderick, E. H. (1981). Solid-St. Electron. 24, 337.
Escher, J. S., James, L. W., Sankaran, R., Antypas, G. A., Moon, R. L. and Bell, R.
L. (1976). J. Vac. Sci Technol 13, 874.
Farrow, R. F. C , CuUis, A . G., Grant, A. J. and Pattison, J. E. (1978). /. Cryst.
Growth 45,292.
Finetti, M., Sumi, I., Bartur, M., Banwell, T. and Nicolet, M.-A. (1984). Solid-St.
Electron. 27,611.
Flores, F. and Tejedor, C. (1987). /. Phys. C: Solid-St. Phys. 20,145.
Fonash, S. J. (1983). Appl Phys. 54,1966.
Fonash, S. J., Ashok, S. and Singh, R. (1981). Appl Phys. Lett. 39,423.
Fowler, R. H . (1931). Phys. Rev. 38,45.
Frankl, D. R. (1967). Electrical properties of semiconductor surfaces, p. 33.
Pergamon Press, Oxford.
Fraser, D. B. (1983). In VLSI Technology (ed. S. M . Sze). Chap. 9. McGraw-Hill,
Singapore.
Freeman, L. B. and Dahlke, W. E. (1970). Solid-St. Electron. 13,1483.
Freeouf, J. L. (1980). Solid-St. Commun. 33,1059.
Freeouf, J. L. (1981). /. Vac. ScL Technol 18,910.
Furukawa, Y. and Ishibashi, Y. (1967). Jap. J. Appl Phys. 6,787.
Gerzon, R H., Barnes; J. W., Waite, D. W. and Northrop, D. C. (1975). Solid-St.
Electron. 18,343.
Ghosh, C , Yenigalla, P. and Atkins, K. (1983). Electron. Device Lett. EDL-4, 301.
Glover, G. H. (1973). Solid-St. Electron. 16, 973.
Gol'dberg, Y. A., Nasledov, D. N. and Tsarenkov, B. V. (1971). Instrum. Exp. Tech.
3,899.
Gol'dberg, Y. A., Posse, E. A . and Tsarenkov, B. V. (1971). Electron. Lett. 7,601.
Gol'dberg, Y. A., Posse, E. A . and Tsarenkov, B. V. (1975). Soviet Phys. Semicond.
9,337.
236
Gol'dberg, Y. A., Rafiev, T. Y., Tsarenkov, B. V. and Yakoulev, Y. R (1972). Sov.
Phys. Semicond. 6,398.
Goodman, A. M . (1963). /. Appl. Phys. 34, 329.
Goodman, A. M . (1964). /. Appl Phys. 35, 573.
Goodman, A. M . and Perkins, D. M . (1964). /. Appl Phys. 35, 3351.
Gordy, W. and Thomas, W. J. O. (1956). /. Chem. Phys. 24,439.
Gossick, B. R. (1963). Solid-St. Electron. 6,445
Gray, K. E. (1973). Solid-St. Commun. 13,1787.
Green, M . A . (1976). Solid-St. Electron. 19,421.
Green, M . A. and Shewchun, J. (1973). Solid-St. Electron. 16,1141.
Grimmeiss, H. G. (1974). Proc. Manchester Conf. Metal-Semiconductor Contacts,
p. 187. Conference Series No. 22. Institute of Physics, London.
Grimmeiss, H . G. and Ovren, C. (1981). /. Phys. E. 14,1032.
Grondahl, L. O. (1926). Phys. Rev. 27,813.
Grondahl, L. O. (1933). Rev. Mod. Phys. 5,141.
Grunthaner, P. J., Grunthaner, F. J. and Mayer, J. W. (1980). J. Vac. Sei. Teehnol
17,924.
Gutknecht, P. and Strutt, M . J. O. (1971). Electron Lett. 7, 298.
Gutknecht, P. and Strutt, M . J. O. (1972). Appl Phys. Lett 21,405.
Gutknecht, P. and Strutt, M . J. O. (1974). IEEE Trans. Electron Devices ED-21,
172.
Gyulai, J., Mayer, J. W., Rodriguez, V., Yu, A . Y. C. and Gopen, H . J. (1971). /.
Appl Phys. 42, 3578.
Haeri, S. Y. and Rhoderick, E . H . (1974). Proc. Manchester Conf Metal-
Semiconductor Contacts, p. 84. Institute of Physics, London.
Hall, R. N. (1952). Phys. Rev. 87, 387.
Hamilton, B. (1974). Proc. Manchester Conf Metal-Semiconductor Contacts, p.
218. Conference Series No 22. Institute of Physics, London.
Hammer, R. (1970). Rev. Seient. Instrum. 40,292.
Hansson, G. v., Uhrberg, R. I. G. and NichoUs, J. M . (1983). Surf. Sei. 132, 31.
Hargrove, M . J. and Anderson, R. L. (1986). Solid-St. Electron. 29, 365.
Hauenstein, R. J., Schlesinger, T. E., McGill, T. C , Hunt, B. D. and Schowalter, L.
J. (1985). Appl Phys. Lett 47, 853.
Heiblum, M., Nathan, M . I. and Chang, C. A . (1982). Solid-St. Electron. 25,185.
Heine, V. (1965). Phys. Rev. A138,1689.
Heine, V. (1972). Proc. R. Soc. A331, 307.
Henisch, H . K . (1957). Rectifying semiconductor contacts. Clarendon Press,
Oxford.
Henisch, H. K. (1984). Semiconductor contacts. Clarendon Press, Oxford.
Henry, C. H., Kukimoto, H., Miller, G. L. and Merritt, F. R. (1973). Phys. Rev. B7,
2499.
Hillegas, W. J. and Schnäble, G. L. (1963). Electroehem. Teehnol 1, 228.
Hiraki, A., Lugujjo, E., Nicolet, M.-A. and Mayer, J. W. (1971). Phys. Status Solidi,
XI, 401.
Hiraki, A., Shuto, K., Kim, S., Kamamura, W. and Iwami, M . (1977). Appl Phys.
Lett. 31, 611.
Hiraki, A., Kim, S., Kamamura, W. and Iwami, M . (1979). Appl Phys. Lett. 34,
194.
Hirose, M., Altaf, N. and Arizunü, T. (1970). Jap. J. Appl Phys. 9,260.
Hökelek, E. and Robinson, G. Y. (1983). J. Appl. Phys. 54, 5119.
Holland, L. (1956). Vacuum deposition of thinfilms.Chapman and Hall, London.
Hollinger, G. and Himpsel, F. J. (1983). Phys. Rev. B28, 3651.
Hooper, R. C , Cunningham, J. A . and Harper, J. G. (1965). Solid-St. Electron. 8,
831.
Houzay, F. Moison, J. M . and Bensoussan, M . (1985). /. Vac. Sei. Technol. B3,
756.
Houzay, F., Bensoussan, M . and Barthe, F. (1986). Surf. Sei. 168, 347.
Hughes, G. J., McKinley, A. and WiUiams, R. H. (1983). J. Phys. C: Solid-St. Phys.
16,239L
Hughes, G. J., McKinley, A., Williams, R. H., and McGovern, I. T. (1982). /. Phys.
C:Solid-St Phys. 15,L159.
Huijser, A., van Laar, J. and van Rooy, T. L. (1977). Surf. Sei. 62,472.
Humphreys, T. R, Hughes, G. I., McKmley, A., Cunningham, E. C. and WiUiams,
R. H. (1985). Surf. Sei. 152,1222.
Ihm, J. and Joannopoulos, D. J. (1981). Phys. Rev. Lett. 47,679.
Ihm, J. and Joannopoulos, D. J. (1982). /. Vac. Sei. Technol. 21,340.
Ihm, J., Louie, S. G. and Cohen, M . L. (1978). Phys. Rev. B18,4172.
Iliadis, A. and Singer, K. E. (1983). Solid-St. Electron. 26,7.
lUadis, A. and Smger, K. E. (1984). Solid-St Commun. 49,99.
Inada, Т., Kalo, S., Нага, Т. and Toyada, N. (1979). J. Appl. Phys. 50,4466.
Inglesfield, J. (1984). In Electronic properties of suфces (ed. M . Prutton). p. 71.
Adam Hilger, Bristol.
Ismail, A., Ben Brahim, A., Palaci, J. M . and Lassabatere, L. (1986). Vacuum 36,
217.
Ismail, A., Palau, J. M . and Lassabatere, L. (1984). Rev Phys. Appl. 19,205.
Jackson, W. В., Nemanich, R. J., Thompson, M . J. and Wacker, В. (1986). Phys.
Rev B33,6936.
Jäger, H. and Kosak, W. (1969). Solid-St Electron. 12,511.
Jäger, H. and Kosak, W. (1973). Solid-St. Electron. 16,357.
James, H. M . (1949). Phys. Rev 76,1602.
Johnson, N . M . (1986). Mat. Res. Soc. Symp. 69, 75.
Johnson, N. M., Magee, T. J. and Peng, J. (1976). /. Vac. Sei. Technol. 13,838.
Johnson, W. C. and Panousis, R T. (1971). IEEE Trans. Electron. Devices EB-IH,
965.
Joyce, B. A. and Neave, J. H. (1971). Surf. Sei. 27,499.
Kahng, D. (1963). Solid-St. Electron. 10,45.
Kahng,D. (1964). Bell Syst Technol. J. 42,215.
Kajiyama, K., Mizushhna, Y. and Sakata, S. (1973). Appl. Phys. Lett 23,458.
Kajiyama, K., Sakata, S. and Ochi, O. (1975). /. Appl. Phys. 46, 3221.
Kamimura, K., Suzuki, T. and Kunioka, A. (1980). J. Appl. Phys. 51,4905.
Kano, G., hioue, M., Matsuno, J. and Takayanagi, S. (1966). /. Appl. Phys. 37,
2985.
Kar, S. (1975). Solid-St. Electron. 18,169.
Keeler, W., Roth, W. J. and Fortin, E. (1980). Can. J. Phys. 58,63.
Kelly, W. M . and Wrixon, G. T. (1978). Electron. Lett. 14,80.
Kendelewicz, Т., Newman, N., List, R. S., Lindau, I. and Spicer, W. E. (1985). J.
Vac. Sei. Technol.B3,1206.

238
Kennedy, D. P., Murley, P. C. and Kleinfelder, W. (1968). IBM J. Res. Dev. 12,
399.
Kleefstra, M . and Herman, G. C. (1980). /. Appl. Phys. 51,4923.
Kim, H. В., Lovas, A . F., Sweeney, G. G. and Heng, Т. М. S. (1976). Proc. 6th
Symp. Gallium Arsenide and Related Compounds. Institute of Physics Confer­
ence Series 3 3b, p. 145.
Kim, H . В., Sweeney, G. G. and Heng, Т. М. S. (1974). Proc. 5th Int. Symp.
Gallium Arsenide, p. 307. Institute of Physics, London.
Kimerling, L. C. (1974). /. AppL Phys. 45,1839.
Koos, V. and Neumann, H. G. (1975). Phys. Status Solidi, A29, K115.
Korwin-Pawlowski, M . L. and Heasell, E. L. (1975). Solid-St. Electron. 18, 849.
Kowalczyk, S. Т., Waldrop, J. R. and Grant, R. W. (1981). J. Vac. Sei. Teehnol 19,
611.
Kroemer, H. and Chien, W. Y. (1981). Solid-St. Electron. 24,655.
Kroemer, H., Chien, W. Y , Harris, J. S. and Edwall, D. D. (1980). Appl Phys. Lett.
36,295.
Kuan, T. S., Batson, P. E., Jackson, T. N., Rupprecht, H. and Wilkie, E. L. (1983).
J. Appl Phys. 54,6952.
Kuech, T. F. and McCaldin, J. O. (1980). J. Vac. Sei. Teehnol 17,891.
Kumar, R. C. (1970). Int J. Electron. 29, 365.
Kurdn, S., McGill, T. C. and Mead, C. A. (1969). Phys. Rev Lett. 22,1433.
Kusaka, M., Matsui, T. and Okazaki, S. (1974). Surf. Sei. 41,607.
Landkammer, F. J. (1967). Solid-St. Commun.5,247.
Lang, D. V. (1974). J. Appl Phys. 45,3023.
Lanyon, H. P. D. and Richardson, R. E. (1971). Phys. Status Solidi, A7,411.
Lassabatere, L., Ismail, A., Palau, J. M . and Ben Brahim, A. (1986). Surf. Sei. 168,
336.
Le Lay, G., Derrien, J. and Sebenne, C. A . (eds.) (1986). Proc. Int Conf Forma­
tion of Semiconductor Interfaces. Surf. Sei. 168.
Lepselter, M . P. and Sze, S. M . (1968). Bell Syst. Teehnol J. 47,195.
Levine, J. D. (1971). /. Appl. Phys. 42, 3991.
Liehr, M., Schmid, P. E., Le Goues, R K. and Ho, P. S. (1985). Phys. Rev Lett. 54,
2139.
Lindau, I., Chye, P. W., Garner, C. M., Pianetta, P., Su, C. Y. and Spicer, W. E.
(1978). /. Vac. Sei. Teehnol 15,1332.
Livingstone, A . W., Turvey, K. and Allen, J. W. (1973). Solid-St. Electron. 16, 351.
Lodge, O. (1890). J. Inst. Eleetr Engrs. 19, 346.
Loeb, L. B. (1961). The kinetic theory ofgases, p. 42. Dover, New York.
Losee, D. L. (1975). J. Appl Phys. 46, 2204.
Louie, S. G. and Cohen, M . L. (1975). Phys. Rev. Lett. 35,866.
Louie, S. G. and Cohen, M. L. (1976). Phys. Rev. В І З , 4172.
Louie, S. G., Chelikowsky, J. R. and Cohen, M . L. (1977). Phys. Rev. B15,2154.
Loveluck, J. M . and Rhoderick, E. H. (1967). Solid-St Electron. 10,433.
Ludeke, R. (1983). Surf. Sei. 132,143.
Ludeke, R. (1986). Surf. Sei. 168,290.
Ludeke, R. and Landgren, G. (1981). J. Vac. Sei. Teehnol 19,667.
Ludeke, R., Chiang, T. C. and Eastman, D. E. (1982). /. Vac. Sei. Teehnol 21,
599.

239
Ludeke, R., Chiang, T. C. and MUler, T. (1983). /. Vac. Sei. Technol. B l , 581.
Maani, C , WilUams, R. H., McKinley, A., Hughes, G. J. and Humphreys, T. P.
(1984). / . Vac. Sei. Technol. B2, 561.
McCaldin, J. O. (1974). /. Vac. Sei. Technol. 11,990.
McCaldin, J. O., McGUl, T. C. and Mead, C. A. (1976). J. Vac. Sei. Technol. 13,
802.
McGilp, J. (1984). /. Phys. C: Solid-St. Phys. 17,2249.
Mcllroy, P. W. A. and Pepper, M . (1985). J. Phys. C: Solid-St. Phys. 18, L87.
McKinley, A., Hughes, G. J. and WiUiams, R. H. (1982). /. Phys. C: Solid-St. Phys.
15,7049.
McKinley, A , Parke, A. W. and Williams, R. H. (1980). J. Phys. C: Solid-St. Phys.
13,6723.
McLean, A . B. (1986). Semicond. Sei. Technol. 1,177.
McLean, A . B., Dharmadasa, I. and Williams, R. H . (1986). Semicond. Set
Technol. 1,177.
Madams, C. J., Morgan, D. V. and Howes, M . J. (1975). Electron. Lett. 11, 574.
Magee, T. J. and Peng, J. (1975). Phys. Status Solidi A32,69,5.
Mailhiot, C , Duke, C. B. and Chadi, D. J. (1985). Surf. Sei. 149,366.
Majerfeld, A . and Wada, O. (1978). Electron. Lett. 14,125.
Many, A., Goldstern, Y. and Grover, N . B. (1965). Semiconductor surfaces. North-
Holland, Amsterdam.
Massies, J.,Dezaly,F. andLinh, N.T. (1980). J. Vac. Sei. Technol. 17,1134.
Massies, J., Chaplart, J., Laviron, M . and Linh, N . T. (1981). Appl. Phys. Lett. 38,
693.
Masszi, F., Stoh, L., Tove, P. A . and Tarnay, K. (1981). Phys. Scripta 24,456.
Mayer, J. W. and Poate, J. M . (1978). In Thinfilms—Interdiffusionand reactions
(ed. J. M . Poate, K. N . Tu and J. W. Mayer) Chap. 6. Wüey, New York.
Mayer, J. W. and Tu, K. N . (1974). J. Vac. Sei. Technol. 11, 86.
Mayer, J. W. and Turos, A . (1973). Thin Solid Films 19,1.
Mead, C. A. and Spitzer, W. G. (1964). Phys. Rev. 134, A713.
Meyer, D. E. (1969). Ohmie contacts to semiconductors, p. 227. Electrochemical
Society, New York.
Michaelson, H . B. (1977). /. Appl. Phys. 48,4729.
Miller, G. L., Lang, D. V. and Kimerlmg, L. C. (1977). Ann. Rev Mat Sei. 7, 377.
Miller, W. L. and Gordon, A. R. (1931). /. Phys. Chem. 35, 2785.
Milnes, A. G. and Feucht, D. L. (1972). Heterojunctions and metal-semiconductor
junctions. Academic Press, New York.
Missous, M . (1987). Unpublished work at UMIST.
Missous, M . and Rhoderick, E. H . (1985). Solid-St. Electron. 28,233.
Missous, M . and Rhoderick, E. H . (1986). Electron. Lett. 22,477.
Missous, M., Rhoderick, E. H. and Singer, K. E. (1986a). Electron. Lett. 22,241.
Missous, M., Rhoderick, E. H . and Singer, K. E. (1986b). /. Appl. Phys. 59, 3189.
Missous, M., Rhoderick, E. H . and Singer, K. E. (1986c). /. Appl. Phys. 60,2439.
Monch, W. (1986). J. Vac. Sei. Technol. B4,1085.
Montgomery, V. and Williams, R. H. (1982). /. Phys. C: Solid-St Phys. 15, 5887.
Montgomery, v., Williams, R. H . and Srivastava, G. P. (1981). /. Phys. C: Solid-St.
Phys. 14,L191.

240
Moreau, Y., Manifacier, J.-C. and Henisch, H. K. (1982). Solid-St. Electron. 25,
137.
Morgan, D.V. (1975). Contemp. Phys. 16, 221.
Morgan, D. v., Howes, M . J. and Devlin, W. J. (1978). /. Phys. D: Appl Phys. 11,
1341.
Mott, N. F. (1938). Proc. Cambr. Phil Soc. 34,568.
Mott, N . F. and Jones, H. (1936). Theory of the properties of metals and alloys.
Oxford University Press, Oxford.
Mottram, J. D., Northrop, D. C , Reed, C. M . and Thanailakis, A. (1979). /. Phys.
D: Appl Phys. \1,ПЪ.
MulUns, F. H. and Brunnschweiler, A . (1976). Solid-St. Electron. 19,47.
Murakami, M., Childs, K. D., Baker, J. M . and Callegari, A . (1986). J. Vac. Sci.
Technol B4, 903.
Murarka, S. P. (1980). J. Vac. Sci. Technol 17,775.
Murarka, S. P. (1983). Silicides for VLSI applications. Academic Press, New York.
Muret, P. (1982). J. Appl Phys. 53,6300.
Muret, P. and Deneuville, A. (1982). /. Appl Phys. 53,6289.
Nathan, M . I. and Heiblum, M . (1982). Solid-St. Electron. 25,1063.
Nemanich, R. J. (1984). In Semiconductors and semimetals (ed. R. K. Willardson
and A . C. Beer) Vol. 21, Part C, Chap. 11. Academic Press, London.
Nemanich, R. J. and Thompson, M . J. (1984). In Metal-semiconductor Schottky
barrier junctions and their applications (ed. B. L. Sharma) Chap. 9. Plenum
Press, New York.
Neshpor, V. S. and Samsonov, G. V. (1960). Sov. Phys. Solid St. 2,1966.
Newman, N., Kendelewicz, Т., Bowman, L. and Spicer, W. E. (1985a). Appl Phys.
Lett. 46,1176.
Newman, N., van Schilfgaarde, M., Kendelewicz, Т., WUliams, M . D. and Spicer,
W.E. (1985b). Phys. Rev. B33,1146.
Newman, N., Spicer, W. E., Kendelewicz, T. and Lindau, I. (1986). J. Vac. Sci.
Technol B4,931.
Newman, N., van Schilfgaarde, M . and Spicer, W. E. (1987). Phys. Rev. В 35,
6298.
Nguyen, P. H., Lepley, В., Nadeau, A . and Ravelet, S. (1975). Proc. lEE 122,
1193.
Nicholls, J. M., Martensson, P. and Hansson, G. V. (1986). Phys. Rev. Lett. 54,
2363.
Nicolet, M.-A. and Lau, S. S. (1983). In VLSI electronics (microstructure science)
(ed. N. G. Einspruch and G. B. Larrabee) Vol. 6, Chap. 6. Academic Press, New
York.
Nicollian, E. H. and Goetzberger, A. (1967). BellSyst. Technol J. 46,1055.
NiU, K. W., Walpole, J. N., Calawa, A . R. and Harman, T. C. (1970). Proc Conf.
Phys. of Semimetals and Narrow-Gap Semiconductors, p. 383. Pergamon Press,
Oxford.
Norde, A . H. (1979). / Appl Phys. 50,5052.
Okamoto, K., Wood, C. E. C. and Eastman, L. F. (1981). Appl Phys. Lett. 38,
636.
Okuto, Y. and CroweU, C. R. (1974). J. Jap. Soc. Appl Phys. 43(Suppl.), 390.

241
Ottaviani, G. (1979). J. Vac. Set Technol. 16,1112.
Ottaviani, G. (1981). J. Vac. Sci. Technol. 18, 924.
Ottaviani, G., Tu, K. N. and Mayer, J. W. (1980). Phys. Rev. Lett. 44,284.
Padovani, F. A. (1967). J. Appl. Phys. 38, 891.
Padovani, F. A . (1968). SoUd-St Electron. 11,193.
Padovani, F. A. (1971). Semiconductors and semimetals (ed. R. K. Willardson and
A. C. Beer) Vol. 7A, Chap. 2. Academic Press, New York.
Padovani, F. A. and Stratton, R. (1966). Solid-St. Electron. 9, 695.
Padovani, F. A. and Sumner, G. G. (1965). J. Appl. Phys. 36, 3744.
Palau, J. M., Ismail, A. and Lassabatere, L. (1985). Solid-St. Electron. 28,499.
Parker, G. H., McGill, T. C , Mead, C. A . and Hoffman, D. (1968). Solid-St.
Electron. 11,201.
Patterson, M . H. and WiUiams, R. H. (1982). /. Cryst Growth 59, 281.
Pauling, L. (1932). J. Am. Chem. Soe. 54, 3570.
Pauling, L. (I960). The nature of the chemical bond (3rd edn). Cornell University
Press, Ithaca.
Pautrat, J. L., Katircioglu, B., Magnea, N., Bensahel, D., Pfister, J. C. and Revoil, L.
(1980). Solid-St. Electron. 23,1159.
Perlman, S. S. (1969). IEEE Trans. Electron. Devices ED-16, 450.
Phillips, J. C. (1974). J. Vac. Sci. Technol. 11,947.
Pickard, G. W. (1906). US patent no. 836531.
Pierce, G. W. (1907). Phys. Rev 25, 31.
Poate, J. M . and Tisone, T. C. (1974). Appl. Phys. Lett 24,391.
Ponpon, J. P. and Siffert, P. (1975). J. Phys. 36, L-149.
Poole, I., Lee, M . E., Missous, M . and Singer, K. E. (1987). Appl Phys. Lett 62,
3988.
Pruniaux, B. R. and Adams, A. C. (1972). J. Appl Phys. 43,1980.
Prutton, M . (1983). Surface physics. Clarendon Press, Oxford.
Pugh, J. H. and Williams, R. S. (1986). /. Mat Res. 1, 343.
Purtell, R. J., Ho, R S., Rubloff, J. W. and Schmid, P. E. (1983). Physica 117/
118B,834.
Rhoderick, E. H. (1972). /. Phys. D.Appl Phys. 5,1920.
Rhoderick, E. H. (1975). J. Appl Phys. 46,2809.
Rhoderick, E. H . (1976). Proc. 5th European Solid-State Device Res. Conf., p.
103. Société Française de Physique, Paris.
Rhoderick, E. H. (1984). Electron. Lett. 20, 868.
Rideout, V. L. (1974). Solid-St. Electron. 17,1107.
Rideout, V. L. (1975). Solid-St. Electron. 18, 541.
Rideout, V. L. and Crowell, C. R. (1970). Solid-St. Electron. 13, 993.
Ritchie, R. H. and Ashley, J. C. (1965). J. Phys. Chem. Solids 26,1689.
Rivière, J. C. (1969). Solid-St Surf. Sci 1,179.
Roberts, G. G. and Pande, K. R (1977). /. Phys. D.Appl Phys. 10,1328.
Roberts, G. I. and Crowell, C. R. (1970). 7. Appl Phys. 41,1767.
Roberts, G. I. and Crowell, C. R. (1973). Solid-St Electron. 16,29.
Robinson, G. Y. (1975). Solid-St Electron. 18, 331.
Robinson, G. Y (1976). /. Vac. Sci Technol 13, 884.
Robinson, G. Y (1980). Thin Solid Films 72,129.
Robinson, G. Y. (1985). In Physics and chemistry of III-V compound semi-

242
 conductor interfaces (ed. C. W. Wilmsen) Chap. 2. Plenum Press, New York.
Roth, J. A. and CroweU, C. R. (1978). J. Vac. Sei. Technol. 15,1317.
Rubloff, G. W. (1982). Phys. Rev. B25,4307.
Rubloff, G. W. (1983). Surf. Sei. 132, 268.
Ryder, E. J. and Shockley, W. (1951). Phys. i?ev. 81,139.
Sa, C. J. and Meiners, L. G. (1986). Appl. Phys. Lett. 48,1796.
Sah, C. T., Chan, W. W., Fu, H. S. and Walker, J. W. (1972). Appl. Phys. Lett 20,
193.
Sah, C. T., Forbes, L., Rosier, L. L. and Tasch, A. F. (1970). Solid-St Electron. 13,
759.
Sah, C. T. and Neugroschel, A. (1976). IEEE Trans. Electron. Devices ED-23,
1069.
Sah, C. X , Noyce, R. N . and Shockley, W. (1957). Proc. IRE 45,1228.
Sah, C. T., Rosier, L. L. and Forbes, L. (1969). Appl Phys. Lett 15,316.
Sahich, J. L. (1969). Ohmic contacts to semiconductors, p. 187. Electrochemical
Society, New York.
SaUich, J. L. and Clark, L. E. (1970). Solid-St Electron. 13, 857.
Saltich, J. L. and Terry, L. E. (1970). Proc. IEEE 58, 492.
Sankey, O. F., AUen, R. E., Ren, S. F. and Dow, J. D. (1985). J. Vac. Sei. Technol
B3,1162.
Saxena, A. N. (1969). Surf. Sei. 13,151.
Scharfetter, D. L. (1965). Solid-St Electron 8, 299.
Schlüter, M . (1978). Phys. Rev. B17, 5044.
Schmid, P. E. (1985). Helv. Phys. Acta 58,371.
Schmid, P. E., Ho, P. S., Foil, H. and Rubloff, G. W. (1981). J. Vac. Sei. Technol 18,
937.
Schmid, P. E., Ho, P. S., Foil, H. and Tan, T. Y. (1983). Phys. Rev. B28,4593.
Schottky, W. (1938). Naturwiss. 26, 843.
Schottky, W. (1939). Z. Phys. 113, 367.
Schottky, W. (1942). Z. Phys. 118,539.
Schottky, W. and Spenke, E. (1939). Wiss. Verojf. Siemens-Werken 18,225.
Schottky, W., Störmer, R. and Waibel, F. (1931). Z. Hochfrequenztech. 37,162.
Schultz, W. (1954). Z. Phys. 138, 598.
Schwarz, R. E and Walsh, J. F. (1953). Proc. IRE 41,1715.
Sebenne, C , Bolmont, D., Guichar, G., and Balkanski, M. (1974). Proc. 12th Int.
Conf Phys. Semiconductors, p. 1313. Teubner, Stuttgart.
Sebestyen, T. (1982). Solid-St. Electron. 25, 543.
Seitz, F. (1940). Modern theory of solids. McGraw-Hill, New York.
Selberherr, S. (1984). Analysis and simulation of semiconductor devices. Springer-
Verlag, Vienna.
Senechal, R. R. and Basinski, J. (1968a). J. Appl Phys. 39,3723.
Senechal, R. R. and Basinski, J. (1968b). J. Appl Phys. 39,4581.
Shaw, J. M . and Amick, J. A. (1970). RCA Rev. 31, 306.
Shannon, J. M . (1974). Appl Phys. Lett. 25, 75.
Shannon, J. M . (1976). Solid-St Electron. 19, 537.
Sharma, B. L. (1981). In Semiconductors and semimetals (ed. R. K. Willardson
and A. C. Beer) Vol. 15, Chap. 1. Academic Press, New York.
Shockley, W. (1939). Phys. Rev. 56, 317.

243
Shockley, W. (1950). Electrons and holes in semiconductors. Van Nostrand, New
York.
Shockley, W. and Read, W. T. (1952). Phys. Rev. 87,835.
Sigurd, D. (1974). Proc. Manchester Conf. Metal-Semiconductor Contacts, p. 141,
Conference Series No. 22. Institute of Physics, London.
Sinha, A. K. and Poate, J. M . (1973). Appl. Phys. Lett. 23,666.
Sinha, A. K. and Poate, J. M . (1978). In Thin films—interdiffusion and reactions
(ed. J. M . Poate, K. N. Tu and J. W. Mayer) Chap. 11. Wiley, New York.
Slowick, J. H., Brillson, L. J. and Richter, H. W. (1986). /. Vac. Sei. Technol. B4,
974.
Smith, B. L. (1968). Electron. Lett. 4, 332.
Smith, B. L. (1969a). Ph.D. Thesis, Manchester University.
Smith, B. L. (1969b). J. Appl Phys. 40,4675.
Smith, B. L. (1973). J. Phys. D:Appl Phys. 6,1358.
Smith, B. L. and Rhoderick, E. H. (1969). /. Phys. D.Appl Phys. 2,465.
Smith, B. L. and Rhoderick, E. H. (1971). Solid-St. Electron. 14,71.
Snell, A. J., Mackenzie, K. D., Le Comber, R G. and Spear, W. E. (1979). Phil
Mag. B40,1.
Spear, W. E., Le Comber, P G. and Snell, A. J. (1978). Phil Mag B38,303.
Spenke, E. (1958). Electronic semiconductors. McGraw-Hill, New York.
Spicer, W. E., Lindau, I., Skeath, P. and Su, C. Y. (1980a). /. Vac. Sei Technol 17,
1019.
Spicer, W. E., Lindau, I., Skeath, P., Su, C. Y. and Chye, P W. (1980b). Phys. Rev
Lett 44,420.
Spicer, W. E., Kendelewicz, T. J., Newman, N., Chin, K. K. and Lindau, I. (1986).
Surf. Scl 168, 240.
Spicer, W. E., Pan, S., Mo, D., Newman, N., Mahowald, P., Kendelewicz, T. and
Eglash, S. (1984). J. Vac. Scl Technol B2,476.
Spitzer, W. G. and Mead, C. A. (1963). / Appl Phys. 34, 3061.
Srivastava, A. K. and Arora, B. M . (1981). Solid-St Electron. 24,1049.
Srivastava, A. K., Arora, B. M . and Guha, S. (1981). Solid-St Electron. 24,185.
Stall, R., Wood, C. E. C , Board, K. and Eastman, L. F. (1979). Electron. Lett. 15,
800.
Stokoe, T. Y and Parrott, J. E. (1974). Solid-St Electron. 17,477.
Stolt, L., BohUn, K., Tove, P. A. and Norde, H. (1983). Solid-St Electron. 26,295.
Stratton, R. (1962). Phys. Rev. 126, 2002.
Strayer, R. W., Mackie, W. and Swanson, L. W. (1973). Su/f. Scl 34,225.
Strikha, V. I. (1964). Radio-EngngElectron. Phys. 4, 552.
Strikha, V. I. and Kil'chitskaya, S. S. (1968). Sov Phys. Semicond. 1, 831.
Sullivan, A. B. J. (1976). Electron. Lett 12,133.
Sullivan, M . W. and Eigler, J. H. (1957). /. Electrochem. Soc. 104, 226.
Svensson, S. P. and Andersson, T. G. (1985). J. Vac. Scl Technol B3, 760.
Sze, S. M . (1981). Physics of semiconductor devices (2nd edn). Wiley, New York.
Sze, S. M . (1985). Semiconductor devices: physics and technology. Wiley, New
York.
Sze, S. M . and Gibbons, G. (1966). Solid-St Electron. 9,831.
Sze, S. M., Coleman, D. J., and Loya, A. (1971). Solid-St Electron. 14,1209.
Sze, S. M., Crowell, C. R., and Kahng, D. (1964). /. Appl Phys. 36,2534.

244
 Szydlo, N. and Poirier, R. (1973). J. Appl. Phys. 44,1386.
' Tamm, I. E. (1932). Phys. Z. Sowjet 1,733.
Tang, J. Y. F. and Freeouf, J. L. (1984). J. Vac. Sci. Technol. B2,459.
Tejedor, C , Flores, F. and Louis, E. (1977). /. Phys. C: Solid-St. Phys. 10,2163.
Tersoff, J. (1984). Phys. Rev. Lett. 52,465.
Tersoff, J. (1985). Phys. Rev B32,6968.
Thanailakis, A. (1974). Proc. Manchester Conf. Metal-Semiconductor Contacts,^.
59. Institute of Physics, London.
Thanailakis, A. (1975). /. Phys. C: Solid-St Phys. 8,655.
Thanailakis, A. and Northrop, D. C. (1971). J. Phys. D: Appl Phys. 4,1776.
Thanailakis, A. and Northrop, D. C. (1973). Solid-St Electron. 16,1383.
Thanailakis, A. and Rasul, A. (1976). J. Phys. C: Solid-St Phys. 9,337.
Thompson, M . J., Johnson, N. M., Nemanich, R. J. and Tsai, C. C. (1981). Appl
Phys. Lett 39,274.
Torrey, H . C. and Whitmer, C. A . (1948). Crystal rectifiers. McGraw-Hill, New
York.
Todd, C. J., Ashwell, G. W. B., Speight, J. D. and Heckingbottom, R. (1974). Proc.
Manchester Conf. Metal-Semiconductor Contacts, p. 171. Conference Series
No. 22. Institute of Physics, London.
Tove, P. A. (1982). J. Phys. D.Appl Phys. 15,517.
Tove, P. A. (1983). Surf. Sci. 132, 336.
Tove, P. A., Hyder, S. A. and Susila, G. (1973). Solid-St Electron. 16,513.
Tsaur, B. Y , Lau, S. S., Mayer, J. W. and Nicolet, M.-A. (1981). Appl Phys. Lett.
38,922.
Tu, K. N. (1975). Appl Phys. Lett 21,221.
Tu, K. N. and Mayer, J. W. (1978). In Thinfilms—interdijfusionand reactions (ed.
J. M . Poate, K. N . Tu and J. W. Mayer) Chap. 10. Wiley, New York.
Tu, K. N., Alessandrini, E. J., Chu, W. K., Kraiitle, H. and Mayer, J. W. (1974). Jap.
J. Appl Phys. Suppl 2-1,669.
Tu, K. N., Thompson, R. D. and Tsaur, B. Y. (1981). Appl Phys. Lett 38,626.
Tugulea, A. and Dascalu, D. (1984). J. Appl Phys. 56, 2823.
Tung, R. T. (1984). Phys. Rev Lett 52, 461.
Tung, R. T., Gibson, J. M . and Poate, J. M . (1983a). Phys. Rev. Lett 50,429.
Tung, R. T., Gibson, J. M . and Poate, J. M . (1983b). Appl Phys. Lett. 42,888.
Tung, R. T, Ng, K. K., Gibson, J. M. and Levi, A. F. J. (1986). Phys. Rev. B33,
7077.
Turner, M . J. and Rhoderick, E. H. (1968). Solid-St. Electron. 11,291.
Uhrberg, R. I. G., Hansson, G. V., Nicholls, J. M . and Flodstrom, S. A. (1982).
Phys. Rev. Lett 48,1032.
van Laar, J. and Sheer, J. J. (1967). Surf. Sci. 8, 342.
van Otterloo, J. D. and de Groot, J. G. (1976). Surf^ Sci. 57, 93.
van Otterloo, J. D. and Gerritsen, L. J. (1978). /. Appl Phys. 49, 723.
Varma, R. R., McKinley, A., Williams, R. H . and Higginbotham, I. G. (1977). J.
Phys. D:AppL Phys. 10, L171.
Vilms, J. and Wandinger, L. (1969). Ohmic contacts to semiconductors, p. 31.
Electrochemical Society, New York.
Vincent, G., Bois, D. and Pinard, P. (1975). J. Appl. Phys. 46, 5173.
Wada, O., Yanagisewa, S., and Takanashi, H. (1976). Appl Phys. Lett. 29,263.
245
Wagner, С. (1931). Phys. Z. 32,641.
Waldrop, J. R. (1984). J. Vac. Sei. Technol. В2,445.
Waldrop, J. R. (1985). J. Vac. Sei. Technol. ВЗ, 1197.
Walker, L. G. (1974). Solid-St. Electron. 17,763.
Walpole, J. N. and NiU, K. W. (1971). J. Appl. Phys. 42, 5609.
Walser, R. M . and Bene, R. W. (1976). Appl. Phys. Lett. 28,624.
Walser, R. M . and Bene, R. W. (1978). Proc. Symp. Thin Film Phenomena (eds J.
E. Baglin and J. M . Poate) p. 21. Electrochemical Society, Princeton.
Wang, W. I. (1983). /. Vac. Sei. Technol. B l , 574.
Weinman, L. S., Jamison, S. A. and Helix, J. (1981). /. Vac. Sei. Technol. 18, 838.
Werner, J., Ploog, K. and Queisser, H. J. (1986). Phys. Rev. Lett 57,1080.
Whiteaway, J. E. A . (1983). Proc. lEE. 130,165.
Wickenden, D. K., Sisson, M . J., Todd, A. G. and Kelly, M . J. (1984). Solid-St.
Electron. 27, 515.
Wilkinson, J. M . (1974). Proc. Manchester Conf. Metal-Semiconductor Contacts,
p. 27. Institute of Physics, London.
Wilkinson, J. M., Wilcock, J. D. and Brinson, M . E. (1977). Solid-St Electron. 20,
45.
Williams, R. (1966). /. Appl. Phys. 37, 3411.
Williams, R. H. (1983). Surf. Sei. 132,127.
Williams, R. H . (1985). In Physics and chemistry of III-V compound semicon­
ductor inteфces (ed. C. W. Wilmsen) Chap. 1. Plenum Press, New York.
Wilhams, R. H., Montgomery, V. and Varma, R. R. (1978). / . Phys. C: Solid-St.
Phys. 11,L735.
Williams, R. H., Srivastava, G. P. and McGovern, I. T. (1980). Rep. Prog. Phys. 43,
1357.
Wilhams, R. H., Srivastava, G. P. and McGovern, I. T. (1984). In Electronic
properties of surfaces (ed. M . Prutton) p. 366. Adam Hilger, Bristol.
Williams, R. H., Varma, R. R. and McKinley, A. (1977). J. Phys. C: Solid-St Phys.
10,4545.
Williams, R. H., Varma, R. R. and Montgomery, V. (1979). /. Vac. Set. Technol. 16,
1418.
Williams, R. H., Dharmadasa, I. M., Patterson, M . H., Maani, C. and Forsythe, N.
M . (1986a). Surf. Sei. 168, 323.
Wilhams, R. H., McLean, A . В., Evans, D. A . and Herrenden-Harker, W. G.
(1986b). /. Vac. Sei. Technol 84,966.
Wilhams, R. H., McKinley, A., Hughes, G. J., Montgomery, V. and McGovern, I. T.
(1982). /. Vac. Sel Technol 21,594.
Wilmsen, C. W. (1985). In Physics and chemistry of III-V compound semi­
conductor inteфces (ed. С. W. Wilmsen) Chap. 7. Plenum Press, New York.
Wilson, A . H. (1932). Proc. R. Soc. A136,487.
Wood, E. A. (1964). /. Appl Phys. 35,1306.
Woodall, J. M . and Freeouf, J. L. (1982). J. Vac. Sel Technol 21,574.
Woodall, J. M., Freeouf, J. L., Petit, G. D., Jackson, T. and Kirchner, P (1981). /.
Vac. Sel Technol 19,626.
Woodall, J. M., Lanza, С. and Freeouf, J. L. (1978). /. Vac. Sel Technol 15,1436.
Wronski, C. R. and Carlson, D. E. (1977). Solid-St. Commun. 23, 421.
Wronski, C. R., Carlson, D. E. and Daniel, R. E. (1976). Appl Phys. Lett. 29,602.

246
Yndurain, F. (1971). J. Phys. CSolid-St. Phys. 4,2849.
Yoder, M . N . (1980). Solid-St. Electron. 23,117.
Yu, A. Y. C. (1970). Solid-St Electron. 13, 239.
Yu, A. Y. C. and Snow, E. H. (1968). J. Appl Phys. 39, 3008.
Yu, A. Y. C. and Snow, E. H. (1969). Solid-St Electron 12,155.
Zang, S. B., Cohen, M . L. and Louie, S. G. (1985). Phys. Rev. 332, 3955.
Zang, S. B., Cohen, M . L. and Louie, S. G. (1986). Phys. Rev B34, 768.
Zettlemoyer, A. C. (1969). Ohmie contacts to semiconductors, p. 48. Electro­
chemical Society, New York.
Ziegler, J. F , Mayer, J. W., Kircher, C. J. and Tu, K. N. (1973). J. Appl Phys. 44,
3851.
Zohta, Y. (1970). Appl Phys. Lett 17, 284.
Zohta, Y. (1973). Solid-St Electron. 16,1029.
Zunger, A. (1981). Phys. Rev. B24, 4372.
Zur, A., McGill, T. C. and Smith, D. L. (1983). Phys. Rev B28, 2060.

247
248
Index

ageing of Schottky barriers 54, 58 for diamond 84

abruptness of interface 29, 49, 64, 66, 79, for GaAs, see gallium arsenide
85, 198-9 for Ga^ln,_,As 82
adsorption (on free surface) 5, 183 forGaP 79
amorphous semiconductors 226-31 for GaSb 80
antisite defects 34, 78, 85 forGe 57
Auger electron spectroscopy 186, 188, forlnAs 80
189 for InP
avalanche breakdown 131 clean 64, 68-9, 74-5, 77-9
oxidized 75-6
forlnSb 80
Bardeen limit 17, 20 for metallic alloys on Si 56, 202
Bardeen model 17-28, 60, 61, 85 for p-type semiconductors 14, 25, 26
band bending, see diffusion potential forPbTe 83
band gap, reduction due to image force 38 forSe 84
barrier for Si, see silicon
effect of image force on 35-8 forSiC 84
formation of 11-15 for silicides on Si 199-202
in amorphous semiconductors 227-8 for ZnS 32, 83
Mott 13-14 for ZnSe 32, 83
Schottky 13-14 forZnTe 83
barrier height for II-VI compounds 83
ageing of 54-8 for III-V compounds 63-82, 85-6
Bardeen limit 17, 20 ultra-thin metal layers on 65-9
bias-dependence of 26-7 for ternary III-V compounds 80-2
control of 202-4 for IV-VI compounds 83
dependence on donor density 56 methods of measurement of 38-46
determination from C / V data 42-3, 51, orientation dependence of 73, 83
69, 72, 145, 147, 148-51 Schottky-Mott limit 12, 20
determination from J/V data 38-41, temperature dependence of 56, 139
201 zero-bias 27
determination by photoelectric emission breakdown voltage 131
spectroscopy 43-6 tables of 214, 215
determination by photoelectric method burn-out in microwave mixers 187
41-2
effect of exposure to oxygen 53-4
effect of penetration of field into metal
27-8 capacitance of Schottky barrier 141-7
effect of surface (interface) states 15-20 absence of diffusion component 179,
effect of two bands of surface states 223
20-3 dependence on bias voltage 145-6, 213
effective 38 effect of deep traps
field dependence of 23-4, 124-6 under forward bias 179-81
flat-band 19 under reverse bias 167-79
for Al,Ga,_,As 80-1 effect of degeneracy 145-6
for Al,Ga,_,Sb 82 effect of incomplete ionization 145-6
forCdS 83 effect of interface states 150-7
forCdSe 83 effect of interfacial layer 150-5
forCdTe 83 effect of light 177-8
forCujO 84 effect of minority carriers 147-50

249
capacitance of Schottlcy barrier (cont.)
effect of non-uniform donor distribution
158- 62
used to measure barrier height 42-3,
145
used to measure impurity distribution
159- 62
thermally stimulated 178
transient measurements 174-7
under forward bias 179-81
charge
on surface of metal 18
in depletion region 18, 145, 212-13
in surface (interface) states 18, 19
chemical reactions at interface 29, 63-4,
68, 77, 85
chemically deposited contacts 185-6
clean surfaces 5, 28-9, 183
cleaved surfaces 5, 9, 33, 50-1, 65, 183
effect of cleavage steps 33
clusters (nuclei) 64
combined thermionic-emission/diffusion
theory 100-4
common anion rule 80-2
conductivity modulation 122-3, 222
contacts
chemically deposited 185-6
effect of heat treatment 186-93
evaporated 182-3
intimate 49-53,61-3, 65-72,75-7,
183
ohraic 14, 116-18
on etched surfaces 53-7, 60-1, 72-4,
76, 182
sputtered 184
control of barrier heights 202-4
covalent-ionic transition 87
current-voltage charcteristic
due to thermionic emission/diffusion
90-109
effect of interfacial layer 133-8
effect of recombination in depletion
region 118-21
effect of thermal generation in depletion
region 132
effect of tunnelling on 109-18, 127-30
under reverse bias 124-32
used to measure barrier heights 38-41
C - V profiling of dopant distributions
159-62
electrochemical technique for 162
errors in 160
deep level transient spectroscopy (DLTS)
177
deep traps
effect on capacitance 167-81
250
occupation probabihty for 163-7,
179-80
transient measurements 174-7
defects at interfaces 33-4
due to cleavage steps 33
due to vacancies 34, 78-9
due to misplaced atoms 34, 78-9, 85-6
energy levels associated with 33-4,
77-9, 85-6
demarcation levels for electrons and holes
164
depletion approximation 211-15
depletion region 11
capacitance of 141-7, 152, 213
tables of 214, 215
generation of electron-hole pairs in 132,
118-21
recombination in 118-21
width of (tables) 214-15
diffusion capacitance 179, 223
diffusion potential 26, 145, 212
diffusion theory 90, 92-4
condition for validity of 97, 102-3
diode theory, see thermionic-emission
theory
dipole layer
surface 10
interfacial 13, 48, 62-3, 202
edge breakdown 130
elimination by use of guard ring 131
electric field in depletion region 212,
216-18
electroless plating 185, 208
electron affinity 10
electron states in solids and at surfaces
5-10
electronegativity 47-8
table of 48
epitaxial contacts 29, 64, 85, 108, 192
etched surfaces, see contacts on etched
surfaces
evaporated contacts 182-3
field emission 109, 112-13, 116-17
under reverse bias 127, 129-30
field penetration in metal 27-8
gallium arsenide
barrier heights on clean surfaces
ultra-thin metal layers 65-8, 76, 78
thick metal layers 32, 69-72, 76-9
barrier heights on oxidized surfaces
72-4
 orientation dependence of barrier height
73
surface state density on cleaved surfaces
9,65
guard ring 131
heat treatment, effect of 186-93
on silicon contacts 186-9
on contacts to Ill-V compounds 189-93
hole-injection in Schottky diodes 121-4,
221-3
hot-electron effects 104
ideality factor (n) 39, 99-100, 113-15,
118, 121, 134-6
image force
effect on barrier height 35-7
effect on band gap 38
effect of interfacial layer 136
effect on J/V characteristic 98-100,
108, 114-15
effective dielectric constant for 35-6
interface states, see surface states
spectroscopy of 155-7
interfacial layer 12, 15, 17, 43, 84
effect on capacitance 150-5
effect on J/V characteristic 133-8
formation of 182-3
intimate contacts 28-34; see also contacts
on clean surfaces
ionization energy (of semiconductor) 11
'lifetime' semiconductor 123
metal-induced gap states (MIGS) 30-3,
79, 86, 87
minority carrier storage 132-3, 222-3
Mott barrier 13
n value, see ideality factor
non-abrupt (graded) interfaces 29, 49, 64,
79, 85, 198-9
numerical analysis (modelling)
of current-flow 139-40
of capacitance 161
ohmic contacts 14, 116-18, 204-9
photocapacitance 178
photoelectric determination of barrier
height 41-2
251
photoelectron emission spectroscopy 43-6
used to measure barrier height 45-6
p-n junctions, comparison with Schottky
diodes 219-23
point contacts 123-4, 182
polycrystalline metal films 29, 64, 72, 192
p-type semiconductors
barrier heights on 14, 25-6, 56-7,
65-7, 69, 71-2, 77-8
quasi-Fermi level (imref) 90
for electrons 90-2, 107-8, 163, 179
for holes 121, 163, 224-5
recombination in depletion region 118-21
recombination velocity 100-1, 104
reconstruction (of free surface) 4
effect on barrier height 71-8
rectifiers, use of Schottky diodes as
100-1, 104
relaxation (of free surface) 3-4, 9
'relaxation' semiconductor 123
reproducibility of barrier heights 43, 57-9
dependence on interface perfection 58
reverse characteristics 124-32
Richardson constant 96
dependence on crystal orientation 97
effect of phonon scattering and
quantum-mechanical reflection on
105-6
values for Si and GaAs 106, 107, 109
Rutherford back-scattering 186, 194-5
saturation current density, reverse 124
Schottky barrier 13
Schottky diodes
comparison with p-n junctions
219-23
recovery time of 132-3
Schottky-Mott limit 13, 20
Schottky-Mott theory 11-15
secondary-ion mass-spectroscopy (SIMS)
186, 190
series resistance, effect on J/V
characteristic 40, 121
silicides 58-9, 193-202
abruptness of the interface 198-9
barrier heights 199-202
reaction kinetics 194-6
sequence of phase formation 196-8
type A and В interfaces (NiSiz) 58-9,
62, 201
silicon effect on barrier height 15-23
barrier height for clean Si 32, 49-53, effect on J/V characteristics 125, 135-7
61-3 intrinsic 7
with ultra-thin metal layers 51, 53 neutral level for 8, 15-17, 19
table of 52 pinning of barrier height by 17
barrier height for cleaved Si, see clean surface stoichiometry 30
silicon
barrier height for etched and oxidized Si
53-6, 60-1
'To'effect 139
dependence on donor density 56
thermally stimulated capacitance 178
barrier height for p-type silicon 56-7
thermionic-emission theory 90, 94-8,
barrier height for metallic alloys on 56
105-6
barrier height for NiSiz on 58, 62, 85,
comparison with experiment 107-9
201
condition for validity of 102-4
barrier height for silicides on 199-202
thermionic-field emission 109, 111-16
surface-states on cleaved surfaces 7-9,
under reverse bias 127-8
60, 61
Thomas-Fermi screening distance 11,
surface-states on etched surfaces 60
27
temperature dependence of barrier
transient effects 132-3
height 56
tunnelling through Schottky barrier
space-charge region, see depletion region
109-18, 126-31
sputtered contacts 184
tunnelling, quantum-mechanical 110
surface states 7-9
and dangling bonds 7-8
bands of 7, 20
charge distribution due to 7 unified defect model (of surface states) 34,
density of, for cleaved GaAs 9, 65 66, 77
density of, for cleaved Ge 9
density of, for etched Ge 57
density of, for cleaved GaP 10
vacuum level 10
density of, for cleaved Si 9, 61
for interfacial layer 17-18
density of, for etched Si 60
density of, for cleaved III-V compounds
9
donor-like and acceptor like 8, 21, 66, work function 10-11
78 table of 48

252