You are on page 1of 89

Surface Science Reports 5 (1985) 199-288 199

North-Holland. Amsterdam

ION BEAM CRYSTALLOGRAPHY OF SURFACES AND INTERFACES

J.F. VAN DER VEEN


FOM Institute for Atomic and Molecular Physics, P.O. Box 41883, 1009 DB Amsterdam,
The Netherlands

Manuscript received in final form 22 July 1985

The current status of Rutherford Backscattering Spectrometry (RBS) of surfaces and interfaces
is reviewed. The reader is made familiar with the use of shadowing and blocking techniques for
surface crystallography and with instrumental aspects of RBS. A formal theory of shadowing and
blocking is presented, along with a variety of Monte Carlo methods for computer simulation of the
shadowing and blocking experiments. The atomic geometries of various relaxed and reconstructed
surfaces - clean and adsorbate covered - are surveyed and the performance of RBS on these
surfaces is evaluated in comparison with other techniques for structure determination, Some
attention is given to RBS investigations of adsorbate geometries and surface dynamical properties.
The results are discussed in the light of recent theoretical predictions of surface structure and
dynamics. The review further treats in-depth analysis of thin films and interfaces, prepared under
UHV conditions. Topics to be addressed are (1) the growth mode of strained heteroepitaxial films,
(2) the composition and morphology of thin metal films on silicon surfaces, (3) the initial stages of
silicide formation, (4) the growth of oxide films, and (5) the atomic structure of bicrystal interfaces.
The use of (near)-monolayer depth resolving power is shown to be essential in most of these
studies. Emphasis is placed on fundamental aspects of interface formation, but possible technologi-
cal applications are briefly mentioned too.

0167-5729/85/$31.50 0 Elsevier Science Publishers B.V.


(North-Holland Physics Publishing Division)
Contents

1. Introduction 201
2 Shadow and blocking cones 203
2.1. Shadowing 203
2.2. Blocking 208
3. Backscattering spectrometry 211
3.1. The energy spectrum 211
3.2. The wrface peak 214
3.3. Instrumentation 215
3.4. Measurement of blockmg mmlma 218
3.5. Beam-induced damage 219
4. Calculation of backscattering intensities 221
4.1. Scattering from a single atomic row 221
4.2. Two-atom model 224
4.3. Double alignment 225
4.4. Correlated thermal motion 227
4.5. “Complete crystal” method 228
5. Atomic geometry and dynamics of metal surfaces 230
S.1. Relaxation 230
5.2. Reconstruction 236
5.3. Chemisorption sites and bond lengths 243
5.4. Dynamics of surfaces 245
6. Sen&?onductor surface structures 249
6.1. Si( 100) 250
6.2. Si(ll1) 252
6.3 Compound semiconductor surfaces 256
7. Thin films and interfaces 261
7.1. Heteroepitaxy 262
7.2. Metal-semiconductor interfaces 266
7.3. Oxide films 216
8. Synopsis 217
Acknowledgements 218
References 278
J.F. oan der Veen / Ion beam crystallography 201

1. Introduction

The atoms in the topmost layer of a crystal usually do not occupy the same
lattice positions as in the bulk. The surface may be in- or outwardly relaxed or
reconstructed. Since many important chemical and electronic properties of
surfaces are related to their atomic structure [l] it is of great interest to
investigate these structural changes in detail. Another topic is the crystallogra-
phy of solid-solid interfaces. Such studies, if performed on metal-semiconduc-
tor contacts, are primarily motivated by the fact that the microscopic structure
and chemical bonding at the interface determine the electrical properties of the
contact [2]. Here a survey is given of recent progress in ion-beam crystallogra-
phy of surfaces and interfaces. During the last decade this technique has
matured to a powerful structure probe, by which atoms in surfaces have been
located with a precision of +O.Ol A in favourable cases.
The review will first deal with the crystallography of metal and semiconduc-
tor surfaces. A semiconductor surface may reconstruct in different ways. For
some surfaces buckling-type distortions have been reported [3], while for other
surfaces new bonds appear to have been formed between atoms [4]. Relatively
little is yet known about the forces that drive the reconstruction. This is not too
surprising, given the facts that reliable structure data are scarce and that some
of the “determined” structures are still subject to controversy. Only very few
semiconductor surface structures have indeed been solved. Despite these
difficulties substantial progress has been made in the theoretical description of
the energetics of semiconductor reconstruction and a number of theoretically
predicted surface structure models have been verified experimentally.
Metal surfaces, if not reconstructed, are almost always relaxed. Low Energy
Electron Diffraction (LEED) data on metal relaxation are abundant [5]. In
many surfaces, however, the relaxation effect, being a rather small percentage
of the interlayer spacing, is only marginally larger than the uncertainty of a
LEED determination. It then becomes particularly difficult to prove the
existence of, e.g., oscillatory multilayer relaxation effects or adsorbate-induced
structural changes.
Of all structure probes LEED has found most widespread use, mainly
because commercial LEED equipment became available some 15 years ago.
During the last decade, however, new and more accurate methods have been
developed that directly image positions of surface atoms in real space. One of
them is Rutherford Backscattering Spectrometry (RBS) of energetic light ions
(50 keV to 2 MeV H+ or He+). When in the early seventies this technique was
applied for the first time to atomically clean surfaces [6], RBS already had a
long and successful history as a tool for in-depth analysis of thin film
composition and structure [7,8] and for bulk lattice location studies [9]. Since
then a steadily increasing number of these conventional RBS facilities have
been adapted to meet the UHV requirements for surface studies and a few
202 J.E “on der Veen / Ion beam ctystallography

have been built specifically for this purpose. Compared to LEED the equipment
for RBS is more complex and costly. A MeV Van de Graaff generator or an
ion implantation machine is needed for production of the analyzing ion beam.
This disadvantage is, however, more than offset by the fact that RBS, due to its
directness, does not suffer from the interpretational difficulties commonly
encountered in LEED.
It may seem rather surprising that ion beams which penetrate deeply into
the bulk of a crystal can be at all useful for surface studies. Yet high surface
sensitivity is obtained if the beam is aligned with a channeling direction of the
crystal, resulting in a dramatic reduction of backscattering from the bulk. In
essence, surface structures are determined by measuring shadowing and block-
ing effects in the yield of ions backscattered from the surface region of the
crystal. More specifically, the positions of single surface atoms in the top layer
can be imaged by accurate measurement of the directions in which these atoms
block off ions backscattered from atoms just below the top layer. In principle,
their positions then follow from simple trigonometry. In this manner changes
in interlayer spacings as small as 1% have been detected in clean and adsorbate
covered surfaces. Blocking effects from the adsorbate atoms themselves allow
an accurate determination of their positions.
Though emphasis is usually placed on surface atomic geometry, surface
vibrational amplitudes and absolute adsorbate coverages can be determined as
well. However, perhaps the most interesting application of RBS is to be found
in the analysis of thin films and interfaces. The highest depth resolution ever
achieved in these studies is 3 to 4 A, permitting a detailed characterization of
crystalline structure and morphology of very thin films.
This review is solely concerned with Rutherford backscattering of H+ or
He+ ions at medium to high energies. The technique of Low Energy Ion
Scattering (LEIS) involving noble gas or alkali ions of a few keV energy, will
not be treated; a review of LEIS is given in ref. [lo]. Here it is only emphasized
that LEIS is different from RBS in several important respects. At low ion
energies the scattering potential is poorly known and neutralization effects
become problematic. Hence, the accuracy of LEIS in quantitative surface
structure analysis is rather low ( - 0.1 A) and, e.g., surface relaxation effects
cannot be measured.
The paper is organized as follows: After a brief outline of ion shadowing
and blocking in section 2, the backscattering technique and instrumentation
are discussed in section 3. In section 4 the shadowing and blocking theory is
formulated. The atomic geometry and dynamics of metal and semiconductor
surfaces are discussed in sections 5 and 6. Section 7 gives a progress report on
thin-film and interface analysis with emphasis on metal-semiconductor con-
tacts. Section 8 presents a brief evaluation of ion beam crystallography in
comparison with other methods and an outlook to possible future develop-
ments.
203

2. Shadow and blocking cones

2. I. Shadowing

When a parallel beam of energetic light ions (H+ or He+) hits an atom, a
shadow is cast as shown in fig. 1. The shadow cone arises from small-angle
deflections of the ions in the screened Coulomb potential of the atom’s nuclear
charge. The scattering is predominantly in the forward direction and only a
few ions, having an impact parameter much smaller than the Thomas-Fermi
electron screening radius a, make a nuclear encounter and backscatter from
the atom into the detector. The screened Coulomb potential commonly used in
RBS analyses is due to Moliere [ll]:

Z,Z,e* 3
V(r) = 7 C a, e-Prrlu, (2.1)
i=l
where {q} = (0.10, 0.55, 0.35}, {p,} = (6.0, 1.2, 0.3}, r is the distance from
the target atom, and Z,e and Z,e are the nuclear charges of projectile and
target atom. The Thomas-Fermi screening length a is given by

a = 0.885~,Z,-“~, (2.2a)
for a completely ionized projectile and

a = 0.885a,( Z;‘* + Z;‘*) -2’3, (2.2b)

for a partially ionized projectile [9,12]. Here a, is the Bohr radius (a,, = 0.529
A). In the momentum approximation the deflection angle v(s) of the ion at
impact parameter s (fig. 1) is given by

(2.3)

Fig. 1. Formation of a shadow cone behind an atom. The radius of the cone cast over the
neighbouring atom is denoted by R.
204 J. F. van der Veen / Ion beam crystallography

where E is the projectile energy and K, is the first order modified Bessel
function of the second kind [13]. The shadow cone radius is determined as
follows. The trajectories deflected at impact parameter s lie on a cone having a
radius R(s) = s + Iv(s) at a distance 1 from the atom. The radius of the
shadow cone R = R(s,) is then formed at the critical impact parameter s0 for
which R(s) is stationary with respect to variation of s [14]:
dR(s)/ds=l +Id~(s)/ds=O. (2.4)
For an unscreened Coulomb potential, eq. (2.3) reduces to

q(s) = Z,Z,e’/Es, (2.5)


and eq. (2.4) can be solved analytically to yield

s0 = J Z,Z,e I/E . (2.6)


The corresponding Coulomb shadow cone radius is

R,=s,+Icp(s,)=2\j’Z,Z;eil/E. (2.7)
For the Moliere potential no such expression exists, but eq. (2.4) can be solved
numerically, resulting in a Moliere shadow cone radius of [15]
R, =tR,. (2.8)
The screening parameter 5 is plotted in fig. 2 against the “reduced” Coulomb
radius R Ja.
In backscattering experiments on single crystals surface sensitivity is ob-
tained by aligning the ion beam with a major crystal direction. In the example
given in fig. 3a for an ideal static lattice, only the top layer atoms are hit by the
beam, since the atoms further alongthe aligned atomic rows are located in the
shadow and, hence, cannot backscatter. The backscattered surface yield L,
expressed as the number of visible atoms per row, is then equal to one. The
thermal motion of the atoms, however, renders the shadowing imperfect and L
becomes larger than one. The thermal vibration period (- lo-l2 to lo-l3 s) is
much longer than the ion transit time through the surface region (- lo-l5 s),
so that the atoms in the row effectively remain frozen in their thermally
displaced positions during the passage of an ion. Fig. 3b shows a collection of
computer generated trajectories of ions incident on a lattice of harmonically
vibrating atoms with displacements that for each trajectory have been ran-
domly sampled from a Gaussian probability distribution [16] around their
equilibrium positions. The backscattering yield may then be written as

L=CPp (2.9)
J

where p, is the individual nuclear encounter (“hitting”) probability of the jth


atom in the atomic row. Because of shadowing, the pi are monotonically
J.F. onn der Veen / Ion beam crystallography 205

0 4 6
R,/a

Fig. 2. Universal curve showing the ratio c of the shadow cone for Molitke scattering to that for
Coulomb scattering as a function of reduced Coulomb shadow cone radius. From ref. [17]. with
permission.

decreasing with atom number j. By definition p, = 1. The backscattering yield


may also be expressed as an effective number of atoms N per unit surface area
by simply multiplying L by the area1 density of atomic rows exposed to the
beam.
Experimentally, the backscattering yield N (or L) is determined by taking
an energy spectrum of backscattered ions and measuring the area of its
“surface peak”. This procedure, that lies at the heart of any structure de-
termination, will be discussed in detail in section 3. A theoretical calculation of
L is cumbersome and often recourse is taken to Monte Carlo computer
simulations of the scattering process. Applying the latter approach to a single
atomic row of a crystal, Stensgaard et al. have shown that L is a monotonically
increasing function of the ratio a/R, [17]. Here (I is the one-dimensional rms
vibrational amplitude, which can be computed from the Debye theory of
thermal vibrations [16], and R, is the shadow cone radius evaluated at the
nearest-neighbour distance in the atomic row. This functional relationship is
borne out by the “universal curve” given in fig. 4 for many different pro-
jectile-target combinations. Clearly, excellent surface sensitivity of better than
3 atoms/row is achieved if the shadow cone radius is tuned to a value larger
than u. The universal curve has been derived under the assumptions of an ideal
bulk-terminated lattice and of independently and isotropically vibrating atoms
206 J. F. uan der Veen / Ion beam crystallography

Fig. 3. Collection of computer-generated trajectories for an ion beam incident on a crystal along a
low-index direction. Note the development of shadows behind the surface atoms and a few
small-impact parameter collisions leading to backscattering. (a) Scattering from an ideal static
lattice. (b) Scattering from a thermally vibrating lattice.

along the row. However, significant deviations from the universal curve occur
for relaxed and reconstructed surfaces and for surfaces with enhanced vibra-
tional amplitude with respect to the bulk. It is just these deviations that are of
interest in a determination of surface structure and dynamical properties.
Fig. 5a illustrates what is expected for a relaxed surface. While for normal
incidence of the beam the amount of shadowing is not affected by a small
change AZ in interlayer spacing, for non-normal incidence the second layer
atoms are no longer fully captured in the centre of the shadow cone and,
hence, contribute more to the backscattering yield. The amount of surface
relaxation can be determined by rocking the crystal axis around the ion beam
direction, while monitoring the effective number of atoms per row L that
J. F. uan der Veen / Ion beam crystallography 207

I”

i_ rHe++Si <too>
7- AHe++AU OlO>
6- vHe++W <llO>

* He+-- MO <lCXJ>
00.2 Mev He+-hlo <jo@ -HIGH T
2 _ + BARRETT’S “L” VANE - W<lll

I I IIll I
0.1 0.3 0.4 0.5 0.6 QB 1.0 2.0 : 3
d/Rhr(

Fig. 4. Universal curve showing the backscattered yield expressed as the number of atoms per row
versus the ratio of the vibrational amplitude to the shadow cone radius. The points in the curve are
based on computer simulations of the scattering process for many different projectile-target
combinations. From ref. [17], with permission.

contribute to backscattering. Such an angular scan then shows an asymmetric


curve with a minimum in the backscattering yield that is slightly displaced
toward the direction in which the second layer atom again becomes maximally
shadowed.
Reconstruction effects may also give rise to an increased surface yield. This
is illustrated in fig. 5b for a surface layer in which the atoms have formed
dimers. The atoms in the second layer are no longer shadowed and are all fully
exposed to the beam. Through variation of the shadow cone radius, i.e. by
changing the beam energy, the magnitude of the atomic displacements can be
determined. Fig. 5c shows the case in which the radius has been widened to the
extent that the second layer atoms are just captured in the shadow. The surface
atom displacement Ay is then equal to the radius for which this occurs. As for
the relaxed surface, the direction of the displacement follows from surface
backscattering measurements for different directions of the incident beam.
Additional contributions to the yield may be related to subsurface distortions
accompanying the reconstruction of the top layer.
Though the shadowing techniques described here can conveniently be used
as a stringent test of various structure models for a given surface reconstruc-
tion, it is much harder to use them for a model-independent ab initio structure
determination. In particular, it can be difficult to distinguish reconstruction
effects from enhanced surface vibrational amplitudes, which also give rise to an
208 J. F. uan der Veen / Ion beam crystallograph_v

A+?+j$fq~.
=..
L Y
4’. . . . I
tilt angle p

Ill
.
fifl$r
..
. . . ..ee

Fig. 5. Shadowing in relaxed and reconstructed surfaces: (a) determination of surface relaxation by
rocking a non-normal crystal axis around the ion beam direction; (b) exposure of extra atoms to
the beam in the surface region of a reconstructed crystal; (c) crystal as in (b). but shadow cones are
widened to the extent that these atoms are shadowed.

increased surface backscattering yield. If, however, shadowing is combined


with blocking, a direct determination of surface atomic positions becomes
possible.

2.2. Blocking

Since the cross section for a nuclear encounter is very small (see section 3.2)
a backscattered ion reaching the detector has undergone just a single close-
impact parameter collision with an atom in the lattice (the probability for
multiple large-angle deflections is negligibly small at the energies of interest). If
the backscattering atom is located beneath the surface layer, the backscattered
projectile may be blocked on its way out by another atom, as is illustrated in
fig. 6 for a static atom configuration. The backscattering atom 1 can be
considered as a point source emitting ions in different directions. The blocking
J.F. oan der Veen / Ion beam uystallography 209

Fig. 6. Shadow and blocking cones for scattering from a pair of atoms. For clarity, the opening
angles of the cones are drawn wider than they would be in reality. From ref. [14].

cone arises from small-angle deflections of the ions in the screened Coulomb
potential of the nuclear charge of atom 2. It can be shown that the opening
angle /I of the cone is given by /3 = R,/l, where R, is given by eq. (2.8) and I
is the distance between emitting and blocking atom [15]. As in the case of
shadowing, blocking along a row of atoms is rendered imperfect by thermal
vibrations. Because of time reversibility [18], the probability for an ion emitted
from atom n along the atomic row to leave the crystal parallel to the row
direction, is equal to the probability p,, for the same atom to be hit by a
parallel beam incident along the same direction. Consequently, a theory of
shadowing describes blocking as well.
At the author’s laboratory the technique of ion blocking has been developed
into a structure probe, by which positions of surface atoms can directly be
imaged [14]. Fig. 7a illustrates how a surface relaxation measurement proceeds
by the combined use of shadowing and blocking. The ion beam is incident
along a direction in which only the top two layers are exposed - the shadow
cone radius is taken sufficiently wide that the backscattering contribution from
deeper layers is negligible. A detector capable of simultaneously detecting a
range of scattering angles around the blocking direction accurately measures
the tilt angle Aa of the surface blocking cone with respect to the bulk axis. The
change in the first interlayer spacing then follows immediately from

4,
-=
tan(a+Aa) _ I
(2.10)
d tan (Y ’

where d is the bulk interlayer spacing and (Yis the angle of the bulk axis with
respect to the surface plane. Contrary to the shadowing technique, this method
essentially relies on simple trigonometry. The angular shift Aa of the blocking
J.F. uan der Veen / Ion beam crystallography

Fig. 7. Principle of surface relaxation measurement by the combined use of shadowing and
blocking, illustrated for the (001) surface of an fee crystal: (a) asymmetric backscattering
configuration with beam incident in the [ZOi] direction and detection around the [301] direction;
(b) symmetric configuration with the beam incident in the [3Oi] direction and detection around the
[301] direction.

minimum is largely insensitive to the screened interaction potential and to the


rms surface vibrational amplitude, provided a suitable scattering geometry and
shadow cone radius are chosen. However, the shape and depth of the blocking
minimum do depend on the latter parameters. Let us first consider the case
that the vibrational amplitude is much smaller than both shadow and blocking
cone radii. For the doubly aligned scattering configuration, in which both
beam and detector are aligned with shadowing and blocking directions, only
the topmost layer effectively contributes to backscattering, since the yield from
the second layer is almost completely blocked. Thus the yield in the blocking
minimum is approximately half that measured in single alignment, i.e. for
shadowing only. If the surface vibrational amplitude becomes comparable to
the blocking cone radius, but remains smaller than the shadow cone radius -
this may be the case when the nearest-neighbour atomic distance along the
shadow direction is larger than along the blocking direction, as in fig. 7a -
J.F. uan der Veen / Ion beam crystallography 211

then the blocking minimum becomes shallower, while its angle Aa remains
largely unaffected. For still higher amplitudes the shadowing in fig. 7a eventu-
ally becomes ineffective. The third and deeper layer hitting probabilities will
then become appreciable and Aa no longer reflects the true change in first
interlayer spacing. Other scattering configurations such as the symmetric one
given in fig. 7b are possible too. Here the cone radius is tuned such that the
second layer hitting probability pz is appreciable (say 30%) while p, - 0 for
n 2 3. Again, the blocking tilt angle ACXthen gives approximately the true
relaxation, while its depth is a sensitive measure of the surface thermal
vibration amplitude (see also section 4).
Blocking analyses of adsorbate geometries [19] and reconstructed surfaces
proceed in a similar fashion and will be discussed’in sections 5 and 6.

3. Backscattering spectrometry

3.1. The energy spectrum

When the ion beam is aligned with a low-index crystal axis the energy
spectrum of backscattered ions exhibits a “surface peak”, containing ions that
have backscattered from the exposed atoms in the surface region (fig. 8a). Ions
backscattered from somewhere in the bulk are detected at lower energies since
they have lost energy due to electronic stopping on their way in and out of the
crystal. The ion yield from the bulk is much reduced because of the well-known
channeling effect [6,9]. This can readily be understood from fig. 9, where an
ion travelling along an aligned direction is seen to make a gentle oscillatory
motion around the centre of the crystal channel without coming close to the
surrounding atomic strings. These channeled ions may penetrate microns deep
into the crystal. The residual backscattering yield from the bulk originates
from de-channeling effects related to thermal vibrations, possible crystal
disorder and multiple scattering [20]. When an energy spectrum is taken in a
backscattering direction aligned with a bulk crystal axis, the bulk yield behind
the surface peak is further reduced because of the blocking effect [21-231. This
geometry, in which both beam and detection direction coincide with a bulk
crystal axis, is commonly referred to as a double-alignment geometry. Since
this review is mostly concerned with analysis of surface peaks, the channeling
and bulk blocking phenomena themselves will not be further discussed.
The energy of the elastically backscattered particles from the surface region
is equal to the energy E, of the incident beam multiplied by the kinematic
factor KM>, where

1,
K _ (M,Z - M: sin%)“*+ M, cos 8 *
-
M2 (3.1)
w+M*
[
212 J. F. oan der Veen / Ion beam crystallography

a
M2 E0.V

A
0 0 0 0 0

0 0 0 0

0 0 0

0 0 0 0

0 0 0 0 0
0 \

0’

KM2Eo
BACKSCATTER ENERGY

(6,
f42 M3
o 0 0 0 0
l
0 0 0 00

0 o”oe.
l
0 ---_
0 0 0
\
0 p2 0 .o
. \
0 0 0 I
\
0 0 0 R\ ‘1

0 0 0

BACKSCATTER ENERGY

Fig. 8. (a) Energy spectrum of particles with incident energy E,, and mass M, backscattered
through an angle 0 from surface and bulk atoms of mass M,. The beam is aligned with a
low-index direction of the crystal. (b) Same as (a), but now for a surface covered with an
amorphous layer of mass M3 (M, z MS) and thickness Ax.

M, and M, are the projectile and target atom masses and 0 is the scattering
angle [8].
Fig. 8b shows how the energy spectrum changes when the crystal surface is
covered by a thin polycrystalline or amorphous film composed of atoms of
heavier mass M3. Since these atoms are randomly distributed with respect to
J. F. uan der Veen / Ion beam crystallography 213

Fig. 9. Channeling of a particle in a crystal. From “Channeling in Crystals”, by W. Brandt.


Copyright 0 (March, 1968) by Scientific American, Inc., all rights reserved.

the incoming beam they are all visible to the beam and so give rise to the
rectangularly shaped backscattering peak with its high-energy edge located at
an energy K,,E,. The thickness Ax of the film can readily be determined from
PI
Ax = AE/[S], (3.2)
where A E is the measured peak width and [S] is the energy loss factor, given
by

(3.3)

The angles 8, and /I, are defined in fig. 8b and d E/dx is the random stopping
power evaluated at the projectile energies along the incident and outgoing
214 J.F. oan der Veen / Ion beam crystallography

paths. Random stopping power values are tabulated as a function of H+ and


He+ projectile energy for many different elements [24]. From the knowledge of
the energy loss factor S and the energy resolution 6E of the detector one can
calculate the corresponding depth resolution 6x. Silicon surface barrier detec-
tors [25], which are widely used because of their practical convenience, have
typically a resolution of - 15 keV for He and - 5 keV for H, resulting in a
rather poor depth resolution of - 100 to 200 A [26]. The effective resolution
can be improved by a factor of - 5 if a backscattering geometry is chosen in
which the incident beam or the detector makes a glancing angle with the
surface plane (see eqs. (3.2) and (3.3)) [27,28]. However, at these glancing
angles and correspondingly long path lengths one must be aware of surface
roughness and energy straggling effects. An alternative is the use of a magnetic
spectrometer [6,29]. Unparalleled depth resolution of 3 to 5 A has been
obtained by electrostatic analysis of backscattered ions [30,31], permitting
detailed analysis of the composition and morphology of thin films (section 7).

3.2. The surface peak

One of the main advantages of the Rutherford backscattering technique is


that the scattering cross section is well known. A quantitative structure
determination of surfaces therefore becomes possible. Using the definition of
the differential cross section da/da one obtains the following expression for
the effective number of atoms per unit area contributing to backscattering:

N = A[ Q(du/dC?) As21 -I, (3.4)

where Q is the number of incident projectiles, A is the number of particles in


the surface peak (“surface peak area”) and A0 is the solid angle in which they
are detected. Eq. (3.4) is valid for a detector which measures all particles, i.e.
ions plus neutral particles. When an electrostatic analyser is used, the right
hand side of eq. (3.4) should be multiplied by l/n+= (A ++ A”)/A +, with A +
and A0 being the number of backscattered ions and neutrals, respectively. The
ion fraction q+ can independently be determined, as will be shown below. The
differential cross section is given by the familiar Rutherford expression cor-
rected for screening of the target nucleus by the surrounding electron cloud:

(3.5a)

where 8 is the scattering angle and F is a screening correction factor, which in


the Moliere approximation is given by [32]:

F = 1 - 0.042 Z, Z;j3/E [keV] , (3.5b)


J.F. oan der Veen / Ion beam crystallography 215

and where g(0, M,, M2) is a transformation factor from the centre of mass to
the laboratory frame of reference:

g( 0, M,, M,) f 1 - 2( M,/M,)* sin4(8/2), for M, -=z M,. (3Sc)


Depending on ion-target combination and ion energy the screening term lies
typically in the range 0.85 G F G 1.0, thus even at impact parameters much
smaller than the Thomas-Fermi screening radius the cross section does not
exactly follow the Rutherford law. For smaller values of F eq. (3.5b) breaks
down and a substantial angle-dependent deviation from the Rutherford expres-
sion results. This scattering regime will not be further discussed, nor the high
energy regime in which non-Rutherford behaviour relates to the occurrence of
nuclear reactions [33].
In practice eq. (3.4) is not used, since the effective solid angle of the detector
is often not accurately known. Instead, a commonly used procedure for the
determination of N is to calibrate the surface peak area against some back-
scattering standard (e.g. Si), in which an accurately known number N, of heavy
atoms (e.g. Bi) is implanted per unit area [32,34], or on which a known amount
of material is deposited [35]. One then obtains

N = (da/dfi)sAN (3.6)
(da/dS2) A, ”

where (da/d&?), is the differential backscattering cross section (eq. (3.5a))


from the implanted (or deposited) species in the standard. In writing eq. (3.6) it
has been assumed that the standard is measured using the same detector and
beam dose. When an electrostatic analyser is used for both sample and
standard the right hand side of eq. (3.6) should be multiplied by the ratio of
ion fractions 11:/n+. The accuracy of this calibration method is 1% to 2%. If
no accurate backscattering standard is available, the surface peak area can be
calibrated against the backscattering intensity in a non-channeling spectrum,
i.e. a spectrum which is taken with the beam incident in a “random” direction
not corresponding to any major crystallographic axis or plane [8]. This
procedure is less accurate (SW), because it is difficult to find a truly random
direction, and it also requires knowledge of the stopping power which may be
in error by 5%.

3.3. Instrumentation

A system for surface backscattering studies comprises a UHV scattering


chamber coupled to an ion accelerator through a differentially pumped beam
line. Fig. 10 gives a schematic overview of an ion backscattering chamber
developed at the author’s laboratory specifically for surface and interface
studies [14,31]. An ion beam collimated to within 0.1” impinges on a sample
mounted on a high precision goniometer [36]. Three independent axes of
wtndow

clraver
asscmhly

Fig. 10. Schematic of the experimental set-up for surface RBS studies at the FOM Institute. The
system is described in the text.

rotation allow a crystal plane to be aligned with the horizontal scattering plane
and a crystal axis in this plane to be aligned with the incident beam direction
with an overall angular accuracy of - 0.1”.
A toroidal shaped electrostatic analyser, which is rotatable around the
sample, permits simultaneous detection of ions in a 20’ range of scattering
angles centred around a selected blocking direction in the aligned crystal plane
[31]. The beam spot on the sample serves as the entrance slit, which combined
with the exit slit fixes the energy resolution at a value of SE = 4 x lo-’ E.
Since the analyser measures ions only, a conversion of surface peak areas into
N (or L) values by the procedure outlined in section 3.2 requires accurate
knowledge of the ion fraction q+, or alternatively the neutralization efficiency
1 - T)+= A”/( A + + A”) for both sample and standard. The latter quantity can
be determined directly by means of a simple detection system located along-
side the electrostatic analyzer (fig. 10). In front of a solid state detector
electrostatic deflection plates are mounted, which can deflect all ions out of the
backscattered beam. The neutralization efficiency then follows immediately
from the ratio of surface peak intensities measured by the detector with the
voltage on the deflection plates on and off.
J.E uan der Veen / Ion beam crystallography 217

In the energy range of 50 < E < 500 keV, the backscattered ion fraction
increases from - 30% at 50 keV to - 90% at 500 keV, its value depending on
type of ion (H+ or He+) and surface composition [37-401. In this energy range
9+ is found to be independent of the backscattering depth and of the
azimuthal and polar angle at which the ion leaves the surface. This greatly
simplifies the analysis of blocking effects using ions. For MeV projectiles n+
approaches 100% but here its determination is irrelevant for calibration
purposes, since solid state detectors are commonly used in this energy regime,
where electrostatic analysis becomes impossible.
A surface RBS system should contain various auxiliary equipment for
surface preparation and characterization. The set-up displayed in fig. 10
features a vacuum interlock system for sample introduction without breaking
the vacuum, a crystal cleavage facility, a sputter ion gun for surface cleaning, a
LEED system and a cylindrical mirror electron analyser for Auger electron
spectroscopy.
Rutherford backscattering spectrometry requires stable monoenergetic ion
beams. These may be produced in an ion implantation machine or a small Van
de Graaff accelerator if MeV energies are required. The accelerator typically
consists of an ion source, a first acceleration stage, an analysis magnet for
selection of the ion mass, and a post-acceleration and beam collimation stage.
Fig. 11 is a schematic drawing of a commercially available RBS system [41]
specifically developed for surface studies. The switching magnet can steer the
beam either into a UHV chamber, which houses a high precision goniometer
and toroidal analyser for surface blocking studies [42], or into another chamber
for ion implantation or routine RBS analyses of bulk samples. The voltages on
the 400 kV accelerator column and the energy analyser are stabilized to within
120 and 4 V, respectively, to fulfill the requirement that the energy drift is

con implan:atlon

LOOkV Ion accelerator

Fig. 11. Schematic of a complete RBS system with beam lines for surface RBS studies and for ion
implantation. Courtesy of High Voltage Engineering BV (Amersfoort. The Netherlands).
218 J. F. “on der Veen / Ion beam crystallograph_v

much less than the energy resolution of the analyser (&E/E = 4 x 10e3). It is
more difficult to stabilize the high voltages on the terminal of a Van de Graaff
type accelerator, but electronic feedback loops have been developed that
reduce voltage fluctuations to only a few hundred volt [43].

3.4. Measurement of blocking minima

The parallel detection scheme outlined in section 3.3 provides an elegant


means of measuring surface relaxation effects very precisely. Fig. 12 shows a
collection of energy spectra obtained for 98 keV H+ ions backscattered from
the Ni(OO1) surface in the shadowing geometry of fig. 7a [44]. These spectra
have been measured simultaneously over a - 13” range of backscattering
angles centred around the [301] crystal axis. The surface peak exhibits a
distinct blocking minimum in a direction, which is slightly tilted away from the
[301] bulk axis. The direction of the bulk axis is readily established from the
same set of spectra: it is found at the angle for which a minimum occurs in the
bulk yield at an energy well below that of the surface peak. Thus the surface
blocking tilt angle ACY and hence the surface relaxation follow from a single

Fig. 12. Collection of energy spectra from Ni(001) for 98 keV H + incident in the [20i] direction.
The spectra were measured simultaneously using the toroidal electrostatic analyzer displayed in fig.
10. The inset shows the correspondmg shadowing-blocking geometry. Surface and bulk blocking
minima are clearly visible.
J. F. uan der Veen / Ion beam ciysmllograph~r 219

111 1 I I 1 ,

10 15 20 25 30
EXIT ANGLE a DEGREE)

Fig. 13. Surface blocking pattern for clean Ni(OO1) measured with 98 keV H+ in the scattering
geometry of fig. 12. Each data point represents the integrated backscattering intensity under the
surface peak at the corresponding exit angle in this figure. The solid curve is a Monte Carlo
computer simulation for a 3.2% contracted surface. The vertical lines indicate the measured
directions of the bulk blocking minima. From ref. [44].

measurement, in which no mechanical movements of goniometer and detector


are involved. At each scattering angle the surface peak area can be converted
into the number of visible Ni layers by use of eq. (3.6) (1 monolayer of Ni(OO1)
is equivalent to 1.6 X 10” atoms/cm2). The resulting surface blocking pattern
is shown in fig. 13, in which the bulk blocking directions are indicated as well.
The [501] and [301] blocking angles are tilted by 0.3” and 0.4” toward lower
scattering angles, indicating a small contraction of the top layer spacing d,, by
3.2% [44] - this contraction is somewhat larger than would follow from eq.
(2.10), since the deeper layers are incompletely shadowed in this scattering
geometry. The solid curve represents a computer simulation of the scattering
process, to be discussed in section 4.

3.5. Beam-induced damage

Upon backscattering of a projectile into the detector a recoil energy of

(3.7)

is imparted to the target atom ( KM2 is given by eq. (3.1)). For a large-angle
backscattering event E, is typically a few percent of the beam energy E,, i.e. a
few keV. This energy is sufficiently large to displace the target atom from its
lattice site, which in turn creates a collision cascade of limited extent in the
surface region. Fortunately, the probability for such an event to occur is very
small, since most of the ions pass the target atom at a large distance and
220 J. F. uan der Veen / Ion beam crysrallography

scatter in the forward direction (see fig. 1). The total number of atoms
displaced per incident ion at the end of-the collision sequence is given by
N, = Na,N,, (3.8,)
where N is the effective area1 density of backscattering atoms at zero beam
dose (eq. (3.6)) a,, is the displacement cross section for a primary recoil atom
and Nd is the total number of displaced atoms in the cascade per primary
knock-on. In the Kinchin-Pease theory [45] ud and Nd are given by

(3.8b)
“’ =

and

Nd =$[I + ln[z I], (3.8~)

where E,““’ = [l - K,*( 19= ISO’)] E is the maximum recoil energy, Ed is the
displacement energy, i.e. the energy which must be transferred to kick the atom
out of a lattice site (typically between 10 and 25 eV), and E, is the Rydberg
energy (13.6 eV). For protons of a few hundred keV energy 3 6 Nd 6 5 and
lO_” < 0 < 10-l’ cm2 depending on target material. Consequently, N, is
only 10md’ to 10e2 per incident ion. This is a very low rate of damage
production. For example, after irradiation by a typical beam dose of 5 X 1014
H+ ions per cm2, only - 1012 to - 10” atoms per cm2 are displaced in the
surface region. This is less than one percent of a monolayer, so that almost
every newly arriving ion within this dose “sees” the surface in a virgin state. If,
on the other hand, excessive beam doses are applied, significant surface
damage will result and a structure determination becomes impossible. The

0 1 2 L
DOSE ( 1016 1ons/cm~1

Fig. 14. The increase in the number of visible Si atoms as a function of 174 keV He+ ion dose on
Si(ll1). The beam was aligned with the [iil] crystal axis. Data were taken on: (a) clean Si(lll), (b)
Si(ll1) covered with a Pd layer of 2.1 x 10” atoms/cm’ thickness and (c) Si(ll1) covered with
2.7 X lOI atoms/cm2 Pd. From ref. [46].
J. F. “on der Veen / Ion beam c~vsrallo~raphy 221

build-up of damage manifests itself as a steady increase in surface peak area,


reflecting the displacement of atoms out of the shadow cones. This is shown in
fig. 14 for high-dose He+ bombardment of a clean Si(lll) surface [46]. From
the measured rate of increase a displacement cross section follows, which is
even smaller than predicted by the Kinchin-Pease theory (eqs. (3.8)). This is
probably due to annihilation of defects at the surface. However, the damage
production rate is seen to be dramatically enhanced when the surface is
covered by an overlayer of a higher Z element. This is related to the much
higher cross section for displacement of overlayer atoms. These atoms all recoil
in the direction of the substrate, where they initiate a collision sequence.

4. Calculation of backscattering intensities

Though the determination of a surface structure could in principle proceed


via a geometric analysis of blocking directions as outlined in the previous
section, such an analysis would neglect the influence of lattice vibrations.
which blur the shadowing and blocking effects and which may introduce
significant backscattering contributions from deeper layers. These effects can
be accounted for by simulating the scattering of ions from the vibrating atoms
in the computer by Monte Carlo methods. Since the ion-atom scattering
potential is well-known (see section 2.1) the backscattered intensity can be
calculated for any desired scattering configuration. For a state-of-the-art
structure determination, it is now common practice to simulate complete
blocking patterns and rocking curves for a number of assumed surface struc-
tures and vibration amplitudes and to seek for the best fit to the experimental
data with the use of some x2 goodness-of-fit criterion (see also section 6.3).
Monte Carlo calculations play a key role in the quantitative analysis of
surface structures. Although one often refers to them as Monte Carlo simufu-
tions, since they involve the tracking of ions in the lattice as if it were a real
experiment, these simulations are not rigorous in the sense that one actually
waits for a real large-angle backscattering event to occur. This would cost an
excessive amount of computation time, because of the small backscattering
cross section (eq. (3.5a)). Instead, the nuclear encounter probability concept
[20] or modifications thereof are generally used. Various computation schemes
based upon this concept are discussed in the following subsections.

4. I. Scattering from a single atomic row

Consider the situation that scattering between adjacent rows of atoms in the
crystal does not contribute significantly to the surface backscattering intensity.
This condition is readily met if the shadow cone radius R remains much
smaller than the distance between neighbouring rows within the backscattering
depth of interest. The ions then sense only the presence of the atoms in the
222 J.F. “an der Veen / Ion beam cystallography

Fig. 15. Ion trajectory along a row containing n atoms. The ion undergoes discrete small-angle
deflections from the thermally displaced atoms in the row. Symbols are defined in the text. From
ref. [47].

nearest row and we just need to consider scattering from an isolated atomic
row in the crystal (“single-row approximation”). The scattering process can be
treated as a sequence of discrete small-angle deflections in the screened
Coulomb potential of the atoms along the row. It is further assumed here that
the atoms are vibrating independently; correlation effects in the thermal
motion will be treated in section 4.4. We now calculate the hitting probability
p,, of atom n along the row for the configuration of fig. 15 [47]. For a
definition of the symbols involved, one specific track - out of many possible
tracks - is shown. The ion is entering parallel to the z-axis at a distance x0
from the origin, deflected from thermally displaced atoms 1 to n - 1 at
positions (xi, xi,. . . , XL_, ) , and finally crossing plane n at a position x0 + A,
(only the coordinates in the (x, r) planes are relevant to the problem). A
nuclear encounter would have occurred if the atom residing in plane n were
located exactly at x0 + A,. The probability density p(x,) for such a track to
occur is given by
J. F. van der Veen / Ion beam crystallography 223

where G,( x;) designates the Gaussian probability density for the thermal
displacement of atom j [16]:

(4.lb)

Here xJ” is the equilibrium position of atom j projected on plane j and uJ its
one-dimensional rms vibrational amplitude. For simplicity isotropic vibrations
have been assumed here, but anisotropic distributions may be chosen as well.
All information on structure and vibrations in the surface region is contained
in the {XI”} and { uJ }, respectively, while the multiple deflections are all
“hidden” in the parameter A,, = A,(xl, x2,. . . , x,_,). To obtain the hitting
probability of atom n, we should integrate eq. (4.la) over all possible positions
{x;} and track starting positions x0:

Pn = J.../
n-l

=/.../G,,(+,+A,((x,, x2,...,x,_,)) ~C,(G+~,) d2xJd2xo,(4.2)


1
where we have substituted XI = x0 + xj and d2xj = d2x$ since integration over
xJ is performed at constant x0. The second integral expression for p,, is
explicitly written-out to emphasize the dependence of A, on the integration
variables (x,, x1,. . . , x,_,). With the exception of p = 1 and p2 (see section
4.2), there exists no exact analytical expression for this integral so that it must
be evaluated by numerical integration. Since there are many integration
variables (2n for a row of n atoms) the most practical way to solve it is by
statistical sampling techniques in 2n-dimensional space, i.e. by Monte Carlo
methods. The most efficient method proceeds as follows. First a random point
x0 is chosen uniformly from a sufficiently broad impact area surrounding the
atomic row. Then XI points are chosen quasi-randomly in accordance with the
Gaussian probability densities (eq. (4.lb)) of atoms 1, 2,. . . , n - 1. After
calculation of the track the probability density for a nuclear encounter
G,(x, + A,,) is evaluated. This procedure is repeated for many different tracks
starting at different x0 values, while accumulating the sum over G,(x, + A,).
After appropriate normalization this sum gives p,,. This computational scheme,
known as the “nuclear encounter probability method”, was pioneered by
Barrett [20], though it was formulated by him somewhat differently. Note that
for each track G,(x, + A,) can be accumulated for each plane j, so that all pJ
probabilities are simultaneously accumulated using the same batch of tracks.
This procedure yields directly the total number of visible atoms per row
L = Epj (eq. (2.9)). L can be obtained for any angle of the incident beam
direction with respect to the atomic row by performing the above calculation
224 J.F. uan der Veen / Ion beam crystallography

for the then different equilibrium positions {x; } of the atoms relative to the
new z-axis.
In the single-row approximation outlined above the atomic rows in the
crystal are treated as completely independent scatterers. If the lattice is
terminated by identical atomic rows, such as is the case for the configuration
of fig. 7b, we only need to calculate L for one type of row, i.e. for one set of
equilibrium positions {x,“}. In many other cases, however, the rows are
non-equivalent. For example, at the relaxed surface shown in fig. 7a the beam
scatters from relaxed rows ending in the top plane and from (nearly) unrelaxed
rows ending in the second plane. In the latter case L must be calculated for
both rows. In general, when referring to the whole crystal, one should redefine
L as the average over the different rows, but its use may then become rather
confusing. Ambiguities are avoided if backscattering yields are expressed as the
number of atoms per unit area N (see section 2.1) which is readily obtained by
calculating for each non-equivalent row its L value, multiplying it by the row’s
area1 density, and finally summing over all non-equivalent rows.

4.2. Two-atom model

The integral equation (4.2) for the hitting probability can be worked out for
the simple case of scattering from the first two atoms of the row [15,48,49]. The
probability for the second atom along the row is given by

PZ =/jG,( x,+A,)G,(x;) d2x; d2x,. (4.3)

Substitution of eq. (4.lb) yields

1 1x,+A212 _ ixo+xl -x:t2 d2x d2x,, (4.4)


exp - 1
p2 = (27r)2(J:(J; 11 i i
Here the equilibrium position xi of atom 2 has been chosen to lie on the
z-axis. Integration over x0 yields

1 _ IX1 -X:-421’

p2= 2a(a:+a;) J iexp 2( 0: + (12)


(4.5)

We now substitute
A, = -4(x,)%, (4.6a)
and
X 1” = reap, (4.6b)
where 1 is the equilibrium distance between atom 1 and 2, @(xi) is the
deflection angle at impact parameter xi (eq. (2.3)), f, is the unit vector along
xi, 8 is the angle between the equilibrium internuclear axis and the beam
J.F. van der Veen / Ion beam crystallography 225

entering parallel to the z-axis, and ap is an arbitrary unit vector in the (x, y)
plane of atom 1.
Introducing polar coordinates
d’x, = x, dx, dol, 2, $’ = cos (Y, (4.7)
replacing the impact parameter x, by s and integrating over cx we obtain

(s + 14(s)P
(4.8)
u: + u;

Here I, is the modified Bessel function of the first kind. The integral (4.8) can
be solved analytically for the simplest case of Coulomb scattering of a beam
aligned with the internuclear axis. Substitution of the Coulomb expression
(2.5) for the deflection angle 4 and integration over s then yield:

(4.9)

where K, is the first-order modified Bessel function of the second kind and
u+ = 2(u: + u,‘) is the two-dimensional transverse vibrational amplitude of
atom 2 relative to atom 1. As expected on the basis of arguments presented in
section 2, p2 is a function of a single parameter, namely the ratio of the
vibrational amplitude to the shadow cone radius. For the Moliere potential
and the condition 8 = 0 there exists no exact solution to the integral (4.8), but
numerical evaluation shows that in this case u,/RM is the proper scaling
parameter [15]. Formulas based on this scaling law have also been constructed
for the hitting probabilities-of the third and following atoms, but none of them
is exact [50,51]. They essentially reproduce the universal curve in fig. 4
obtained by the Monte Carlo method [17].

4.3. Double alignment

Using a simple extension of the method described in section 4.1 the case of
double alignment can be solved [47,52]. Consider the shadowing/blocking
configuration of fig. 16, in which ions enter the crystal parallel to the z’-axis
along path 1, then backscatter from the atom labeled a and finally leave the
crystal parallel to the z2-axis along path 2. For a given ion track atom a
occupies the same thermally displaced position xa, thereby correlating the
small-angle scattering events along the incident path to those along the exit
path. Let the three-dimensional probability density for the thermal displace-
ment of atom a with respect to its equilibrium position xf be given by an
isotropic Gaussian distribution

G,(s) = -4- ‘x”;;“‘2jj


(2T$,2ui (4.10)
226 J.F. van der Veen / Ion beam c~ystallogruph_~

Fig. 16. Ion trajectory in shadowing-blocking geometry. The black circles indicate the frozen
thermally displaced positions of the atoms, the open circles the equilibrium positions. From ref.
1471.

by analogy with the two-dimensional distribution (4.lb). We define F’(x,) as


the normalized probability for an ion track incident along the z’-axis to
intersect x, and Fz( x,) as the probability for an ion, when emitted from atom
a at xa, to leave the crystal parallel to the z2-axis. The double-alignment
probability p, da, that an ion backscatters from atom a and reaches the vacuum
is then given by the probability density F’(x,) . G,(x,) for atom a to be hit,
multiplied by F’(x,) and integrated over all possible positions x,:

p,“” = (4.11)
J F’(x,) G,(x,) F2(x,) d3x,.

Various Monte Carlo methods for the computation of the conditional prob-
abilities F’(x,) and F2(x,) and efficient schemes for the evaluation of (4.11)
have been discussed by Tromp and Van der Veen [47]. Note that by virtue of
time reversibility F2(x,) is equal to the probability for an ion track incident
along the z2-axis to intersect x, [18].
The connection to the single-alignment probabilities of section 4.1 is easily
made. The hitting probability of atom a is given by

p: =
Jf”(q) G,(q) d3X,, (4.12)

and its equivalence to eq. (4.2) for pk (atom a is the kth atom along path 1)
has been demonstrated [47]. Similarly, the single-alignment detection prob-
J.F. uan der Veen / Ion beam crystallography 227

ability is given by

p: =jf.‘(x,) G,(q) d3xa, (4.13)

which in turn is equivalent to (4.2), but now evaluated for atom a being the i th
along path 2. The double-alignment probability can very well be approximated
by

P,“”“PlP,‘. (4.14)

This approximation amounts to independent treatment of incident and exit


path with complete neglect of the above correlation effect and is accurate to
within - 2% for most scattering configurations.
In analogy with eq. (2.9) the number of atoms per row in double alignment
is finally obtained by summation over all backscattering contributions:

Ld"= cp;“. (4.15)

4.4. Correlated thermal motion

So far the thermal vibrations of the lattice atoms have been assumed to be
uncorrelated. In reality, however, correlations are always present and they
should be accounted for in a surface peak calculation [53-581. Consider for
simplicity a row of correlated atoms, which vibrate isotropically. The integral
over the product of Gaussian distributions in eq. (4.2) should then be replaced
by

Pn = J...J@(xi, x;, . . ., x;_,, x0 + A,) d2x; d2x; . . .d2x;_, d2x,, (4.16)

where + is the joint 2n-variate normal probability distribution function

1.
[16,59,60]

exp - + 5 2 (S-‘),kU,$ (4.17)


j=l k=l

Here u is a 2n-dimensional vector and i and j are generalized indices


representing the atom number along the row and the coordinate axis (x or v).
S,, is the equal-time correlation tensor S,, = ( u,u,) and 1S ( its determinant.
The diagonal elements S,, = (u,?) = u, are one-dimensional mean-square
vibrational amplitudes. Often the degree of correlation in the lattice motion is
expressed by the coefficient c,, = ( uiu,)/u,a,. In the Debye theory c,, = 0.36
for nearest neighbours in a cubic bulk lattice at high temperature [61,62]. The
effect of correlations in the surface region is to lower the surface peak
intensity, since it tends to smooth out the atomic row. It therefore opposes the
228 J. F. uan der Veen / Ion beam ctyystallography

effect of enhanced surface vibrations, making it difficult to identify experimen-


tally the presence of correlations. Nonetheless, evidence for correlations in
surfaces have been found in recent ion scattering studies [63,64].
The evaluation of the integral equation (4.16) is a formidable task. Again,
Monte Carlo procedures have to be used to construct random samples from
the thermal displacements, but now the {x:} can no longer be sampled from
independent Gaussian distributions, as was the case for uncorrelated vibra-
tions. A well-known strategy in statistics is to transform the 2n-variate
distribution (4.17) by a change of variables into a product of 2n-univariate
distributions [55,59]:

US’ u=v.v, v= Tu. (4.18)

Hence

FT= S-‘, (4.19)

where F is the transpose of matrix T. T is of triangular form and a method for


its construction is given in refs. [55,65]. To evaluate the integral first 2n
samples are drawn from independent Gaussian distributions and assembled
into the 2n-dimensional vector v. Then the transformation u = T-b is made to
generate a sample for the correlated 2n-dimensional vectors (xi, x;, . . . , xA_ ,).
A random starting position x0 is chosen for the ion and a track is calculated,
so that x0 + A,, becomes known. Finally, the conditional probability for atom
n to be hit at x0 + A, is evaluated. This process is repeated for many tracks,
while the sum over the latter probability is accumulated. After normalisation
p, is obtained for a row of correlated atoms. Similar schemes may also be
devised for double alignment.

4.5. “Complete-crystal” method

The Monte Carlo schemes outlined in the previous sections contain various
approximations, the most critical one being the assumption of scattering from
isolated atomic rows. However, if there is a non-negligible fraction of scatter-
ing events, giving deflections of ions onto neighbouring rows within the
considered depth interval, then the single-row approximation fails. The most
flagrant case of failure occurs when the beam is not aligned with the row, or
when it is incident along a high-index direction of the crystal, i.e. when the
distance between neighbouring rows is small. A partition of the crystal into
isolated rows is then no longer meaningful. This difficulty can be overcome by
the following “complete-crystal” approach. Consider a crystal slab having the
thickness of the backscattering depth under consideration. The slab is parti-
tioned into identical building blocks of the same depth, each having the lateral
dimension of the (n x m) reconstructed surface unit cell. Hitting probabilities
J.F. oan der Veen / Ion beam crystallography 229

are accumulated for all atoms in the building block, using the methods of the
previous sections, but with the important difference that the sequence in which
the atoms are encountered, is no longer that of the atoms in the row, but is to
be determined for every ion individually. Cross-overs from one block to a
neighbouring one are accounted for by imposing periodic boundary conditions
on the particle trajectories. Every atom in the block may have any desired
atomic number, equilibrium position and (an)isotropic vibration amplitude. A
single building block may contain as many as - 1000 atoms, if it represents a
reconstructed surface with a large unit cell (e.g. Si(ll1) - (7 X 7)) and apprecia-
ble lattice distortions in the subsurface region. Very efficient computational
schemes for the calculation of the collision sequence in such a large building
block have been developed by Frenken et al. [67]. The use of “complete-crystal”
methods is now becoming standard practice. They are essential for simulation
of all those experiments in which the shadow cone radii are a sizeable fraction
of the row-to-row distance.
There are various refinements to the Monte Carlo schemes, which may or
may not be important, depending on the experimental condition to be simu-
lated and on the desired accuracy of the structure determination [67]. These
are:
(1) Correction for non-zero impact parameter in the large-angle backscattering
collision. So far, the backscattering (hitting) probability was treated as the
probability for a head-on collision and the nucleus of the backscattering atom
was assumed to be a point source of particles. While this is a very good
approximation in the high-energy regime (OS-2 MeV H+ or He+), some
caution is required when medium energies (50-500 keV) are used. Fig. 6 gives
a schematic illustration of the effect: the source of backscattered ions does not
quite coincide with the nucleus of atom 1 but lies somewhat above it. As a
result, the direction of the blocking minimum in fig. 6 is slightly tilted. At
worst the impact parameter is - 0.01 A and the resulting blocking distortion is
about 0.1”.
(2) Correction for angular variation of the Rutherford cross section. The
particle flux emerging from the backscattering atom follows the angular
dependence of the Rutherford cross section (see eq. (3.5)) and this influences
the shape of the blocking minima in a doubly-aligned scattering configuration.
The effect is accounted for by weighting the probability density for each track
in and out of the crystal with the factor sin4(f?/2)/sin4(8’/2). 8 in this factor
is the (fixed) angle between beam and detector direction and 8’ is the true
backscattering angle for that particular track (fig. 16).
(3) Corrections for the divergence of the incident beam and for the finite
acceptance angle of the detector. In practical situations the ion beam is never
perfectly collimated, but has a certain angular spread (typically only a few
hundredths of a degree). The same angular spread should be imposed on the
Monte Carlo generated ions in the computer code. Similarly, the acceptance
230 J. F. um der Veen / Ion beam crystallography

angle can be accounted for in the time-reversed calculation of the outgoing ion
trajectories.
Computer codes in which all of the above features are combined generally
yield excellent fits to the experimental data, provided of course that the correct
surface structure and vibrational amplitudes are assumed. In the following
sections comparisons of experimental and simulated data will be made
throughout. Not only are the simulations used for surface structure analyses
but also for structural investigations of interfaces and for calculation of
dechanneling in thin-film systems.

5. Atomic geometry and dynamics of metal surfaces

The structure of a metal surface is often observed to deviate from that of a


truncated bulk lattice, because of changes in atomic coordination and electron
distribution in the surface region. These structural changes fall into two
categories. If the atomic displacements leave size and shape of the 2D surface
unit cell unchanged with respect to that of a truncated bulk lattice, we refer to
that surface as being refuxed and designate its unit cell dimension by (1 X 1).
The other category consists of all those arrangements that result in a new
periodic structure (super-structure) having a larger unit cell than (1 X 1). We
refer to the latter type of structural change as reconstruction.
In this section we review some.accomplishments of RBS pertinent to the
structure determination of metal surfaces. Some aspects of surface dynamics
will be discussed as well. The thermal displacements of surface atoms are
expected, and are indeed observed by RBS, to be very different from those in
the bulk. Their magnitude is found to be dependent on surface atomic
structure.
Where applicable, RBS results will be compared with theoretical predictions
of surface structure and dynamical properties and with data obtained by
LEED or other techniques.

5.1. Relaxation

The absence of reconstruction in a relaxed surface generally implies that the


atoms in the surface layer are in registry with those in deeper layers. Hence, the
problem of finding the atomic positions reduces to determining the distance
between atomic planes in the surface region. Till now, most investigations have
been concerned with finding the first interlayer spacing only, but the most
recent analyses have interrogated the second and even the third interlayer
spacings as well.
The very first shadowing and blocking studies of surface relaxation were
performed on Pt(ll1) [66] and Cu(ll0) [14], but some of these results (no
J.F. van der Veen / Ion beam crystallography 231

Table 1
Comparison of (multilayer) relaxation values determined by RBS and LEED for various metal
surfaces

Surface RBS LEED

Ah/d Ad,,/d Ref. Ad,,/d Adz/d Ref.


(%I (%) (W) (%)
Pt(lll}, clean +1.5+1 1631 0+5 PO31
<I21 1741 < (2.51 I1041
+1.3?0.4 [751
< 10.5 ( I781
H *-covered +1.3*0.4 [781

Pt(lOO}-(1 X 1). clean < 10.2 1 I791 0 (1051


CO- or H ,-covered +0.8 1791

W{lOO)-(1 x l)H <I41 [851

Ni( ill), clean < Ill f86.871 Ok5 (1141


-1*1 11071
(6 x J5)R30”-0 +7*1 wi871

Ni(lOO), clean -3.2kO.5 [44,881 Ok5 [1141


+3+6 11151
+1*1 11161
+3 ill71
+1*1 I1071
0+4 Pl81
0 v191
0 (1201
0 [I211
c(2 x 2)O +5.2+1.5 wm

Ni( llO), clean -4 +1 PO1 -5kl PO71


-4 *1 [82] -7 [108,109]
-4.8k1.7 +2.4+1.2 [81]
c(2 x 2)s +6 k3 1191
+ l/3 ML oxygen +1 +1 1801
+l MLCO <Ill [821

Cu{ llO), clean - 5.3 f 1.6 + 3.3k 1.6 [77] - 10 + 2.5 0 k2.5 [70]
-8 +3 1711
- 8.5kO.6 +2.3+0.8 [llO]

Ag(llO), clean -7.8+2.5 +4.3+2.5 [83] -7 [llll


- 10 11121
-8 (1131
- 5.7 + 2.2 11061
Pb(llO), clean -16.6+2.0” [841

a) (Ad,,+0.7Ad,,)/d= %0.6+3%.
232 J. F. cwn der Veen / Ion hrum ciyadlo~ruph~

relaxation for Cu( 110) and a 15% expansion for Pt(l11)) disagreed strongly
with LEED data [68-711 and were later contradicted by other RBS studies
[72277]. The initial difficulties were related to the use of too large beam doses
damaging the surface (section 3.5) or to surface contamination, but they were
soon overcome by more careful experimentation and the implementation of
various auxiliary equipment for surface cleaning and characterization (see
section 3.3). In table 1 we present a synopsis of RBS relaxation values. only
including those obtained on undamaged and well characterized surfaces.
The closest-packed surfaces (e.g. fee (111)) all exhibit very little or no
relaxation of the first interlayer spacing, while the most open surfaces (e.g. fee
(110)) are strongly ( > 4%) contracted. This dependence on packing density
reflects a common trend among metal surfaces and is in accordance with
theoretical predictions [89.90] and LEED observations [91,92]. For simple
metals (e.g. Al, Na, Pb) this trend has been related by Finnis and Heine [89] to
a smoothing of the atomic surface roughness by a spatial redistribution of
electronic charge. This results in flattened WignerrSeitz (WS) cells as depicted
in fig. 17 and the creation of a dipole layer. For each single WS cell to remain
electrically neutral, the positive ion cores in the first layer should move to the
electrostatic centre of each cell. i.e. the surface should contract. This simple
model correctly predicts the largest relaxation for the most open (atomically
rough) surfaces. Later. energy minimization schemes were developed for simple
metals which provide quantitative estimates of (multilayer) lattice relaxation
[93-951. These calculations account for the screening response of the electrons
to the shift of the top layer ions in varying levels of approximation. So far, the
most realistic treatment of electronic response has been given by Barnett et al.
[96698]. For Al, Na and Pb surfaces they predict damped oscillatory multilayer
relaxations, with an oscillation period equal to the stacking period of the
crystal. Experimental confirmation has come from LEED dynamical analyses
(Al(110) [99] and Na(lOO) [loo]) and very recently from an RBS investigation
of Pb(ll0) [84]. The RBS values for dd,, and deeper-layer relaxations in
Pb(ll0) (see table 1) agree very well with the predicted values.
For transition metals the amount of relaxation is usually smaller than for
simple metals, which probably relates to their much greater hardness. Exten-

s--s__’

Fig. 17. Smoothtng of electronic charge a! the surface of a simple metal. The related electron
transfer produces a surface dlpole layer. The top layer contracts and the second layer expands
slrghtly. a\ indicated by the ;1rrow\. After refs. [89.90].
J. F. oan der Veen / Ion beam crystallogmphy 233

sive RBS studies have been performed on the near-noble metals Pt and Ni
(table 1). Obviously, the theories formulated for sp-bonded (simple) metals do
not apply to these materials, since their cohesive properties are determined to a
large extent by d-bonding. Instead, semi-empirical tight-binding models have
proved quite valuable [loll. Using the latter approach, Allan and Lopez have
predicted top-layer relaxations of -6.5%, -2.7% and - 1.5% for the (110)
(100) and (111) surfaces of Ni [102]. Given the rather coarse simplifications
made in this model, these values compare surprisingly well with the RBS
results listed in table 1. Oscillatory multilayer relaxation is predicted too by
this model, and indeed appears to have been observed [81], but probably it
dampens out more quickly than for simple metals.
Table 1 includes a comparison with LEED relaxation values, listed for each
surface in chronological order. The overall agreement is fair, with the exception
of Ni(lOO), for which LEED indicates an expansion instead of a contraction.
We note, however, that the earlier LEED determinations for Ni(lOO), which
disagree most with RBS, are probably unreliable, since they were based on
analysis of too few diffracted beams. On the other hand, the reportedly more
refined LEED analyses for Ni(ll0) [108,109], yield, for yet unknown reason, a
few percent larger top layer contraction than the value of 4 to 5% found in
three independent RBS studies [80-821. In the RBS investigation by Feiden-
hans’l et al. [Sl] the amount of Ni(ll0) relaxation was determined by a rocking
scan as depicted in fig. 5a, i.e. by monitoring the effective number of atoms per
row as a function of tilt angle between the ion beam direction (300 keV He+)
and a specific crystal axis (no blocking effects were used). The rocking curve
around the [loll axis was found to be highly asymmetric, indicative of a
surface contraction (fig. 18a). However, from this rocking curve alone the first
layer contraction Ad,, could not be determined, since its effect on the
asymmetry is opposed by a possible expansion of the second layer. To be more
sensitive to the second layer, additional scans were measured around the [loo]
axis (fig. 18b). Feidenhans’l et al. performed Monte Carlo simulations of these
rocking curves for various assumed relaxation values Ad,, and Ad,, and for
different vibrational amplitudes, and compared these simulations with their
experimental data. Using an R-factor analysis similar to that used in LEED
[122], they concluded that Ad,,/d = -4.8 + 1.7% and Ad&d = + 2.4 f 1.2%.
Best fits were obtained for a Debye temperature of 325 K, which amounts to a
20% enhancement of the surface atom vibrational amplitude 0, with respect to
the bulk value.
Most intriguing is the observation that the surface expands upon chemisorp-
tion of foreign atoms. Half a monolayer of sulphur atoms adsorbed in a
configuration of c(2 X 2) symmetry on Ni(ll0) even reverses the sign of the
relaxation [19] and so does oxygen on Ni(100) (88,441. These RBS studies were
the first to establish unambiguously that the distance between the first and
second layer changes upon chemisorption. Chemisorption-induced expansions
234 J. F. uan der Veen / Ion beam ciy:allogruphy

2.33
a
2.40. 300 keV He -> NL(110)11011

;
LZZ.30.
\
In

;2.2w
11

,2.10-

~2.00-

%
1.90.

I .ElO’ I
-1.50 -I .oo -0.50 0.00 0.50 I .oo I .50

RNGLE (Degrees)

2.30-
b
2.20.
300 ReV He -> Nl(110)l1001

:2.10-
,”
;2.00-

:
c I .90.
(L

; I .BO.

d
_ 1 .70.
>
[L I .60-
ln

I .so-

1.40’
-1.50 -1 .oo -0.50 0.00 0.50 1 .oo
!
1.50

RN6LE (Oegrees)

Fig. 18. (a) Determination of surface relaxation in Ni(ll0) by rocking the crystal around the [loll
axis. The angle is positive away from the normal. The solid curve is the computer-simulated
rocking curve for the optimum model (see text). (b) Same as (a) but now for rocking along the
[loo] axis. From ref. 1811. with permission.

have also been reported for Ni(ll0) [80], Pt(lll) [78] and Pt(lOO) [79]. Again,
this trend confirms theoretical predictions based on the tight-binding model
[102]. The effect was most dramatically demonstrated in an ion blocking study
by Frenken et al. on Ni(OO1) [44]. By employing the symmetric scattering
geometry of fig. 7b for a measurement of the surface blocking tilt angle Aa in
both clean and oxygen covered Ni(OO1) they could directly determine the
change in relaxation (fig. 19). To achieve almost perfect shadowing of the third
235

10 15 20 25 30
EXIT ANGLE a (DEGREE)

Fig. 19. Blocking patterns for clean and oxygen covered Ni(001) taken with 52 keV Hf in the
shadowing-blocking geometry of fig. 7b. The solid curves represent computer simulations with
relaxation values of - 3.2% for the clean surface and +5.2% for the c(2X2)O surface, for a
number of vibrational amplitudes P,~ as defined in eq. (5.10): (a) p,* = 0.16 A (bulk value); (b)
P,~ = 0.18 A; (c) P,~ = 0.20 A; (d) p,* = 0.23 A; (e) p ,2 = 0.25 A. In the simulations for the
c(2 x 2)0 surface 0”. = 0.13 A and p ,* = 0.22 A were used. In the curve labeled 0 = 0 the oxygen
atoms have been removed without changing the parameters for the nickel substrate. Labels A, B
and C indicate the oxygen blocking directions, shown in fig. 20. From ref. [44].

l l l l l

oeoeoeoeo
l

oeoeoeoeo
l l l l L-
5
B
I1001

l l l l l

0: mckel
0: oxygen
Ni (001) - ~(2x2) 0

_L 1.76A
--• l l l l

Fig. 20. The Ni(OO1) surface with a c(2 x 2) oxygen overlayer in top and side view. Weak blocking
by oxygen atoms occurs in directions labeled A, B and C. The determined oxygen height and
substrate interlayer distances are indicated. From ref. [44].
236 J.F. uan der Veen / Ion beam crystallography

and deeper layers, the Hi beam energy was tuned to 52 keV. A 0.75 + 0.97 =
1.72” shift of the blocking minimum is observed after adsorbing half a
monolayer of oxygen in a configuration of c(2 x 2) symmetry (see fig. 20 and
section 5.3). This corresponds to an expansion of 8.4%. Thus the -3.2%
inward relaxation found for the clean surface (see section 3.4) changes into a
net outward relaxation of - 3.2 + 8.4 = + 5.2 f 1 .O%. The solid curves in fig.
19 represent Monte Carlo computer simulations of blocking in these relaxed
surfaces, to be discussed in section 5.4.
Note that these blocking measurements are very sensitive to oscillatory
multilayer relaxation, since they directly yield the change Ad,, in first inter-
layer spacing, virtually independently of the position taken by the second
layer. Additional measurements, taken in scattering geometries which also
probe the deeper layer relaxations, yielded Ad&d = + 1 + 1% for the clean
surface. By contrast, relaxation measurements based on rocking scans [77,78],
are rather insensitive, since the opposing Ad,, and Ad,, always reduce the
asymmetry of the rocking curve, making its shape resemble that of a nearly
unrelaxed surface.
Until now the relaxation effect has been investigated systematically only for
the simplest case of a low-index surface in which the movement of atomic
planes is restricted to the direction along the surface normal. We note that
higher-index surfaces such as bcc(211) and bcc(310) are in fact expected to
relax in a more complex fashion. Because of the broken symmetry the top layer
in these surfaces may be shifted not only along the normal but also parallel to
the surface plane. Recent LEED experiments have indeed given indications of
such an effect [92]. Here RBS could be very helpful, because of its higher
sensitivity to parallel displacements.

5.2. Reconstruction

While the origin of surface relaxation is now reasonably well understood,


this is not the case for reconstruction phenomena. Complicated structural
rearrangements may occur in a clean metal surface or may be induced by
adsorption of gases. Temperature-dependent reconstructions have been re-
ported too. A fair number of LEED investigations have been carried out
[122,123], but only a modest amount of success has been achieved and many
reconstructions are not yet well understood. In recent years, some structures
that have been subject to considerable controversy in the LEED literature,
have been solved with the help of other structure probes. An example is the
(1 x 2) reconstruction of Au(l10). To explain the (1 x 2) LEED pattern, which
is due to a doubling of the periodicity along the [OOl] direction, various models
have been proposed: the “paired rows”, the “missing rows”, and the “saw
tooth” models (fig. 21). The combined results of LEED [124,125], surface core
level spectroscopy [126], low-energy ion scattering spectroscopy [127] and He
237

paired rows

missmg rows “‘saw tooth” model

Fig. 21. Structure models proposed for the (1 ~2) reconstructed (110) surfaces of Ir. Pt and Au.
Hatched circles denote atoms in the top layer.

atom diffraction [128] suggested a modified missing rows model with top layer
relaxation and possibly the existence of sub-surface distortions. Later electron
microscopy [129] and scanning tunneling microscopy [130] studies showed that
indeed every other [ilO] row is missing. The other models were ruled out.
Recently, glancing X-ray diffraction gave evidence for a not too well-ordered
missing row structure with lateral pairing displacements in the second layer
and a vertically relaxed top row (1311. The existence of these distortions was
later confirmed by RBS [132]. Here the use of RBS was restricted to measuring
the number of laterally displaced atoms and the magnitude of these displace-
ments (0.18 _+0.04 A).
The Ir(ll0) and Pt(ll0) surfaces exhibit (1 x 2) reconstructions very similar
to that of Au(ll0). LEED slightly favoured missing rows [133,134]. though the
238 J.F. van der Veen / Ion beam crystallography

corresponding R-factor of 0.4 for Pt(ll0) [134] did not in fact constitute
acceptable agreement between calculations and experiment. Models for Pt(ll0)
involving significant lateral displacements such as the paired atom or the
hexagonal overlayer models were ruled out in a RBS experiment by Jackman et
al. [135]. Crystal rocking scans around the normal and around off-normal
directions were all found to be perfectly symmetrical. It was concluded that
any displacement of the surface atoms from bulk-like lattice sites must be
smaller than 0.02 A laterally and 0.07 A vertically, in marked contrast with the
distortions found for Au(ll0) [132]. The data were consistent either with
missing rows in an unrelaxed surface or with a very weak buckling. More
definite conclusions cannot be drawn at present. Various adsorbate-induced
structural changes on the Pt(ll0) surface have been reported as well [135,136],
but a satisfactory explanation of the underlying mechanisms is still lacking.
An interesting phase transition occurs on the (100) surfaces of Ir, Pt and
Au. These surfaces are all stable in a hexagonally close-packed arrangement
[137] (fig. 22). Their complicated LEED patterns (“(5 X 20)” for Pt and Au,
“(1 x 5)” for Ir) are therefore simply referred to as “hex”. These surfaces can
also be prepared in a bulk-like metastable (1 X 1) phase by NO or oxygen
adsorption and subsequent removal of most of the adsorbate by chemical

1x5 1x1
Fig. 22. Structure model for the (1x5) reconstructed (100) surface of Ir. Atoms in the
hexagonally-close-packed top layer are indicated by heavy circles.
J.F. van der Veen / Ion beam crystallography 239

0.61 , , , I ,j
400 450 500 550 600
TEMPERATURE (IO

Fig. 23. The CO-induced hex c~ (1 xl) phase transition in Pt(100) monitored by work function
change and RBS. From ref. [142], with permission.

reduction techniques [138-1401. A hex + (1 X 1) transition in Pt(lOO) is also


induced by CO adsorption [141]. In a comprehensive study using RBS, nuclear
microanalysis, LEED and work function measurements Jackman et al. [142]
followed the latter phase transition in detail by monitoring the number of Pt
atoms off bulk lattice sites and the corresponding CO coverage, while cooling
the sample at fixed CO pressure (fig. 23). The hex + (1 x 1) transition was
found to start abruptly when 0.08 f 0.05 monolayer of CO had accumulated
on the surface and to go to completion at 0.5 monolayer coverage. The
transition was thought to occur via nucleation of CO clusters of high local
coverage, causing the underlying Pt surface to convert to its bulk-like (1 x 1)
structure. The hysteresis effect upon reversal of the transition (fig. 23) was
attributed to the heat of CO adsorption being larger on the (1 x 1) surface than
on the hex surface.
240 J.F. oan der Veen / ion beam ctysrallography

Reconstructions resulting from interaction between Ni surfaces and ad-


sorbates have been the subject of numerous investigations, mainly aimed at
understanding surface catalytic properties or incipient oxidation. The in-depth
oxidation of Ni surfaces is preceded by a Langmuir chemisorption stage at low
coverages [143]. The saturation coverages for chemisorption were measured by
RBS to be - 0.3 monolayer for Ni(ll0) [144] and 0.46 + 0.04 monolayer for
Ni(lOO) [44], giving rise to fully developed (2 X 1) and c(2 X 2) LEED patterns,
respectively. The atomic structure of the oxygen covered Ni(llO)-(2 x 1) surface
has raised a substantial controversy which persisted for almost two decades.
Germer and MacRae originally proposed a structure model in which every
other [OOl] row of Ni atoms is missing [145]. However, subsequent LEED [146]
and ion scattering [80] studies gave evidence for a periodic (2 X 1) overlayer
structure on an unreconstructed surface. The first Low Energy Ion Scattering
(LEIS) studies by Heiland and Taglauer were inconclusive [147], but later
LEIS results obtained by Van den Berg et al. [148] again gave support to the
missing row model. Unambiguous proof for alternate missing rows was given
by Smeenk et al. [149], who used the shadowing-blocking geometries shown m
fig. 24a (the oxygen atoms, to which these measurements were insensitive, are
not shown). In geometry “A” the incoming ion beam was aligned with the
[?lO] direction. Trajectories of type “b” hit 100% of the second layer but were

Fig. 24. (a) The (001) scattering plane perpendicular to the Ni(llO)-(2 x 1)0 surface. The surface is
reconstructed according to the missing row model. Oxygen atoms are not shown. Panel A: beam
incident along (3701 direction. Panel B: beam incident along [ZiO] direction.
J.F. oan der Veen / Ion beam ctystollography 241

efficiently blocked along the outgoing [120] direction. If the surface layer were
unreconstructed, i.e. no missing rows, the ions following path “b” would be
backscattered mainly by the first layer and the second layer would have a
hitting probability much smaller than one. Thus alternate missing rows should
give rise to extra blocking by top row atoms along the [lzO] direction. The
same arguments apply to geometry “B”, where additional blocking by the top

20 25

exit angle (degree)

Fig. 24 (b) Surface blocking patterns for the scattering geometries of fig. 24(a). Solid and broken
curves are discussed in the text. From ref. [149].
242 J. F. uan der Veen / Ion beam crystallograph>

rows is expected to occur in the [130] and [350] directions. The measured
blocking patterns are shown in fig. 24b. The solid curves represent the data
taken from a clean unreconstructed surface. The additional blocking effects
were indeed observed, but not as strongly as expected. for a fully reconstructed
surface (broken curves in fig. 24b). Smeenk et al. therefore concluded that the
(2 X 1) missing row structure exists as islands covering about half of the
surface. In this experiment, the (2 X 1) structure had been prepared by adsorb-
ing 0.33 monolayer of oxygen at a substrate temperature of 470 K.
After all, the Ni(llO)-(2 x 1)0 controversy may have been caused by
different methods of surface preparation used in the different experiments.
Support for this view has been given in a Low Energy Electron Loss Spec-
troscopy (LEELS) study by Masuda et al. [150]. These authors found that
exposure to oxygen at 300 K produces a (2 x 1) overlayer structure on a
unreconstructed surface, which coexists with a few (2 X 1) reconstructed is-
lands, whereas subsequent heating to - 540 K converts the surface to the
reconstructed phase. It may not be surprising then, that the surfaces that were
reported to be reconstructed [148,149], were prepared by oxygen dosage at
elevated temperatures, whereas the reportedly unreconstructed surfaces [80,146]
had oxygen adsorbed at 300 K.
Evidence for a similar missing row structure in Cu(llO)-(2 x 1)0 has been
found in LEIS [151] and He diffraction [152] studies, but these results were
recently questioned in an RBS study by Feidenhans’l and Stensgaard [153].
In general, the origin of “missing row” reconstructions is not well under-
stood. The situation is all the more puzzling, since their formation requires
diffusion of atoms over the surface over long distances. This difficulty was
recognized by Bonzel and Ferrer [154], which led them to propose the “saw
tooth” model for the clean (llO)-(1 X 2) surfaces of Pt, Ir and Au (fig. 21) as
an alternative to the simple “missing row” model. The saw tooth can be
generated by single diffusion jumps of atoms, but despite its appealing aspects
it has so far been found incompatible with experimental data. Recently a very
similar, 90°-rotated, saw tooth model has been proposed for the oxygen
induced Ni(llO)-(2 x 1) reconstruction [155,156], but this model appears in-
consistent with the ion blocking data displayed in fig. 24 (the blocking dips
A-[lZO], B-[350] and B-[l?O] would become deeper than observed). This leads
us back to “missing rows” and the associated problem of long-range mass
transport. Presumably, nearby surface defects (e.g. ledges and kinks) act as
sources or sinks of atoms. Scanning tunneling micrographs indeed show that
such defects are abundant on these surfaces [155]. It is finally noted that the
hexagonally close-packed atomic arrangement on the (100) surfaces of Ir, Pt
and Au also requires long-range migration of surface atoms.
A reconstruction which does not involve mass transport is the (fi X a>-
R45” reconstruction of the clean W(100) surface. This surface has attracted
much interest, since it reversibly transforms to (1 X 1) at elevated temperatures
J. F. van der Veen / Ion beam crystallography 243

[157-1591. The now commonly accepted model for the reconstructed phase
involves alternate shifts of the top layer atoms parallel to the surface producing
“zig-zag” rows running along the (11) directions of the surface unit cell
[159,160]. RBS experiments by Feldman et al. showed that at 300 K in fact
only - 50% of the surface atoms are displaced from lattice sites [161]. This
would be consistent with a model proposed by Debe and King [159], in which
reconstructed domains coexist with unreconstructed domains, the latter being
stabilized by surface defects, e.g. steps. Note here that RBS and LEED are
complementary, in that RBS provides surface averaged structural information,
while LEED probes essentially the ordered regions. The origin of the tempera-
ture dependent phase transformation is not well understood. It has been
conjectured that it is a periodic lattice distortion associated with charge density
waves [158,162]. Very complicated phase transitions are induced in W(100) by
hydrogen adsorption. A quarter of a monolayer stabilizes the c(2 X 2) phase,
but higher coverages destabilize it again, leading to intermediate incom-
mensurate phases and finally to a reordering into a (1 x 1) phase at saturation
with hydrogen [163-1661. The last process involves the gradual return of
displaced atoms back to bulk-like lattice sites, as has been elegantly demon-
strated in RBS measurements by Stensgaard et al. [85].
Clearly RBS has been of great value in establishing the nature of some of
the complex reconstruction phenomena occurring in metal surfaces. Data are
steadily accumulating, but much remains yet to be learned about the driving
forces of the observed structural rearrangements.

5.3. Chemisorption sites and bond lengths

The effect of adsorption on relaxation and reconstruction being so clearly


established, the question immediately arises of where on the surface the
adsorbate atoms responsible for these structural changes are located. Because
of their small scattering cross sections adsorbates such as C and 0 do not
efficiently shadow or block. Nevertheless, half a monolayer of oxygen ad-
sorbed on Ni(lOO) gave rise to a,measurable blocking dip (A in figs. 19 and 20)
and a overall change in shape of the blocking pattern, from which the
symmetry of the adsorption site and the height of the oxygen atom above the
5.2% expanded top plane of Ni atoms could be determined [44]: adsorption in
the four-fold symmetric hollows and d, = 0.86 f 0.04 A (fig. 20) in excellent
agreement with surface EXAFS determinations [167,168]. The structure model
given in fig. 20 is the only one compatible with the observed c(2 x 2) LEED
pattern and the measured oxygen coverage of 0.46 _t 0.04 monolayer [44].
Stronger blocking was observed for sulphur on Ni(llO), because of its
higher Z-number [19]. Measurements taken in the scattering geometry of fig.
25 showed that the adsorbed sulphur atoms reside in the deep hollows. A
blocking pattern taken for 0.5 monolayer (fig. 26) exhibits a distinct blocking
244 J. F. uon der Veen / Ion beam crystallography

l/2 monolayer S on tourtold site

101 heV H+-Ni (110)


0.5 ml of adsorbed sulphur
beam direction: [loll
scattermg plane: (iii)

0
-u
second layer

L5 50 55 60 65 70
-scattering angle 8 (degree)

Fig. 25. The combined use of shadowing and blocking for adsorbate position determination.
schematically illustrated for the sulphur-covered Ni(llO)-c(2 x 2) surface. From ref. 1191.

Fig. 26. Blocking pattern for Ni(ll0) surface covered with half a monolayer of sulphur. The
scattering geometry is that of fig. 25. The position of the Ni-S blocking cone in accordance with
the LEED d.-spacing of ref. [169] is indicated too. From ref. [19].

minimum due to sulphur (cone a in fig. 25) at a scattering angle of 52.0”, from
which, by triangulation, a height of d I = 0.87 + 0.03 A was deduced. Also
indicated in fig. 26 is the scattering angle at which blocking would occur if the
sulphur atoms were to occupy the LEED position of d, = 0.93 A [169]. The
horizontal bar indicates the quoted LEED uncertainty of +O.l”, converted to
scattering angle. This comparison not only shows good agreement between the
two methods, but also demonstrates the superior precision of RBS in determin-
ing adsorbate atom positions. The angular shift of ACX= 2.5” of the main Ni
surface blocking dip in fig. 26 (cone b in fig. 25) corresponds to a substrate
expansion of 10 Jo 3% relative to the clean surface relaxation value. The clean
surface is contracted by 4%, thus the sulphur covered surface shows a net
expansion of - 4 + 10 = 6 f 3%. A hard sphere model of the complete chem-
isorbed structure is shown in fig. 27. The sulphur atom radius of 0.94 A
derived from it is only slightly smaller than the standard single-bond covalent
radius of 1.04 A [170]. In general, adsorbate-metal bond lengths are close to
J.F. oan der Veen / Ion beam crystallography 245

. nickel
Q : sulphur

Fig. 27. Hard-sphere structure models for the clean and sulphur covered Ni(ll0) surfaces. Dashed
lines: Ni spheres in the top layer of clean Ni(ll0). showing a contraction of 4%. Solid lines: Ni
spheres in the sulphur covered surface, corresponding to a net expansion of 6%. From ref. 1191.

known bond lengths in molecules and bulk compounds, and to detect any
variation a high precision is required. Many earlier LEED studies and so far all
SEXAFS [171] and Normal Photoelectron Diffraction [172] analyses implicitly
assumed the substrate to be unperturbed, resulting in larger error margins in
d I -spacings than is often claimed.
The sulphur and oxygen blocking experiments were the first and still are the
only available RBS determinations of adsorbate atom positions. The use of ion
blocking is certainly one of the most attractive methods for this purpose.
Unlike the shadowing method the blocking method does not require the
adsorbate to be close to an atomic row of the substrate to be located.

5.4. Dynamics of surfaces

The lattice dynamics of surfaces has been an active topic of theoretical


study during the last decade 11731, but only few experimental data yet permit a
meaningful quantitative comparison with lattice dynamics theories. Inelastic
electron [174,175] and atom [176] scattering studies have now made it possible
to determine w versus k,, dispersion relations for the various surface vibra-
tional modes. Early temperature-dependent measurements of LEED intensities
have provided information on the mean-square amplitude of surface atoms
[177,178], though it was later recognized that amplitudes thus extracted could
be seriously in error [123].
Recently, ion shadowing and blocking have also contributed to this interest-
ing field by providing quantitative information on rms surface vibrational
amplitudes and correlation functions. First some relevant aspects of the theory
246 J. F. oan der Veen / Ion beam ctysrallography

of lattice dynamics in the harmonic approximation [16] are treated, then


contact is made with available RBS data.
Besides the atomic coordinates, the crucial parameters entering a Monte
Carlo simulation of shadowing or blocking are the correlation functions
Sap(f, f’) = (u,(f)u,(f’)) (eq. (4.17)), where u,(f) is the cuth Cartesian compo-
nent of the displacement of the atom at site I. A convenient method for their
calculation is the use of continued fractions [179-1821. For a definition of the
quantities involved some general relationships will first be given, following the
notation of Black et al. [183]. The equations of motion for the atoms in the
crystal are given by [16]

where Dap(l, l’) is the dynamical matrix, containing as its elements the
inter-atomic force constants. The force constants in the bulk are well known
from elastic properties [184]. At the surface, however, these constants are
expected, and indeed found, to be different from their bulk value. Models for
force constant changes at relaxed surfaces have been formulated by Allan for
transition metals [loll and by Barnett et al. for simple metals [185], allowing
the dynamical matrix to be constructed.
The eigenfrequency w, of the s th normal mode of the semi-infinite crystal
and the associated eigenvector e;‘(l) can be extracted from the set of
eigenvalue equations:

w2e’“‘(f)
5 n = c xDa,(f, I’) eg”‘(f’). (5.2)

Because Dap(f, I’) = Dpa(f’, I), the frequencies w, are real and the e’“‘(l)
satisfy the orthonormality and closure relations. For a system in thirmal
equilibrium at temperature T the correlation functions may be written as

(u,(f) up(f’)) =F &e?)(f) eF’(f’) (1 + 2n,), (5.3)


s
where jj, = (ehwr/kr - I)-’ is the Bose-Einstein function which gives the
number of thermally excited phonons associated with the mode of frequency
0,. Rather than trying to solve the normal mode eqs. (5.2) and (5.3) directly
for, e.g., a slab with free surfaces, which is very tedious, we define the spectral
density function

Pa& I’; a>=c e;)(f) ef’(f’) S(w -as), (5.4)

which enables the correlation function to be written as

(u,(f) am) =/,, da&--(1 + =,) p&f, f’; 0). (5.5)


J.F. uan der Veen / Ion beam crysrallography 247

In the generalized form of eq. (5.4) the spectral density provides information
on the characteristic frequencies which contribute to correlations between
displacements u,(f) and up(l’). It can be calculated by use of the highly
efficient method of continued fractions [179-1821. For this the Green function
is introduced, defined by

el;“( I) ek)( I’)


G,,(I, I’; z) =c Ld2 - z
2 ’ (5.6)
s s

so that, using eq. (5.4),


-2 .
~(1, I’; a)=- hm,,, Im[wG+(I, I’, wti~)]. (5.7)
rr
The expansion of Green functions in continued fractions has been discussed
extensively in the literature [181]. For example, the Green function used to
form the spectral density for mean-square displacements in the surface can be
expanded into
1
G&s, I,; z) = (5.8)
IB212 ’
z%4, -
z2-~2-Ly)

. . .

where I, designates an atom in the top layer. The series of coefficients A, and
B,, depend on the inter-atomic force constants coupling the atoms in a cluster
surrounding the atomic site in question. A,, and B, can be generated by a set
of recursion relations following the procedures described by Haydock et al.
[181]. Cluster sizes have to be at least - 1000 atoms for the calculation to be
accurate. Similar schemes exist for the computation of off-diagonal elements.
Using the above methods in a combined theoretical and experimental study,
Frenken et al. have investigated the influence of relaxation on surface dynami-
cal properties of Ni(OO1) [88]. A calculation of the dynamical matrix based on
a tight-binding model fit to bulk properties, showed that the 3.2% contraction
of the clean surface enhances the force constants between first and second Ni
atoms by - 20%, whereas the 5.2% expansion of the c(2 x 2) oxygen covered
surface results in a reduction by 30%. A calculation of the rms displacements
using the continued fraction technique then yielded for the clean surface

U] I = (l$)” = 0.102 A, Ul,, = (z&.)‘” = (&)“” = 0.097 A, (5.9a)


u* I = (&)“2 = 0.090 A, u2,/ = (z&)” = (u;,.)“* = 0.080 A, (5.9b)

to be compared with a bulk value of ubulk = 0.080 A. Thus the surface


amplitude u, I is significantly enhanced with respect to the bulk value, but not
quite as strongly as the factor of - 1.4 expected on the basis of missing
neighbours alone. Clearly the bond stiffening related to the surface contraction
reduces the enhancement and causes it to be less anisotropically distributed
over the components u, I and u,,, [88.186]. It is also seen that the predicted
second layer amplitudes nearly assume the bulk value. The theoretical ampli-
tudes thus obtained have been compared with the results of the ion blocking
experiments on Ni(OO1) discussed in section 5.2. The depth and shape of the
blocking minimum displayed in fig. 19 is largely determined by the 2D rms
relative amplitude p,: of the first and second layer atoms perpendicular to the
(301) shadowing and blocking strings (see fig. 7b):

p12=[((7fL +a:,) COs’,+(0<,+(~~,,)(1 tsin’a)] ‘, (5.10)

where a is the angle between string and surface plane. In writing eq. (5.10) use
is made of the fact that neighbouring atoms along this string, being second-next
nearest neighbours in the lattice, vibrate almost uncorrelated. Substituting the
amplitudes of eqs. (5.9) in expression (5.10) we obtain a predicted value of
p,: = 0.19 A. This value is to be compared with the results of fitting the surface
blocking data to Monte Carlo simulations of the experiment. The fitting
procedure yields p,: = 0.20 I 0.01 A (solid curve c in fig. 19). in excellent
agreement with the calculated value. Note that for the shadowing-blocking
conditions used here the parameters dd,,/d and p,? could be determined
completely independently, the former indicating a mere shift of the complete
blocking minimum. the latter causing a change in depth. Such a condition
cannot easily be met in RBS experiments based on shadowing only.
The dynamics of the oxygen covered c(2 X 2) Ni(OO1) surface has been
analyzed by Frenken et al. along the same lines [88]. In both calculation and
experiment the amplitude plr was found to be hardly affected by oxygen
adsorption. However, an important difference from the clean surface result is
that the interlayer force constants at the expanded surface are weakened with
respect to their bulk crystalline value. This weakening may well be a precursor
to oxidation, i.e.. penetration of oxygen into the surface.
Enhanced vibrational amplitudes have further been reported in RBS investi-
gations of Pt(lOO) [79]. Pt(ll1) 180.751, Ni(ll0) [81]. Ni(ll1) [86,87], and
Cu(ll0) [77]. A substantial enhancement initially reported for W(100) by
Feldman et al. [187] was later attributed by these authors to a reconstruction
effect [85] (see section 5.3). Often the enhancement is simply expressed as an
effective lowering of the Debye temperature 0. In the high-temperature limit
of the Debye model the 1D mean-square amplitude of an atom with mass M is
given by [16]
(u3) = .?h-‘T/MkO’. (5.11)
where k is the Boltzmann constant. Anisotropy in the thermal displacements
can be taken into account by introducing different surface Debye temperatures
0, and O,, for the perpendicular and in-plane components.
J. F. uan der Veen / Ion hemn c~sstollogrcrph> 249

Evidently, for an RBS determination of vibrational amplitudes to be relia-


ble, either nearly uncorrelated atomic strings (i.e. strings with large atom
spacing) must be selected in the experiment or correlations must explicitly be
taken into account in the Monte Carlo simulation, using the formalism
outlined in section 4.4. Yet there often remain considerable uncertainties
regarding the exact distribution of the enhancement over the first atomic
layers, the degree of anisotropy, and the magnitude of the correlation coeffi-
cients at the surface, which may well be different from those in the bulk. In
fact, an “ab initio” determination of all these vibrational parameters by RBS
would be a formidable, if not impossible, task. Occasionally, one of these
parameters can be isolated. Anomalously low backscattering yields from
Pt(ll0) could only be explained by invoking correlations at the surface, equal
in strength to those in the bulk [64]. Rocking scans taken on Pt(lOO) indicated
enhancement factors of 2.0 and 1.1 for the first and second (third) layer
amplitudes [79]. In many cases, however, a more realistic approach is to test
predictions of lattice dynamical models, as was done for Ni(OO1) [88].

6. Semiconductor surface structures

Much experimental and theoretical work has been devoted to the determina-
tion of semiconductor surface structures [3,188]. Of all semiconductors, silicon
is undoubtedly the most studied, because of its importance in modern technol-
ogy and also because of the intriguing complexity of some of its surface
reconstructions. In general, reconstruction phenomena in semiconductors are
intricately coupled to often drastic changes in the surface electronic structure.
The electrons in tetrahedral semiconductors (e.g. Si, GaAs) reside in highly
directional bond orbitals, pointing along the four tetrahedral directions. At a
bulk-like surface these bonds are left “dangling” and remain partially oc-
cupied, which is energetically unfavourable. Two types of reconstruction have
been proposed, by which the surface energy can be lowered: (I) buckling
distortions and (2) various rearrangements, involving the formation of new
bonds at the surface. The reconstruction is usually not restricted to the top
layer only; the bond angle strain associated with the reconstruction is partially
relieved by lattice distortions in deeper layers.
In comparison to LEED, RBS investigations of semiconductor surface
structures have been relatively few in number. Even so, they have yielded a
substantial amount of information on the atomic geometries of the recon-
structed (111) [189-1921 and (100) [193-1981 surfaces of Si and on the
buckling distortions occurring in the (110) surfaces of GaAs [199], GaSb
[200,201] and InAs [202]. While the structure of the latter surfaces could be
directly determined by ion blocking, the very complex atomic rearrangements
in the Si surfaces have so far precluded an ab initio structure determination.
The RBS approach taken there is similar to that commonly used in LEED
analyses; various plausible structure models, including those resulting from
energy minimization calculations, are tested for their ability to yield an
acceptable fit to the experimental data. In RBS, such a test usually consists of
simulating ion blocking patterns or crystal rocking scans (see section 4) for a
series of assumed models and then passing a judgment based on some x’
goodness-of-fit criterion. This procedure, though cumbersome and time-con-
suming, has helped resolve some persistent controversies concerning the nature
of some of the reconstructions occurring in Si. Besides providing a rigorous test
of structure models, RBS investigations have enabled model-independent
conclusions to be drawn on the total number of displaced Si atoms, the
magnitude of their displacements, the number of distorted layers, the effect of
adsorbates and the rms thermal vibration amplitude of surface atoms.

6, I. Si(100)

The clean Si(100) surface exhibits a (2 x 1) LEED pattern. Energy depen-


dent RBS measurements by Feldman et al. [193,194] and Tromp et al.
[195-l 981 have revealed large lateral displacements in the top layer and
extensive sub-surface distortions. Fig. 28 shows the extra number of visible Si
layers above that expected for a bulk-like surface as a function of the Moliere
shadow cone radius (eq. (2.8)). Even for a radius as large as 0.4 A about a full
extra monolayer remains visible, indicating that the atoms in the top layer are
on average displaced by - 0.4 A. As the shadow cone is tuned to smaller radii
(by increasing the energy of the primary beam). the number of visible mono-
layers rises until the equivalent of - 4 monolayers is seen at a shadow cone
radius of 0.06 A. This is clear evidence of sub-surface distortions.
A number of reconstruction models, such as the missing row [203], the
conjugated chain [204] and a model involving (111) microfacets [205] could be
ruled out, since they gave either a smaller or larger number of displaced
monolayers than measured [198]. Models involving dimerization of surface
atoms [206-2113 were found to describe the data best. To test further the
merits of symmetric [206,207] and asymmetric [208-2111 dimer models (fig.
29) Tromp et al. have measured blocking patterns in various scattering
geometries and found that only an asymmetric dimer configuration including
sub-surface distortions yielded satisfactory fits [197.198]. The distortions in
deeper layers were calculated by minimizing the elastic strain. using a bulk
force constant model after Keating [212] and imposing the first layer atomic
positions as a boundary condition.
The dimer model also explains the LEED pattern in a natural way. An
intense two-domain (2 x 1) pattern is seen, which correlates with the presence
of equally probable 90°-rotated domains being one atom step apart in height.
Weak quarter-order spots have also been reported by LEED [205] and He
251

0
0 01 04 0.5
SHAD% CONE F&S Cfil

Fig. 28. Number of extra monolayers as a function of the Moliere shadow cone radius, for the
Si(lOO)-(2x 1) surface (full points) and the Si(lOO)-(1X 1)2H surface (open points). From ref. [35].
Circles have been taken from refs. [189,193.194].

diffraction [213]. They are believed to relate to a secondary reconstruction


phenomenon, i.e. different arrangements of thl: tilted dimers in the surface
plane.
The dimerization essentially re-establishes a strong sp3-bond between two
surface atoms, leaving them with only one half-filled dangling bond (fig. 29a)
instead of two half-filled dangling bonds that they would have had in a
bulk-like configuration. The resulting energy gain with respect to the ideal
(1 X 1) surface is calculated to be - 1.5 eV/dimer [210]. Additional energy
(0.1-0.2 eV) is gained by tilting the dimer and allowing for subsurface
distortions [210,211]; the associated charge transfer in the asymmetric dimer
then completely fills the dangling bond on the “ up”-atom and empties the one
on the “down’‘-atom (fig. 29b), yielding a semconducting surface with band
dispersions in agreement with photoemission data [214]. We conclude that the
RBS data [198] are consistent with these theore-.ical insights.
Upon saturation of the Si(100) surface with atomic hydrogen the LEED
pattern becomes (1 x 1) and the effective number of displaced Si layers is seen
by RBS to decrease dramatically [195,196] (fig. Z!S). This means that the dimers
are broken up, resulting in a more bulk-like arrangement in which (supposedly)
two hydrogen atoms saturate the two dangling bonds of each surface atom.
(Nuclear microanalysis of deuterium adsorption by Stensgaard et al. [194]
revealed a somewhat lower saturation coverage, but this could be related to
Fig. 29. Top and side view of dimer models for the Si(lOO)-(2 x 1) surface. The open circles in the
lower half of the figure give bulk positions. Atom displacements are drawn to scale. Surface states
originating from dangling and bridge bonds are schematically shown. The hatchings indicate the
filling of the associated surface bands. Structure models: (a) symmetric dimer and (b) asymmetric
dimer.

unintentional coadsorption of some hydrogen during exposure of the surface to


deuterium.) The hydrogenated surface layer is weakly contracted and has a
strongly enhanced rms thermal vibration amplitude [196].

6.2. Si(ll I)

When a Si crystal is cleaved in UHV along the [ii21 direction, the exposed
(111) surface exhibits a single-domain (2 x 1) LEED pattern with the two-fold
periodicity along the direction of cleavage. Until recently the buckling model
due to Haneman was widely accepted [3,215,216]. In this model alternate rows
are slightly raised and lowered, as shown in fig. 30a. However, recent self-con-
sistent total energy calculations by Pandey have shown buckling and its
associated charge transfer to be unstable for this surface [217]. Furthermore,
ionic buckling is incompatible with photoemission results from valence states
[218] and core levels [219,220]. These inconsistencies led Pandey to propose a
stabilization of the surface by a “m-bonded chain” reconstruction [4] (fig. 30b).
In this configuration the bonding topology is completely changed, with the
effect that the dangling-bond orbitals which otherwise reside on next-nearest
253

~1 bonded chain

,- upper’chaln

- lower chain

Fig. 30. Structure models for the Si(lll)-(2x 1) surface with corresponding shadow effects: (a)
weak buckling distortion, (b) n-bonded chains and (c) tilted n-bonded chains.

neighbours, now reside on nearest neighbours. The proximity of these bonds


allows them to interact and form a a-bond, causing a reduction of the energy
by - 0.30 eV per surface atom. Alternative reconstruction models have been
conceived by others, in which the surface is thought to be stabilized by the
formation of m-bonded “molecules” [221] or by an antiferromagnetic arrange-
ment of the dangling bonds on the surface [222-2241.
Though the m-bonding model may explain photoemission data, no unique
conclusion on the atomic geometry of the (2 X 1) surface could be drawn on
the basis of these data alone [218,225,226]. An ion blocking investigation by
Tromp et al. [192] has finally resolved this controversial issue. The scattering
254 J. F. oan der Veen / Ion beam crysrallography

IC
I I / r 1 ” 1 t
95 loo 105 110 115 120
SCATTERING ANGLE [DEGREES]

Fig. 31. Panel a: blocking patterns for the cleaved Si(lll)-(2 X 1) surface, taken with 99 keV H+ in
the scattering geometry shown in the right-hand half of fig. 30 (squares) and in the 2’ rotated
geometry (triangles) (see text). Panels b and c: comparison with simulated blocking patterns for a
bulk-like surface (B), a buckled surface (BC), the n-bonded chain model (PC) and the T-bonded
molecule model (PM), for the two different backscattering geometries. From ref. [192].

geometry chosen to distinguish between buckling and n-bonding is shown in


fig. 30. Obviously, the r-bonding arrangement leaves many more sub-surface
atoms exposed to the ion beam than the buckled surface. Fig. 31 shows
blocking patterns for the scattering geometry of fig. 30 and in the same
geometry, but 2’ rotated around the surface normal. A comparison with
Monte Carlo simulations for the different models gave strong support for
Pandey’s r-bonded chain model and essentially ruled out the other models.
The sub-surface distortions concomitant with n-bonding were taken into
J.F. oan der Veen / Ion beam crystallography 255

account, but minor discrepancies between simulations and measurements


remained. These could be removed by tilting the first and second layer chains
[227], as in fig. 30~. This finally reconciles the a-bonding model with LEED
1-I’ analysis, which earlier discriminated against m-bonding [216], but now
appears to be consistent with tilted r-bonded chains [228]. The question of
how r-bonds form has been addressed by Northrup and Cohen [229]. They
argued that the cleavage process supplies more than enough energy to induce
the necessary changes in bond topology. The transformation to r-bonding
likely involves a shear motion along the cleavage direction accompanied by
flipping of bonds.
The (2 X 1) structure irreversibly transforms to the (7 X 7) structure by
thermal annealing [230]. The (7 X 7) structure also forms after cleaning proce-
dures such as ion bombardment and subsequent annealing. The geometry of
this surface has been a continuing challenge ever since the first observation of
the (7 x 7) LEED pattern by Schlier and Farnsworth in 1959 [231]. Numerous
structure models for its extraordinarily large unit cell have been proposed
(reviewed in ref. [3]), but most of them are rather speculative and not
supported by much experimental evidence. However, since the scanning tun-
neling microscope has provided the first “real-space” image of this surface
[232], there is much less room for speculation. In addition, RBS results have
imposed further restrictions on the number of possible models. Measurements
of the surface peak intensity for normal beam incidence indicated a nearly
bulk-like surface, while for non-normal incidence the equivalent of two extra
monolayers were seen to be displaced from regular lattice sites by at least 0.4 A
[189-1911. This result is surprising since it is hardly conceivable for a hard
material such as Si that bond lengths could change by so much. This notion led
Bennett et al. [191] to propose a model based on stacking faults, i.e., stacking
sequences of atom layers in the selvedge which are different from the cubic
structure. When viewed along the normal, such a faulted surface would indeed
appear bulk-like, but when viewed along a non-normal direction, it would
exhibit atoms off cubic lattice sites, the number of atoms depending on the
type and particular arrangement of stacking faults within the unit cell. How-
ever, non-cubic stacking alone cannot explain the tunneling microscope images
nor the observed blocking patterns. Therefore, McRae [234] proposed an
arrangement of stacking faults, dividing the (7 X 7) unit cell in two differently
stacked triangular units, with dimers of atoms lying on the sides of each unit
and with deep holes at the corners, precisely where minima are found in the
tunneling microscopy images. The model further assumes adatoms on positions
corresponding to maxima in these images. Himpsel and Batra [235] on the
other hand, using a similar arrangement of stacking faults, attributed these
maxima to trimers of neighbouring surface atoms, which, they reason, could
easily be produced by a bond flipping motion similar to that yielding the
a-bonded chain structure in the (2 x 1) surface. Recent ion blocking analyses
256 J.F."cmderVeen/Ion heom ciyrstullo~raph,

by Trom and Van Loenen [233] appear to be consistent with these models and
rule out other ones, such as the milk stool [236], pyramidical cluster [237] and
the Binnig ad-atom [232] models. Here it should be emphasized that other
models could be conceived that describe the blocking data equally well.
Evidently, a unique assignment of the (7 x 7) structure is still lacking.
We finally mention the Si(lll)-(1 x 1) structure, which can be obtained by
adsorbing a minute quantity of impurity atoms [238,239] or by thermally
quenching to room temperature a Si(ll1) surface melted with a pulsed laser
beam [240]. The atomic geometry yielding this seemingly simple (1 x 1) pattern
has been a matter of considerable debate. LEED analyses gave “acceptable”
R-factors for a 15% to 25% contraction of the first interlayer spacing and
otherwise a bulk-like termination of the crystal [239,241]. Others considered
this structure to be merely disordered (7 x 7) [242] or (2 x 1) [243]. Ion
blocking showed the structure to be strikingly similar to the (7 x 7) structure,
confirming the disordered nature of the surface [190]. It is still not understood
why the LEED results are so much at variance. but the lack of long-range
order must undoubtedly have compromised the LEED dynamical analyses.
Clearly the (7 X 7) reconstruction of Si(ll1) involves complex atomic re-
arrangements at the surface, making it quite stable. Though the (7 x 7) LEED
pattern may disappear quickly upon impurity adsorption or thermal quench-
ing, the atomic displacements cannot be undone easily by these treatments. A
similar conclusion was reached in a recent ion channeling study by Gossmann
et al. [244]. In this experiment a few monolayers of Ge were deposited onto the
Si(lll)-(7 x 7) surface at room temperature. Ge is known to interact strongly
with Si. Again, the (7.x 7) LEED pattern disappears - in fact no LEED
pattern is observed for this highly disordered overlayer system - but the
atomic displacements persist at the Ge-Si(l11) interface. It is only after
heating to - 650 K that the reconstruction is relieved. By contrast, adsorption
of Ge onto Si(100) at 300 K was found to readily transform the surface into a
bulk-like lattice, presumably by breaking up the dimers. These experiments
illustrate nicely the different nature of the reconstructions on the (100) and
(111) surfaces of Si.

6.3. Compound semiconductor surfaces

Of all compound semiconductors, GaAs is the most important one. Its polar
(111) and (100) surfaces exhibit a variety of complex reconstructions which are
dependent on surface stoichiometry, i.e., the abundance of As and Ga in the
top layer. It appears that these faces can only be prepared properly by a UHV
crystal growth technique known as Molecular Beam Epitaxy (MBE) [245].
Difficulties with surface preparation have limited the number of reports on
their atomic structure to only a very few [246-2481. By contrast, the very stable
(110) face has received much attention, because it shows a simple (1 X 1)
J. F. uan der Veen / Ion beam crystallography 251

LEED pattern and it can easily be produced by cleavage in UHV. The same
holds for the (110) faces of other compound semiconductors having zinc-blende
structure (e.g. InAs, GaAs, GaP, etc.). Their surface structures have been
extensively reviewed [188,249], so that we may limit ourselves here to a critical
comparison of some very recent RBS and LEED analyses.
It is now generally accepted that the (110) surfaces are all relaxed, with an
upward displacement of the anions and a downward displacement of the
cations in the top layer. Reasonably good fits to the LEED data have been
reported for models in which the bond lengths in the selvedge are approxi-
mately conserved, while the planes of the atomic chains in the top layer are
rotated with respect to the surface plane by an angle o in the range 25” c w <
31’ (fig. 32) (1881. However, Duke et al. [250] subsequently re-evaluated the
fits to their LEED data for GaAs(ll0) in terms of a squared-deviations
parameter known as the X-ray R-factor, and found that a rotation by only 7’
could fit the LEED I-V curves equally well. This might have been regarded as
an artifact of the R-factor analysis if recent ion channeling results by Goss-
mann and Gibson [251] had not indicated that top-layer atomic displacements
parallel to the surface are rather small and therefore favoured the w = 7O
model or an essentially bulk-like surface. Alternatively, an o = 30” model

rotational relaxation
0 catlo”
. s”lOn

bond relaxation

Fig. 32. The relaxed (110) surface of a III-V compound, the relaxation is characterized by a
rotation of the plane of the anion-cation bonds in the top layer by an angle o with respect to the
surface plane.

Fig. 33. Bond rotation and bond relaxation models for the (110) surface of a III-V compound.
258 J.F. uan der Veen / Ion beam ciytallogruphy

could still be valid if large bond length changes were permitted, so as to


maintain near-zero parallel displacements. Henceforth we refer to the latter
model as the “bond relaxation” (BR) model, as opposed to the above “rota-
tional relaxation” (RR) model (fig. 33). To resolve this issue, Smit et al. [199]
performed blocking experiments on cleaved GaAs(ll0) surfaces. They mea-
sured a value of w = 29 + 3”, in agreement with a refinement of the original
LEED analyses by Duke et al. [250]. Furthermore, they concluded that the
relaxation is purely rotational, i.e. all bond lengths in the top layer are
conserved. This result again confirms the original LEED analyses, which
marginally favoured rotational relaxation [252], but disagrees strongly with the
ion channeling results of Gossmann et al. [251]. It was concluded [199] that the
low surface peak yield which was measured by Gossmann et al. and interpreted
by them as evidence for small parallel displacements, was in fact caused by
inadequate surface preparation.
The same bond length conserving rotation of w = 29” was measured on the
(110) surfaces of GaSb [200,201] and InAs [202]. Here the mass difference
between cations and anions was sufficiently large that their backscattering
signals could be separated. Thus the blocking patterns for backscattering from
the cation and anion sublattices were independently obtained, each of them
containing all the information necessary to derive the complete surface geome-
try. The method is illustrated for GaSb(ll0) in figs. 34-36. A side view of the
(710) scattering plane, in which the top layer rotation occurs, is given in fig. 34.
Note that there are two inequivalent (110) crystal planes, one going through
top-layer Sb atoms (top panel) and one going through top-layer Ga atoms
(bottom panel). The angular distributions in fig. 34 represent Monte Carlo
computer simulations of backscattering from the different Sb atoms in a
rotationally relaxed crystal (backscattering from the Ga sublattice is not
considered here). The sum of these distributions gives the simulated Sb
blocking pattern labeled “RR” in fig. 35, where the experimentally determined
pattern is shown as well. The various overlapping blocking minima can be
traced back to specific directions labelled a-e in fig. 34. With the exception of
the subsurface blocking axis e they are all characteristic of the relaxation of the
surface. A blocking pattern simulated for the RR model is seen to fit the data
very well, whereas the BR and bulk-termination models are ruled out. In a
refined fitting procedure Monte Carlo simulations were performed for a large
series of bond rotation angles and enhanced surface vibrational amplitudes.
The results are summarized in a contour plot of the goodness-of-fit xt between
data for Sb and simulations (fig. 36). The goodness-of-fit criterion used here is
defined by

(6.1)
where q ohs is the number of backscattering counts observed in the i th angular
J. F. uon der Veen / ion beam crystallography 259

174 keV He’

. .
0 0

. .

0 0

. .

L
\
,’

I”; /\

\o;,~+/-&5+#~~-~;~-~~
L
0 0’0 0 0 0 0 0 0

. . . . . . . . .

Fig. 34. Computer simulations of ion blocking in the rotationally relaxed GaSb(ll0) surface, for a
174 keV He+ beam incident in the [ii21 direction. Only backscattering from the Sb sublattice is
considered here. A decomposition is made into backscattering contributions from inequivalent
scattering planes (top and bottom panels) and from different layers.

channel (total N), y,,,, is the number predicted by the computer simulations,
and u, is the experimental standard deviation of the number of counts. The
number of degrees of freedom Y equals N - 2 because there are two indepen-
dent parameters to be fitted: the bond rotation angle and the surface vibra-
tional enhancement. The best fit was obtained for a bond rotation angle of
o = 29’;: and an enhancement factor of 1.5 _t 0.2 in the vibration amplitude
of Ga and Sb surface atoms. For the Ga blocking data (not shown) the best fit
was independently obtained at precisely the same values.
We conclude that ion blocking is a very attractive method for the structure
determination of these compound semiconductor surfaces. It is superior to
LEED with regard to its high sensitivity for parallel displacements. It also
provides much more detailed information on surface atomic positions than the
crystal-rocking method commonly used in ion channeling experiments. Though
mass resolution offers the advantage of having two independent blocking
patterns available for analysis, it is not essential to the method. In fact, the
bond rotation angle in the GaAs(ll0) surface, for which the Ga and As
J.F. uan der Veen / Ion hewn ctystullography

0
?

z
d

i
I ” ” I”” 1 ” ” i ” ” 1
45 50 55 60 ES

Scattering angle (degree)


Fig. 35. Surface blocking pattern for backscattering from the Sb sublattice. The partially overlap-
ping minima labeled a-e can be traced back to specific blocking directions labeled a-e m fig. 34.
Solid curves represent computer simulattons for a bulk-like surface (B) and for the rotational
relaxation (RR) and bond relaxation (BR) models. The curve (RR) is the sum of the different
contributions shown in fig. 34. From ref. (2001.

BOND ROTATION ANGLE w fdeg)

Fig. 36. Contour plot of the goodness-of-fit criterion x, * between data and simulation for
backscattering from the Sb sublattice in a rotationally relaxed crystal. From ref. 1201).
J. F. uan der Veen / Ion beam crystallograph~v 261

backscattering signals could not be resolved, was determined with even greater
accuracy than for the GaSb(ll0) surface.

7. Thin films and interfaces

The structure and chemistry of thin films and synthetic superlattices are
subjects of considerable interest in the search of new materials with unusual
solid-state properties and in semiconductor technology. Anomalous magnetic
effects [253,254] and elastic properties [255] have been reported for metallic
sandwich-layer structures and superlattices [256]. Multilayer stacks of alternat-
ing high and low Z elements are serving as high-efficiency reflectors of X-rays
[257,258]. Lattice-matched semiconductor heterostructures find widespread
application in optoelectronics [259] and are used for the development of novel
devices with special transport properties [260-2631. Furthermore, a consider-
able amount of research has gone into formation techniques and various
properties of thin metallic films for use as contacts in integrated circuits
[2,7,264-2661. The usual contact material to Si is Al, but with increasing
packing density of electronic circuits there is a drive to employ new metalli-
zation schemes giving more reliable ohmic and Schottky barrier contacts.
Silicides of transition metals are highly attractive as contact material, because
of their low resistivity, high thermal stability, and the relative ease with which
they can be formed and patterned.
The electronic properties of thin-film structures are intimately dependent on
the atomic composition, structure and abruptness of the interface, about which
relatively little is known. RBS is one of the few techniques suited for in-depth
structure analysis of interfaces. With the use of shadowing and blocking effects
the degree of crystalline order in the film and at the interface can be examined.
In particular, epitaxial crystalline films with lattice constants (nearly) matching
to the substrate can be characterized in detail. It can be determined whether
the thin-film structure is pseudomorphic or not, i.e., whether or not its atomic
rows are commensurate with the substrate rows, and the lattice distortion
resulting from the elastic strain in the pseudomorphic film can be measured as
a shift in the blocking or shadowing directions. With the additional use of high
depth resolution - through electrostatic energy analysis of backscattered ions,
as discussed in section 3.1 - important questions can be addressed not only
concerning the morphology of the film (smooth or rough) but also regarding
the abruptness and atomic structure of the interface.
There is a large body of information from conventional RBS analyses of
thin films [7]. In most of these studies films of a few thousand Angstrom
thickness have been deposited under relatively poor vacuum conditions, giving
rather ill-defined interfaces on an atomic scale. Here we limit ourselves to
262 J.F. uan der Veen / Ion beam crysrallographv

systems prepared in-situ in UHV and to the initial stage of interface formation
following deposition of ultra-thin films (typically of less than 50 A thickness).

7. I. Heteroepitaxy

Epitaxial growth of thin films may occur in different modes [267-2701. The
simplest growth mode is the so-called Frank-Van der Merwe (FM) mode, in
which the film is built up layer by layer, while the other modes (named after
their initial investigators the Volmer-Weber and Stranski-Krastanov modes)
involve nucleation of three-dimensional islands. The FM growth mode is
generally expected for reasonably lattice-matched systems, in which the over-
layer has a surface energy lower than or comparable to that of the substrate. A
good example is Au on Ag(lll), both lattices having fee structure and a
mismatch of only 0.2%. The first indication of epitaxy in this system was given
in a RBS study by Bogh [6], while the FM growth mode was established by
Culbertson et al. [270]. Backscattering energy spectra measured by Culbertson
et al. are shown in fig. 37 for Au film thicknesses ranging from zero to four
monolayers. As expected for epitaxial pseudomorphic growth, the Ag peak
area decreases rapidly, as more Au atoms in the overlayer shadow the Ag
substrate atoms. For Au coverages beyond one monolayer there is of course
also Au-Au shadowing along the (110) incident beam direction. A direct
measure of this effect is the ratio x,,,,” (Au) of the Au scattering yield in the
aligned direction to that in a nonaligned (random) direction. The coverage
dependence of both Ag surface peak and x,,,,“(Au) is shown in fig. 38. Direct
evidence for monolayer-by-monolayer growth comes from the observation that
Au-Au shadowing is completely absent (i.e. x,,,(Au) = 1.0) up to a coverage
of precisely one monolayer, whereupon xmln(Au) decreases sharply. By con-
trast, the formation of three-dimensional islands would have resulted in an
immediate decrease of x,,,,“(Au). Monte Carlo simulations of the scattering
process based on the layer-by-layer growth model indeed fit the ion scattering
data very well (broken and solid curves in fig. 38).
The epitaxial system Au on Pd(ll1) (lattice mismatch of 4.8%) behaves
similarly up to - 1 monolayer, but quite differently beyond this coverage, as is
shown in fig. 39. The Pd surface peak area is seen to decrease initially at the
rate predicted for pseudomorphic growth (solid curve), but already above a
“critical” thickness of - 1 monolayer it increases again, indicating that Au
and Pd atoms are moving out of registry. Kuk et al. [271], who performed these
measurements, attributed this behaviour to a sharp transition from the pseudo-
morphic state with maximum strain to an incommensurate state with misfit
dislocations and lower strain. Below the critical thickness in the pseudomor-
phic state, the lattice of the Au overlayer is compressed so as to match the Pd
lattice exactly, while the first Pd lattice spacings in turn are slightly expanded
along the normal. Above the critical thickness, part of the strain energy in the
J. F. uan der Veen / ion beam crystallography 263

4 AU I MeV He+

ENERGY IkeV)

Fig. 37. Backscattering spectra for 1.0 MeV He+ incident in a (110) direction of (a) clean Ag(lll),
(b) Ag+0.7 monolayer (ML) of Au, (c) Ag+2.9 ML of Au and (d) Ag+ 3.8 ML of Au. From ref.
[270], with permission.

3.0 * Au/Aq (III) <Oil >


I MeV He+, 140°K

-7. Ag S.P.
!.O 4’
--_ 99 0.6
Au Xmin
.5

Lu
1.0 .. .
0.4 x’
---;m . .
0.2

TGL=G--
---mm

0
0 2 4 a 6 lb
[MONOLAYERS]

Fig. 38. Ag surface peak area and xmin(Au) as a function of Au coverage. The full and dashed
curves are derived from computer simulations assuming monolayer-by-monolayer growth. From
ref. [270], with permission.
*
*
+
l +

2
Au
4
COVERAGE
6 8
(MONOLAYERS
IO 12
I
14

Fig. 31). Pd surface peak area ver~uh Au coverage for 1.X MeV He+ incident in the [OOi] direction
of the Pd( 111) substrate. The full curve 1s calculated for pseudomorphic growth. H, is the
predlcted crItical thlcknes for the formation of miaflt dislocations. From ref. [271],with permls-
sion

overlayer is relieved by the spontaneous generation of dislocations and a


concomitant relaxation of the in-plane Au lattice constant toward its bulk
value. The lattice relaxation effects were measured by angular scans of the type
discussed in section 2.1. These data by Kuk et al. are in qualitative agreement
with the equilibrium theory of dislocation formation [272], which predicts a
critical thickness of only 5 A ( - 2 monolayers) for this system. By contrast, the
much better-matched Au on Ag(lll) system is predicted to remain in the
pseudomorphic state for at least 300 A. This then explains the absence of the
transition in fig. 38. The slight expansion of the Pd lattice spacing in the highly
strained Au on Pd system may explain the strongly enhanced magnetic
susceptibility observed in very thin Pd films sandwiched between Au [253].
Another class of epitaxial systems, which is particularly interesting from a
technological point of view, are those formed from non-lattice matched semi-
conductors having different band gaps [263,273-2781. In general, only those
systems are of interest, in which the lattice mismatch is accommodated by
strain without dislocations at the interface. Let us examine first Ge on Si(lll)
for which the lattice misfit is 4%. Narusawa and Gibson [279], using ion
channeling, reported pseudomorphic growth up to a coverage of - 3 mono-
layers of Ge deposited on Si(ll1) at 650 K (the (7 X 7) reconstructed surface
reorders to a nearly bulk-terminated structure upon formation of the interface
[244]). Above this critical thickness the overlayer atoms start to move out of
registry: resulting in defected films. Clearly. as for the metallic Au on Pd(ll1)
system, the lattice mismatch is too large for growing dislocation-free epitaxial
films of appreciable thickness. More successful has been the growth of Ge, Si, ,
alloys. Smooth and pseudomorphic layer growth has been reported for a range
of alloy compositions between 0 < x’ < 0.5. at a deposition temperature of 820
K [277]. The lattice mismatch is accommodated by a tetragonal distortion of
J. F. eon der Veen / Ion beam crystallography 265

the overlayer lattice. Channeling measurements of the concomitant angular


shift of crystal channels have been used to establish the transition from the
pseudomorphic state to the non-registered state, in which much of the tetrago-
nal distortion has disappeared but dislocations are present (fig. 40). The critical
thickness at which the transition occurs is - 100 A for the Ge,,SSi,,, alloy,
while for the better matched Ge,,,Si,,, alloy this thickness is at least 2500 A.
These critical thicknesses are significantly larger than predicted by the equi-
librium theory for the formation of misfit dislocations [272]. A similar observa-
tion has been made for Si layers grown on GaP(lOO) at a temperature of 840 K
[280]. The theory assumes thermodynamic equilibrium between an array of
misfit dislocations along the interface and the strained film. Thus apparently
there is a kinetic barrier to the formation of dislocations at the relatively low
temperatures at which the layers have been grown. Use of low temperatures

1-x

iiiiiI
Fig. 40. Schematic illustration of transition from (a) a strained pseudomorphic overlayer to (b) a
non-registered overlayer containing misfit dislocations and negligible strain. After ref. [277].
266 J. F. ucln der Veen / Ion beam ctysrallograph~y

also limits the migration of atoms over the surface and therefore it prevents the
film from growing in three-dimensional islands. Thus smooth films can be
grown and even a Ge,Si,-,/Si superlattice has been fabricated in this way
[277].
Other strained-layer superlattices have been grown in the GaAs,P, _,/GaP
[273,263] and In,Ga,_,As/GaAs [274] systems. For the former system the
amount of strain was determined from the amount of dechanneling of an ion
beam in the zig-zagging channels of the superlattice [281]. The channeling
behaviour of MeV ions is so sensitive to these small directional misalignments
that anomalous dechanneling has even been reported for the only slightly
mismatched InAs-GaSb superlattice [282,283]. Other heteroepitaxial structures
such as Si/CoSi,/Si [284], NiSi,/Si [285] and Si/CaF,/Si [286] have also
been grown with reasonable crystal quality as regards channeling behaviour.

7.2. Metal-semiconductor interfaces

This section deals mainly with the interfacial structure of metal-Si contacts.
Silicides of transition metals are very useful as contact material in integrated
circuits, since they lead to low-resistivity ohmic contacts and interconnects or
reliable and reproducible Schottky barriers [264-2661. There is ample evidence
in the literature that the electrical properties of contacts are largely controlled
by chemical reactions at the interface [2,266]. In particular, the barrier height
appears to be determined by the local chemical bonds between metal and Si
atoms at the intimate interface; it is often found that the Schottky barrier
height has already evolved to its final value after deposition of a few mono-
layers of metal [266]. These findings are of an empirical nature and are not
easily explained. Despite a significant amount of theoretical work [287-2941
which followed the pioneering studies of Schottky [295] and Bardeen [296]
some 40 years ago, we are still far from understanding the mechanism of
barrier formation. None of the theoretical models has yet given an explicit
account of all available data on Schottky barrier heights. One of the main
factors that appears to have limited progress in this field is the general lack of
knowledge of the interface microstructure and composition. It is in this area
that RBS has made valuable contributions.
Transition metals react with Si and form silicides by contact reaction, i.e. by
a thermally induced reaction at the interface between silicon and a deposited
metal film, whereas noble metals such as Ag and Au do not form silicides. The
growth of a silicide requires both mass transport of atoms through the silicide
film already formed and atomic motion across the metal-silicide and silicide-Si
interface. The silicide phase M,YSi y produced is found to be largely kinetically
controlled. The phase depends on temperature, reaction time and supply of
metal atoms (i.e. thin or thick film). Furthermore, the kinetics of the initial
reaction at the interface differs from that of further silicide growth. This is
J. F. uan der Veen / ion beam crystallography 267

because little mass transport is required in the former case. Also important in
this respect is the morphology of the film during deposition of the first few
monolayers [297]: atoms can migrate more easily in a clustered film than in a
continuous one. Thus the initial stage of silicide formation, during which the
Schottky barrier develops, may differ significantly from the stage of in-depth
growth. Also, the composition of the film may be graded along the direction of
the surface normal.
A variety of surface science tools has been applied to the study of metal-Si
and silicide-Si interface formation. These include: (1) Auger spectroscopy
[298,299] which qualitatively identifies the atomic composition and bonding
characteristics, (2) photoemission [300-3071 which reveals the electronic struc-
ture, Schottky barrier height, and indirectly the composition at the interface,
(3) cross-sectional TEM [298,308,309] which provides information on the
morphology of the film, and on the interface microstructure and abruptness,
(4) SEXAFS [310,311] which probes the local atomic environment of metal
atoms at the interface, and (5) RBS [297,312-3201 which quantitatively identi-
fies structure and composition of the (reacted) layer, and, if performed with
sufficiently high depth resolution [297,320], the properties mentioned under
point (3).
RBS, combined with high (3-5 A) depth resolution, is one of the most
powerful interface probes available. Fig. 41 illustrates how films of varying
composition and morphology give rise to very different energy spectra. Beam
alignment with a crystal axis in the Si substrate yields a single backscattering
peak when the surface is clean (panel a). Panel b shows how the spectrum
changes when a reactive metal is deposited and a uniform polycrystalline (or
amorphous) silicide is formed. The reacted metal and Si atoms all become fully
visible to the beam and give rise to broadened backscattering peaks. The
silicide phase M,Si, and the film thickness can be deduced from the peak area
and width, respectively, using the procedures described in section 3. If the
metal does not react (panel c),’ the same spectrum is observed as in fig. 8. If the
metal layer consists of islands as depicted in d then a spectrum results which is
a linear combination of a and c, (neglecting edge effects in the backscattering
process). Similarly, for silicide islands (not shown in fig. 41) the corresponding
spectrum can be constructed out of a and b. Finally, in the case of islands with
varying thickness (panel e) the Si interface peak broadens because of the
different path lengths traversed by the ions; its peak shape directly reflects the
distribution of island thicknesses. Of course, also the metal peak shape is
affected. The first systematic RBS study of interface formation which made
explicit use of high depth resolution, was performed by Van Loenen et al. [320]
on the non-reactive system Ag on Si(ll1). The experiment was aimed at
solving some controversies concerning the growth mode of room-temperature
deposited Ag films [321-3231 and the possibility of mixing at the interface
[321-3261. Fig. 42 shows a collection of energy spectra taken for Ag coverages
268 J. F. uan der Veen / Ion beam crysfallograph~

Zl Si
PI-

l-7

gJJl-y$J~
BACKSCATTER ENERGY

Fig. 41. Films of varying composition and morphology on Si. and their corresponding backscatter-
ing energy spectra. The panels a-e are discussed in the text. From ref. [320].

increasing from 0 to 8.4 X lOI atoms/cm2. The depth resolution in this


experiment was 3-4 A. As the Ag coverage increased the spectra became
strikingly similar to the one shown in fig. 41e, indicating the formation of
three-dimensional (3D) islands. Analysis of the peak shapes revealed three
contributions to the Si signal: (A) from bare parts of the Si substrate, (B) from
Si underneath the islands and (C) a dechanneling contribution in the Si
substrate, resulting from angular straggling in the islands. It was shown by Van
Loenen et al. that in fact first a two-dimensional layer forms up to a coverage
of 6 X lOI atoms/cm2, beyond which nucleation of 3D islands occurs. They
concluded that Ag exhibits a Stranski-Krastanov growth mode when de-
posited on Si(ll1) at room temperature (at a rate of (1-4)~ 1O’j atoms cm-’
s-l). Clear evidence that the Ag on Si system is non-reactive is given in fig. 43,
where the Si peak area (the sum of contributions (A) and (B)) is plotted against
the Ag coverage. The nearly constant signal shows that no Si atoms are
displaced at the interface upon deposition of Ag. We note here that the Si
J. F. eon der Veen / Ion beam crystallography 269

150 160 170


BACKSCATTER ENERGY [ keV I
Fig. 42. Backscattering energy spectra for different Ag coverages (as indicated), deposited on
Si(ll1) at 300 K. The backscattering contributions labeled A, B and C are discussed in the text. A
depth interval of 10 A is indicated to emphasize the high depth resolution of the experiment. From
ref. 13201.

displacements reported in an earlier ion channeling study by Narusawa et al.


[314] for this system, were probably caused by beam dosage effects of the type
discussed in section 3.5 [315,320].

Fig. 43. The Si surface peak area as a function of Ag, Ni and Ti coverage. BULK denotes the value
expected for a bulk-like Si(ll1) surface. Depositions were made on a clean Si(ll1) substrate at
300 K.
270 J.F. uan der Veen / Ion beam crysrallograph~

Quite different behaviour is observed for Ni [297,312] or Ti [327]. These


metals react with Si even at a temperature as low as 300 K, which is far below
the usual silicide formation temperature. The number of visible Si atoms
increases at a rate of one Si atom per two deposited Ni atoms, indicating
nucleation of the Ni,Si silicide phase. Deposition of Ti at 300 K leads to the
formation of an ultrathin film with a composition close to TiSi. From fig. 43 it
is also apparent that the silicide forming reactions slow down dramatically
beyond a certain coverage. Furthermore, a detailed analysis of backscattering
peak shapes revealed that the silicide films grow out from 3D clusters. From
these results emerged the model for silicide formation at low temperatures
shown schematically in fig. 44. Metal atoms arrive on the Si surface and form
small clusters. The energy released upon clustering is sufficient to overcome the
activation barrier for a silicide forming reaction (panel b). Since these reactions
are exothermic, additional energy becomes available to promote Si diffusion
over the islands. The Si atoms react with newly arriving Ni atoms and the
islands grow laterally (c) until they coalesce into a continuous film (d). At that
stage surface diffusion is blocked and further reaction requires diffusion of
metal or Si atoms through the film. Since bulk diffusion is very slow at 300 K,
further reaction is blocked and unreacted metal builds up on top (e). Indeed.
for both Ti and Ni the formation of silicide is observed to slow down precisely
at the coverages for which the islands have just coalesced [297,327].
The above cluster-induced reaction model fully explains the strong chemical
interaction between transition metal atoms and Si. However, other models
have been proposed as well. One of these models, due to Tu [328], involves
interstitial diffusion of metal atoms into the Si selvedge, with the effect that the
covalent bonds between Si atoms are weakened. The bond weakening then
leads to disruption of the Si lattice and formation of the silicide. In another
model, proposed by Hiraki [329], the bonding is weakened in the presence of a
metallic overlayer, inducing metallic screening in the substrate. According to
this model, there must be a critical overlayer thickness (of the order of the
electron mean free path) before reaction takes place.
Photoemission data on Ni,Si formation have been interpreted to give
support for the interstitial diffusion model [307], but the presence of Ni

M
M,Si,

Si

0 b c d e

Fig. 44. Cluster-induced reaction model for the initial stage of silicide formation at low tempera-
ture. The reaction steps a-e are discussed in the text.
interstitials in the Si selvedge has sofar only been detected in a single ion
channeling study by Cheung and Mayer on Si(100) [313]. Other RBS investiga-
tions of Ni and Ti on Si(ll1) were inconclusive in this respect; blocking effects
on the metal backscatter yield, which could be attributed to an interstitial
position, were absent [327,330]. This rules out the presence of any appreciable
amount of metal atoms in interstitial positions below the first two Si double-
layers (it proves difficult to test whether they are located in the top two layers
of the Si lattice). It could also be envisaged that the supply of metal interstitials
is pre-empted rapidly by formation of the silicide. In that case the interstitial
causing the bond breaking is never seen, since only the end phase of the
reaction is detected.
The RBS results by Van Loenen et al. [297,327] do not provide support for
the Hiraki screening model [329]. In the case of Ti on Si(lll), electron energy
loss measurements [331] were interpreted by Hiraki as to give evidence for the
presence of an unreacted Ti film up to a critical thickness of 2.5 monolayer.
However, fig. 43 shows quite clearly that this is not the case; in fact, the
reaction is almost completed at this coverage.
The above observations leave us with the cluster-induced reaction model as
the most plausible one. A similar model was formulated earlier by Zunger [332]
to account for the chemical interaction between Al and GaAs surfaces.
Recently, clusters were invoked by Grioni et al. [333] for explaining silicide
formation in the system Ce on Si(ll1).
All of the above silicide films are polycrystalline (or amorphous) but other
silicides such as CoSi, [334,335], Nisi, [285,286] and Pd,Si [298] can be
grown epitaxially on Si in single-crystal form. In particular the silicides of Co
and Ni are ideal choices for epitaxy as they have the cubic CaF, structure with
bulk lattice parameters within 1.2% of Si. Tung et al. [308,336], using TEM
lattice imaging, have shown that continuous and pseudomorphic (111) oriented
films of single-crystal Nisi, can be grown on Si(ll1) with a high degree of
perfection. They showed that the film may either grow in the so-called type-B
orientation in which the film is rotated 180” about the Si surface normal (fig.
45b) or in the unrotated type-A orientation (fig. 45a), depending on the
method of preparation. Very thin films were prepared by cold deposition of Ni
on atomically clean Si(ll1) and subsequent heating to 450-500°C. An average
thickness of deposited Ni of less than 6 A resulted in a coherent B-type
orientated film, whereas a thickness in the range of 16-20 A gave a purely
A-type silicide (fig. 46). For other thicknesses the film contains mixed grains of
both A- and B-type. Though at present it is still unclear why the pinning of a
particular orientation depends on the amount of Ni, the effect can be utilized
for growing thick defect-free Nisi, crystals of single orientation, using these
thin A- or B-type films as “templates” [308]. (In turn, high-quality Si may be
grown on top.) The epitaxial silicide films so formed make well-defined and
reproducible contacts to Si, and are especially promising as buried conducting
272 J. F. uan der Veen / Ion beam crystallography

a b
type-A Nisi, type -B Nisi2

Si (111) Si (111)
Fig. 45. Two possible orientations for a NiSi, film epitaxially grown on Si(ll1): (a) type-A
orientation; (b) type-B orientation, in which the film is rotated 180” about the surface normal. In
both cases the Ni atoms (large circles) are assumed to be seven-fold coordinated at the interface.

NICKEL COVERAGE (X 10%-*)

___/ 100%
10 20 30 40 50
AVERAGE THICKNESS OF DEPOSITED NICKEL Ci\,

Fig. 46. The percentage of Nisi, volume with type-A orientation in films grown by annealing
room-temperature deposited Ni on Si(ll1). From ref. 13081, with permission.
J. F. uan der Veen / Ion beam crystallography 213

layers in monolithic Si/metal/Si structures. The latter structures may be


important for the development of high-speed devices such as the metal base
transistor [337] and ultimately for the realization of three-dimensionally in-
tegrated circuits.
The epitaxial Nisi, on Si(ll1) system, with its atomically abrupt interface,
obviously serves as an ideal system for testing theoretical models [287-2961 of
Schottky barriers. These models in turn often require detailed knowledge of the
interface atomic structure, which we will now examine more closely. For a
given silicide orientation (type-A or type-B) there are two different possibilities
for the registry of the Nisi, overlayer with respect to the substrate: one with
five-fold coordinated and one with seven-fold coordinated Ni atoms at the
interface. This is shown in fig. 47 for the type-B interface. The seven-fold
coordinated arrangement was favoured by cross section TEM lattice imaging
[309,338], but a definite structure assignment by this technique was hampered
by uncertainties in sample thickness and other input parameters for TEM
image calculations [309]. On the other hand, five-fold coordination was favoured

Fig. 47. Scattering geometry for determination of the type-B Nisi,-Si(lll) interface structure.
showing (110) planes with (a) seven-fold and (b) five-fold coordinated Ni atoms at the interface.
From ref. [340].
by recent X-ray standing wave measurements [339]. The issue was resolved in a
high-resolution RBS experiment by Van Loenen et al. [340]. They used the
backscattering geometry of fig. 47 to analyze an ultra-thin B-type film ( - 30 A
thickness). The interface was probed by directing a 100 keV H + beam into the
[OOi] channels of the Nisi, film. Upon entering the substrate, the focused ion
flux, which is peaked near the centre of the silicide channels, may hit the
substrate Si atoms marked 4 and 5 in fig. 47. Atom 1 and the other substrate
atoms are shadowed by the atomic rows in the silicide. By choosing slightly
different incident beam directions within the [OOi] channels, it proved possible
to direct the focused ion flux onto different substrate atoms. For the higher
angles of incidence (Y(35.0” < cy < 35.5”) atoms 2 and 3 are hit, while 4 and 5
are shadowed, whereas for the lower angles (34.3” < (Y< 34.8”) atoms 4 and 5
are hit and 2 and 3 are shadowed (fig. 47). The backscattered ions emerge from
the NiSi,[l lo] channels in well-defined ranges of exit angles. For seven-fold
coordinated epitaxy the location of atoms 4 and 5 with respect to the latter
channels is such that this angular range is very narrow, whereas for atoms 2
and 3 this range is rather wide. For the five-fold structure the situation is
reversed. Thus the overlayer registry follows directly from measurement of the
angular distribution of emergent ions for different tilt angles of the incident
beam on either side of the [OOi] axis. Angular distributions around the [ii01
exit channel were measured for three different angles of incidence (Y and are
shown in fig. 48. Each data point (solid circles in fig. 48) represents the
integrated yield of the Si interface peak in a backscattering energy spectrum
taken at the corresponding in- and outgoing angles (Yand p. (Here the use of
335 A depth resolution turned out to be essential for isolating the interface
peak from the other backscattering contributions.) As (Y is swept through the
[OOi] channel from lower to higher values. the initially narrow angular distribu-
tion was found to broaden substantially. This is precisely the behaviour
predicted for the seven-fold interface. For comparison the channeling experi-
ment was simulated on a computer for both five- and seven-fold coordinated
structures. Typically. for each 1yand p some 1Oi ions were tracked through the
thin-film structure, using the Monte Carlo method discussed in section 4.5. The
simulations for the seven-fold structure agree very well with the data. whereas
those for the five-fold structure do not at all. Not only could the structure be
identified as seven-fold, but also a small contraction was seen in the bond
lengths across the interface [340]. This is the first time that atomic positions at
a bicrystal interface have been determined. The method of ion focusing in
crystal channels. as outlined above, is generally applicable to epitaxial thin-film
systems and constitutes a very powerful probe of interface structures.
Finally, the interaction of metals with compound semiconductor surfaces
will be considered briefly. Thick-film contacts have received some attention [7].
but virtually no data exist on the initial stage of interface formation. A single
ion channeling study by Gossmann and Gibson [341] on metal-GaAs inter-
J. F. oan der Veen / Ion beam ctyvsrallography 215

faces was aimed at gaining information about the structural changes in a


GaAs(ll0) substrate upon deposition of disordered layers of Pd and Au at 300
K. Au overlayers did not induce any lateral displacements in the substrate up
to a coverage of 5 monolayers. Above this coverage there is some evidence of
substrate disordering. By contrast, deposition of Pd immediately leads to a
strong reaction and disruption of the substrate lattice at the interface. Another
structural study was reported by Narusawa et al. [342] for Au on an MBE-grown
GaAs(OO1) surface - the (001) orientation is the one used for device fabrica-
tion. The results show a release of As atoms, indicative of a reaction occurring
at room temperature. Annealing the film to 600 K causes the film to break up
into islands, leaving large parts of the GaAs(OO1) surface bare. Some ordering
of the Au overlayer in one dimension was detected in an angular scan of the
Au scattering yield. We finally note that Au-GaAs contacts (important for
device application) are rather unstable upon annealing. There is substantial

a
6_ 7-FOLD

46 50 5L 50 L6 50 5L 56

EXIT ANGLE f3 ldegreel

Fig. 48. Angular distributions of ions backscattered from Si atoms labeled 2 to 5 in fig. 47. Both
measurements (solid dots) and computer simulations (solid curves) were performed with a 100 keV
H+ beam in the scattering geometry of fig. 47, for different incidence angles a. Panel a shows the
simulations for the seven-fold coordinated interface, panel b for the five-fold coordinated interface.
The dashed and dash-dotted curves indicate that the backscattering contributions from atoms 2
and 3 are almost completely resolved. From ref. [340].
diffusion of Au, resulting in a highly defected GaAs substrate [343]. A highly
reactive thin film (e.g. Pd) sandwiched between Au and GaAs may act as a
diffusion barrier [7].

7.3. Oxide films

Oxide films on metal and semiconductor surfaces have been investigated


with a variety of surface sensitive techniques. These studies were either aimed
at understanding the initial stage of oxide growth or concentrated on a precise
determination of near-interface stoichiometry of oxide films. To the first
category belongs high-resolution RBS work by Smeenk et al. on the interaction
of oxygen with (110) and (100) surfaces of Ni [144,344]. The first oxygen atoms
arriving on the surface are chemisorbed, then oxidation starts at l/3 to l/2
monolayer coverage. The onset of oxidation is observed in RBS as a sudden
displacement of Ni atoms off lattice sites, giving a dramatic increase in the Ni
surface peak area. By measuring the number of Ni atoms taking part in the
oxidation versus the oxygen exposure conclusions could be drawn on the
mechanism of incipient oxidation. The oxide growth was found to proceed in
the way described by Holloway and Hudson [143]. NiO islands of - 2
monolayers thickness grow across the surface from a fixed number of pre-exist-
ing nucleation sites by capturing diffusing surface oxygen molecules or directly
impinging oxygen at the island perimeters. According to the Holloway-Hud-
son model the oxidation rate first rises as the islands grow in size and then falls
off sharply as they coalesce. This is precisely the behaviour found in the RBS
studies of Smeenk et al. [344]. The oxide islands were identified by them in a
backscattering peak shape analysis similar to the one given in fig. 41. The
“final” NiO thickness was found to be - 2 monolayers ~ in fact the oxide
keeps growing in depth, but at a very slow rate at the temperatures (300-420
K) at which these experiments were performed.
One of the most important oxides is undoubtedly SiO,, being an essential
insulator in MOS technology. There is a need for ultra-thin oxide films of high
quality. Though the most relevant test of the quality of such films is of course
an electrical characterization, it is of great interest to know more about their
interface structure. Haight and Feldman [345], using ion channeling, de-
termined the strain in the first layers of a Si(ll1) crystal, induced by a
thermally grown SiOz film on top. Their results revealed a stoichiometric oxide
up to the interface and a thin (1 to 2 monolayers) region of strained crystal
underneath. An upper limit of one monolayer of non-stoichiometric oxide was
found. Among other possibilities, bond strain in Si is known to have an effect
on the electronic density of states [346]; it could explain a density of 4 X
lO”/eV. cm’. which is found by various workers [347] in the centre of the Si
band gap.
J. F. oan der Veen / Ion beam crystallography 217

8. Synopsis

This review has largely been devoted to the precise determination of atomic
positions in the top layer of single-crystal surfaces. Of all experimental
structure probes presently available, Rutherford Backscattering Spectrometry
(RBS) proves to be one of the most accurate. In particular, if used in
conjunction with “blocking”, it is unequaled for its sensitivity to small bond
length changes in surfaces. Surface relaxation values can now be measured
almost routinely and complex surface reconstruction models can also be tested
with relative ease. Even the bonding geometry of light chemisorbed atoms can
be determined.
Interpretation of the data in terms of a structure is greatly facilitated by the
fact that atom positions are directly “imaged” in real space, an advantage that
RBS has in common with a technique such as Scanning Tunneling Microscopy
(STM) [348]. Note, however, that RBS measures positions of atomic nuclei,
whereas STM essentially maps the electronic charge distribution far out from
the plane of nuclei terminating the crystal. In a sense both these techniques are
complementary and combined use on the same surface would form the
ultimate test of electronic structure theory.
Gratifying as it may be to know “where the atoms are”, it is perhaps more
important to know “to where they move” when the surface is brought in
contact with foreign atoms. After all, this is how real surfaces come about. The
substrate atoms may move over only small distances, as is the case if the
chemical interaction with the foreign atom is weak, or a reaction may take
place which significantly perturbs the surface region. Examples of the latter are
incipient oxidation following chemisorption of oxygen and the chemical reac-
tion taking place when a transition metal overlayer is deposited on a semicon-
ductor surface. Our present state of knowledge on the responsible reaction
mechanisms have in both cases greatly benefited from ion channeling and
blocking investigations. The use of ions in the medium-energy range opens up
the possibility for probing the interface region of these systems nearly with
monolayer depth resolution. For the first time atomic positions have been
determined at a bicrystal interface.
A modern RBS set-up for surface and interface analysis should contain
besides the common auxiliary tools for surface characterization, adequate
facilities for in-situ deposition of overlayers. The preparation of heteroepitaxial
multi-layered systems even requires complicated techniques for crystal growth
such as Molecular Beam Epitaxy (MBE). To achieve these goals, an RBS
system should ideally comprise a UHV scattering chamber communicating
with a separate UHV chamber for MBE crystal growth which contains an
array of Knudsen cells and electron evaporators. Such a system is presently
under construction in various laboratories.
The strength of shadowing and blocking effects is sensitively dependent on
the vibration amplitude of surface atoms. This relationship led to systematic
investigations of surface dynamic properties. Contrary to spectroscopies such
as LEELS and inelastic atom scattering which give detailed information on,
e.g., surface phonon dispersion relations, RBS essentially measures thermally
averaged quantities such as rms amplitudes and correlation functions. It is
generally found that the surface vibrational amplitude is significantly enhanced
with respect to the bulk value. Given the observation, that a solid melts when
the vibrational amplitude exceeds a critical percentage (- 10%) of the nearest-
neighbour distance [349], one may then wonder what would happen to the
surface as the temperature of the crystal is raised to the melting point. Since
surface atoms have higher amplitudes, the condition for melting of the surface
is expected to be met at a temperature below the bulk melting point T,, [350].
That this is indeed the case has been demonstrated very recently by Frenken
and Van der Veen [351]. Shadowing and blocking measurements on a Pb(ll0)
surface indicated that the topmost layer starts to melt already at - 40 K below
T,,,. The measurements showed further that melting of the bulk is a surface-
initiated process. This is the first direct observation of a melting-point depres-
sion at the surface of a three-dimensional crystal.

Acknowledgements

The author is greatly indebted to Professor F.W. Saris for his stimulating
interest. He has made me familiar with the use of ion channeling techniques for
surface analysis. I wish to thank my colleagues for many years of fruitful and
enjoyable collaboration: Drs. J.W.M. Frenken, E.J. van Loenen, R.G. Smeenk,
L. Smit, R.M. Tromp and W.C. Turkenburg. The excellent technical assistance
of J. ter Beek, J.W. Derks. S. Doorn, H.H. Kersten and A.P. de Jongh is
gratefully acknowledged. Finally, thanks are due to Dr. T.E. Derry for critical
reading of the manuscript. This work is part of the research program of the
Stichting voor Fundamenteel Onderzoek der Materie (Foundation for Funda-
mental Research on Matter) and was made possible by financial support from
the Nederlandse Organisatie voor Zuiver-Wetenschappelijk Onderzoek
(Netherlands Organization for the Advancement of Pure Research).

References

[l] T.N. Rhodm and G. Ertl. Eds.. The Nature of the Surface Chemical Bond (North-Holland,
Amsterdam, 1978).
[2] L.J. Brillson. Surface Sci. Rept. 2 (1982) 123.
[3] D. Haneman. Advan. Phys. 31 (1982) 165. and references therem.
[4] K.C. Pandey, Phys. Rev. Letters 47 (1981) 1913.
(51 F. Jona. J.A. Strozier, Jr. and W.S. Wang. Rept. Progr. Phys. 45 (1982) 527. and references
therein.
J.F. uan der Veeti / Ion heum cystallogruph~ 279

[6] E. Bogh, in: Channeling, Ed. D.V. Morgan (Wiley-Interscience. New York. 1973). ch. XV.
(71 J.W. Mayer and J.M. Poate. in: Thin films. Interdiffusion and Reactions. Eds. J.M. Poate.
K.N. Tu and J.W. Mayer (Wiley-Interscience. New York, 1978) p. 119.
[8] W.K. Chu, J.W. Mayer and M.A. Nicolet. Backscattering Spectrometry (Academic Press.
New York, 1978) ch. IV.
[9] D.S. Gemmel, Rev. Mod. Phys. 46 (1974) 129. and references therein.
[lo] M. Aono, Nucl. Instr. Methods B2 (1984) 374, and references therein.
(111 G. Moliere. Z. Naturforsch. 2a (1947) 133.
[12] O.B. Firsov, Zh. Eksperim. Teor. Fiz. 33 (1957) 696 [Soviet Phys-JETP 6 (1958) 5341.
[13] C. Varelas and R. Sizmann, Radiation Effects 25 (1975) 163.
[14] W.C. Turkenburg, W. Soszka. F.W. Saris, H.H. Kersten and B.G. Colenbrander. Nucl. Instr.
Methods 132 (1976) 587.
[15] O.S. Oen, in: Proc. 7th Intern. Conf. on Atomic Collisions in Solids, Vol. 2, Eds. Yu.V.
Bulgakov and A.F. Tulinov (Moscow State University Publishing House. Moscow, 1981) p,
124;
OS. Oen. Surface Sci. 131 (1983) L407.
[16] A.A. Maradudin, E.W. Montroll, G.H. Weiss and I.P. Ipatova, in: Theory of Lattice
Dynamics in the Harmonic Approximation, Solid State Physics Series, Suppl. 3. Eds. H.
Ehrenreich, F. Seitz and D. Turnbull (Academic Press. New York, 1971).
(171 1. Stensgaard, L.C. Feldman and P.J. Silverman, Surface Sci. 77 (1978) 513.
[18] J. Lindhard, Kgl. Danske Videnskab. Selskab. Mat.-Fys. Medd. 34. No. 14 (1965).
[19] J.F. van der Veen. R.M. Tromp, R.G. Smeenk and F.W. Saris. Surface Sci. 82 (1979) 468.
[20] J.H. Barrett, Phys. Rev. B3 (1971) 1527.
[21] VS. Kuliskauskas, M.M. Malov and A.F. Tulinov. Zh. Eksperim. Teor. Fiz. 53 (1968) 487
[Soviet Phys.-JETP 26 (1968) 3211.
[22] E. Bsgh, Can. J. Phys. 46 (1968) 653.
[23] L.C. Feldman and B.R. Appleton, Appl. Phys. Letters 15 (1969) 305.
[24] H.H. Andersen and J.F. Ziegler, The Stopping and Ranges of Ions in Matter, Vols. 3 and 4
(Pergamon, Oxford, 1977).
[25] P.J. Duseph, Introduction to Nuclear Radiation Detectors (Plenum. New York, 1975).
[26] Ref. [S], pp. 200, 201.
[27] J. Williams, Nucl. Instr. Methods 149 (1978) 207.
[28] Y. Tamminga, M.F.C. Willemsen and R. van Silfhout, Nucl. Instr. Methods 218 (1983) 107.
[29] J.K. Hirvonen and G.K. Huber, Nucl. Instr. Methods 149 (1978) 457.
(301 A. Feuerstein, H. Grahmann, S. Kalbitzer and Oetzman, Nucl. Instr. Methods 149 (1978)
471.
[31] R.G. Smeenk, R.M. Tromp,‘H.H. Kersten, A.J.H. Boerboom and F.W. Saris, Nucl. Instr.
Methods 195 (1982) 581.
[32] J. I’Ecuyer, J.A. Davies and N. Matsunami. Radiation Effects 47 (1980) 229.
[33] Ref. [8], pp. 210-218.
[34] J. I’Ecuyer, J.A. Davies and N. Matsunami, Nucl. Instr. Methods 160 (1979) 337.
[35] R.M. Tromp, J. Vacuum Sci. Technol. Al (1983) 1047.
[36] W.C. Turkenburg, E. de Haas, A.F. Neuteboom, J. Ladru and H.H. Kersten, Nucl. Instr.
Methods 126 (1975) 241.
[37] T.M. Buck, G.H. Weatley and L.C. Feldman, Surface Sci. 35 (1973) 345.
[38] A. Chateau-Thierry and A. Gladieux. in: Proc. 5th Intern. Conf. on Atomic Collisions in
Solids, Gatlinburg, 1973, Eds. S. Datz, B.R. Appleton and C.D. Moak (Plenum, New York,
1975) p. 307.
[39] R. Behrisch, W. Eckstein, P. Meischner, B.M.U. Scherzer and H. Verbeek. in: Proc. 5th
Intern. Conf. on Atomic Collisions in Solids, Gatlinburg. 1973, Eds. S. Datz, B.R. Appleton
and C.D. Moak (Plenum, New York, 1975) p. 315.
[40] D.A. Thompson and W.F.S. Poehlman, Nucl. Instr. Methods 168 (1980) 63.
280 J. F. uan der Veen / Ion beam crysrallography

141) High Voltage Engineering Europe BV, P.O. Box 99, 3800 AB Amersfoort, The Netherlands.
[42] R.M. Tromp, H.H. Kersten, E. Granneman, F.W. Saris, R. Koudijs and W.J. Kilsdonk,
Nucl. Instr. Methods B4 (1984) 155.
(431 Ref. IS], p. 160.
[44] J.W.M. Frenken, R.G. Smeenk and J.F. van der Veen, Surface Sci. 135 (1983) 147.
[45] G.K. Kinchin and Pease, Rept. Progr. Phys. 18 (1955) 1.
[46] M. Iwami, R.M. Tromp, E.J. van Loenen and F.W. Saris, Physica 116B (1983) 328.
(471 R.M. Tromp and J.F. van der Veen. Surface Sci. 133 (1983) 159.
1481 O.S. Oen, Phys. Letters 19 (1965) 358.
[49] J.F. van der Veen, J.B. Sanders and F.W. Saris, Surface Sci. 77 (1978) 337.
[50] D.P. Jackson, J.A. Davies, T.E. Jackman, P.R. Norton and W.N. Unertl, Nucl. Instr.
Methods 194 (1982) 143.
[51] K. Makoshi and M. Hatada, Surface Sci. 114 (1982) 673.
(521 J.F. van der Veen, R.M. Tromp, R.G. Smeenk and F.W. Saris, Nucl. Instr. Methods 171
(1980) 143.
[53] V.M. Agranovitch and V.V. Kirsanov, Fiz. Tverd. Tela 11 (1969) 674 [Soviet Phys-Solid
State 11 (1969) 5401.
[54] V.A. Ryabov, Fiz. Tverd. Tela 12 (1970) 2747 [Soviet Phys-Solid State 12 (1971) 22161.
[55] D.P. Jackson and J.H. Barrett, Computer Phys. Commun. 13 (1977) 157.
[56] D.P. Jackson and J.H. Barrett, Phys. Letters 71A (1979) 359.
[57] J.H. Barrett and D.P. Jackson, Nucl. Instr. Methods 170 (1980) 115.
[58] D.P. Jackson and J.H. Barrett, Nucl. Instr. Methods B2 (1984) 318.
[59] H. Cramer, Mathematical Methods in Statistics (Princeton University Press, Princeton, 1971)
p. 310.
[60] V.A. Ryabov, Phys. Status Solidi 38 (1970) 63.
(611 R.S. Nelson, M.W. Thompson and M. Montgomery, Phil. Mag. 1 (1962) 1385.
(621 C. Scheringer, Acta Cryst. A29 (1973) 70.
(631 J.F. van der Veen! R.G. Smeenk, R.M. Tromp and F.W. Saris, Surface Sci. 79 (1979) 219.
[64] D.P. Jackson, T.E. Jackman, J.A. Davies, W.N. Unertl and P.R. Norton, Surface Sci. 126
(1983) 226.
[65] D.K. Faddeev and V.N. Faddeeva, Computational Methods of Linear Algebra (Freeman,
San Francisco, 1983) p. 144.
[66] J.A. Davies, D.P. Jackson, J.B. Mitchell, P.R. Norton and R.L. Tapping, Nucl. Instr.
Methods 132 (1976) 609.
(671 J.W.M. Frenken, R.M. Tromp and J.F. van der Veen, to be published.
[68] L.L. Kesmodel and G.A. Somorjai, Phys. Rev. Bll (1975) 630.
[69] J.R. Noonan, H.L. Davis and L.H. Jenkins, J. Vacuum Sci. Technol. 15 (1978) 619.
(70) H.L. Davis, J.R. Noonan and L.H. Jenkins, Surface Sci. 83 (1979) 559.
1711 J.R. Noonan and H.L. Davis, Surface Sci. 99 (1980) L424.
[72] J.F. van der Vcen, R.G. Smeenk and F.W. Saris, in: Proc. 7th Intern. Vacuum Congr. and
3rd Intern. Conf. on Solid Surfaces, Vienna, 1977, p. 2515.
[73] J.A. Davies, D.P. Jackson and P.R. Norton, in: Proc. 7th Intern. Vacuum Congr. and 3rd
Intern. Conf. on Solid Surfaces, Vienna, 1977, p. 2527.
[74] E. Bogh and I. Stensgaard, Phys. Letters 65A (1978) 357.
(751 J.A. Davies, D.P. Jackson, N. Matsunami, P.R. Norton and J.U. Andersen, Surface Sci. 78
(1978) 274.
[76] D.L. Adams, H.B. Nielsen, J.N. Andersen, I. Stensgaard, R. Feidenhans’l and J.E. Sorensen,
Phys. Rev. Letters 49 (1982) 669.
[77] I. Stensgaard, R. Feidenhans’l and J.E. Sorensen, Surface Sci. 128 (1983) 281.
[78] J.A. Davies, D.P. Jackson, P.R. Norton, D.E. Posner and W.N. Unertl, Solid State Commun.
34 (1980) 41.
[79] J.A. Davies, T.E. Jackman, D.P. Jackson and P.R. Norton, Surface Sci. 109 (1981) 20.
J. F. oan der Veen / Ion beam crwallograph~ 281

[80] J.F. van der Veen, R.G. Smeenk, R.M. Tromp and F.W. Saris. Surface Sci. 79 (1979) 212.
[81] R. Feidenhans’l, J.E. Sorensen and I. Stensgaard. Surface Sci. 134 (1983) 329.
[82] E. Tornquist, E.D. Adams, M. Copel, T. Gustafsson and W.R. Graham, J. Vacuum Sci.
Technol. A2 (1984) 939.
[83] Y. Kuk and L.C. Feldman, Phys. Rev. B30 (1984) 5811.
[84] J.W.M. Frenken, J.F. van der Veen. R.N. Barnett and U. Landman, to be published.
[85] I. Stensgaard, L.C. Feldman and P.J. Silverman, Phys. Rev. Letters 42 (1979) 247.
[86] T. Narusawa, W.M. Gibson and E. Tornquist, Phys. Rev. Letters 47 (1981) 417.
[87] T. Narusawa, W.M. Gibson and E. Tornquist, Surface Sci. 114 (1982) 331.
1881 J.W.M. Frenken, J.F. van der Veen and G. Allan, Phys. Rev. Letters 51 (1983) 1876.
[89] M.W. Finnis and V. Heine, J. Phys. F (Metal Phys.) 4 (1974) 237.
[90] J. Friedel, Ann. Physique 1 (1976) 257.
lo;] F. Jona, J.A. Strozier, Jr. and W.S. Yang, Rept. Progr. Phys. 45 (1982) 527.
[92] J. Sokolov, F. Jona and P.M. Marcus, Solid State Commun. 49 (1984) 307.
[93] J.P. Perdew and R. Monnier, J. Phys. F (Metal Phys.) 10 (1980) L287.
[94] J.P. Perdew, Phys. Rev. B25 (1982) 6291.
[95] K.P. Bohnen, Surface Sci. 115 (1982) L96.
[96] R.N. Barnett, U. Landman and C.L. Cleveland. Phys. Rev. B27 (1983) 6534.
[97] R.N. Barnett, U. Landman and C.L. Cleveland, Phys. Rev. B28 (1983) 1685.
(981 R.N. Barnett, U. Landman and C.L. Cleveland, Phys. Rev. Letters 51 (1983) 1359.
[99] H.B. Nielsen, J.N. Andersen, L. Petersen and D.L. Adams, J. Phys. Cl5 (1982) L1113.
IlOO] S. Andersson, J.B. Pendry and P.M. Echenique, Surface Sci. 65 (1977) 539.
[loll G. Allan, Surface Sci. 89 (1979) 142.
[102] G. Allan and J. Lopez, Surface Sci. 95 (1980) 214.
[103] L.L. Kesmodel and G.A. Somorjai, Phys. Rev. Bll (1975) 630.
[104] L.L. Kesmodel, P.C. Stair and G.A. Somorjai. Surface Sci. 64 (1977) 342.
[105] R. Feder, Surface Sci. 68 (1977) 229.
[106] H.L. Davies and J.R. Noonan, Surface Sci. 126 (1983) 245.
[107] J.E. Demuth, P.M. Marcus and D.W. Jepsen, Phys. Rev. Bll (1975) 1460.
[108] Y. Gauthier, R. Baudoing. C. Gaubert and L. Clarke, J. Phys. Cl5 (1982) 3223.
[109] Y. Gauthier, R. Baudoing and L. Clarke, J. Phys. Cl5 (1982) 3231.
(1101 D.L. Adams, H.B. Nielsen and J.N. Andersen, Surface Sci. 128 (1983) 294.
[ill] E. Zanazzi, F. Jona, D.W. Jepsen and P.M. Marcus, J. Phys. Cl0 (1977) 375.
[112] M. Maglietta, E. Zanazzi, F. Jona, D.W. Jepsen and P.M. Marcus. J. Phys. Cl0 (1977) 3287.
[113] M. Alff and W. Moritz, Surface Sci. 80 (1979) 24.
[114] G.E. Laramore, Phys. Rev. B8 (1973) 515.
[115] S. Andersson and J.B. Pendry. J. Phys. C6 (1973) 601.
[116] U. Landman and D.L. Adams, J. Vacuum Sci. Technol. 11 (1974) 195.
[117] J.E. Demuth and T.N. Rhodin, Surface Sci. 42 (1974) 261.
[118] W.N. Unertl and M.B. Webb, Surface Sci. 59 (1976) 373.
[119] J.H. Onuferko and D.P. Woodruff, Surface Sci. 91 (1980) 400.
[120] G. Hankk, E. Lang, K. Heinz and K. Muller, Surface Sci. 91 (1980) 551.
[121] R. Feder, S.F. Alvarado, E. Tamura and E. Kisker. Surface Sci. 127 (1983) 83.
[122] F. Jona, J. Phys. Cl1 (1978) 4271.
[123] D.P. Woodruff, in: The Chemical Physics of Solid Surfaces and Heterogeneous Catalysis,
Eds. D.A. King and D.P. Woodruff, Vol. 1 (Elsevier, Amsterdam, 1981) p. 82.
[124] D. Wolf, H. Jagodzinski and W. Moritz, Surface Sci. 77 (1978) 265, 283.
[125] W. Moritz and D. Wolf, Surface Sci. 88 (1979) L29.
[126] P. Heimann, J.F. van der Veen and D.E. Eastman, Solid State Commun. 38 (1981) 595.
[127] S.H. Overbury, W. Heiland, D.M. Zehner, S. Datz and R.S. Thoe. Surface Sci. 109 (1981)
239.
282 J. F. oan der Veen / Ion heum oy_mtllogruphy

[128] K.H. Rieder. T. Engel and N. Garcia, in: Proc. 4th Intern. Conf. on Solid Surfaces, Cannes.
1980, p. 861.
(1291 L.D. Marks, Phys. Rev. Letters 51 (1983) 1000.
[130] G. Binnig, H. Rohrer, Ch. Gerber and E. Weibel, Surface Sci. 131 (1983) L379.
[131] I.K. Robinson. Phys. Rev. Letters 50 (1983) 1145.
[132] Y. Kuk, L.C. Feldman and I.K. Robinson, Surface Sci. 138 (1984) L168.
[133] C-M. Chan, M.A. Van Hove, W.H. Weinberg and E.D. Williams, Surface Sci. 91 (1980) 440.
[134] D.L. Adams, H.B. Nielsen, M.A. Van Hove and A. Ignatiev, Surface Sci. 104 (1981) 47.
[135] T.E. Jackman. J.A. Davies, D.P. Jackson, W.N. Unertl and P.R. Norton, Surface Sci. 120
(1982) 398.
[136] T.E. Jackman, J.A. Davies, D.P. Jackson. P.R. Norton and W.N. Unertl, J. Phys. Cl5 (1982)
L99.
[137] P. Heilmann, K. Heinz and K. Muller, Surface Sci. 83 (1979) 487, and references therein.
[138] T.N. Rhodin and G. Brodtn, Surface Sci. 60 (1976) 466.
(139) G. Broden, G. Pirug and H.P. Bowel, Surface Sci. 72 (1978) 45.
[140] K. Griffiths, T.E. Jackman, J.A. Davies and P.R. Norton. Surface Sci. 138 (1984) 113.
[141] P.R. Norton, J.A. Davies, D.K. Creber, C.W. Sitter and T.E. Jackman, Surface Sci. 108
(1981) 205.
11421 T.E. Jackman, K. Griffiths, J.A. Davies and P.R. Norton, J. Chem. Phys. 79 (1983) 3529.
(143) P.H. Holloway and J.B. Hudson, Surface Sci. 43 (1974) 123, 141.
(1441 R.R. Smeenk, R.M. Tromp, J.F. van der Veen and F.W. Saris, Surface Sci. 95 (1980) 156.
[145] L.H. Germer and A.H. MacRae, J. Appl. Phys. 33 (1962) 2923.
(1461 J.E. Demuth, J. Colloid Interface Sci. 58 (1977) 184.
11471 W. Heiland and E. Taglauer. Surface Sci. 68 (1977) 96.
11481 J.A. van den Berg. L.K. Verheij and D.G. Armour, Surface Sci. 91 (1980) 218.
[149] R.G. Smeenk, R.M. Tromp and F.W. Saris, Surface Sci. 107 (1981) 429.
[150] S. Masuda, M. Nishijima, Y. Sakisaka and M. Onchi, Phys. Rev. B25 (1982) 863.
11511 R.P.N. Bronckers and A.G.J. de Wit, Surface Sci. 112 (1981) 133.
[152] J. Lapujoulade, Y. le Cruer, M. Lefort, Y. Lejay and E. Maurel, Phys. Rev. B22 (1980) 5740.
(1531 R. Feidenhansl and I. Stensgaard, Surface Sci. 133 (1983) 453.
[154] H.P. Bonzel and S. Ferrer, Surface Sci. 118 (1982) L263.
[155] A.M. Barb, G. Binnig, H. Rohrer, Ch. Gerber, E. Stall, A. Baratoff and F. Salvan, Phys. Rev.
Letters 52 (1984) 1304.
1156) M. Schuster and C. Varelas, Nucl. Instr. Methods B2 (1984) 299.
[157] M.K. Debe and D.A. King, Phys. Rev. Letters 39 (1977) 708.
[I581 T.E. Felter, R.A. Barker and P.J. Estrup, Phys. Rev. Letters 38 (1977) 1138.
[159] M.K. Debe and D.A. King, Surface Sci. 81 (1979) 193.
11601 R.A. Barker, P.J. Estrup, F. Jona and P.M. Marcus, Solid State Commun. 25 (1978) 375.
(161) L.C. Feldman, P.J. Silverman and I. Stensgaard, Surface Sci. 87 (1979) 410.
[162] E. Tosatti, Solid State Commun. 25 (1978) 637.
11631 D.A. King and G. Thomas, Surface Sci. 92 (1980) 201.
[164] M.R. Barnes and R.F. Willis, Phys. Rev. Letters 41 (1978) 1729.
11651 R.F. Willis, Surface Sci. 89 (1979) 457.
[166] J.F. van der veen, F.J. Himpsel and D.E. Eastman, Solid State Commun. 40 (1981) 57.
1167) J. Stohr, R. Jaeger and T. Kendelewicz, Phys. Rev. Letters 49 (1982) 142.
[168] M. de Crescenzi, F. Antonangeli, C. Bellini and R. Rosei, Phys. Rev. Letters 50 (1983) 1949.
(1691 J.E. Demuth, D.W. Jepsen and P.M. Marcus, Phys. Rev. Letters 32 (1974) 1182.
[170] L. Pauling, The Nature of the Chemical Bond, 3rd ed. (Cornell University Press, Ithaca, NY,
1960).
(1711 See e.g., S. Brennan, J. Stbhr and R. Jaeger, Phys. Rev. B24 (1981) 4871.
[172] See e.g., D.H. Rosenblatt, S.D. Kevan, J.G. Tobin, R.F. Davis, M.G. Mason, D.R. Denley,
D.A. Shirley, Y. Huang and S.Y. Tong, Phys. Rev. B26 (1982) 1812.
J.F. van der Veen / Ion beam ctysmllography 283

[173] R.F. Wallis, Progr. Surface Sci. 4 (1973) 233.


[174] S. Lehwald, J.M. Szeftel, H. Ibach, T.S. Rahman and D.L. Mills, Phys. Rev. Letters 50
(1983) 518.
[175] J.M. Szeftel, S. Lehwald, H. Ibach, T.S. Rahman, J.E. Black and D.L. Mills, Phys. Rev.
Letters 51 (1983) 268.
[176] G. Brusdeylins, R.B. Doak and J.P. Toennies, Phys. Rev. Letters 46 (1981) 437.
[177] D. Tabor, J.M. Wilson and T.J. Bastow, Surface Sci. 20 (1971) 471.
[178] M.G. Lagally, in: Surface Physics of Materials, Vol. 2, Ed. J.M. Blakely (Academic Press,
New York, 1971) p. 419.
(179) F. Cyrot-Lackmann, Surface Sci. 15 (1969) 535.
[180] J.P. Gaspard and F. Cyrot-Lackmann, J. Phys. C6 (1973) 3077.
[181] R. Haydock, V. Heine and M.J. Kelly, J. Phys. C5 (1972) 2845; C8 (1975) 2591.
[182] M. Mostoller and U. Landman, Phys. Rev. B20 (1979) 1755.
[183] J.E. Black, B. Laks and D.L. Mills, Phys. Rev. B22 (1980) 1818.
[184] See, e.g., J.J. de Launay, in: Solid State Physics, Vol. 2. Eds. F. Seitz and D. Turnbull
(Academic Press, New York, 1956) p. 219.
(185) R.N. Barnett, R.G. Barrera, C.L. Cleveland and U. Landman, Phys. Rev. B28 (1983) 1667.
[186] K. Masuda, 2. Naturforsch. 36a (1981) 454.
[187] L.C. Feldman, R.L. Kauffman, P.J. Silverman, R.A. Zuhr and J.H. Barrett, Phys. Rev.
Letters 39 (1977) 38.
[188] A. Kahn, Surface Sci. Rept. 3 (1983) 193.
[189] R.J. Culbertson, L.C. Feldman and P.J. Silverman, Phys. Rev. Letters 45 (1980) 2043.
(1901 R.M. Tromp, E.J. van Loenen, M. Iwami and F.W. Saris, Solid State Commun. 44 (1982)
971.
[191] P.A. Bennett, L.C. Feldman, Y. Kuk, E.G. McRae and J.E. Rowe, Phys. Rev. B28 (1983)
3656.
[192] R.M. Tromp, L. Smit and J.F. van der Veen, Phys. Rev. Letters 51 (1983) 1672.
[193] L.C. Feldman, P.J. Silverman and I. Stensgaard, Nucl. Instr. Methods 168 (1980) 589.
[194] I. Stensgaard, L.C. Feldman and P.J. Silverman, Surface Sci. 102 (1981) 1.
[195] R.M. Tromp, R.G. Smeenk and F.W. Saris, Phys. Rev. Letters 46 (1981) 939.
[196] R.M. Tromp, R.G. Smeenk and F.W. Saris, Surface Sci. 104 (1981) 13.
(1971 R.M. Tromp, R.G. Smeenk and F.W. Saris, Solid State Commun. 39 (1981) 755.
[198] R.M. Tromp, R.G. Smeenk, F.W. Saris and D.J. Chadi, Surface Sci. 133 (1983) 137.
[199] L. Smit, T.E. Derry and J.F. van der Veen, Surface Sci. 150 (1985) 245.
(2001 L. Smit, R.M. Tromp and J.F. van der Veen, Nucl. Instr. Methods B2 (1984) 322.
[201] L. Smit, R.M. Tromp and J.F. van der Veen, Phys. Rev. B29 (1984) 4814.
(2021 L. Smit and J.F. van der Veen, in: Proc. 17th Intern. Conf. on Physics of Semiconductors,
San Francisco, 1984, Eds. J.D. Chadi and W.A. Harrison (Springer, Berlin, 1985) p. 81
12031 R.E. Schlier and H.E. Farnsworth, in: Semiconductor Surface Physics (University of
Pennsylvania Press, 1957) p. 3.
[204] R. Seiwatz, Surface Sci. 2 (1963) 473.
[205] T.D. Poppendieck, T.C. Gnoc and M.W. Webb, Surface Sci. 75 (1978) 287.
[206] J.D. Levine, Surface Sci. 34 (1973) 90.
[207] J.A. Appelbaum and D.R. Hamann, Surface Sci. 74 (1978) 21.
[208] D.J. Chadi, Phys. Rev. Letters 43 (1979) 43.
[209] W.S. Verwoerd, Surface Sci. 99 (1980) 581.
[210] M.T. Yin and M.L. Cohen, Phys. Rev. B24 (1981) 2303.
[211] J. Ihm, M.L. Cohen and D.J. Chadi, Phys. Rev. B21 (1980) 4592.
[212] P.N. Keating, Phys. Rev. 145 (1966) 637.
[213] M.J. Cardillo, Phys. Rev. B21 (1980) 1497.
(2141 J. Pollmann, A. Mazur and M. Schmeits, Physica 117B/118B (1983) 771.
[215] D. Haneman, Phys. Rev. 121 (1961) 1093.
284 J.F. uan der Veen / Ion beam crystallographj

[216] R. Feder, Solid State Commun. 45 (1983) 51.


[217] K.C. Pandey, Phys. Rev. Letters 49 (1982) 223.
(2181 F.J. Himpsel, Th. Fauster and G. Hollinger, Surface Sci. 132 (1983) 22.
[219] F.J. Himpsel, P. Heimann, T.-C. Chiang and D.E. Eastman, Phys. Rev. Letters 45 (1980)
1112.
[220] S. Brennan, J. Stohr, R. Jaeger and J.E. Rowe, Phys. Rev. Letters 45 (1980) 1414.
(2213 D.J. Chadi, Phys. Rev. B26 (1982) 4762.
1222) G. Allan and M. Lannoo, Surface Sci. 63 (1977) 11.
[223] J.E. Northrup, J. Ihm and M.L. Cohen, Phys. Rev. Letters 47 (1981) 1910.
[224] A. Redondo, W.A. Goddard III and T.C. McGill, J. Vacuum Sci. Technol. 21 (1982) 649.
(2251 G.V. Hansson, RIG. Uhrberg and J.M. Nicholls, Surface Sci. 132 (1983) 31.
12261 F. Houzay, G. Guichar, R. Pinchaux, G. Jezequel, F. Solal, A. Barski, P. Steiner and Y.
Petroff, Surface Sci. 132 (1983) 40.
[227] R.M. Tromp, L. Smit and J.F. van der Veen, Phys. Rev. B30 (1984) 6235.
[228] R. Feder and W. Month, Solid State Commun. 50 (1984) 311.
[229] J.E. Northrup and M.L. Cohen, Phys. Rev. Letters 49 (1982) 1349.
[230] P.P. Auer and W. Month, Surface Sci. 80 (1979) 45.
(231) R.E. Schlier and H.E. Farnsworth, J. Chem. Phys. 30 (1959) 917.
[232] G. Binnig, H. Rohrer, Ch. Gerber and E. Weibel, Phys. Rev. Letters 50 (1983) 120.
[233] R.M. Tromp and E.J. van Loenen, to be published.
(2341 E.G. McRae, Phys. Rev. B28 (1983) 2305.
[235] F.J. Himpsel and I.P. Batra, J. Vacuum Sci. Technol. A2 (1984) 952.
12361 L.C. Snyder, Z. Wasserman and J.W. Moskowitz, J. Vacuum Sci. Technol. 16 (1979) 1266.
[237] M. Aono, R. Souda, C. Oshima and Y. Ishizawa, Phys. Rev. Letters 51 (1983) 801.
(238) J.V. Florio and W.D. Robertson, Surface Sci. 24 (1971) 173.
[239] H.D. Shi, F. Jona, D.W. Jepsen and P.M. Marcus, Phys. Rev. Letters 37 (1976) 1622.
12401 D.M. Zehner, C.W. White and G.W. Ownby, Surface Sci. 92 (1980) L67; Appl. Phys. Letters
37 (1980) 456.
(2411 D.M. Zehner, J.R. Noonan, H.L. Davis and C.W. White, J. Vacuum Sci. Technol. 18 (1981)
852.
(2421 D.E. Eastman, F.J. Himpsel and J.F. van der Veen, Solid State Commun. 35 (1980) 345.
(2431 Y.J. Chabal, J.E. Rowe and S.B. Christmann, J. Vacuum Sci. Technol. 20 (1982) 763.
[244] H.-J. Gossmann, L.C. Feldman and W.M. Gibson, Phys. Rev. Letters 53 (1984) 294.
[245] See e.g., C.T. Foxon and B.A. Joyce, in: Current Topics in Materials Science, Vol. 7, Ed. E.
Kaldis (North-Holland, Amsterdam, 1981) ch. 1, p. 1.
1246) P.K. Larsen, J.F. van der Veen, A. Mazur, J. Pollmann, J.H. Neave and B.A. Joyce, Phys.
Rev. B26 (1982) 3222.
(2471 P.K. Larsen and J.F. van der Veen, Surface Sci. 126 (1983) 1, and references therein.
[248] S.Y. Tong, G. Xu and W.N. Mei, Phys. Rev. Letters 52 (1984) 1693.
(249) C.B. Duke, R.J. Meyer and P. Mark, J. Vacuum Sci. Technol. 17 (1980) 971.
[250] C.B. Duke, S.L. Richardson, A. Paton and A. Kahn, Surface Sci. 127 (1983) L135.
[251] H.-J. Gossmann and W.M. Gibson, Surface Sci. 139 (1984) 239.
[252] C.B. Duke, Appl. Surface Sci. 11/12 (1982) 1.
[253] M.B. Brodsky and A.J. Freeman, Phys. Rev. Letters 45 (1980) 133.
[254] M. Grimsditch, M.R. Khan. A. Kueny and I.K. Schuller, Phys. Rev. Letters 51 (1983) 498.
(2551 A. Kueny, M. Grimsditch, K. Miyano, I. Banerjee, C.M. FaIco and I.K. Schuller, Phys. Rev.
Letters 48 (1982) 166.
[256] I.K. Schuller, Phys. Rev. Letters 44 (1980) 1597.
[257] E. Spiller, in: Low Energy X-Ray Diagnostics, Monterey, 1981, AIP Conf. Proc., Vol. 75, p.
124.
[258] M.P. Bruijn, J. Verhoeven and M.J. van der Wiel, Nucl. Instr. Methods 219 (1984) 603.
J.F. oan der Veen / Ion beam ctystollography 285

(2591 See, e.g., H.C. Casey, Jr. and M.B. Panish, Eds., Heterostructure Lasers (Academic Press,
New York, 1978).
[260] L.L. Chang and L. Esaki, in: Molecular Beam Epitaxy, Ed. B.R. Pamplin (Pergamon,
Oxford, 1980) p. 15.
[261] H.L. Stormer, Surface Sci. 132 (1983) 519.
[262] G.H. Dohler and K. Ploog, in: Molecular Beam Epitaxy, Ed. B.R. Pamplin (Pergamon,
Oxford, 1980) p. 145.
[263] G.C. Osboum, J. Vacuum Sci. Technol. 21 (1982) 469.
(2641 S.P. Murarka, J. Vacuum Sci. Technol. 17 (1980) 775.
[265] P.S. Ho, J. Vacuum Sci. Technol. Al (1983) 745.
[266] G.W. Rubloff, Surface Sci. 132 (1983) 268.
[267] E. Bauer and H. Poppa, Thin Solid Films 12 (1972) 167.
[268] R. Kern, G. Le Lay and J.J. Metois, in: Current Topics in Materials Science, Vol. 3, Ed. E.
Kaldis (North-Holland, Amsterdam, 1979) ch. 3, p. 131.
[269] J.A. Venables and G.L. Price, in: Epitaxial Growth, Part B, Ed. J.W. Matthews (Academic
Press, New York, 1975).
(2701 R.J. Culbertson, L.C. Feldman, P.J. Silverman and H. Boehm, Phys. Rev. Letters 47 (1981)
657.
[271] Y. Kuk, L.C. Feldman and P.J. Silverman, Phys. Rev. Letters 50 (1983) 511.
[272] J.H. van der Merwe and C.A.B. Ball, in: Epitaxial Growth, Part B, Ed. J. Matthews
(Academic Press, New York, 1975).
(2731 J.W. Matthews and A.E. Blakeslee, J. Crystal Growth 27 (1974) 118; 32 (1976) 265.
[274] M. Ludowise, W. Dietz, C. Lewis, M. Camras, N. Holonyak, B. Fuller and M. Nixon, Appl.
Phys. Letters 42 (1983) 487.
[275] E. Kasper, H.J. Herzog and H. Kibbel, Appl. Phys. 8 (1975) 199.
[276] E. Kasper and H.J. Herzog, Thin Solid Films 44 (1977) 357.
[277] J.C. Bean, L.C. Feldman, A.T. Fiory, S. Nakabara and I.K. Robinson, J. Vacuum Sci.
Technol. A2 (1984) 436.
[278] T. de Jong, W.A.S. Douma, J.F. van der Veen, F.W. Saris and J. Haisma, Appl. Phys. Letters
42 (1983) 1037.
[279] T. Narusawa and W.M. Gibson, J. Vacuum Sci. Technol. 20 (1982) 709.
[280] P.M.J. Ma&, R.I.J. Olthof, J.W.M. Frenken, J.F. van der Veen, PC. Zalm, C.W.T.
Bulle-Lieuwma and M.P.A. Viegers, J. Appl. Phys., in press.
[281] W.K. Chu, J.A. Ellison, S.T. Picraux, R.M. Biefeld and G.C. Osboum, Phys. Rev. Letters 52
(1984) 125.
[282] W.K. Chu, F.W. Saris, C.-A. Chang, R. Ludeke and L. Esaki, Phys. Rev. B26 (1982) 1999.
[283] J.H. Barrett, Phys. Rev. B28 (1983) 2328.
[284] S. Saito, H. Ishiwara and S. Furukawa, Appl. Phys. Letters 37 (1980) 203.
(2851 K.C.R. Chiu, J.M. Poate, L.C. Feldman and C.J. Doherty, Appl. Phys. Letters 36 (1980) 544.
[286] H. Ishiwara and T. Asano, Appl. Phys. Letters 40 (1982) 66.
[287] V. Heine, Phys. Rev. Al38 (l%S) 1689.
12881 S. Kurtin, T.C. McGill and CA. Mead, Phys. Rev. Letters 22 (1969) 1433.
[289] M. Schlttter, Phys. Rev. B17 (1978) 5044.
[290] J.C. Phillips, J. Vacuum Sci. Technol. 11 (1974) 947.
12911 L.J. Brillson, Phys. Rev. Letters 40 (1978) 260.
[292] S.G. Louie and M.L. Cohen, Phys. Rev. B13 (1976) 2461.
[293] W.E. Spicer, I. Lindau, P.R. Skeath, C.Y. Su and P.W. Chye, Phys. Rev. Letters 44 (1980)
420.
[294] J. Tersoff, Phys. Rev. Letters 52 (1984) 465.
(2951 W. Schottky, 2. Physik 113 (1939) 367.
[296] J. Bardeen, Phys. Rev. 71 (1947) 717.
(2971 E.J. van Loenen, J.W.M. Frenken and J.F. van der Veen, Appl. Phys. Letters 45 (1984) 41.
286 J.F. “an der l’een / Ion beam crystailograph~

(2981 P.E. Schmid, P.S. Ho, H. FBI1 and G.W. Rubloff, J. Vacuum Sci. Technol. 18 (1981) 937.
12991 J.A. Roth and C.R. Crowell, J. Vacuum Sci. Technol. 15 (1978) 1317.
[300] R. Butz, G.W. Rubloff and P.S. Ho, J. Vacuum Sci. Technol. Al (1983) 771.
[301] J.G. Clabes, G.W. Rubloff and T.Y. Tan, Phys. Rev. B29 (1984) 1540.
[302] G. Rossi, I. Abbati, L. Braicovitch, 1. Lindau and W.E. Spicer, Phys. Rev. B25 (1982) 3627.
13031 Y.-J. Chang and J.L. Erskine, J. Vacuum Sci. Technol. Al (1983) 1193.
1304) A. Franciosi, D.J. Peterman, J.H. Weaver and V.L. Moruzzi, Phys. Rev. B25 (1982) 4981.
[305] P.J. Grunthaner, F.J. Grunthaner and J.W. Mayer, J. Vacuum Sci. Technol. 17 (1980) 924.
[306] Y. Shiraki, K.L.I. Kobayashi, H. Daimon, A. Ishizuka. S. Sugaki and Y. Murata. Physica
117B/118B (1983) 843.
(3071 0. Bisi, L.W. Chiao and K.N. Tu, Phys. Rev. B30 (1984) 4664.
[308] R.T. Tung, J.M. Gibson and J.M. Poate, Phys. Rev. Letters 50 (1983) 429.
[309] D. Cherns, G.R. Anstis, J.L. Hutchinson and J.C.H. Spence, Phil. Mag. 46A (1982) 849.
(3101 J. Stohr and R. Jaeger, J. Vacuum Sci. Technol. 21 (1982) 619.
[311] F. Comin, J.E. Rowe and P.H. Citrin, Phys. Rev. Letters 51 (1983) 2402.
[312] N.W. Cheung, R.J. Culbertson, L.C. Feldman, P.J. Silverman, K.W. West and J.W. Mayer.
Phys. Rev. Letters 45 (1980) 120.
13131 N.W. Cheung and J.W. Mayer, Phys. Rev. Letters 46 (1981) 671.
[314] T. Narusawa, W.M. Gibson and A. Hiraki, Phys. Rev. B24 (1981) 4835.
[315] T. Ito and W.M. Gibson, J. Vacuum Sci. Technol. A2 (1984) 561.
[316] R.M. Tromp, E.J. van Loenen, M. Iwami, R.G. Smeenk and F.W. Saris, Thin Solid Films 93
(1982) 151.
[317] R.M. Tromp, E.J. van Loenen, M. Iwami, R.G. Smeenk, F.W. Saris, F. Nava and G.
Ottaviani, Surface Sci. 124 (1983) 1.
[318] R.M. Tromp, E.J. van Loenen, M. Iwami, R.G. Smeenk, F.W. Saris, F. Nava and G.
Ottaviani, Surface Sci. 128 (1983) 224.
[319] E.J. van Loenen, M. Iwami, R.M. Tromp, J.F. van der Veen and F.W. Saris, Thin Solid
Films 104 (1983) 9.
(320) E.J. van Loenen, M. Iwami, R.M. Tromp and J.F. van der Veen, Surface Sci. 137 (1984) 1.
(3211 J.A. Venables, J. Derrien and A.P. Janssen, Surface Sci. 95 (1980) 411.
[322] G. Le Lay, M. Manneville and R. Kern, Surface Sci. 72 (1978) 405.
[323] M. Saitoh, F. Shoji, K. Oura and T. Hanawa, Surface Sci. 112 (1981) 306.
[324] G. Rossi, I. Abbati, L. Braicovitch, I. Lindau and W.E. Spicer, Surface Sci. 112 (1981) L765.
[325] T. Narusawa, W.M. Gibson and A. Hiraki, Phys. Rev. B24 (1981) 4835.
(3261 H. Tokutaka, K. Nishimori, S. Nomura, A. Tanaka and K. Takashima, Surface Sci. 115
(1982) 79.
(3271 E.J. van Loenen, A.E.M.J. Fischer and J.F. van der Veen, Surface Sci. 155 (1985) 65.
[328] K.N. Tu, Appl. Phys. Letters 27 (1975) 221.
[329] A. Hiraki, Surface Sci. Rept. 3 (1984) 357.
[330] E.J. van Loenen, J.F. van der Veen and F.K. Legoues, Surface Sci. 157 (1985) 1.
[331] M. Iwami, S. Hashimoto and A. Hiraki, Solid State Commun. 49 (1984) 459.
(3321 A. Zunger, Phys. Rev. B24 (1981) 4372.
[333] M. Grioni, J. Joyce, S.A. Chambers, D.G. O’Neill, M. de1 Giudice and J.H. Weaver, Phys.
Rev. Letters 53 (1984) 2331.
[334] S. Saitoh, H. Ishiwara, T. Asano and S. Furukawa, Japan. J. Appl. Phys. 20 (1981) 1649.
[335] R.T. Tung, J.M. Poate, J.C. Bean, J.M. Gibson and D.C. Jacobson, Thin Solid Films 93
(1982) 77.
1336) R.T. Tung, J.M. Gibson and J.M. Poate, Appl. Phys. Letters 42 (1983) 888.
[337] S.M. Sze, Physics of Semiconductor Devices (Wiley, New York, 1969) p. 587.
[338] J.M. Gibson, R.T. Tung and J.M. Poate, in: Defects in Semiconductors II, Boston, 1982,
Proc. Materials Research Society Symp., Vol. 14, Eds. S. Mahajan and J.W. Corbett
(North-Holland, New York, 1983) p. 395.
J. F. uan der Veen / Ion beam crystallography 287

[339] K. Akimoto, T. Ishikawa, T. Takahashi and Seishi Kikuta, Japan. J. Appl. Phys. 22 (1983)
Ll98.
[340] E.J. van Loenen, J.W.M. Frenken, J.F. van der Veen and S. Valeri, Phys. Rev. Letters 54
(1985) 827.
[341] H.-J. Gossmann and W.M. Gibson, Surface Sci. 139 (1984) 239.
[342] T. Narusawa, N. Watanabe, K.L.I. Kobayashi and H. Nakashima, J. Vacuum Sci. Technol.
A2 (1984) 538.
[343] J. Gyulai, J.W. Mayer, V. Rodriguez, A.Y.C. Yu and H.J. Gopen, J. Appl. Phys. 42 (1971)
3578.
[344] R.G. Smeenk, R.M. Tromp, J.W.M. Frenken and F.W. Saris, Surface Sci. 112 (1981) 261.
[345] R. Haight and L.C. Feldman, J. Appl. Phys. 53 (1982) 4884.
(3461 K.L. Nagi and CT. White, J. Appl. Phys. 52 (1981) 320.
[347] Y.C. Cheng, Progr. Surface Sci. 8 (1977) 181.
[348] G. Binnig, H. Rohrer, Ch. Gerber and E. Weibel, Phys. Rev. Letters 49 (1982) 57.
[349] F.A. Lindemann, Physik. Z. 14 (1910) 609.
[350] L. Pietronero and E. Tosatti, Solid State Commun. 32 (1979) 255.
[351] J.W.M. Frenken and J.F. van der Veen, Phys. Rev. Letters 54 (1985) 134.

You might also like