Home Search Collections Journals About Contact us My IOPscience

Interface states at semiconductor junctions

This content has been downloaded from IOPscience. Please scroll down to see the full text.

1999 Rep. Prog. Phys. 62 765

(http://iopscience.iop.org/0034-4885/62/5/203)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 136.186.1.81
This content was downloaded on 18/05/2015 at 13:27

Please note that terms and conditions apply.

Rep. Prog. Phys. 62 (1999) 765–808. Printed in the UK PII: S0034-4885(99)21756-3

Interface states at semiconductor junctions

G Margaritondo
Institut de Physique Appliquée, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland

Received 5 January 1999

Abstract

The experimental and theoretical progress in understanding the electronic structure and the
related parameters of Schottky interfaces and heterojunctions is reviewed. Particular emphasis
is devoted to the solution of several historical controversial points, to the impact of novel ab
initio theoretical approaches, to new experimental techniques based on synchrotron light and
free electron lasers, to the efforts towards controlled modifications of interface parameters
and to the foreseeable future developments of this vigorously progressing and technologically
crucial field.

0034-4885/99/050765+44$59.50 © 1999 IOP Publishing Ltd 765

766 G Margaritondo

Contents

Page
1. Historical background 767
1.1. The early models 768
1.2. The first surface-sensitive experiments 769
2. Experimental techniques 771
2.1. Transport and optical techniques 772
2.2. Electron spectroscopy techniques 773
2.3. BEEM 778
3. Older results and past controversies 779
3.1. Identification of the general issues 780
3.2. More recent experimental results and evolution in theory 782
3.3. After the controversies, a general picture emerges—the wonders of ab initio
calculations 784
4. The search continues: recent results 786
4.1. Investigating fundamental issues 787
4.2. The initial stages of interface formation 787
4.3. Technology-oriented work 788
4.4. New experimental techniques and new applications 789
5. From understanding to controlling 790
5.1. Empirical approaches: ultrathin intralayers 790
5.2. Advanced theoretical gap engineering and its practical implementation 790
6. The new frontiers: lateral variations, free-electron-laser techniques 793
6.1. The issue of lateral variations 794
6.2. Laterally resolved experimental techniques 794
7. Summary and future developments 801
7.1. Present trends and open problems 801
7.2. New experimental opportunities 802
7.3. Final remarks 804
 Interface states at semiconductor junctions 767

1. Historical background

The electronic structure of semiconductor interfaces is one of the oldest and most fascinating
problems of condensed-matter science (Brillson 1982, 1992, 1993, Franciosi and Van de Walle
1996, Rhoderick and Williams 1988, Yu et al 1992, Mönch 1990, Kahn 1994, Williams 1991,
Brillson and Margaritondo 1988, Capasso and Margaritondo 1987, Margaritondo 1986, 1993,
McKinley and Margaritondo 1993). The initial interest in this problem was obviously triggered
by a practical need: understanding the behaviour of the early solid-state microelectronic
devices. Quite soon, however, it became evident that the problem was quite complicated and
involved some of the most fundamental aspects of solid-state physics (Margaritondo 1993,
Peressi et al 1998).
The key issue, i.e. the link between the interface parameters and the local electronic and
chemical structure, is not yet entirely clarified. From the point of view of theory, substantial
progress was made possible by the advent of suitable approaches to numerically solve semi-
realistic interfaces (Peressi et al 1998). This led, in particular, to predictions on how one can
modify and therefore control the interface parameters as required for bandgap engineering
(Capasso and Margaritondo 1987). The experimental implementation of such predictions is
one of the milestones towards a complete understanding and exploitation of semiconductor
interfaces (Brillson 1982, Margaritondo 1993, Peressi et al 1998).
In recent years, however, a new problem has emerged (Gozzo et al 1993, 1995), for
which a complete clarification has not yet been achieved. Virtually all theoretical models
of semiconductor interfaces—such as metal–semiconductor and heterojunction systems—
assume that the interface electronic structure and the interface parameters are constant over
the entire plane of the interface. For example, parameters such as the Schottky barrier and the
heterojunction valence and conduction band discontinuities are assumed not to significantly
change from place to place.
This assumption is in sharp contrast to the results of the most recent experiments (Gozzo
et al 1993, 1995). For the first time, spatially resolved techniques were used to locally measure
the interface parameters and to explore the factors that determine them. In most cases, the
results unmistakably show that significant lateral changes do occur.
This requires a new way of thinking about semiconductor devices. Consider, for example,
the transport behaviour of a metal–semiconductor junction: a realistic model cannot assume
that the real device is equivalent to a single diode, and must use, instead, a parallel combination
of microscopic diodes with different Schottky barriers. One can intuitively guess that the overall
behaviour will be dominated by the weakest Schottky barriers; a complete and detailed model,
however, is not yet available.
The implications are quite relevant and worrying. For example, standard barrier measure-
ments could yield results that are irrelevant to the actual transport properties. Unfortunately,
efforts to clarify this issue are negatively affected by the scarcity of the experimental data.
Experiments in this field require highly specialized techniques such as photoemission spec-
tromicroscopy (Margaritondo and Cerrina 1990, Margaritondo 1995c, Ade 1997), ballistic
electron emission microscopy (BEEM) (Kaiser and Bell 1988, 1996, Prietsch 1995) or scan-
ning near-field optical microscopy (SNOM) with a suitable photon source such as a free electron
laser (Almeida et al 1996, Cricenti et al 1998, Davy et al 1999). None of these techniques is
widely available, thus the experiments on this specific problem are still quite scarce.
Because of the new challenges and open issues, semiconductor interface research remains
an extremely active field, even after several decades of work. The first objective of this review
is to briefly summarize the historical development of this field, in particular by discussing
some of the experimental approaches that provided its foundations. Then, we discuss the open
768 G Margaritondo

Figure 1. (a) A schematic explanation of the ‘Schottky model’ for metal–semiconductor interfaces
(Schottky et al 1931). Top: the metal and the semiconductor sides, ideally separated before forming
the junction. EC is the conduction band edge; EV is the valence band edge; EF is the Fermi level;
χM is the the metal work function; and χ is the semiconductor electronegativity. Bottom: after
the junction is formed, according to the model, the n-type Schottky barrier height 8SB is simply
given by the difference between χM and χ . (b) Localized interface states, however, can ‘pin’ the
Fermi level in the gap, making the Schottky model invalid. In this case, the Schottky barrier height
is simply given by the distance, right at the interface, between the conduction band edge EC and
the pinning position of EF .

issues, again presenting the experimental methods and some representative results.
Several excellent reviews (Brillson 1982, 1992, 1993, Franciosi and Van de Walle 1996,
Rhoderick and Williams 1988, Yu et al 1992, Mönch 1990, Kahn 1994, Williams 1991)
already exist in this field, therefore we did not duplicate them; on the contrary, we reduced
the discussion of issues treated by other reviews to a minimum—the minimum required for
understanding the relevant problems. In particular, an excellent review was recently published
by Peressi et al (1998) on the theoretical aspects of semiconductor interface research; the
reader should consult it for a detailed picture of the present theoretical status.

1.1. The early models

The first attempts to understand the rectifying properties of metal–semiconductor interfaces
(Schottky et al 1931) did not take into account the local electronic structure. No theoretical
tools were, in fact, available to attack or even to correctly formulate this problem. Schottky
et al (1931) was nevertheless able to justify the rectifying action by using bulk or pseudo-bulk
quantities such as the work function 8M of the metal and the semiconductor electron affinity, χ .
These should be considered as ‘pseudo-bulk’ rather than bulk quantities because, although
based on the bulk electronic structure of the two materials, they are measured by performing
experiments on their surfaces. Furthermore, it is not clear how these surface measurements
relate to interface properties.
Neglecting these rather important problems and in particular ignoring the interface
 Interface states at semiconductor junctions 769

electronic structure, one obtains the so-called Schottky model (Schottky et al 1931), in which
the rectifying barrier 8SB (for n-type semiconductors) is given by (see figure 1(a)):
8SB = χM − χ . (1)
The most relevant feature of equation (1) is the linear dependence on the metal work function
χM . Many years after the formulation of the Schottky model, this prediction turned out to be in
direct contrast to the measured barriers, which exhibited little or no dependence on the metal—
in particular for diodes involving the technically relevant covalent semiconductors (Bardeen
1947, Kurtin et al 1970).
This prompted Bardeen (1947) to propose that the local electronic structure at the interface
plays an important role, leading to the concept of Fermi-level (EF ) pinning. Basically, the
Schottky model assumes that as the interface is formed the local semiconductor band structure
is free to move with respect to EF because the Fermi level is in the semiconductor gap, which
does not have electronic states. One usually describes this property by saying—somewhat
incorrectly—that ‘EF is free to move in the gap’.
The absence of states from the gap is a consequence of the formation of chemical bonds
in the bulk semiconductor and of the consequent hybridization of atomic states. Assume, on
the other hand, that the local interface chemical bonds are substantially different from those
in the bulk: there is no longer a guarantee of a gap free of states. If there are states in the gap,
then the Fermi level is no longer free to move—somewhat like the Fermi level in a metal. One
then says that EF is ‘pinned’.
If the Fermi level is pinned, then the Schottky barrier (figure 1(b)) is not determined by
equation (1), but by the EF pinning position in the gap:
8SB = EC − EF , (2)
where EC is the conduction-band edge position, and 8SB can be independent of the metal.
Note that Bardeens EF -pinning concept does not explain how the Schottky barrier is
created: it merely states that to understand the Schottky barrier one must first understand the
local electronic structure. What are, from this point of view, the key features of the local
electronic structure? Answering this crucial question was the objective of decades of studies
accompanied by much controversy.
A similar conflict between the Schottky model and the notion of Fermi-level pinning
exists for semiconductor-semiconductor heterojunctions – see figure 2. In this case, the
relevant interface parameters are the valence and conduction band discontinuities, 1EV and
1EC , which accommodate the difference in gap width 1Eg = Eg1 − Eg2 between the two
semiconductors—and determine the junction transport and optical properties. The extension
of the Schottky model—known as the ‘Anderson model’ (Anderson 1962)—would predict that
1EC = χ2 − χ1 . (3)
On the other hand, if the Fermi level is locally pinned and therefore its distances at the interface
from the conduction-band edges EC1 − EF and ECC2 − EF are fixed, the conduction band
discontinuity is simply given by the differences of these two fixed distances:
1EC = (EC2 − EF ) − (EC1 − EF ) = EC2 − EC1 . (4)
Once again, the problem is thus not solved but only redefined: the solution requires a complete
understanding of the local electronic states.

1.2. The first surface-sensitive experiments

The need to analyse local electronic states and their role in pinning the Fermi level was already
clear in the early 1950s. However, experiments in this domain posed a formidable problem: the
770 G Margaritondo

Figure 2. (a) The Anderson (1962) model, which is the equivalent of the Schottky model in
the case of semiconductor–semiconductor heterojunction interfaces. The relevant parameters for
heterojuction interfaces are the valence and conduction band discontinuities 1EV and 1EC , which
accommodate the difference in gap width 1Eg = Eg1 − Eg2 between the two semiconductors.
The Anderson model gives 1EC = χ2 −χ1 . (b) If the Fermi level is locally pinned, 1EC is simply
given by EC2 − EC1 , the difference between the conduction-band edge positions with respect to
the Fermi level, at the interface.

local electronic states are superimposed on a huge background of bulk states. How could one
filter out bulk states and explore the properties of local states? The answer was provided
by surface-sensitive electron spectroscopy experiments, and primarily by photoemission
techniques (Capasso and Margaritondo 1987).
Electrons in the energy range 10–500 eV have, in fact, a very short mean free path in
solids (Margaritondo 1988a, b, c, 1995a). Therefore, all electron spectroscopies performed in
this energy range probe the near-surface region, at a depth not exceeding 5–30 Å. Consider, for
example, the case of photoemission techniques. The photon beam used to excite photoelectrons
can penetrate relatively deep into the solid—the penetration depth being determined by the
reciprocal of the absorption coefficient. However, most of the excited electrons lose energy
when they are far from the surface, and they are not able to cross the surface; even if they can
cross the surface, their energy is less than it was right after the excitation, thus they are emitted
as ‘secondary’ photoelectrons. The ‘primary’ photoelectrons originate from a very thin slab
near the surface.
Therefore, photoemission experiments—and electron spectroscopy experiments in
general—are automatically blind to most of the bulk electronic states. This is exploited for
interface formation studies, typically (Capasso and Margaritondo 1987) in the following way.
The study begins with the photoemission analysis of the ultraclean surface of the material
that will eventually form one side of the interface, under ultrahigh vacuum to prevent
contamination. Then, the interface is formed step-by-step, by subsequent depositions of the
second material—which produce an overlayer of gradually increasing thickness. The surface-
sensitive analysis is repeated at each step.
 Interface states at semiconductor junctions 771

Figure 3. An early example of a photoemission study of

metal–semiconductor interface formation: the experiments
of Margaritondo et al (1976) on Ga overlayers on Si(111)7×
7. The clean-Si spectrum (dashed curve) is compared with
the spectra for three different Ga coverages. Note the Ga-
induced shift of the Fermi-level position with respect to the
Si spectral features; the final distance position of EF in the
Si gap corresponds to the Schottky barrier height. Also note
that this was the ‘old way’ to show such data: later, it was
preferred to keep EF in a constant position and show the shift
of the spectra, as is done for all remaining photoemission
spectra in this review. Finally, note that the spectral features
due to clean-Si surface states (the weak shoulders near EF
in the clean-Si spectrum) are removed by Ga: this implies
that the clean-Si surface states do not coincide with the
Ga–Si interface states, as was then believed for metal–
semiconductor interfaces.

Quite paradoxically, surface sensitivity was a major problem in the early development
of photoemission techniques, due to the rapid contamination of solid surfaces (Margaritondo
1988a). Only after the seminal work (Siegbahn et al 1967, 1969) of Kai Siegbahn—five
decades after the conceptual background was provided by Einstein (1905)—photoemission
spectroscopy could be performed on contamination-free surfaces.
The localized states hypothesized by Bardeen (1947) as the main players in the interface
formation process were, of course, a prime target for the early photoemission studies of
semiconductor interface formation. Two independent landmark experiments at Stanford
(Wagner and Spicer 1972) and IBM (Eastman and Grobman 1972) detected, for the first
time, the localized states of clean Si surfaces—confirming the earlier studies by Chiarotti’s
group in Rome (Chiarotti et al 1971) of the corresponding local optical transitions. Shortly
afterwards, photoemission (Margaritondo et al 1976) and electron energy-loss spectroscopy
(Rowe et al 1975) were applied to the study of metal–semiconductor interfaces. The study
of heterojunction interfaces was then initiated by Perfetti and co-workers at Berkeley (Perfetti
et al 1978). Figures 3–5 show some of the first results of that early period.

2. Experimental techniques

At this point, we must dedicate a bit of time to consider the main experimental techniques used
in semiconductor interface research. We will not provide an exhaustive treatment here, but only
enough information to understand the data that will be discussed later. Additional details can
be found in several previous reviews (Brillson 1982, 1992, 1993, Franciosi and Van de Walle
1996, Rhoderick and Williams 1988, Yu et al 1992, Mönch 1990, Kahn 1994, Williams 1991,
Brillson and Margaritondo 1988, Capasso and Margaritondo 1987, Margaritondo 1988a, b, c,
1993, McKinley and Margaritondo 1993).
772 G Margaritondo

Figure 4. An early example of the electron energy-loss spectroscopy study (Rowe et al 1975) of
metal–semiconductor interface formation: Ga on Ge(111). The spectral features correspond to an
optical-like excitation having the Ge 3d level or the Ga 3d level for initial state. The comparison of
clean-Ge and Ga-covered-Ge spectra confirms the metal-induced removal of intrinsic Ge surface
states (the shaded area). On the other hand, the lineshape comparison for Ga 3d excitations for
pure Ga and for the interface reveals the creation of ‘extrinsic’ interface localized states.

2.1. Transport and optical techniques

The interface electronic structure and the related interface parameters influence the transport
and optical properties. It is possible, therefore, to study the electronic states with transport and
optical techniques. The problem, however, is the lack of surface sensitivity, for the reasons
discussed above. A transport experiment, for example, simultaneously probes the interface
(Capasso and Margaritondo 1987), the two bulk materials and the other contacts: deconvolving
from all this information the elements that are pertinent to the interface is a rather difficult task.
The two main transport methods to measure interface properties are based on
measuring current–voltage (I –V ) curves or capacitance–voltage (C–V ) curves (Capasso and
Margaritondo 1987). In each case, simple models predict the curve properties as a function
of the interface energy barriers. Therefore, from each curve one can, in principle, extract the
corresponding barrier value—such as, for example, the Schottky barrier height.
The problem with this approach is that the link between the I –V or C–V curves and the
 Interface states at semiconductor junctions 773

Figure 5. An early example of the photoemission study of

semiconductor–semiconductor heterojunction interface formation:
spectra of clean and Ge-covered Si(111)7 × 7 from Margaritondo
et al (1980). The Ge overlayer thickness θ is shown in equivalent
monolayers. Note again the disappearance of the clean-surface states.
The vertical lines emphasize the Ge-induced shift of the Si spectral
features. From this information and from additional analysis of the
spectra, the Si–Ge valence band discontinuity 1EV was derived.

extracted information is established with rather simplified models of the interfaces. Consider,
for example, the case of C–V curves, which are typically measured for a given frequency of
the bias voltage modulation. The simplest model (Capasso and Margaritondo 1987) predicts a
reciprocal square-root dependence of C on the difference between the bias voltage V and the
interface barrier.
Thus, one could easily extract the barrier from a C −2 versus V plot—if the model
assumptions are valid. This, however, could raise serious questions: interface states may
or may not influence the C–V curves depending on their response time and on the modulation
frequency. To achieve some reliability, one should at least perform measurements at different
frequencies over a wide range.
Optical techniques constitute a better approach. A typical approach (Capasso and
Margaritondo 1987) to measure interface barriers is based on optical measurements on
artificially fabricated quantum wells. The quantum well exhibits quantized energy levels whose
energies can be derived from infrared absorption or photoluminescence spectra. In turn, the
energy levels depend on the interface energy barriers—for example, on the heterojunction
conduction band discontinuity between the two materials forming the quantum well. Thus,
from the spectra, one can derive the barriers.
This method, however powerful and widely used, is not immune from problems. First of
all, it relies on a somewhat idealized model of the quantum well. In order to obtain quantum
wells reasonably close to this idealized situation, the fabrication technique must be quite
sophisticated. Because of this, the approach is not suitable for all interfaces, and is typically
confined to III–V materials and their ternary alloys.
A common problem affects both the transport and the optical approaches: the difficulty
in applying them to laterally resolved studies. This, as we discuss later, is a very important
trend in the present evolution of semiconductor interface research. Thus, the lack of lateral
resolution is a rather serious handicap.

2.2. Electron spectroscopy techniques

Several electron-based techniques have been used in semiconductor interface studies.
Specifically, low-energy electron diffraction (LEED) experiments (Brillson 1982, Kahn 1994)
were extensively applied to the identification of overlayer reconstructions and to relate them
to the local electronic structure.
Another widely used technique is electron energy-loss spectroscopy (EELS) (Brillson and
774 G Margaritondo

Margaritondo 1988, Rowe et al 1975). This technique consists of sending a monochromatic

beam of electrons onto the system under investigation and in detecting the energy distribution
spectrum of the scattered electrons. The spectral features reveal the characteristic losses caused
by the interaction of the electrons with the system under investigation.
Roughly speaking, the bulk loss function is given (Margaritondo and Rowe 1986,
Margaritondo and Weaver 1985) by the imaginary part Im (1/(ε(ω)) of the reciprocal of the
dielectric function (ε(ω)), and the surface loss function by Im (1/(ε(ω)) + 1)). In both cases
maxima occur at zeros of the denominator, |(ε(ω))| or of |(ε(ω)) + 1|, or at maxima of (ε2(ω)).
The first two cases correspond to the excitation of surface or both plasmons, and the third to
optical-like excitations (note, however, that the exciting wave is not transverse, like photons).
For interface research, EELS can be used (Margaritondo and Rowe 1986) as an alternate,
surface-sensitive version of optical spectroscopy, primarily to detect surface phonons and low-
energy electronic excitations. It has been used, for example, to study (Rowe et al 1975) optical-
like transitions associated to interface states at metal–semiconductor interfaces—see figure 4.

2.2.1. Photoemission techniques. Photoemission is, by far, the most widely used class of
electron spectroscopy technique for interface research. The reason for this is not only its
surface sensitivity, but also the possibility of studying interface electronic states, interface
chemical properties and interface parameters with the same technique.
Figure 6 schematically shows a photoemission study of a heterojunction interface, and
some of the results in the case of ZnSe–Ge (Margaritondo et al 1984). A photoelectron
spectrum or energy distribution curve (EDC) typically contains six types of features (see
figure 6(a)):
(1) A vacuum-level low-electron-energy cut-off reflecting the work function of the analysed
system (or that of the electron analyser).
(2) A low-energy peak due to secondary photoelectrons, i.e., electrons which lose part of
their energy after absorbing a photon and before leaving the sample and being emitted in
a vacuum. The secondary peak has a long tail extending to high energies.
(3) Core-level peaks, whose presence reflects that of the corresponding elements. The energies
depend on the atomic core-level energy of the relevant element, corrected for the interaction
with the valence electronic charge distribution reflecting the formation of chemical bonds.
Thus, core-level peaks provide information on the qualitative chemical composition and
on the chemical binding status of each component (Margaritondo 1995a).
(4) Valence-electron peaks which provide direct information on the chemical bonds that
determine them (Margaritondo 1995a).
(5) A high-energy cut-off which reflects either the Fermi edge (for metals) or the valence band
edge (for semiconductors and insulators).
(6) Features due to localized states such as the clean-surface or interface states. The results
of figure 3 show a nice example of weak but clearly visible surface-state features near the
Fermi level.
In the case of a semiconductor or an insulator (figure 6(b), top), there might be a band
bending between the bulk and the surface (or the interface with another material). The
band bending occurs over a distance of the order of the Debye length. Because of the high
surface sensitivity of photoemission, the measured photoelectron energies reflect the electronic
energies at the surface rather than in the bulk.
Consider now the results for the heterojunction interface ZnSe–Ge (figures 6(b) and (c)).
Most importantly, the valence-electron features reflect the two valence bands at the two sides
of the interface. Two edges are visible, and from their distance one can directly derive the 1EV
 Interface states at semiconductor junctions 775

Figure 6. (a) A schematic explanation of a photoemission experiment in the case of a metal:

the density of (occupied) states—including core, valence and surface states—is reflected in the
photoelectron spectrum, with a superimposed secondary electron distribution and the two cut-offs
caused by the Fermi level and by the vacuum level. (b) photoemission study of a semiconductor–
semiconductor heterojunction interface formation. Top: an initial study of a clean semiconductor
substrate; bottom: a subsequent study of the same substrate, covered with a thin overlayer of
another semiconductor. Note the double edge in the photoemission spectrum, which corresponds
to the valence band discontinuity. (c) Actual data for ZnSe–Ge from Margaritondo et al (1984).

discontinuity (Margaritondo et al 1984). Furthermore, the valence-electron features include

those related to localized states, and can be used to analyse their nature.
This approach can be, mutatis mutandis, extended to all types of semiconductor interfaces.
In the case of heterojunction, small values of 1EV often make it impossible to resolve the two
valence-band edges of the two sides of the junction. Even then, 1EV can be derived from the
spectra, by separately measuring the distance between each edge and a reference core level
and the core-level energies, and then combining these values after correcting them for possible
band-bending changes during the interface formation. The method is discussed in detail, for
example, by Capasso and Margaritondo (1987).
As for metal–semiconductor junctions, the semiconductor (valence) edge position and the
Fermi level can be directly derived from the spectra (Margaritondo et al 1976). From these one
can easily extract the p-type Schottky barrier height and, after combining it with the gap width,
the n-type Schottky barrier height. Once again, the localized electronic state contributions are
directly visible in the spectra and their properties can thus be analysed (Margaritondo et al
776 G Margaritondo

Figure 6. (Continued.)
 Interface states at semiconductor junctions 777

1976).

2.2.2. Role of synchrotron light. The photon source is a major ingredient in all photoemission
techniques (Margaritondo 1995a). Until the late 1960s, only conventional photon sources were
available, with severe performance limitations. The advent of synchrotron sources was a major
factor in unlocking the potential capabilities of photoemission, in particular when applied to
the study of interfaces.
The progress in synchrotron sources over the past 30 years was marked by three subsequent
generations (Margaritondo 1995a). The most recent third generation includes facilities of
unprecedented brightness levels, primarily based on insertion devices such as the undulators
(Margaritondo 1995a). A fourth generation already appears on the horizon (Weyer and
Margaritondo 1995).
Each generation of synchrotron sources opened the way to new photoemission techniques
relevant to surface and interface research. In the historical sequence we note (Margaritondo
1988a, c): EDCs at different photon energies, constant-initial-state and constant-final-
state spectroscopy, partial-yield spectroscopy, photon-polarization techniques, angle-resolved
photoemission and band mapping, photoemission resonances, photoelectron diffraction,
ultrahigh-resolution spectroscopy and photoemission spectromicroscopy.
A detailed discussion of these techniques can be found in several books and reviews—see,
for example, Margaritondo (1988a, b, c, 1995a). Later, we will concentrate our attention on the
latest synchrotron-related developments relevant to semiconductor interface research—most
notably, laterally resolved studies of the fluctuations of interface parameters (Gozzo et al 1993,
1995).

2.2.3. The problem of surface photovoltage. However powerful, photoemission techniques

are not immune from problems. A few years ago, the discovery of surface photovoltage effects
(Hecht 1990, Waddil et al 1990, Alonso et al 1990, Mao et al 1990, 1991a, b, c, Margaritondo
et al 1980) in photoemission spectra caused substantial damage to the credibility of several
photoemission studies of semiconductor interfaces.
The illumination of a non-conductive surface with an intense photon beam can lead to
the creation of electron–hole pairs in the valence and conduction bands. The band bending
may separate these charges, prevent recombination and result in strong electrostatic effects that
modify the band bending. As a result, the measured energy positions of the electronic structure
features can be significantly affected by the measurement process itself, and the results can
become unreliable.
Such surface photovoltage effects (Hecht 1990, Waddil et al 1990, Alonso et al 1990)
do not always occur: they may be eliminated, for example, by a rapid surface electron–hole
recombination rate. On the other hand, a temperature decrease may reduce the recombination
rate and enhance the impact of surface photovoltage.
Quite luckily, spurious effects of this kind do not occur beyond the initial stages of the
interface formation process. In the latter stages, a significant fraction of the semiconductor
substrate is covered by a semiconductor or metal overlayer: the recombination rate becomes
fast enough to eliminate the problem. Thus, the vast majority of the results of photoemission
studies of semiconductor interfaces, which concern the final interface structure, are not affected
by surface photovoltage problems and are unquestionably reliable.
The surface photovoltage problems affected a specific group of studies which tried to
investigate the early stages of the interface formation process as a function of temperature,
As it turned out, many of the low-temperature results were affected by surface photovoltage
778 G Margaritondo

Figure 7. A schematic explanation of the BEEM technique (Kaiser

and Bell 1988, 1996, Prietsch 1995).

effects (Hecht 1990, Waddil et al 1990, Alonso et al 1990). Thereafter, studies of the early
interface formation steps have been considered with much scepticism, which, in most cases,
is not really justified.
In order to avoid uncertainties, we decided not to include the potentially affected studies in
this review. It should be noted, on the other hand, that surface photovoltage effects have been
exploited before as useful ingredients of specialized techniques—see Mao et al (1991a, b, c).
For example, intense illumination with a secondary source and the subsequent electron–
hole creation could lead to ‘band flattening’, i.e., to the complete or partial elimination of
the band bending. The comparison of spectra taken with and without illumination could
then directly lead to a quantitative evaluation of the band bending (Margaritondo et al 1980).
Unfortunately, one can never guarantee that the bands have become completely flat: when
shifting in the gap, the Fermi level may encounter localized states capable of pinning it, even
in the presence of strong illumination. Consequently, this approach has not yet reached an
acceptable level of reliability.
One should also note another problem that can potentially affect photoemission interface
studies: charging effects. In a metal connected to earth, the charge carried away by
photoelectrons is immediately restored—and no charging effects occur. In an insulator or
a semiconductor this may not be true, and the surface may become charged. This can cause
an electrostatic shift of the electronic structure and of the corresponding spectral features.
The situation is further complicated by the recently discovered fact that charging effects
can significantly change from place to place on a microscopic scale (Coluzza et al 1994).
When averaged in a non-laterally resolved experiment, such changes can simulate non-
existing lineshape features. Standard procedures to eliminate charging effects in photoemission
experiments do not appear capable of removing this problem. It is therefore prudent to conclude
that no safe and general cure exists for charging problems: the possibility of charging effects
must always be considered and should only be ruled out after careful tests and analysis.

2.3. BEEM
We already mentioned the importance of lateral resolution, which is discussed at length later
in this review. We would like to immediately introduce, however, the technique that, for the
first time, brought lateral resolution into this domain: BEEM (Kaiser and Bell 1988, 1996,
Prietsch 1995).
BEEM measurements are performed on ad hoc structures using a technique derived from
the scanning tunnel microscope (STM). In figure 7, for example, we see a thin (typically,
10–300 Å) metal film on a semiconductor substrate. The STM tip injects low-energy electrons
into the structure, 1–1.5 eV above the Fermi level of the metal. These electrons may or may
not be able to cross the Schottky barrier, depending on their energy, when they reach the barrier
and on the barrier height. If they do cross the barrier, then they can be detected as a current by
using a contact to the semiconductor substrate.
In this way, one can ‘map’ the Schottky barrier height with a lateral resolution that ap-
 Interface states at semiconductor junctions 779

proaches the STM levels without equalling them. Usually, BEEM measurements are accompa-
nied by standard STM measurements performed with a second contact to the metal overlayer.
Although apparently simple, a BEEM experiment is conceptually complex because the
phenomenon includes several different steps whose theoretical description is quite complex
(Kaiser and Bell 1988, 1996, Prietsch 1995). The steps specifically include tunnelling from
the tip to the metal film, propagation through the metal film and the crossing of the Schottky
barrier. Numerical simulation can reasonably handle the first and third step. The second
step is really a combination of different elastic and inelastic interactions with other electrons,
phonons, defects, the periodic lattice of the metal, etc. A complete description of all these
factors is not a simple task.
The major impact of BEEM experiments, in this authors opinion, has been the indication
that lateral fluctuations of interface barriers can occur. This paved the way to several other
laterally resolved experiments to analyse this problem, as we discuss later.
Among some of the recent results obtained with BEEM, we would like to note: the study
by Meyer and Von Kanel (1997) of individual point defects located at the CoSi2 –Si (111)
interface formed by thin silicide films epitaxially grown on silicon substrates—which gave
clear evidence for the trapping of point defects at dislocations; the extensive and detailed
study of Alx Ga1−x –As interfaces by Cheng et al (1997) and the study by O’Shea et al
(1996) of conduction band offsets in ordered-GaInP–GaAs heterostructures which investigated
heterostructures simultaneously grown on misoriented substrates to determine the effect of
ordering on the conduction band offset—and observed a lower conduction band offset for
highly ordered GaInP than for the less ordered sample. Other interesting results will be
discussed in section 6.2.3.

3. Older results and past controversies

The research on semiconductor interfaces took a critical turn in the early 1970s after a seminal
work by Kurtin et al (1970). These authors suggested a general correlation between the
Schottky barrier height of different interfaces and the chemical properties of the same interfaces.
Specifically, they described the dependence or independence of the Schottky barrier height on
the metal work function by introducing the so-called ‘S-parameter’, defined as the derivative
of 8SB with respect to (8M − χ).
In the extreme Schottky limit of equation (1), S would be equal to one, whereas in the
extreme Bardeen limit of equation (2) it would be equal to zero. Kurtin et al (1970) presented
a plot of the S-parameter as a function of the component electronegativity difference for a
large number of binary semiconductors. From such a plot, the authors derived a transition
from a Schottky regime in the case of ionically bound materials to a Bardeen regime for
covalent materials. A similar conclusion was later reached by Brillson (1994) using the heat
of formation rather than the electronegativity difference.
In the light of the subsequent theoretical work, the assumptions of Kurtin et al (1970)
appear oversimplified. But one cannot possibly overestimate the historical impact of this work:
it clearly raised the possibility that complicated systems like solid-state contacts, presumably
affected by a variety of complicated factors, could nevertheless exhibit simple and general
trends valid for many different semiconductor families.
This was an intriguing possibility with very interesting and fundamental implications.
Not surprisingly, it attracted the attention of many theorists and experimentalists of several
generations, and convinced them to participate to the research effort in semiconductor interface
research.
780 G Margaritondo

3.1. Identification of the general issues

The discussion of the historical impact of the work of Kurtin et al (1970) brings this review
to an important question. Rather than considering each interface as a separate problem, could
one hope to identify general properties valid for several different interfaces—or even for all
of them? It is necessary to guide our future discussion by clearly stating what the general
issues are in this field. These issues are the underlying problems that have been attacked by
thousands of experimental and theoretical studies spanning several decades.
In this authors opinion, the fundamental dilemma in semiconductor interface research is
the following:
Is it or is it not possible to express the interface parameters—such as the Schottky
barrier height and the heterojunction band discontinuities—as universal functions of
the properties of the two interface components?
One can simplify the initial discussion by adopting two extreme points of view, i.e.
(1) The Schottky barrier heights and the heterojunction band discontinuities only depend on
the parameters of the interface components.
(2) The Schottky barrier height and the heterojunction band discontinuities only depend on
the specific properties of the interface and cannot be directly and universally explained in
terms of the parameters of the interface components.
In order to understand the practical and fundamental impact of these two extreme points
of view, one should note the following additional points:
• These are, indeed, extreme cases: a realistic approach could be based on an intermediate
point of view, in which both the component parameters and the specific interface properties
play an important role. There could actually exist a continuum of practical situations
between the two extreme cases.
• The first extreme point of view would imply that is not possible to modify the Schottky
barrier height or the heterojunction band discontinuities for a given pair of materials. This
would thus imply that it is not possible to implement the so-called ‘bandgap engineering’
(Capasso and Margaritondo 1987)—that is, the tailoring of interface parameters to specific
needs for specific devices. In contrast, all other points of view—from the extreme case
of interface barriers that are independent of the component parameters to all intermedi-
ate cases—open the way to controlled barrier modifications and therefore to ‘bandgap
engineering’.
• It should be noted that the first extreme point of view does not necessarily rule out a role
for the interface properties. In fact, the interface properties themselves, in principle, could
be completely determined by the parameters of the two interface components.
In order to illustrate this last point, consider, for example, interface electronic states which
can pin the Fermi level, thus determining the interface barriers. The interface states can change
from interface to interface even if the components do not change; for example, different local
chemical phases could be formed depending on the specific interface formation process. One
could thus be tempted to conclude that, if the Fermi level is pinned by interface states, then
one can rule out a universal relation between the interface barriers and the parameters of the
interface components.
But this is not necessarily true: the relevant EF -pinning states could be directly linked to
the bulk electronic structures of the two interface components. For example, Tersoff (1984)
proposed the concept of MIGs (metal-induced gap states) for metal–semiconductor interfaces,
 Interface states at semiconductor junctions 781

assuming that the tailing of bulk metal wavefunctions produces a large density of interface-
localized states in the semiconductor gap. As a consequence, the Fermi-level pinning position
should be primarily determined by the so-called ‘midgap energy’ value of the semiconductor,
which is in fact a bulk property of one of the components of the interface. Similarly, in Tersoff’s
approach, the band discontinuity for a given pair of semiconductors should be determined by
their midgap-energy values (Tersoff 1984, Flores and Tejedor 1979, Margaritondo 1985).
There exist several versions of the first extreme point of view, ranging from the original
Schottky model (Schottky et al 1931, Anderson 1962) to the tight binding calculations
of absolute valence-band positions by Harrison (1977) and to other historically relevant
approaches (Wei and Zunger 1987, Frensley and Kroemer 1977). Similarly, there are several
versions of the second point of view (Brillson and Margaritondo 1988). The assessment of
the relative merit of each of the two extreme points of view and the discrimination between
them and between different versions of each of them was, for many years, a key objective of
semiconductor interface research.
Consider, in the first place, the discrimination between two such extreme points of view.
In the late 1970s and early 1980s it was realized that empirical tests are possible without even
needing to consider a specific version of each extreme point of view (Katnani and Margaritondo
1983). There exists in fact at least two classes of such empirical tests.
First of all, one can take a given metal–semiconductor or semiconductor–semiconductor
pair and form several different interfaces with such a pair under different conditions, for
example by artificially manipulating the local chemistry on a microscopic scale (Niles et al
1986). According to the first extreme point of view, the measured interface barrier values
should always be the same for all interfaces based on the same pair. If this is not true, then the
first extreme point of view is not valid.
The second class of tests is based on the following arguments. If the first extreme point
of view is valid, then one must be able to express the interface barriers for a given pair of
materials as a linear function of the interface barriers between each component of the pair and
a third material. For example, given three semiconductors S1, S2 and S3, the corresponding
valence band discontinuities should be linearly related by the equation:
1EV (S1–S2) + 1EV (S2–S3) + 1EV (S3–S1) = 0, (5)
so that given two discontinuities one can calculate the third one. Equation (5) is a specific
consequence, for example, of the Anderson–Schottky models (equation (3)) (Schottky et al
1931, Anderson 1962). Similarly, given two semiconductors S1 and S2 and a metal M, the
two n-type Schottky barriers and the conduction band discontinuity for the corresponding
interfaces should be linearly linked (Niles et al 1986) by the equation:
1EC (S1–S2) = 8SB (S1–M) − 8SB (S2–M), (6)
Note that equation (6) could be immediately derived from equations (1) and (3).
Several empirical tests in the two classes were conducted beginning in the late 1970s
(Katnani and Margaritondo 1983, Niles et al 1986). Their results can be summarized as
follows: the prediction of the first extreme point of view is valid only to a certain level of
accuracy (typically, 100–200 meV); beyond that level, significant deviations are observed.
For example, microscopic manipulations are able to significantly change the Schottky barrier
for a given metal–semiconductor pair (Niles et al 1986). Similarly, the ‘transitivity rules’
of equations (5) and (6) are only valid in the first approximation (Katnani and Margaritondo
1983), and the observed deviations are clearly beyond the experimental errors.
The conclusions of these empirical tests, therefore, are quite clear: first of all, the first
extreme point of view is not strictly valid, and the interface barriers for a given pair of materials
are not completely determined by the parameters of the materials. Nevertheless, the empirical
782 G Margaritondo

tests of the first extreme point of view are not a complete failure, since the observed deviations
do not typically exceed 100–200 meV. Thus, one cannot simply replace the first extreme point
of view with the second.
It appears reasonable to conclude in favour of an intermediate point of view: Schottky
barriers and heterojunction band discontinuities are determined in the first approximation by the
component parameters—but they are also significantly affected by specific interface properties
which are not directly linked to the component properties. This point of view appears to agree
with the most recent theoretical thinking (Peressi et al 1998), and opens the way to Capasso’s
‘bandgap engineering’.

3.2. More recent experimental results and evolution in theory

Perhaps the most positive recent development in semiconductor interface research is the
progress in theory. The theoretical picture is much simpler than in the early days of this domain.
Achieving this simplification was a difficult task which required more than two decades of
experimental and theoretical work. Here, we briefly review some landmarks of this evolution;
the reader should consult the work of Peressi et al (1998) for a more detailed discussion.

3.2.1. First steps beyond the early models. Roughly speaking, the theoretical approaches to
semiconductor interfaces fall in three groups:
(1) Theories which are based—at least in the first approximation—on the first extreme point
of view of the previous section, and which try to identify and calculate the component
material parameters that determine the interface barriers.
(2) Theories which try to model specific interface effects and their impact on the interface
parameters.
(3) Approaches based on the empirical identification of general trends, most notably those
related to the chemical properties of the interface components (Brillson and Margaritondo
1988).
As to the first group, there were several early attempts to identify an ‘absolute’ energy
for each material, and to use this ‘absolute’ energy for calculating interface barriers. We can
note, for example, the pioneering work of Frensley and Kroemer (1977) and the already
mentioned tight-binding approach of Harrison (1977). Note that the Schottky–Anderson
models themselves (Schottky et al 1931, Anderson 1962) belong to this group.
The first group also includes empirical efforts which tried to predict interface barriers
without a full theoretical framework. For example, Katnani and Margaritondo (1983)
developed an empirical table of ‘absolute’ energies based on measured barriers of different
materials with respect to Si and Ge. This table did not explain the nature of the empirical
‘absolute’ energies, although they were later related (Margaritondo 1985) to Tersoff’s midgap
energies (Tersoff 1984, Flores and Tejedor 1979). The approach was moderately successful
in predicting barrier values; on the other hand, its accuracy limitations were being primarily
related to the general limitations of the first extreme point of view.
Realistically modelling specific interface properties is a more complex task. Such a task
could only be tackled after the hardware and software for numerical interface modellization
became available (Peressi et al 1998). Among the early efforts in this group of theories (Pickett
and Cohen 1978, Van de Walle and Martin 1987) we note the ‘effective work function model’
by Freeouf and Woodall (1986). This theory analysed the role of interface chemical phases by
assuming the Schottky model as the starting point, but then utilizing the work function values
of the interface phases for barrier calculations.
 Interface states at semiconductor junctions 783

The aforementioned work by Kurtin et al (1970) is a nice example of the efforts in the
third group of theories. This work inspired several scientists to search for general trends.
Discussions of these attempts can be found in the reviews of Brillson (1992, 1993).
Among the efforts along this direction, we would particularly like to note in particular
the work by Brillson et al (1990): from cathodoluminescence experiments, these authors
inferred that by controlling the atomic-scale interface chemistry one can affect the creation
of deep levels near the junction—thereby altering the EF pinning position as discussed in
section 5. This conclusion was then corroborated by photoemission spectroscopy and internal
photoemission measurements of barrier heights; examples include metals on clean, ordered
GaAs, InP and CdTe and InAIAs–InP heterojunctions.

3.2.2. The defect model. In the late 1970s, a major event polarized the attention of scientists
active in semiconductor interface research: the proposal by Stanford scientists (Spicer et al
1979) that the interface parameters could be entirely explained by the effects of interface
defects. This proposal stimulated a very strong interest and caused much controversy over
more than a decade.
Its conceptual background can be traced back to some serious difficulties affecting
Bardeens (1947) proposal about the role of localized states and the Fermi-level pinning at
metal–semiconductor interfaces. Localized states had indeed been observed (Wagner and
Spicer 1972, Eastman and Grobman 1972, Chiarotti et al 1971) for clean semiconductor
surfaces, and many scientists interpreted the Bardeen hypothesis by attributing the EF -pinning
to the clean-surface states. This would imply that the clean-surface states are somewhat able
to survive during the interface formation.
In spite of the arguments that were presented to justify this assumption, it was hard to
believe that the interface formation process would allow the survival of the clean-surface
electronic structure. A ‘real’ interface formation process is a rather messy affair, influenced by
a large number of parameters and conditions—many of which are not even controlled. How
could one, then, hope to shield a tiny minority of states from its effects? How could one, in
fact, hope to find any general property in the products of such messy fabrication procedures?
Yet, general properties could be observed; for example, the Schottky barrier height of
covalent semiconductor appeared, at least in the first approximation, almost independent of
the metal and of its work function. Before the advent of surface-sensitive experiments, one
could guess that the semiconductor surfaces before interface formation were very contaminated,
and that their contamination—which was independent of the subsequent interface formation
process—determined the final interface properties. But this ‘realistic’ point of view could not
explain the general trends observed with surface sensitive techniques on ultraclean interfaces
(Brillson 1992, 1993).
The hypothesis of the defect model by Spicer et al (1979) was that the ‘universal’ factor
determining the interface barriers was the formation of interface defects. The interface barriers
were, in fact, explained as being due to Fermi-level pinning by the localized defect states.
General trends could be justified by the corresponding general trends in the defect properties.
Many experimental results were used to argue either in favour or against this point of
approach, which polarized the attention of the entire field for several years. The conclusion, in
this authors opinion, is as follows: the defect model failed to provide the ‘unified’ or ‘universal’
theoretical framework to treat semiconductor interfaces, since it did not take into account other
factors whose importance in determining interface properties is now solidly demonstrated—as
discussed later.
On the other hand, the defect model had a fundamental merit: it stressed the unquestionable
fact that local defects play an important role in determining the semiconductor interface
784 G Margaritondo

properties. The repercussions of the defect model can be noticed even in the more recent
and most sophisticated theoretical approaches (Peressi et al 1998). For example, numerical
calculations of the interface electronic structure routinely adopt several different atomic
configurations for each interface, following the basic philosophy of the defect model.

3.2.3. The MIGs model. The major problem of the defect model was the same one that affected
the notion of EF -pinning by surviving clean-surface states. Consider a metal–semiconductor
interface: a dominating role of local defect is quite plausible at the early stages of interface
formation, when the semiconductor is covered by a submonolayer of metal atoms.
However, as the interface formation process continues one introduces a large density
of metal-related electronic states at energies corresponding to the forbidden gap of the
semiconductor. In the long run, the density of these metal-induced states becomes much
larger than the density of local defect states. It was quite reasonable, therefore, to shift ones
attention from defects to metal-induced gap states (MIGs).
This was stimulated by the two already mentioned seminal works of Tersoff (1984) and
Flores and Tejedor (1979). How do the bulk metal states interact with the semiconductor after
the interface is formed? Tersoff argued that they tail into the semiconductor gap, giving rise
to a high density of localized electronic states—the MIGs—on the semiconductor side, and
determining both the position of the Fermi level in the gap and the Schottky barrier height. The
MIGs are also supposed to screen out and thus reduce the importance of specific phenomena
occurring at the interface, such as the formation of local defects.
The MIGs can be constructed using the bulk states of the semiconductor as an eigenfunc-
tion basis. At least in the first approximation, the EF position in the semiconductor gap must
coincide (Tersoff 1984) with the energy for which the basis content of the MIGs changes from
primarily valence-band states to primarily conduction-band states. This energy was called the
‘midgap energy point’ by Tersoff, and is related to the concept of the charge-neutrality point
independently developed by Flores and Tejedor (1979). Thus, the Schottky barrier height
should be determined, at least in the first approximation, by the midgap energy point of the
semiconductor.
The concept of midgap energy can be extended to heterojunction interfaces, based on the
tailing of valence or conduction band states of one semiconductor into the gap of the other,
beyond the band discontinuities. The band lineup and the band discontinuities are determined,
again at the first approximation, by the alignment of the midgap energy points of the two
materials. The predictions of this approach (Tersoff 1984) were found (Margaritondo 1985)
to be consistent with the band discontinuity trends observed by Katnani and Margaritondo
(1983)—lending support to the conceptual background of the MIGs model.
Tersoff’s MIGs model and the related work of Flores and Tejedor (1979) had the
fundamental merit of putting on solid conceptual ground the intuition (Frensley and Kroemer
1977, Katnani and Margaritondo 1983) that each semiconductor possesses an absolute energy
reference level. Specifically, it replaced the rather naive concepts of the Schottky model
(Schottky et al 1931) with the theoretically sound notion of midgap energy. It was, in a sense,
as far as the idea of justifying the interface barriers by using bulk material properties could go.

3.3. After the controversies, a general picture emerges—the wonders of ab initio calculations
The appearance in the late 1970s and early 1980s of the defect model and of the MIGs model,
together with the continuing search for general chemical trends, led to a period of extreme
and complex controversy. When the dust settled, it became clear—for example, on the basis
of empirical tests of the type described in section 3.1—that none of the two extreme points
 Interface states at semiconductor junctions 785

of view about the interface barriers could be valid. And it also became clear that neither the
MIGs nor the defect-induced states could be ruled out as important barrier-influencing factors.
A more realistic attitude emerged: many different factors, including MIGs and defect
states, can play a role and must be taken into account as potentially important. Their relative
importance depends on the characteristics of the interface and, in particular, on the details of
the preparation process.
A major factor facilitating this realistic point of view was the development of ab initio
calculations of the interface electronic structure (Peressi et al 1998). Energy-minimization
approaches actually reduce the need for experimental input: the theory can directly identify
the most probable atomic configuration for each interface. Ab initio calculations can attack
virtually every issue of semiconductor interface physics, and, in particular, explain the nature
of interface parameters.
The ab initio methods described by Peressi et al (1998), based on linear response theory
concepts, are an excellent example of this relatively novel and powerful approach. The best
way to illustrate this point is to reconsider the previously stated basic question in semiconductor
interface research: do or do not the interface barriers depend only on bulk material properties?
The ab initio approach finally provides a clear-cut answer. The answer is positive for the
band discontinuities of lattice-matched isovalent heterojunctions: 1EC and 1EV only depend
on bulk material properties for these interfaces. Apart from this one case, the specific interface
properties also play a significant role.
In particular, the band discontinuities of lattice-matched heterovalent heterojunctions—
including simple systems like GaSe–Ge—crucially depend on the interface orientation and on
other details of the microscopic interface structure. This opens up the possibility of modifying
and tuning the discontinuities, i.e., to implement ‘bandgap engineering’.
In the case of lattice-mismatched heterojunction interfaces, one must consider the effects
on the discontinuities of the interface strain (Cardona and Christensen 1987, Ohler et al 1998).
The conclusions are quite intriguing: the strain strongly affects the bulk contribution to the
discontinuities and only to a much smaller extent the interface contributions. In that sense,
the variations of the band discontinuities with strain are mainly a bulk effect. For isovalent
interfaces, it is possible to treat the effects of strain by simple correction, as discussed by
Peressi et al (1998).
The situation is more complex for heterovalent interfaces for which interface effects are, as
we have seen, non-negligible. Ab initio calculations analysed both strain effects and ‘chemical’
effects, i.e., the variations of the band discontinuities with the interface termination. The results
show that both the microscopic strain effects and the local ‘chemical’ effects are non-negligible.
The problem of metal–semiconductor interfaces is somewhat more complex, in particular
because of the variety of interface morphologies. Ab initio calculations primarily treated
epitaxial systems, which are to some extent a counterpart of lattice-matched heterojunctions.
The results (Bardi et al 1996, Berthod et al 1996) for the GaAlAs–Al interface, and, in
particular, the following two points are quite important.
First of all, the dependence of the Schottky barrier height on pressure. This was a critical
issue for the following reason. We have seen that empirical tests, such as the transitivity
rule of equation (6), yield results that are only approximately positive—and that significant
deviations are observed. Several authors, however, considered the deviations as negligible and
the tests as positive evidence of some kind of universal mechanism, as previously discussed.
The top candidates were, as we have seen, EF -pinning by MIGs or EF -pinning by a specific
and universal type of defect states.
In order to discriminate between MIGs pinning and defect pinning, Shan et al (1988)
and Dobaczewski et al (1993) proposed to analyse the pressure dependence of the band
786 G Margaritondo

Figure 8. Composition dependence of the (p-type) Schottky

barrier height for anion-terminated and cation-terminated interfaces.
The lines are theoretical results by Berthod et al (1996). The
experimental points are from Revva et al (1993) (closed circles, open
and closed squares) and from Missous et al (1990) (open circles).

discontinuities. The results seemed to favour defect pinning instead of MIGs pinning. But
this conclusion was proven unfounded (Peressi et al 1998) by the ab initio calculations: the
pressure dependence could be justified without invoking any kind of defect pinning.
On the other hand, ab initio calculations (Berthod et al 1996) provided a much stronger
argument against any kind of ‘universal’ mechanism, at least for the GaAlAs–Al interface:
the calculations yielded different Schottky barrier heights for cation-terminated and anion-
terminated GaAlAs(100)–Al interfaces (see figure 8, with the results of Berthod et al (1996)
and some experimental points from Revva et al (1993) and Missous et al (1990)), in sharp
contrast, for example, to the predictions of the MIGs approach (Tersoff 1984).
This also demonstrates that the conclusions valid for lattice-matched isovalent
heterojuctions cannot be automatically applied to epitaxial metal–semiconductor interfaces:
whereas the band discontinuities in the first case only depend on bulk semiconductor properties,
the Schottky barrier in the second case is not a bulk semiconductor property and depends instead
on interface-specific features. On the other hand, the variations of the Schottky barrier height
with pressure and compositions are primarily due to bulk mechanisms (Peressi et al 1998).
Unfortunately, these conclusions cannot be extended beyond the somewhat idealized case
of epitaxial metal–semiconductor interfaces. Specifically, there is no reason for ruling out a
priori an important role of defects in non-epitaxial and/or reactive interfaces, even if defects
are not needed to explain the idealized systems explored by ab initio calculations. It remains
to be seen how much these calculations can be extended beyond such idealized systems.
In summary, ab initio calculations provided clear-cut answers for some of the most
controversial issues in semiconductor interface science. However, many questions still remain
for the systems for which this approach is difficult or perhaps impossible—and theoretical
efforts continue in this and other directions (Ossicini and Bernardini 1992). The excellent
results obtained so far certainly justify the effort to apply ab initio methods to more complicated
and more realistic interfaces.

4. The search continues: recent results

The remaining unsolved issues stimulate not only additional work in theory, but also much
experimental activity. The next section of this review is dedicated to the specific and crucial
problem of the lateral variations of energy barriers, which is at present the subject of many
studies. In the present section, we review some recent interesting experimental results in other
directions.
 Interface states at semiconductor junctions 787

4.1. Investigating fundamental issues

Throughout its history, semiconductor interface research had two different motivations: finding
solutions for technological problems and clarifying fundamental issues. This ‘twin personality’
still exists, and a substantial amount of research work continues to be inspired by fundamental
scientific issues. For example, the interface-control efforts described in section 5, although
primarily motivated by the technological interest, also constitute tests to better understand the
nature of the interface barriers. Similarly, the work on lateral fluctuations of interface barriers
not only touches on technological issues but also on fundamental problems concerning the
nature and behaviour of semiconductor interfaces.
Among the recent experiments which attacked fundamental issues, we would like to
mention the work by Sorba et al (1993) on AlAs–Ge–GaAs(001) and GaAs–Ge–AlAs(001)
single quantum well structures. These authors demonstrated that the observed deviations
from the commutativity and transitivity rules of the heterojunction band offsets are consistent
with the establishment of inequivalent, neutral IV/III–V and III–V/IV interfaces, providing an
interesting additional input for the general theoretical predictions based on ab initio calculations
(see the previous section).
Among the many experiments dedicated to EF -pinning issues and other related
fundamental interface problems, we note the research line by Soukiassian and his co-workers
(Bonnet et al 1993, Aristov et al 1995, Spiess et al 1996 and the references therein) on alkali
metal chemisorption. These authors argued, for example, that the p-GaSb(110)–Rb band
bending could be explained by the existence of donor levels located at about 0.3 eV above the
valence-band maximum.
On the other hand, the general issue of pinning or not was attacked by the photoreflectance
and photoluminescence experiments of Alperovich et al (1995). For oxygen and caesium
chemisorption on GaAs, these authors found a generally unpinned behaviour not consistent,
for example, with a universal application of the defect model.
Finally, we would like to note that even for the most intensively investigated systems
fundamental results continue to be obtained—primarily due to improvements in the
instrumentation. One good example is the recent work by De Padova et al (1998) which
analysed with extremely high resolution the Si2p photoemission peaks at silicon surfaces and
interfaces (an example is shown in figure 9). This is one of the most extensively studied
spectral features; nevertheless, De Padova et al (1998) found that the intrinsic linewidth was
overestimated by previous studies, and a revision of the Si2p lifetime was required.

4.2. The initial stages of interface formation

The unfortunate problems caused by surface photovoltage (Hecht 1990, Waddil et al 1990,
Alonsoet al 1990) had a noticeable and negative impact on efforts focusing on the early stages
of interface formation. Nevertheless, many recent studies deal with this problem; one may
hope that the surface photovoltage issue is now definitely a thing of the past.
The most intriguing, in this authors opinion, is the theoretical analysis by Flores and
his co-workers (Flores et al 1997), of possible many-body effects during the initial stages of
interface formation. In particular, the predicted appearance of Kondo-like spectral features
in the early stages of interface formation between certain alkali metal overlayers and III–V
substrates is an intriguing and still unclarified issue.
In the same general area, the aforementioned work on alkali metal overlayers by
Soukiassian and co-workers (Bonnet et al 1993, Aristov et al 1995, Spiess et al 1996) is largely
dedicated to the early interface formation stages. We would like to mention, in particular, the
788 G Margaritondo

Figure 9. An example of the results that are made Figure 10. An example of intralayer-induced mod-
possible by the recent instrumentation advances: very ifications of Schottky barrier height. In this case,
high resolution Si2p photoemission spectra obtained the Au–GaAs Schottky barrier (8)—measured from
on the synchrotron source Elettra (Trieste) for a Si the zero-current extrapolation of internal photoemission
surface covered by one monolayer of antimony and 2.5 (photocurrent) yield curves—varies when a thin silicon
monolayers of germanium by De Padova et al (1998). nitride layer is inserted at the interface. Results from
Almeida et al (1997).

use Spiess et al 1993) of photoemission extended x-ray fine structure (EXAFS) (Margaritondo
and Stoffel 1979) to probe the local atomic structure. Also interesting in the general domain of
alkali metal overlayers is the field-ion scanning tunnelling microscope study of Na deposition
on the GaAs(110) surface by Sakurai’s group (Bai et al 1993).
Much of the experimental work on the early stages of interface formation is focused on
group-IV substrates. We note, in particular, the discovery by Weitering (1996) of new Ba-
induced reconstructions of Si(111); scanning tunnel microscopy studies like that of Dong et al
(1997) for In on Si(100) and of Gothelid et al (1995) for Sn on Ge; photoemission studies of
binary-semiconductor overlayers on Si, notably by Williams’ group for CdTe (Bennett et al
1996). As for other substrates, we note the work of Chen et al (1994) on Au on p-ZnSe(100).

4.3. Technology-oriented work

The borderline between ‘fundamental’ work and ‘technology-oriented’ work in semiconductor
interface research is not well defined: as we have already mentioned, many technology-oriented
studies also touch fundamental issues. One can note, on the other hand, a recent tendency to
extend the studies from the traditional ‘simple’ interfaces to more complex systems, whose
selection is clearly and primarily justified by technological interest—see, for example, Angelo
et al (1995), Horng et al (1992) or Klein et al (1994). One should reserve judgement on this
generally positive tendency: this author believes that, in the long run, the quality assessment
will be primarily based on the capability of experiments to contribute to the clarification of
fundamental issues.
Quite interesting from the technological point of view is the use of intralayers (Katnani
et al 1982, 1983, Faraci et al 1994) to enhance oxidation and passivation processes, including
 Interface states at semiconductor junctions 789

nitridation. Spectacular results in this domain were presented by Soukiassian and co-workers
(Soukiassian et al 1985, Starnberg et al 1992). In the general domain of passivation one should
note the experiments of Riehl-Chudoba et al (1994) and of Huttel et al (1996).
A major development in recent years has been the growing interest on organic semiconduc-
tors. This would be a suitable area for a separate review; we regret to be unable to do more than
just cite some of the recent literature—see Hirose et al (1994, 1996, 1997), Kendrick et al (1996)
and Wu et al (1997). Other areas of growing interest are those of diamond-based interfaces—
see, for example, Pickett and Erwin (1990)—and SiC-based interfaces (Semond et al 1996).
Also technologically inspired is the work on ohmic contacts and contact problems in
general (Lazzarino et al 1996a, b, Malacky et al 1994). Finally, one should mention the renewed
interest in interfaces for high-photon-energy visible-light emitters—and the remarkable
progress both in the technology of the corresponding devices and in their understanding—
see, for example, Herve et al (1995a, b) and Bonard et al (1997a, b).

4.4. New experimental techniques and new applications

Since the very beginning of semiconductor interface research, this domain has profited from
the development of new and advanced instruments. We discuss at length, in section 6, the
impact of laterally resolved experiments which are related to novel experimental techniques
such as BEEM and photoelectron spectromicroscopy. In the present section, we mention some
other noteworthy developments concerning the experimental techniques.
We note, for example, the renewed effort to develop ‘capping’ and ‘decapping’ procedures,
with the objective of being able to transfer samples from an ultrahigh-vacuum preparation
system to a different analysis system also under ultrahigh vacuum. Although not completely
new (Le Lay et al 1991), this technique was enhanced, in recent years, by careful fine-tuning of
new approaches (Karpov et al 1995) and an extension to new interfaces (Dumas et al 1992, Yu
et al 1997). In practice, the ‘capping–decapping’ method opens up interfaces that are prepared
with the most sophisticated deposition techniques for advanced analysis, without the need for
an integrated fabrication-analysis apparatus.
Another noteworthy area is that of optical techniques. They were actually one of the
starting points of semiconductor surface and interface research (see Chiarotti et al 1971),
but were later overshadowed by surface-sensitive electron spectroscopy and microscopy
techniques. However, interesting efforts with optical techniques are still underway. We note,
for example, the combined reflectance–photoemission approach of Gusev et al (1997) and the
fast time-resolved experiments such as that of Christianen et al (1994). We would also like
to mention the surface vibrational experiments based on electron energy-loss spectroscopy
(Margaritondo and Rowe 1986, Margaritondo and Weaver 1985); they did not reach the
status that had been projected by some authors in the late 1980s, but do provide interesting
contributions to semiconductor interface research (Scamarcio et al 1992).
Finally, we would like to mention techniques capable of investigating deep electronic
levels, thereby establishing a bridge from surface-sensitive studies to bulk experiments—
and also to real device interfaces. Once again, these are not entirely new techniques, but,
in recent years, they have been substantially renewed, becoming capable of yielding novel
contributions. We note, in particular, the studies of the deep-level electronic structure of
ZnSe/GaAs heterostructures by Raisanen et al (1995a, b, c).
790 G Margaritondo

5. From understanding to controlling

This section deals with a fundamental development which turn took place (Margaritondo 1986)
in the early 1980s: the attempts to actively modify interface parameters. The background
was the following. On the one hand, surface-sensitive experimental techniques had revealed
that semiconductor interface barriers are very complicated—making it more difficult to
theoretically treat them. On the other hand, the realization that interface barriers depend
on many factors suggested the possibility of modifying them in a controlled way.
This was an extraordinarily interesting idea for technological applications: in most cases,
the ‘natural’ interface barriers are not optimized for specific device applications. We would
also like to recall the aforementioned fundamental implications of the successful efforts to
modify interface barriers. They dealt, in fact, a fatal blow to the idea that interface barriers are
only determined by the parameters of the two interfaces, components, and independent of the
specific interface properties.

5.1. Empirical approaches: ultrathin intralayers

The first steps in the direction of artificially modifying interface parameters were purely
empirical. In 1979–80, Brillson proposed using thin intralayers to modify the interdiffusion and
chemical processes at metal–semiconductor interfaces (Brillson et al 1980a, b, 1981). Brillson,
being a leading advocate of the theories which emphasized the chemical factors in the barrier
formation mechanism, also implied the possibility of modifying the interface barriers.
The empirical tests of this suggestion were very successful: it was quite clear that
semiconductor interfaces could be manipulated by acting on the local composition. This
approach was shortly afterwards extended from Schottky barriers to heterojunction interfaces
(McKinley et al 1991 and references therein).
In subsequent years, efforts to control interface barriers were expanded in many different
directions. We note two important milestones: first, Capasso’s introduction of the concept of
‘bandgap engineering’ (Capasso and Margaritondo 1987), which oriented the efforts towards
specific objectives in device technology. The second was the use of realistic numerical theories
(Peressi et al 1998) of semiconductor interfaces to predict specific mechanisms for controlled
modification of the interface parameters. These theories directed many of the subsequent
experimental efforts, thereby providing a suitable alternative to purely empirical approaches.

5.2. Advanced theoretical gap engineering and its practical implementation

The transition from the empirical stage to efforts based on sound theoretical planning pro-
duced successful results in three different areas: the controlled modifications of Schottky
barrier heights, the controlled modification of heterojunction band lineups and the creation of
artificial band discontinuities at semiconductor homojunctions. We briefly review here some
of relevant work.

5.2.1. Modifications of Schottky barriers. After the early successful attempts to modify
Schottky barriers, several other authors extended them to a variety of metal–semiconductor
interfaces. In 1992, Williams’ group (Spaltmann et al 1992) was able to modify the
Ag–GaAs(110) n-type Schottky barrier by inserting thin Mn intralayers. I –V and C–V
measurements demonstrated that the barrier changed from 0.92 eV for pure Ag (Mn interlayer)
to 0.76 eV for a 15 Å thick Mn intralayer (in comparison, the barrier for pure Mn on GaAs(110)
 Interface states at semiconductor junctions 791

is 0.74 eV). For interlayers above 1 Å the barrier height was found to decrease exponentially
with increasing Mn thickness.
In the same year, Hwu et al (1992) extended, for the first time, the Schottky barrier control
to the case of semiconductor overlayers on metal substrates. The impact of the interface
preparation technique was assessed for Al–GaAs(100) by Vitomirov et al (1992). These
authors found that modifications in surface chemical composition and reconstruction with
annealing temperature produce systematic changes in a set of interface states spanning the
energy range ≈ 0.8–1.2 eV, and also alter the interface barrier height.
Then, the same group extended the approach to other GaAs-based interfaces (Vitomirov
et al 1993) and found, for example, that for Au/GaAs(100) contacts the interface Fermi-level
position shows little sensitivity to either substrate growth technique or the type of doping, and
lies in the 0.37–0.47 eV range above the valence band maximum. In contrast, at Al/GaAs(100)
interfaces, it is highly sensitive to substrate growth method for n-type GaAs, but shows no
significant difference between the epitaxial and melt-grown p-type GaAs.
In 1993, Kolnik and Ivanco analysed the barrier modifications in the case of metal–
semiconductor contacts with plasma deposited silicon nitride interfacial layers. Tungsten and
titanium on both p-type and n-type silicon and gold on n-type gallium arsenide were analysed,
and the observed shift of the Schottky barrier heights towards the Schottky limit was explained
in terms of semiconductor surface passivation.
Shortly afterwards, Dell’Orto et al (1994b) successfully used calcium fluoride intralayers
to modify the Au–Si Schottky barrier height. Quite recently, the same group extended their
investigations of this system to the use of silicon nitride intralayers—see figure 10 in Almeida
et al (1997)
Significant differences in the Ag–Si(111) barrier height were found in 1995 by the group of
Mönch (Schmitsdorf et al 1995) for different Si(111) surface reconstructions—thus suggesting
another possible way to manipulate the interface parameters. A Schottky barrier control method
based on hydrogenation was proposed and successfully tested by Hara et al (1996) for silicon
carbide substrates.
One important issue in the techniques for modifying and controlling interface barriers is
stability: are the engineered interfaces stable and therefore suitable for real devices? This issue
was analysed for Al–Si–GaAs(110) structures by Sorba et al (1996). The results indicated
that substantial Si-induced local interface dipoles persisted after annealing, for annealing
temperatures up to 450 ◦ C. Diodes engineered to achieve an increased barrier heights remained
stable for annealing temperatures up to 300 ◦ C; a gradual degradation in the 350–450 ◦ C range
corresponded to Si dissolution in the Al overlayer. Schottky diodes engineered to achieve
low barrier heights were less stable, and exhibited a gradual increase in the barrier with the
annealing temperature.

5.2.2. Modifications of heterojunction band lineups. The earliest report of controlled

modifications of band lineups at heterojunction interfaces appeared in 1985 (Niles et al 1985).
Shortly afterwards, Perfetti et al (1986) presented huge modifications for the SiO2 –Si interface
which definitely established the feasibility of using intralayers to alter interface parameters—
see figure 11.
Empirical approaches dominated the subsequent years, yielding remarkable results (Niles
et al 1986, 1988, McKinley et al 1990, Margaritondo et al 1992). Gradually, however, a
theory-driven approach emerged. In parallel, there were quite a few examples of interface
control by ultra-localized ‘delta doping’ (Almeida et al 1995). For example, Shen et al (1992)
presented a successful delta-doping approach based on Be to control the band lineup for the
InAs–GaAs interface.
792 G Margaritondo

Figure 11. A rather spectacular case of intralayer-induced modifications of

heterojunction valence band discontinuities, from Perfetti et al (1986). Near-edge
photoemission spectra reveal H or Cs intralayer-induced changes of the silicon–silicon
dioxide discontinuity by up to 0.5 eV.

The efforts to control heterojunction band lineups were considerably expanded very
recently, using other innovative approaches and extending the efforts to new interfaces.
Ceccone et al (1991) announced the successful use of thin Si intralayers to change the band
lineup of the binary–binary interface AlAs–GaAs. The same group achieved a similar result
with Ge intralayers for the ZnSe–GaAs interface (Bratina et al 1993). These investigations
continued in subsequent years, yielding a series of additional important results (Pellegrini et al
1996); specifically, a 0.26–0.75 eV tunability of the ZnSe–GaAs conduction band offset was
reported by Lazzeri et al (1997).
The theoretically-driven approach for modifying band discontinuities is based on
numerical calculations of the interface electronic structure with and without an intralayer
(Peressi et al 1998). One variation of this approach is the use of a binary intralayer which
introduces an artificial dipole and thus changes the band lineup. A description of this approach
is reported, for example, by Wilks and Williams (1995).

5.2.3. Creation of band discontinuities at homojunctions. The transition from the purely
empirical approach to heterojunction band lineup modifications to a method guided by
theory led, in 1991, to a particularly significant achievement: the creation of artificial band
discontinuities at homojunction interfaces (McKinley et al 1992). In order to understand
the mechanism, we must analyse the possible effects on already existing heterojunction
discontinuities of a thin intralayer.
In essence, the insertion of a thin intralayer between the two sides of a junction can
accomplish four things. First, it can act as a barrier against microdiffusion—or trigger it.
Secondly, it can allow and/or stimulate microchemical reactions which lead to the formation
of new interface microphases. Thirdly, it can provide atoms to replace individual atoms at the
two sides of the interface, either by filling vacancies or by exchange interactions (note that the
distinction between such reactions and microchemical processes is rather narrow). Fourthly, it
can simply add new chemical bonds at the interface—the bonds between the intralayer atoms
and the atoms at the two sides of the junction.
Each one of these mechanisms, which can also coexist together, can lead to a redistribution
of the interface electronic charge—and to the modification of the interface dipole. This is, in
essence, the mechanism changing the heterojunction band lineups.
This analysis also leads to an obvious suggestion: by inserting a double intralayer, one
can create a dipole where it does not exist, i.e. at a homojunction interface—see figure 12. This
suggestion was transformed into reality in 1992—as seen in figure 13 for Ge–Ge homojunctions
 Interface states at semiconductor junctions 793

Figure 12. A schematic diagram illustrating the Figure 13. Application of the method illustrated in
creation of artificial band discontinuities at a Ge– figure 12 to a Ge–Ge homojunction. The photoemission
Ge homojunction. Top: the homojunction has no spectra of the Ge3d level reveal the position of the band
discontinuities. Bottom: the insertion of an As– structure of the Ge substrate and that of the Ge overlayer.
Al double intralayer creates an interface dipole and Before depositing the Ge overlayer, a double intralayer
discontinuities in the conduction and valence bands. consisting of one monolayer of As and one monolayer of
Al was evaporated on the Ge substrate. Consequently,
the spectra reveal a clear shift between the substrate
and overlayer electronic structures. The corresponding
artificial conduction and valence band offsets are 0.4 eV.
Results from Marsi et al (1992).

(McKinley et al 1992, Marsi et al 1992, Dell’Orto et al 1994a).

Such results substantially expanded the potential boundaries of bandgap engineering. But
they also revealed some limitations in the theoretical reasoning that guided the creation of
homojunction band discontinuities.
In fact, by adding several double intralayers, a simple electrostatics reasoning would
predict an increase in the artificial homojunction band discontinuity. Intuition suggests that
this cannot be true. And in fact, the increase was observed to saturate (McKinley et al 1992,
Marsi et al 1992, Dell’Orto et al 1994a) after the first double intralayer; the causes of the
saturation, however, are not yet clear.

6. The new frontiers: lateral variations, free-electron-laser techniques

This section is primarily dedicated to one of the most important new avenues in semiconductor
interface research: the investigation of the lateral variations of interface properties. Until
recently, this problem was almost ignored, largely because of the lack of experimental results.
Even now, the lateral fluctuations of the interface barriers are not even mentioned in the
standard textbooks on microelectronic devices—and are ignored by most theoretical models
of semiconductor interfaces.
Space averaging was, in fact, and until quite recently, one of the major limitations of the
experimental studies of semiconductor interfaces. The transport techniques and the electron
spectroscopy experiments were oblivious to microscopic-scale properties and were averaged
over large portions of the interface. In the case of photoemission-based techniques, this
794 G Margaritondo

limitation was overcome with the recent development of spectromicroscopy (Gozzo et al

1993, 1995, Margaritondo and Cerrina 1990, Margaritondo 1995b, 1997, Ade 1997, Almeida
et al 1996, Margaritondo et al 1995, Margaritondo and Hwu 1996), the combination of
photoemission spectroscopy and high lateral resolution. In parallel, techniques like BEEM
(Kaiser and Bell 1988, 1996, Prietsch 1995) also broke the lateral-resolution barrier, providing,
for the first time, a picture of the interface properties on a microscopic scale.

6.1. The issue of lateral variations

How important is it to analyse semiconductor interfaces on a local scale? The answer from
the most recent experiments is quite clear: it is extremely important.
In sharp contrast to this conclusion, most studies of semiconductor interfaces still ignore
this crucial issue. In general, they implicitly adopt the assumption that each interface
barrier is a constant characteristic of the entire interface. Concepts like ‘Schottky barrier’
or ‘heterojunction band discontinuities’ are used assuming that such quantities are constant for
the entire metal–semiconductor or semiconductor–semiconductor interface. This assumption
is fundamental to virtually all models of semiconductor devices.
This assumption, however, is potentially very dangerous. Let us imagine that it is wrong;
let us assume, for example, that a given metal–semiconductor barrier height fluctuates from
point to point. A device model based on an ‘average’ barrier would be completely misleading,
since the barrier exponentially affects the most important interface properties, and therefore
the weakest-barrier regions dominate the device behaviour.
Is there, on the other hand, a conceptual basis for assuming that the barriers do not fluctuate
from one place to another? Not really: the realistic (numerical) models of the interface barriers
can easily justify fluctuations. In short, no guarantee can be given a priori that the fluctuations
do not exist.
This problem, therefore, cannot be simply ignored. It must be attacked, and the best
way is the experimental way. For some time, a search has been underway (Gozzo et al 1993,
1995, Margaritondo 1995b, c) to identify semiconductor barrier fluctuations with suitable
experimental techniques. The main message of these searches is simple: fluctuations do exist
and are in fact quite common.

6.2. Laterally resolved experimental techniques

In this section, we briefly review some of the techniques used to investigate the lateral
fluctuations of interface barriers, beginning with laterally resolved experiments based on
photoemission.

6.2.1. Photoelectron spectromicroscopy. There exist (Margaritondo and Cerrina 1990,

Margaritondo 1995b, c) two ways to obtain lateral resolution in a photoemission experiment:
either by focusing the photon beam which stimulates the emission of photoelectrons or by
processing the photoemitted electrons with an electron optics system similar to an electron
microscope. The best approach for studying semiconductor interfaces is the first one (Gozzo
et al 1993, 1995, Margaritondo 1995). In fact, the use of an electron optics system
often interferes with the photoelectron energy analysis which is essential for measuring
semiconductor interface barriers and their fluctuations (Gozzo et al 1993, 1995, Margaritondo
and Cerrina 1990, Margaritondo 1995b, c, 1997, Ade 1997, Almeida et al 1996, Margaritondo
et al 1995, Margaritondo and Hwu 1996). On the other hand, the recent addition of electron
 Interface states at semiconductor junctions 795

Figure 14. A schematic diagram illustrating the

operation of a focusing-scanning spectromicroscope
(Margaritondo and Cerrina 1990).

energy resolution to instruments based on the second approach may change this conclusion in
the future.
Figure 14 shows the schematic diagram of a focusing-scanning spectromicroscope. The
photon beam is focused by a suitable device, which could be, for example, a multilayer-coated
Schwarzschild objective or a Fresnel zone plate (Margaritondo and Cerrina 1990, Margaritondo
1995b, c). The electron energy analyser measures the photoelectron spectra corresponding to
the small sample surface area which is illuminated by the focused x-ray beam.
The standard procedure in experiments of this kind (Margaritondo and Cerrina 1990,
Margaritondo 1995b, c) is to begin by taking photoelectron intensity microimages. To do this,
the analyser is set at a fixed energy, corresponding to the photoelectrons excited from one of the
electronic states of the sample—in most cases, a given core level of one of its elements. The
sample is then scanned with respect to the photon focusing spot. The photoelectron intensity
plot during the scanning produces an image which reveals, in particular, the lateral distribution
of the photoemitting element.
Scanning photoelectron microimages taken for different elements are used to identify the
most interesting parts of the surface; subsequently, one takes small-spot photoelectron spectra
on these areas. As we saw in section 2.2.1, the photoelectron energies reflect the initial-state
energies on the surface rather than in the bulk—and for a semiconductor they are affected by
the band bending. A scanning photoelectron microimage, therefore, can contain lateral fluctu-
ations not only in the chemical composition but also in the band bending. In turn, such band-
bending fluctuations can reflect lateral variations of the interface barriers (Cerrina et al 1993).
Figure 15 shows one of the very first examples of applying this technique (Cerrina et al
1993). The experiment studied the lateral fluctuations of the band bending of a cleaved
GaAs(110) substrate. The microimages on the left-hand side of figure 15 show intensity
variations; in principle, such variations could be due to fluctuations in the local chemical
composition—for example, to microprecipitates of a given element—or to variations in the
band bending.
How can one distinguish between these two factors? The answer is provided (figure 15,
right-hand side) by photoemission spectra taken in microscopic spots on different sample
surface areas. They reveal clear shifts in energy from place to place, of equal value for both
the Ga3d and As3d core level peaks.
The message is clear: the electronic structure shifts rigidly from point to point, and this
reflects fluctuations in the band bending. Therefore, even for a clean cleaved surface, the
surface energy barrier corresponding to the band bending cannot be assumed to be constant
from point to point!
796 G Margaritondo

Figure 15. A scanning-focusing photoelectron spectromicroscopy study of a clean, freshly cleaved

GaAs(110) surface (data from Cerrina et al (1993)). Left: photoelectron intensity micrographs
(80×80 µm) taken at three different photoelectron energies (70.3, 70.8 and 71.3 eV), in the spectral
region of the Ga3d level (photon energy: 95 eV). Right: Ga3d and As3d microspot-spectra taken
in the regions A and B of the micrographs. The rigid shift of the core level peaks reveals that the
micrographs’ intensity fluctuations are primarily due to band-bending changes from place to place.

6.2.1.1. Interface barrier variations: metal–semiconductor junctions, heterojunctions. Early

experiments, like that of as figure 15, stimulated the basic questions: do similar fluctuations oc-
cur for interface barriers like the Schottky barriers and the heterojunction band discontinuities?
The answer was provided by a series of experiments of Gozzo et al (1993, 1995).
Figure 16 shows the relevant data (Gozzo et al 1993) for a metal–semiconductor interface,
in this case Au on GaSe. The parallel analysis of scanning photoelectron micrographs and
small-area EDCs provides an unmistakable answer: the metal–semiconductor barrier height
fluctuates from point to point on the interface. The bottom part of figure 16, in particular, shows
the rigid shifts of the Ga3d and Se3d core level peaks due to the changes in band bending.
The search for lateral barrier fluctuations by Gozzo et al (1995) was then extended to
semiconductor–semiconductor interfaces and to their band discontinuities. Once again the
results were quite clear, as one can see in figure 17. The data here refer to the interface
between a GaSe substrate and a Ge overlayer. Data from the substrate core levels reveal
coordinated shifts in energy from point to point, corresponding to fluctuations in the substrate
 Interface states at semiconductor junctions 797

Figure 16. A scanning-focusing photoelectron spectromicroscopy study of a cleaved GaSe surface

covered by a thin Au overlayer (data from Gozzo et al (1993)). Top: Photoelectron intensity
micrographs (80 × 80 µm) taken at a photon energy of 95 eV, in the spectral range of the Au3d
states. Bottom: Ga3d and Se3d microspot-spectra taken in the regions A and B of the micrograph.
The rigid shift of the core level peaks reveals a change in the interface barrier.

band bending. In contrast, the Ge3d peak does not follow the same shifts.
Therefore, the relative position of the electronic structure of the overlayer with respect to
the band-bending changes from point to point. This implies fluctuations in the band lineup
and in the corresponding valence and conduction band discontinuities.
Results such as those of figures 16 and 17 deal a fatal blow to the notion that semiconductor
interface barriers are global properties of the interfaces, with no changes from place-to-place
fluctuations. In contrast, barrier fluctuations not only occur but are quite common. This
requires a re-thinking of device modelling and of the notion of semiconductor energy barriers.

6.2.1.2. Chemical fluctuations: the Schottky diode does not exist! Results like those of
figures 16 and 17 are particularly important as far as the notion of ‘Schottky barriers’ is
concerned. For many years, II–VI interfaces were considered (Daniels et al 1984, 1985) as
the only solid example of ‘Schottky’ systems: do the data of figures 16 and 17 imply that this
notion must be abandoned?
As we saw in section 1.1 (equations (1) and (3)), the notion of ‘Schottky interface’ relies
on the absence of chemical interactions between the two sides of the junction. The interface
barriers are determined by bulk or ‘pseudobulk’ properties of the two component materials,
and there is no way of modifying their values from place to place.
As we have seen, many experimental results have revealed interface behaviours
inconsistent with such a notion of ‘Schottky interface’. Does then any interface behave like a
Schottky system?
Interfaces based on a II–VI semiconductor could be particularly promising candidates
798 G Margaritondo

Figure 17. A scanning-focusing photoelectron spectromicroscopy study of a cleaved GaSe surface

covered by a thin Ge overlayer (data from Gozzo et al (1995)). Top left: Photoelectron intensity
micrographs (50 × 50 µm) taken at a photon energy of 95 eV, in the spectral range of the Ge3d
states. Bottom left and right: Ge3d, Ga3d and Se3d microspot-spectra taken in the regions A and
B of the micrograph. There is a rigid shift of the Ga3d and Se3dcore level peaks, and no shift for
the Ge3d peak. This reveals a change in the band lineup from point A to point B, and therefore a
change in the band discontinuities.

for a positive answer to this question. Their chemical bonding structures, corresponding to
the layered crystal structure, includes strong intralayer bonds and very weak interlayer bonds
(Daniels et al 1984, 1985). The cleavage exposes a layer surface which is quite unreactive
from the chemical point of view. Thus, one could suspect that the corresponding interfaces
behave like true Schottky systems, with no chemical interactions.
Indeed, for many years, the experimental results seemed to suggest (Daniels et al 1984,
1985) that II–VI interfaces behave as ideal Schottky systems. The results included both metal–
semiconductor and semiconductor–semiconductor interfaces.
Recently, however, these conclusions had to be revised in light of new laterally resolved
data. For example, the high-lateral-resolution spectra of figure 18 (Almeida et al 1997) reveal
evidence of chemical interactions for the GaSe–Ge interface, which was previously considered
as a prime example of a true Schottky system.
These new results deal a fatal blow to the very notion of ‘Schottky interfaces’: to the best of
 Interface states at semiconductor junctions 799

Figure 18. A scanning-focusing photoelectron spectromi-

croscopy curves taken on different microspots of a Ge-covered
GaSe substrate (data from Almeida et al (1997)). The curves
reveal different spectral components with different weights. In
turn, these results reveal the presence of unexpected chemical
reactions for this interface, and their changes from place to
place.

this authors knowledge, no system exists with unquestionable evidence of such a behaviour. It
might very well be, therefore, that no Schottky barrier behaves like an ideal ‘Schottky interface’!

6.2.2. Free-electron-laser (FEL) techniques: FELIPE and SNOM-FEL. The main limitation
of photoemission spectromicroscopy, in dealing with the problem of interface barrier
fluctuations, is its limited accuracy in measuring such barriers. In general, photoemission
experiments cannot measure (Katnani and Margaritondo 1983) semiconductor interface
barriers with an accuracy better than 50–100 meV. Smaller-scale fluctuations can be detected,
since the accuracy in measuring relative changes is typically better than the absolute accuracy.
Even then, an accuracy better than 30–50 meV does not appear feasible.
This is unfortunate, since even a barrier change of a few meV can strongly influence the
behaviour of an interface and of the corresponding devices. New techniques, on the other
hand, are capable of providing better accuracy.
Figure 19 explains the philosophy of one of these techniques, which was baptized FELIPE
(free-electron-laser internal photoemission) (Coluzza et al 1992). The technique is based on
photocurrent measurements. The photocurrent thresholds correspond to different interface
barriers and, in particular, to the heterojunction conduction band discontinuity.
Note that to measure the conduction band discontinuity one must use a suitable infrared
photon source, with wide enough tunability in the photon energy range of interest. This is why
the technique is implemented with a widely tunable FEL. FELIPE experiments constituted one
of the very first materials science research applications of the FELs.
The technique is quite straightforward and yields discontinuity measurements with very
good accuracy of the order of a few meV. Quite recently, the same approach was implemented
(Almeida et al 1996) in a laterally resolved version to analyse lateral barrier fluctuations. The
lateral resolution was achieved using a small-tip optics fibre. Figure 20 shows (Almeida et al
1996) one particularly significant result. The shifts from point to point in the photocurrent
threshold reveal very small changes in the corresponding Schottky barrier height. The accuracy
in measuring such barrier fluctuations can reach 1 meV or better, thus better matching the needs
800 G Margaritondo

Figure 19. A schematic explanation of the ‘FELIPE’ Figure 20. A laterally resolved version of the FELIPE
technique (Coluzza et al 1992). An infrared photon technique (data from Almeida et al (1996)). The figure
emitted by a tunable FEL excites an electron over the shows two photocurrent spectra obtained on two different
barrier corresponding to a heterojunction conduction microspots, A and B, of a Pt-covered GaP substrate. The
band discontinuity. The electron then contributes to change in threshold reveals a small but clearly detectable
the photocurrent measured by the picoammeter (pA). A change in the Schottky barriers height between the two
threshold in the photocurrent, as a function of the photon sites.
energy, reveals the value of the discontinuity.

Figure 21. One of the first SNOM micrographs obtained with

a FEL (data from Cricenti et al (1998)). The reflectivity image
was obtained with a photon wavelength of 1.2 µm on a PtSi/Si
interface. The FEL photons are brought to the sample surface
by a narrow-tip optics fibre. The distance between sample and
fibre tip was small enough to achieve near-field conditions (Pohl
1992).

for device researchthan other laterally resolved techniques.

The use of small-tip optics fibres to achieve high lateral resolution is similar to the stan-
dard scanning near-field optical microscopy (SNOM) technique for visible light (Pohl 1992).
The results of figure 19 were obtained not with an FEL but with a conventional laser. Quite
recently, in fact, Cricenti et al (1998) obtained the first SNOM microimages using a FEL,
using a similar approach.
One of the first SNOM-FEL microimages is shown in figure 21. These results could have
a strong and very positive impact on semiconductor interface research, opening the way to the
routine detection of very small lateral barrier fluctuations with unprecedented resolution and
accuracy.

6.2.3. Additional results on lateral fluctuations. Semiconductor interface barrier fluctuations

are being discovered and analysed (Almeida et al 1998, 1999, Zacchigna et al 1998) not only
with spectromicroscopic techniques based on photoemission and internal photoemission, but
 Interface states at semiconductor junctions 801

also with other approaches. The most productive of these is, of course, BEEM. Concerning
recent BEEM results in this domain we note, for example, the results by Fowell et al (1992)
on the conduction band discontinuity of InAs–GaAs. In 1996 the same group found that the
InAs–GaAs barrier height decreases with the InAs thickness and that the detailed variation is
in accordance with the transition/relaxation of the InAs layer (Ke et al 1996).
As to Schottky barriers, Talin et al (1994) used the BEEM technique to analyse lateral
variation in the Schottky barrier height at Au–PtSi–Si(100) and Au–GaAs(100) interfaces. All
of the investigated contacts exhibited spatial inhomogeneities in the Schottky barrier height.
The most severe variations observed were 0.09 eV over 0.7 nm for Au–(100)GaAs and 0.08 eV
over 14 nm for Au–PtSi–(100)Si.
Based on the lateral maps of the barrier height at each interface, the difference between
the locally averaged barrier height and the globally averaged barrier height was computed.
This analysis showed that there is a critical diode lengthscale below which the Schottky barrier
height deviates significantly from the averaged value over a macroscopic length scale. The
authors argued that this result implies that the uniformity of the electrical characteristics of
arrays of small devices (e.g., PtSi–Si photodetectors and GaAs FET gates) can be expected to
deteriorate significantly when device dimensions decrease below the critical length.
In the same year, Clausen et al (1994) used BEEM to study Au–Ni and Au–Cr contacts to
InP. They found that the interfacial properties are complex with many possible phases at the
interface. The electrical properties were described in terms of Schottky barrier inhomogeneities
at the interface and using a parallel conduction model. From this model, the authors found
that the observed reduction in the contact resistance is closely connected to a lowering of the
effective Schottky barrier towards 0 eV.
In addition to lateral barrier fluctuations, different kinds of microscopies were recently used
to investigate other microscopic aspects of the semiconductor interfaces. The corresponding
literature is quite extensive, and we note here only two significant examples. The first one is
the cross sectional transmission electron microscopy study of ohmic contacts on (In, Ga)As
and GaAs by Klein et al (1996). The second is the transmission electron microscopy study
by Bonard et al (1997) of native extended defects in pseudomorphic ZnSe–GaAs(001) and
lattice-matched ZnSe–(In, Ga)As(001) heterostructures.

7. Summary and future developments

It is quite difficult to summarize the status of a healthy and very active research field such as
semiconductor interfaces. We will limit our final comments to two important questions. First,
what are the current general trends, in light of the previously discussed results? Secondly,
what are the major new experimental tools that could help this field in the future?

7.1. Present trends and open problems

As far as theory is concerned, the most important advances are being produced by the combined
use of advanced experiments and ab initio calculations (Peressi et al 1998) of the interface
electronic states, as outlined in section 3.5. This approach makes it possible to explore the
effects of different factors on the interface barriers.
As we have seen, no one factor which could influence the barriers can be safely ruled
out a priori: the era of ‘universal models’ is long gone. But ab initio theories can, under
certain circumstances, identify one prevailing factor. Specifically, for lattice-matched isovalent
heterojunctions 1EC and 1EV , which only depend on bulk material properties. The band
discontinuities of lattice-matched heterovalent heterojunctions depend instead on interface
802 G Margaritondo

properties such as the interface orientation.

The heterojuction band discontinuities of lattice-mismatched isovalent heterojunction
interfaces are also mainly affected by bulk properties, which extend their influence to strain-
related phenomena (Berthod et al 1996). For lattice-mismatched isovalent heterojunction
interfaces, both strain effects and local ‘chemical’ effects appear important. On the other
hand, the situation for metal–semiconductor interfaces is more complex and much less clear.
The evidence that non-bulk factors play an important role for many interfaces puts
the empirical approaches for changing the interface parameters on a solid conceptual
ground. We have seen many interesting examples of these efforts, including modifications
of Schottky barriers heights and heterojunction band discontinuities as well as the creation
of artificial discontinuites at homojunction interfaces. Understanding the theoretical basis
of such procedures—which are still primarily empirical—is a major challenge for today’s
semiconductor interface research.
The transition from ‘understanding’ semiconductor interfaces to ‘controlling’ their
properties is one of the major present trends in this domain. Another major trend, as we
have seen, is the tendency to analyse the interface properties on a microscopic scale. The
corresponding discovery of lateral barrier fluctuations is forcing a general revision of well-
established semiconductor device concepts and models. Examples of discovered lateral
fluctuations include both Schottky barriers and heterojunction band discontinuities.
Understanding such fluctuations and, in particular, their relation to the local chemical
and electronic structure is one of the main present challenges in this field. Meeting this
challenge will require new experiments to establish empirical relations and more ab initio
calculations. We note, in particular, the possible use of the recently implemented FEL-based
SNOM techniques (Cricenti et al 1998).

7.2. New experimental opportunities

Throughout its history, semiconductor interface research has been heavily influenced by the
development of new experimental techniques. This domain would not exist—at least not in
the present form—without techniques such as synchrotron-radiation photoemission, BEEM
and FELIPE. The development of new experimental approaches is certainly not exhausted.
In particular, many research opportunities are opened up by the new synchrotron sources of
the third generation—such as the advanced light source (ALS) in Berkeley, Elettra in Trieste,
Max-II in Lund and Bessy II in Berlin (Margaritondo 1995a).
The most relevant characteristic of these new facilities is their extremely high brightness,
which is obtained by means of a low-emittance electron beam in the storage rings and by using
the insertion devices known as ‘undulators’ (Margaritondo 1995a). The high brightness is
exploited for a variety of new techniques applicable to semiconductor interface research.
In particular, high brightness makes it possible to push laterally resolved photoemission
spectromicroscopy experiments to unprecedented performance levels, such as 20 nm lateral
resolution or better (De Stasio et al 1999, Bauer 1998). Furthermore, fast data taking,
made possible by the high brightness, opens up studies of interface formation in real time
(Baraldi et al 1996). This approach already revealed the complex chemical evolution of other
types of interfaces, and will undoubtedly yield very interesting results when it is applied to
semiconductor interfaces.
We believe, however, that the major impact of the third generation synchrotron sources—
and even more of newer facilities like the Swiss Light Source (SLS) (Weyer and Margaritondo
1995)—will be related to their coherence (Margaritondo 1998a, b). Throughout the first
century of the history of x-rays, spatially coherent sources were never available. This is
 Interface states at semiconductor junctions 803

Figure 22. An image of a nanostructure taken

with a photoelectron spectromicroscope. The
illumination with coherent x-rays produces a
series of clear diffraction fringes. Results from
Schmidt et al (1998).

no longer true: spatial coherence is a welcome byproduct of high brightness (Margaritondo

1998a, b). For modern synchrotron sources, high brightness is primarily achieved by improving
the geometrical characteristics of the source—size and angular spread. These are the same
characteristics that determine the spatial coherence. Thus, high brightness for synchrotron
sources automatically implies high spatial coherence.
On the other hand, the high source brightness also makes it easier to achieve high
longitudinal (time) coherence. Longitudinal coherence requires a narrow spectral bandwidth,
which is typically obtained with a monochromator. And the effectiveness of a monochromator
is typically enhanced when the brightness is high.
How will semiconductor interface research benefit from x-ray coherence? We are forced
to speculate, but such speculations can be based on solid results obtained in other fields. For
example, high spatial coherence is exploited for ‘phase-contrast radiology’ (Margaritondo
1998b, Snigirev et al 1995a, b). This technique achieves a remarkable quality enhancement
of radiological images by using the source coherence to create visible edge diffraction fringes.
In practice, techniques of this kind exploit, for the first time in x-ray image formation, not only
the imaginary part of the diffraction index but also the real part.
The impact on interface research is likely to be quite important. We note, in particular,
that the source tunability could be exploited to perform phase-contrast radiology in a
‘resonant’ regime, near an absorption edge of a given element. This could produce high-
quality radiological images of the spatial distribution of specific elements, even when their
concentration is limited. The potential applications in interface research are very numerous
and interesting. One could, for example, obtain chemical maps of buried interfaces—showing
microprecipitates, local chemical phases and other chemical and morphological factors that
can strongly influence the interface parameters.
Figure 22 shows another interesting example of the applications of x-ray coherence
(Schmidt et al 1998). One can clearly see the diffraction fringes produced by a coherent beam
of x-rays that reaches, at near-grazing incidence, an object of nanoscopic scale (the roughly
triangular feature). In figure 22, the fringes are visible in a photoemission micrograph taken
with the ‘SPE-LEEM’ (spectroscopic low-energy electron microscope) instrument constructed
and operated by Bauer and co-workers (Schmidt et al 1998).
Results like those of figure 22 are likely to become quite relevant to semiconductor interface
research. In fact, one of the main technological trends is this field is the fabrication and
exploitation of artificial nanostructures, which introduce new degrees of freedom in the design
of devices—and could potentially lead to a new technological revolution. These developments
are creating an increasing demand for new characterization tools.
X-ray diffraction fringes like those of figure 22 could be potentially exploited to gain
information on the morphological properties of artificial nanosystems—including high-
precision quantitative size measurements. There are many potential techniques in this direction;
804 G Margaritondo

Figure 23. CPD curves obtained by Shikler et al

(1999) with a combination of the atomic force
microscope and of the Kelvin force method. The
CPD curves correspond to the potential distribution
across a GaP p–n junction (exposed by cleavage)
junction. The label for each curve shows the value
of the external bias applied to the junction.

the extreme cases are x-ray interfereometry and x-ray holography. These are still techniques
under development, very promising but certainly not mature. Nevertheless, this author believes
that they will have a strong impact on semiconductor interface research—and, in general, on
studies of solid interfaces.
Finally, we would like to note another interesting area of instrumentation development
that is quite relevant to semiconductor interface research. The combined use of an atomic
force microscope (AFM) and of the Kelvin force technique was used to obtain very interesting
maps of potential measurements across semiconductor barriers (Kikukawa et al 1995, Vatel
and Tanimoto 1995, Shikler et al 1999).
Figure 23 shows a nice example of the results of this approach by Shikler et al (1999)
We see two contact potential difference (CPD) curves plotted as a function of the position
across a p–n GaP junction (exposed by cleavage). The CPD curves correspond to the potential
distribution across the junction. Note that each curve was taken while an external bias was
applied to the junction (the value is specified for each curve). Thus, this approach can show the
voltage profile for operating devices. Note that, for increasing forward bias, the junction built-
in voltage (the effective energy barrier) decreases and, after reaching a flat-band condition, is
inverted.
Results like those of figure 23 clearly illustrate how much the combination of modern
experimental techniques can bring to this field. Many other examples could be presented, and
many more are conceived and implemented every year.

7.3. Final remarks

The research on semiconductor interfaces continues with increasing interest after decades
of effort. These efforts were initiated in a somewhat naive way as an attempt to find simple
solutions to the problems affecting semiconductor devices. As it turned out, no simple solutions
existed because the problem of semiconductor interfaces touches the very foundations of solid-
state physics.
Hopes of finding general and easily manageable solutions disappeared a long time
ago. But they were replaced by something much more exciting: the realization that
semiconductor interface research is one of the most challenging and interesting fields of solid-
state research. With the new developments in the theoretical techniques and in the experimental
instrumentation, the expansion of this field is quite likely to continue for many years. In fact,
the possible end of the ‘silicon technology’ era could stimulate an even faster rate for this
expansion.
 Interface states at semiconductor junctions 805

Acknowledgments

We would like to thank the authors of all references I used to support my conclusions. My own
work in this domain is supported by the Fonds Nation al Suisse de la Recherche Scientifique,
by the US Office of Naval Research NR and by the EPFL.

References

Ade H (ed) 1997 Spectromicroscopy with VUV Photons and X-Rays (Amsterdam: Elsevier)
Almeida J, Coluzza C, Terrasi A, Dell’Orto T, Ivanco J and Margaritondo G 1997 J. Appl. Phys. 81 292–6
Almeida J, Dell’Orto T, Coluzza C, Fasso A, Baldereschi A, Margaritondo G, Rudra A, Buhlmann H J and Ilegems M
1995 J. Appl. Phys. 78 3258–61
Almeida J, Dell’Orto T, Coluzza C , Margaritondo G, Bergoss O, Spajer M and Courjon D 1996 Appl. Phys. Lett. 69
2361–3
Almeida J, Margaritondo G, Coluzza C, Davy S, Spajer M and Courjon D 1998 Appl. Surf. Sci. 125 6–10
Almeida J, Sirigu L, Margaritondo G, Da Padova P, Quaresima C and Perfetti P 1999 J. Phys. D: Appl. Phys. 32 191–4
Almeida J, Vobornik I, Berger H, Kiskinova M, Kolmakov A, Marsi M and Margaritondo G 1997 Phys. Rev. B 55
4899–902
Alonso M, Cimino R and Horn K 1990 Phys. Rev. Lett. 64 1947–51
Alperovich V L, Paulish A G and Terekhov S S 1995 Surf. Sci. 331–333 1250–5
Anderson R L 1962 Solid-State Electron. 5 341
Angelo J E, Gerberich W W, Bratina G, Sorba L and Franciosi A 1995 Thin Solid Films 271 117–21
Aristov V Yu, Mangat P S, Soukiassian P and Le Lay G 1995 Surf. Sci. 331–333 641–5
Bai Chunli, Hashizume Tomihiro, Jeon Dong-Ryul and Sakurai Toshio 1993 Japan. J. Appl. Phys. 32 1401–4
Baraldi A, Comelli G, Lizzit S, Cocco D, Paolucci G and Rosei R 1996 Surf. Sci. 367 L67–70 and references therein
Bardeen J 1947 Phys. Rev. 71 717
Bardi J, Binggeli N and Baldereschi A 1996 Phys. Rev. B 54 R11 102–5
Bauer E 1998 Private communication
Bennett M R, Cafolla A A, Cairns J W, Dunscombe C J and Williams R H 1996 Surf. Sci. 360 187–99
Berthod C, Bardi J, Binggeli N and Baldereschi A 1996 J. Vac. Sci. Technol. B 14 3000–7
Bonard J-M, Ganiere J-D, Heun S, Paggel J J, Rubini S, Sorba L and Franciosi A 1997 Phil. Mag. Lett. 75 219–26
Bonard J-M, Ganiere J-D, Vanzetti L, Paggel J J, Sorba L, Franciosi A, Herve D and Molva E 1997 Phil. Mag. Lett.
76 181–7
Bonnet J J, Soonckindt L, Schirm K M, Nishigaki S, Hricovini K and Soukiassian P 1993 Appl. Surf. Sci. 68 427–31
Bratina G, Vanzetti L, Nicolini R, Sorba L, Yu X, Franciosi A, Mula G and Mura A 1993 Physica B 185 557–65
Brillson L J 1982 Surf. Sci. Rep. 2 123–326
——1992 Handbook on Semiconductors vol 1 ed P T Landsberg (Amsterdam: North-Holland) p 281
—— (ed) 1993 Contacts to Semiconductors (Park Ridge: Noyes)
——1994 Surf. Sci 299–300 909–27
Brillson L J, Brucker C F, Stoffel N G, Katnani A D and Margaritondo G 1981 Phys. Rev. Lett. 49 838–42
Brillson L J, Chang S, Shaw S J and Viturro R E 1990 Vacuum 41 1016–20
Brillson L J and Margaritondo G 1988 Surface properties of electronic materials The Chemical Physics of Solid
Surfaces and Heterogeneous Catalysis vol 5 ed D A King and D P Woodruf (Amsterdam: Elsevier) pp 119–82
Brillson L J, Margaritondo G and Stoffel N G 1980 Phys. Rev. Lett. 44 667–71
Brillson L J, Margaritondo G, Stoffel N G, Bauer R S, Bachrach R Z and Hansson G 1980 J. Vac. Sci. Technol. 17
880–4
Capasso F and Margaritondo G (eds) 1987 Heterojunction Band Discontinuities: Physics and Device Applications
(Amsterdam: North-Holland)
Cardona M, Christensen N E 1987 Phys. Rev. B 35 6182–94
Ceccone G, Bratina G, Sorba L, Antonini A and Franciosi A 1991 Surf. Sci. 251–252 82–6
Cerrina F et al 1993 Appl. Phys. Lett. 63 63–6
Chen W, Gaines J, Ponzoni C, Olego D, Mangat P S, Soukiassian P and Kahn A 1994 J. Cryst. Growth 138 1078
Cheng X-C, Collins D A and McGill T C 1997 J. Vac. Sci. Technol. A 15 2063–8
Chiarotti G, Nannarone S, Pastore R and Chiaradia P 1971 Phys. Rev. B 4 3398
Christianen P C M, Bluyssen H J A, van Hall P J and Wolter J H 1994 Semicond. Sci. Technol. 9 707–9
Clausen T, Leistiko O, Chorkendorff I and Larsen J 1994 Appl. Surf. Sci. 74 287–95
Coluzza C et al 1992 Phys. Rev. B 46 12834–6
806 G Margaritondo

——1994 J. Appl. Phys. 76 3710–13

Cricenti A, Generosi R, Perfetti P, Gilligan J M, Tolk N H, Coluzza C and Margaritondo G 1998 Appl. Phys. Lett. 73
151–3
Daniels R R, Margaritondo G, Quaresima C, Capozi M, Perfetti P and Lvy F 1984 Sol. State Commun. 51 495–9
Daniels R R, Margaritondo G, Perfetti P, Quaresima C and Lévy F 1985 J. Vac. Sci. Technol. A 3 979–83
Davy S et al 1999 J. Microscopy at press
De Padova P, Larciprete R, Quaresima C, Ottaviani C, Ressel B and Perfetti P 1998 Phys. Rev. Lett. 81 2320–4
De Stasio G, Perfetti L, Gilbert B, Fauchoux O, Capozi M, Perfetti P, Margaritondo G and Tonner B P 1999 Rev. Sci.
Instrum. at press
Dell’Orto T, Almeida J, Coluzza C, Baldereschi A, Margaritondo G, Cantile M, Yildirim S, Sorba L and Franciosi A
1994 Appl. Phys. Lett. 64 2111–14
Dell’Orto T, Almeida J, Terrasi A, Marsi M, Coluzza C and Margaritondo G 1994 Phys. Rev. B 50 18189–93
Dobaczewski L, Langer J M and Missous M 1993 Acta Phys. Pol. 84 741
Dong Z-C, Yakabe T, Fujita D, Jiang Q D and Nejo H 1997 Surf. Sci. 380 23–30
Dumas M, Nouaoura M, Bertru N, Lassabatere L, Chen W and Kahn A 1992 Surf. Sci. 262 L91–5
Eastman D E and Grobman W D 1972 Phys. Rev. Lett. 28 1376–80
Einstein A 1905 Ann. Phys., Lpz. 17 932
Faraci G, La Rosa S, Pennisi A R and Margaritondo G 1994 Phys. Rev. B 49 2943–6
Flores F and Tejedor C 1979 J. Phys. C: Solid State Phys. 12 731
Flores F, Yeyati A L, Martin-Rodero A, Ortega J and Rincon R J 1997 Prog. Surf. Sci. 54 229–40 and references
therein
Fowell A E, Cafolla A A, Richardson B E, Shen T -H, Elliott M, Westwood D I and Williams R H 1992 Appl. Surf.
Sci. B 56–58 622–7
Franciosi A and Van de Walle C G 1996 Surf. Sci. Rep. 25 1
Freeouf J L and Woodall J M 1986 Surf. Sci. 168 518
Frensley W R and Kroemer H 1977 Phys. Rev. B 16 6242–52
Gothelid M, Grehk T M, Hammar M, Karlsson U O and Flodstrom S A 1995 Surf. Sci. 328 80–94
Gozzo F et al 1993 Phys. Rev. B 48 17163–7
Gozzo F, Berger H, Collins I R, Margaritondo G, Ng W, Ray-Chaudhuri A K, Liang S, Singh S and Cerrina F 1995
Phys. Rev. B 51 5024–27
Gusev A O, Paget D, Aristov V Yu, Soukiassian P, Berkovits V L and Thierry-Mieg V 1997 J. Vac. Sci. Technol. A
15 192–5
Hara S, Teraji T, Okushi H and Kajimura K 1996 Proc. SPIE 2779 802–6
Harrison W A 1977 J. Vac. Sci. Technol. 14 1016–21
Hecht M H 1990 Phys. Rev. B 41 7918–21
Herve D, Accomo R, Molva E, Vanzetti L, Paggel J J, Sorba L and Franciosi A 1995 Appl. Phys. Lett. 67 2144–6
Herve D, Molva E, Vanzetti L, Sorba L and Franciosi A 1995 Electron. Lett. 31 459–61
Hirose Y, Chen W, Haskal E I, Forrest S R and Kahn A 1994 Appl. Phys. Lett. 64 3482–4
Hirose Y, Kahn A, Aristov V and Soukiassian P 1996 Appl. Phys. Lett. 68 217–9
Hirose Y, Wu C I, Aristov V, Soukiassian P and Kahn A 1997 Appl. Surf. Sci. 113–114 291–8
Horng S, Hirose Y, Kahn A, Wrenn C and Pfeffer R 1992 Appl. Surf. Sci. 56–58 855–60
Huttel Y, Soukiassian P, Mangat P S and Hurych Z 1996 Surf. Sci. 352–354 845–9
Hwu Y, Marsi M, Alméras Ph and Margaritondo G 1992 Phys. Rev. B 46 1835–7
Kahn A 1994 Surf. Sci. 299–300 469–86
Kaiser W J and Bell L D 1988 Phys. Rev. Lett. 60 1406–10
——1996 Ann. Rev. Mater. Sci. 26 189
Karpov I, Venkateswaran N, Bratina G, Gladfelter W, Franciosi A and Sorba L 1995 J. Vac. Sci. Technol. B 13 2041–8
Katnani A D and Margaritondo G 1983 Phys. Rev. B 28 1944–56
Katnani A D, Perfetti P, Zhao Te-Xiu and Margaritondo G 1982 Appl. Phys. Lett. 40 619–22
——1983 J. Vac. Sci. Technol. A 1 650–4
Ke Mao-long, Westwood D I, Matthai C C and Williams R H 1996 Surf. Sci. 352–354 861–4
Kendrick C, Kahn A and Forrest S R 1996 Appl. Surf. Sci. 104–105 586–94
Kikukawa A, Hosaka S and Imura R 1995 Appl. Phys. Lett. 66 3510–13
Klein A, Pettenkofer C, Jaegermann W, Lux-Steiner M and Bucher E 1994 Surf. Sci. 321 19–31
Klein A, Urban I, Ressel P, Nebauer E, Merkel U and Osterle W 1996 Mater. Charact. 37 143–51
Kolnik J and Ivanco J 1993 J. Cryst. Growth 126 156–62
Kurtin S, McGill J C and Mead C A 1970 Phys. Rev. Lett. 22 1433–7
Lazzarino M, Ozzello T, Bratina G, Paggel J J, Vanzetti L, Sorba L and Franciosi A 1996a Appl. Phys. Lett. 68 370–2
 Interface states at semiconductor junctions 807

——1996b J. Cryst. Growth 159 718–22

Lazzeri M, Pellegrini V, Beltram F, Lazzarino M, Paggel J J, Sorba L, Rubini S, Bonanni A and Franciosi A 1997 J.
Cryst. Growth 175–176 603–7
Le Lay G, Mao D, Kahn A, Hwu Y and Margaritondo G 1991 Phys. Rev. B 43 14 301–4
Malacky L, Klockenbrink R, Darmo J, Wehmann H-H, Zwinge G and Schlachetzki A 1994 J. Phys. D: Appl. Phys.
27 2414–17
Mao D, Kahn A, Marsi M and Margaritondo G 1990 Phys. Rev. B 42 3228–30
——1991a Appl. Surface Sci. 48/49 324–8
——1991b J. Vac. Sci. Technol. A 9 898–902
Mao D et al 1991c J. Vac. Sci. Technol. B 9 2083–8
Margaritondo G 1985 Phys. Rev. B 31 2526–7
——1986 Solid-State Electron. 29 123–32
——1988a Introduction to Synchrotron Radiation (New York: Oxford)
——1988b Electronic Structure of Semiconductor Heterojunctions (Dordrecht: Kluwer)
——1988c Physics Today 41 66–72
——1993a J. Vac. Sci. Technol. B 11 1362–6
——1993b Rendiconti CXVII Scuola Internazionale di Fisica Enrico Fermi (Bologna: Editrice Compositori) pp 25–38
——1995a Riv. Nuovo Cimento 18 1–24
——1995b Surf. Rev. Lett. 2 305–13
——1995c Scanning Spectromicrosc. 9 949–63
——1997 Chemical, Structural and Electronic Analysis of Heterogeneous Surfaces on Nanometer Scale ed R Rosei
(Dordrecht: Kluwer) pp 43–52
——1998a Nuovo Cimento D 20 1083–90
——1998b Phys. World 11 28–32
Margaritondo G, Brillson L J and Stoffel N G 1980 Solid State Commun. 35 277–81
Margaritondo G and Cerrina F 1990 Nucl. Instrum. Methods A 291 26–35 and references therein
Margaritondo G, Christman S B and Rowe J E 1976 J. Vac. Sci. Technol. 13 329–33
Margaritondo G, De Stasio G and Coluzza C 1995 J. Electron Spectr. 72 281–7
Margaritondo G and Hwu Y 1996 Appl. Surf. Sci. 92 273–81
Margaritondo G, McKinley J T, Rioux D and Niles D W 1992 Appl. Surf. Sci. 56–58 713–17
Margaritondo G, Quaresima C, Patella F, Sette F, Capasso C, Savoia A and Perfetti P 1984 J. Vac. Sci. Technol. A 2
508–12
Margaritondo G and Rowe J E 1986 Treatise on Analytical Chemistry ed I M Kolthoff and P J Elving (New York:
Wiley) pt 1, vol 8, ch 17
Margaritondo G and Stoffel N G 1979 Phys. Rev. Lett. 42 1567–71
Margaritondo G, Stoffel N G, Katnani A D and Patella F 1980 Solid State Commun. 36 215–9
Margaritondo G and Weaver J H 1985 Methods of Experimental Physics—vol 22 (Surfaces) ed R L Park and M J Lagally
(New York: Academic) ch 3
Marsi M, La Rosa S, Hwu Y and Margaritondo G 1992 J. Vac. Sci. Technol. A 10 741–5
McKinley J T, Hwu Y, Rioux D, Terrasi A, Zanini F, Margaritondo G, Debska U and Furdyna J K 1990 J. Vac. Sci.
Technol. A 8 1917–21
McKinley J T, Hwu Y, Koltenbah B E C, Margaritondo G, Baroni S and Resta R 1991 J. Vac. Sci. Technol. A 9 1917–21
——1992 Appl. Surf. Sci. 56–58 762–6
McKinley J and Margaritondo G 1993 Contacts to Semiconductors ed L J Brillson (Mill Road, NJ: Noyes) pp 600–661
Meyer T and von Kanel H 1997 Phys. Rev. Lett. 78 3133–6
Missous M, Truscott W S and Singer K E 1990 J. Appl. Phys. 68 2239–45
Mönch W (ed) 1990 Electronic Structure of Metal–Semiconductor Contacts (Dordrecht: Kluwer)
Niles D W, Margaritondo G, Colavita E, Perfetti P, Quaresima C and Capozi M 1986 J. Vac. Sci. Technol. A 4 962–6
Niles D W, Margaritondo G, Perfetti P, Quaresima C and Capozi M 1985 Appl. Phys. Lett. 47 1092–5
Niles D W, Tang Ming, Margaritondo G, Quaresima C and Perfetti P 1988 J. Vac. Sci. Technol. A 6 1337–41
Ohler C, Dalniels C, Föster A and Lüth H 1998 Phys. Rev. B 58 7864–71
O’Shea J J, Reaves C M, DenBaars S P, Chin M A and Narayanamurti V 1996 Appl. Phys. Lett. 69 3022–4
Ossicini S and Bernardini F 1992 Solid State Commun. 82 863–7
Pellegrini V et al 1996 Appl. Phys. Lett. 69 3233–5
Peressi M, Binggeli N and Baldereschi A 1998 J. Phys. D: Appl. Phys. 31 1273–1300 and references therein
Perfetti P, Denley D, Mills K A and Shirley D A 1978 Appl. Phys. Lett. 33 66–9
Perfetti P, Quaresima C, Coluzza C, Fortunato C and Margaritondo G 1986 Phys. Rev. Lett. 57 2065–9
Pickett W E and Cohen M L 1978 Phys. Rev. B 18 939–44
808 G Margaritondo

Pickett W E and Erwin S C 1990 Superlattices Microstruct. 7 335–9

Pohl D W 1992 Advances in Optical and Electron Microscopy ed C J Sheppard and T Mulvey (London: Academic)
vol 12
Prietsch M 1995 Phys. Rep. 253 164
Raisanen A D, Brillson L J, Franciosi A, Nicolini R, Vanzetti L and Sorba L 1995a J. Electron. Mater. 24 163–9
Raisanen A D, Brillson L J, Vanzetti L, Bonanni A and Franciosi A 1995b J. Vac. Sci. Technol. B 13 1705–10
Raisanen A D, Brillson L J, Vanzetti L, Sorba L and Franciosi A 1995c J. Vac. Sci. Technol. A 13 690–95
Revva P, Langer J M, Missous M and Peaker A R 1993 J. Appl. Phys. 74 416–25
Rhoderick E H and Williams R H 1988 Metal–Semiconductor Contacts ed P Hammond and R L Grimsdale (Oxford:
Clarendon)
Riehl-Chudoba M, Surnev L and Soukiassian P 1994 Surf. Sci. 306 313–26
Rowe J E, Christman S B and Margaritondo G 1975 Phys. Rev. Lett. 35 1471–5
Scamarcio G, Spagnolo V, Molinari E, Tapfer L, Sorba L, Bratina G and Franciosi A 1992 Superlattices Microstruct.
12 429–32
Schmidt Th, Heun S, Prince K C and Bauer E 1998 unpublished
Schmitsdorf R F, Kampen T U and Mönch W 1995 Surf. Sci. 324 249–56
Schottky W, Stormer R and Waibel F 1931 Z. Hoch Frequentztechnik 37 162
Semond F, Soukiassian P, Mangat P S, Hurych Z, Cioccio L and Jaussaud C 1996 Appl. Surf. Sci. 104–105 79–87
Shan W, Li M F, Yu P Y, Hansen W L and Walukiewicz W 1988 Appl. Phys. Lett. 53 974–7
Shen T-H, Elliott M, Williams R H, Woolf D A, Westwood D I and Ford A C 1992 Appl. Surf. Sci. 56–58 749–55
Shikler R, Meoded T, Fried N, Mishori B and Rosenwaks Y 1999 unpublished
Siegbahn K et al 1967 ESCA: Atomic, Molecular and Solid State Structure Studied by Means of Electron Spectroscopy
(Uppsala: Almqvist and Wiksell)
Siegbahn K et al 1969 ESCA Applied to Free Molecules (Amsterdam: North-Holland)
Snigirev A, Snigireva I, Kohn V, Kuznetsov S and Schelokov I 1995a Rev. Sci. Instrum. 66 5486–92
Snigirev A, Snigireva I, Suvorov A, Kocsis M and Kohn V 1995b ESRF Newsletters 24 23
Sorba L, Biasiol G, Bratina G, Nicolini R and Franciosi A 1993 J. Cryst. Growth 127 93–7
Sorba L, Yildirim S, Lazzarino M and Franciosi A 1996 Appl. Phys. Lett. 69 1927–9
Soukiassian P, Starnberg H I and Kendelewicz T 1985 Appl. Surf. Sci. 56–58 772–6
Spaltmann D, Geurts J, Esser N, Zahn D R T, Richter W and Williams R H 1992 Semiconductor Sci. Technol. 7 344–6
Spicer W E, Chye P W, Skeath P R , Su C Y and Lindau I 1979 J. Vac. Sci. Technol. 16 1418–22
Spiess L, Mangat P S, Tang S-P, Schirm K M, Freeman A J and Soukiassian P 1993 Surf. Sci. 289 L631–7
Spiess L, Wimmer E and Soukiassian P 1996 Appl. Surf. Sci. 92 501–6
Starnberg H I, Soukiassian P, Kim S T, Papageorgopoulos A and Kapoor S 1992 Surf. Sci. 269 934–7
Talin A A, Ngo T, Williams R S, Morgan B A, Ring K M and Kavanagh K L 1994 Materials Research Soc. Symp.
Proc. 337 319–24
Tersoff J 1984a Phys. Rev. B 30 4874–7
——1984b Phys. Rev. Lett. 52 465–68
Van de Walle C and Martin R M 1987 Phys. Rev. B 35 8154–65
Vatel O and Tanimoto M 1995 J. Appl. Phys. 77 2358–62
Vitomirov I M, Raisanen A D, Brillson L J, Kirchner P D, Pettit G D and Woodall J M 1992 Solid State Commun. 84
61–5
Vitomirov I M, Raisanen A, Chang S, Viturro R E, Brillson L J, Rioux D F, Kirchner P D, Pettit G D and Woodall J M
1993 J. Electron. Mater. 22 111–18
Waddil G D, Komeda Tadahiro, Yang Y N and Weaver J H 1990 Phys. Rev. B 41 10 283–6
Wagner L F and Spicer W E 1972 Phys. Rev. Lett. 28 1381–5
Wei S-H and Zunger A 1987 Phys. Rev. Lett. 59 144–8
Weitering H H 1996 Surf. Sci. 355 L271–7
Weyer H J and Margaritondo G 1995 Synchrotron Radiat. News 8 19–21
Wilks S P and Williams R H 1995 J. Electron Spectr. 72 49–54
Williams R H 1991 Surf. Sci. 251–252 12–21
Wu C C, Wu C I, Sturm J C and Kahn A 1997 Appl. Phys. Lett. 70 1348–50
Yu E T, McCaldin J O and McGill T C 1992 Solid State Physics vol 46, ed H Ehrenreich and D Turnbull (Boston,
MA: Academic) p 1
Yu D Y W, Kahn A, Cavus A and Tamargo M C 1997 Surf. Sci. 373 350–6
Zacchigna M, Sirigu L, Almeida J, Berger H, Gregoratti L, Marsi M, Kiskinova M and Margaritondo G 1998 Appl.
Phys. Lett. 73 1859–62