Heteroepitaxial dimer structures on the silicon (100) surface


Material Information

Heteroepitaxial dimer structures on the silicon (100) surface
Physical Description:
vii, 146 leaves : ill. ; 29 cm.
Grant, Mark Wayne, 1965-
Publication Date:


bibliography   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1994.
Includes bibliographical references (leaves 137-145).
Statement of Responsibility:
by Mark Wayne Grant.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
oclc - 32801337
System ID:

Full Text













Surface Structures. .
Transmission Ion Channeling .

2 EQUIPMENT .......

UHV Chambers .
Introduction Chamber .. .
Preparation Chamber .
Ion Scattering Chamber .
Accelerator .


Rutherford Backscattering Spectroscopy .
Channeling .. .


Thin Si Windows .
Chemical Preparation .
Surface Cleanliness .
Crystallinity .
Thin Ge Windows ..
Fabrication. .
Characterization of Crystallinity
Surface Cleaning .


. 30
. 30
. 30
S 33
. 35
. 35
. 44

Calculated Yields .


Assumptions 48
Execution .. 56
Crystallographic Projections ..... 60

6 DATA REDUCTION .... . 66

Experimental Angular Scan Extraction.. . 66
Comparison of Experimental and Calculated Angular Scans 74

7 THE Si(100) SURFACE ..... .83

8 Sb ON THE Si(100) SURFACE ..... 88

M otivation. . 88
Experiment . 90
Results and Discussion . 93
Annealed Surfaces. ...... .93
Unannealed Surfaces ..... 95


M otivation. 101
Experiment 103
Results and Discussion . 103
Adatom Adsorption Sites (0.6 ML) .... .103
Coverage Dependence. .... 109


Motivation. .. 114
Experiment . 117
Results and Discussion . 119


Sb on Si(100) .. 127
GeonSi(100) 129
SbonGeonSi(100) .. 130







There are many to whom I owe a debt at this point in my life. Paramount

among these are my parents and sisters; no person could expect a more loving and

supportive family. They have been a constant and reliable source of encouragement.

I also wish to acknowledge the support of my advisor, Liz Seiberling. She has

shown a wealth of patience, always providing a challenging yet comfortable

environment in which to work. It is largely do to her guidance that I was able to finish

this thing. I am also indebted to Paul Lyman for getting me started in the lab, and for

doing all of the hard work before I began. He has been a good friend.

In the way of colleagues, I have been lucky. My life has been greatly enriched

by the opportunity to interact with two Dutch students, Jan Hoogenraad and Danny

Dieleman, and a Pole, Andrez Dygo. All have expanded my view of the world (not to

mention, Danny and I have the same taste in bars). It has also been a pleasure to share

a lab and office with Mark Boshart. He and I have had quite a few laughs. I have, as

well, enjoyed and benefited from my friendships with Allison Bailes and Rick Picullio.

I wish to thank Dave Wilmess for keeping me sane. Also, I thank the many others, too

numerous to name, who have made my stay here as pleasant as could be expected.

Finally, I wish to acknowledge the love and understanding of Wendy this last


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy


Mark W. Grant

August 1994

Chairman: Lucy Elizabeth Seiberling
Major Department: Physics

I have used transmission ion channeling to characterize thin heteroepitaxial

films on the Si(100) surface. The studies have been structural in nature, designed to

determine the adatom bonding sites for three separate but related systems: Sb

deposited on the Si(100) surface at coverages 5 1 monolayer (ML), Ge below 1 ML

on Si(100), and Sb deposited on 1 ML pseudomorphic Ge on Si(100). These studies

illustrate the power and utility of transmission channeling for adatom site


For the Sb/Si(100), I have determined the overlayer geometry for samples

annealed at 5500C after Sb deposition. I find this surface to be terminated by

symmetric Sb-Sb dimers having a bond length of 2.80.1 A. The site compares well

with previous studies for the annealed surfaces. Surprisingly, I find that the surface

formed upon deposition at room temperature also is terminated by symmetric dimers,

in spite of the presence of significantly less long-range order on the surface.

I have characterized the Ge-adsorbed Si(100) surface for a range of coverages.

The data suggest the presence of some coverage dependence on the surface. The

bonding geometry of the Ge has been determined for a Ge coverage of 0.6 ML,

yielding asymmetric dimers, having a bond length of 2.60.1 A and a dimer tilt of


Having characterized these systems, I next studied the system of 1 ML of Sb

adsorbed on 1 ML Ge-terminated Si(100). Two structural issues were addressed.

First, the effect of the Sb overlayer on the Ge has been determined. Upon deposition

of Sb, the reconstruction in the Ge layer is lifted, rendering the Ge near bulk-like.

Second, the geometry of the Sb overlayer on strained Ge has been determined. By

characterizing both the Ge and Sb layers, I conclude that the Sb dimers are

asymmetric, having a bond length of 2.760.07 A and a dimer tilt of 73.

The main focus of this dissertation concerns the deposition, and subsequent
formation of monolayer (1 monolayer, ML, is defined as 6.78x1014 atoms/cm2) films,

of foreign chemical species on the clean Si(100)-2xl surface (heteroepitaxy).

Specifically, three systems are addressed: germanium adsorbed at coverages below 1
ML on the clean Si(100) surface {Ge/Si(100)}, near 1 ML of antimony adsorbed on

the clean Si(100) surface (Sb/Si(100)}, and 1 ML of antimony adsorbed on 1 atomic
layer of Germanium on the Si(100) surface {Sb/Ge/Si(100)}. These studies, though

conducted on technologically important systems, were motivated by an interest in the

basic physical mechanisms governing the interactions of the adsorbates (adatoms) with

the surface and with each other. They were designed to determine the structures of
these surfaces on an atomic scale, and thereby to elucidate the processes leading to

epitaxy and interface formation. While it is often the case that the application of any

single experimental technique does not lead to a complete understanding of a given

physical system, it is true that through the combined efforts of many techniques and
with the contemplation of many scientists, often, collectively, such an understanding

may be achieved. Studies of semiconductor surfaces are, more often than not,

characterized by this collective approach. In this spirit, the following structural

investigations were undertaken.

In the interest of clarity, the discussions of the three systems studied (Sb/Si,
Ge/Si, Sb/Ge/Si) are presented separately and the relevant connections are made in the

conclusions. These three topics constitute chapters 8, 9, and 10. For each, remarks

on the motivation for the research are presented along with a review of the important

results from other researchers. In no way are these intended to be a complete reviews

of the topics. In general, only work directly impacting the interpretation or

understanding of the present results is included. Preceding those specific chapters are

ones describing the experimental procedures and data reduction, including a discussion

of the main technique and computer calculations involved in the analysis. Those topics

covered in detail were chosen to facilitate the continued effort by the research group

on such studies. For instance, a moderate amount of space has been devoted to the

production and characterization of alternate substrates (Ge, chapter 4), as this

information should be useful for future studies of adsorbates on the Ge(100) surface

using transmission ion channeling. In contrast, little space is devoted to the

experimental setup, since this, while important, has been covered thoroughly elsewhere


The results presented here should be of interest to many theorists and

experimentalists concerned with reconstructed surface structures in relation to

electronic properties, epitaxy on strained layers, surfactant-mediated growth, adatom

mobility, and many other topics. In addition, ion-beam enthusiasts should be

interested in these studies as an example of the power and utility of MeV ion beams.

These two aspects of this study are introduced below.

Surface Structures

When a crystalline surface is created (by cleaving or cleaning) in the absence of

contaminants (i.e. in a vacuum), the resultant two crystals are terminated by atoms that

are representative of the bulk constituents of the material, but that reside in a physical

and chemical environment which has been altered drastically. As a result, the

outermost atomic layer, as well as several layers directly adjacent to it, often exhibits

structural characteristics not observed in the bulk [Zang88]. In order to compensate

for their new environment, the surface atoms move from their nominal, bulk-

terminated lattice positions to a configuration of lower energy. For Si surfaces, due

partly to the directional nature of the Si-Si bonds and to the dangling bonds left by the

act of cleaving, the effect is remarkable [Need87; Robe90; Chad79; Jaya93]. For this

and other semiconductors, such movement ("reconstruction") typically occurs in the
plane of the surface and the phenomenon is often accompanied by altered bonding

states, orbital rehybridization, and lattice strain [Mead89; Redo82; Appe78a;
Appe78b; Ters92]. As a dramatic example, consider the clean Si(111) surface. The
ideal bulk-terminated surface has a simple lxl symmetry (the "mxn" nomenclature
indicates the number of in-plane bulk lattice vectors needed to describe the

reconstructed surface). However, the surface in equilibrium at room temperature

exhibits a superstructure having a 7x7 symmetry, where more than 100 atoms per unit

cell take place in the reconstruction over 4 atomic layers [Zang88, p. 49]. This

amazing structure is illustrated in figure 1-1. Note the complexity and substantial size
of the surface unit cell.

The surface of interest in this study, Si(100), commonly exhibits a considerably
less complicated reconstruction. The surface unit cell contains only 2 top-layer atoms,

and the symmetry of the surface (2x1) [Schl59] is related simply to the symmetry of

the bulk. The surface, though, still exhibits striking structural features which are not

completely transparent, despite years of intense study by experimentalists and
theorists. Perhaps because of the relative simplicity and the technological importance

of this particular Si surface, it has become a model one to study and somewhat of a

gauge with which to measure the current state of the science. Although the

reconstructed surface is presumably in an energetically-optimized configuration for

clean Si, the surface is still highly reactive, possessing one dangling bond per surface
atom [Chad79; Redo82]. Because of this reactivity, most impurities introduced on the
surface will interact in such a way as to form chemical bonds with the topmost Si

atoms chemisorbb), and hence adhere to the surface in preferred bonding sites

Figure 1-1. Si(11l) 7x7 surface reconstruction. Shown is the surface unit cell. This
figure is from Zangwill [Zang88, p. 49].

[Tang92; Zang88, p. 185]. For this reason, structural studies on such systems are of


Since the primary thrust of these structural studies has been to elucidate the

ordering, on an atomic level, of chemisorbed overlayers on the Si(100) surface, the

studies concentrate on thin films. This focus entails the deposition of monolayers (a

few A) of the adsorbate, and hence involves the initial stages of interface formation

and the chemistry of the initial stages of epitaxial growth. At this level, structural

studies address the positions of adsorbed atoms with respect to the underlying Si

lattice and/or the positions of the adatoms with respect to each other. Also of interest

is the effect of the adsorbed overlayer on the underlying, adjacent Si. Issues affecting

the surface region in these systems can involve such variables as the size of the adatom

[Cope90], the availability and mobility of the adatoms [Ters94; Mo90; Eagl93],

epitaxy-induced surface stress [Trom92a; Ters91; Trom93], the preferred valence

electron state of the adatom [Mead89; Uh86], the relative strength of the adsorbate-

adsorbate to adsorbate-substrate interaction [Mo92; Barn86], and the interaction of

the adatoms with defects (such as steps and surface vacancies) on the surface [Mo89;


It has been through intense study that detailed features of such systems

typically are clarified. From the beginning, advances in experimental studies have

accompanied advances in theoretical treatments, as the experimentally determined

bonding geometries and subsurface lattice distortions were compared to those

predicted by total energy calculations, molecular dynamics simulations, or some other

treatment. The usual result was a clearer understanding of the mechanisms leading to

reconstruction and a more certain interpretation of experimental data [Appe78b;

Chad79]. In conjunction with the electronic properties of such surfaces, provided by

probes such as photoemission spectroscopy [Rowe74; Land92; Yang92], both

structural studies and theories foster insight into the bonding characteristics of these
increasingly difficult systems. It is hoped that the work presented herein will aid in this

Transmission Ion Channeling
The development of the technique of transmission ion channeling has occurred
over the last 30 years. The initial discovery of the channeling phenomenon sprang

from the observation of anomalous sputtering yields that varied with incident-beam
angle [Rol60] and anomalous ranges of ions in crystals when the crystals were aligned
in certain directions [Davie60]. There was then a tremendous amount of experimental

and theoretical work on the subject (there are several early review articles on the

subject including that by Gemmell [Gemm74]). For instance, one important work was

published in 1965 by Lindhard [Lind65], detailing the continuum model (chapter 5)

and making several other lasting contributions to the field. Also, Barrett, using
computer simulations of the channeling process, provided early insights into the
interpretation of channeling data [Barr71]. It was realized early on that the channeling

effect could be used to locate impurities in crystalline materials. Experiments in this

vein were first conducted around 1967 [B0gh67, Matz67]. Now such lattice location

experiments in the bulk are common. Examples include a wide range of impurities in a

variety of crystals, from W implanted in Cu at various levels [Bord76] to D implanted

[Bech88a] in Si.

The first use of transmission ion channeling for interface studies was that of
Feldman et al. [Feld78] in studies of the Si-SiO2 interface. It was first used to

determine the adsorption site of an impurity on a surface in 1981 with the study by

Cheung and Mayer of Ni on the chemically clean Si surface [Cheu81]. Similar studies

were carried out by Jin, Ito and Gibson for Au on Si(100) [Jin85]. In the 1980's, a

group at the University of Aarhus, Denmark, began using channeling in the

transmission geometry for studies of D, H, O and Te on metal surfaces [Mort88,
Jens90]. Then, in 1985, the first ultrahigh vacuum (UHV) adsorption-site

determination was carried out on a UHV-cleaned surface using transmission ion
channeling on a metal surface {D on Ni(100)} [Sten85].

Continuing to expand the range of application of this technique, the results
presented in this dissertation represent the first use of the technique for the
determination of adsorbate bonding sites on the UHV-clean Si(100)-2x1 surface.
Also, channeling in the transmission geometry (such that the beam passes through the
sample, with scattered particles detected at forward angles), or transmission
channeling, is currently practiced at only a few laboratories in the world. At the
University of Florida Van de Graaff accelerator, where the present studies were

undertaken, many of the clean-surface-preparation and equipment-related problems
were solved prior to the present work, and studies had been conducted using
transmission ion channeling on the psuedomorphic growth of Ge thin films on Si(100)
[Lyma9lb]. However, no structural studies on an atomic level had been performed.

A secondary focus of the work detailed in this dissertation has therefore been to refine
the existing techniques, leading to the ability to use transmission ion channeling for

adatom site determinations on clean semiconductor surfaces [Seib93]. Finally, the
techniques developed were tested on physical systems. Specifically, agreement was
sought with previous studies for the systems of Sb/Si(100) annealed at 5000C, which

had been structurally characterized using surface extended x-ray adsorption fine
structure (SEXAFS) and scanning tunneling microscopy (STM) [Richt90], and
Ge/Si(100), which had been structurally characterized using x-ray standing waves
(XSW) [Font93].

The quantity and complexity of the equipment used for the studies presented

here has much to do with the intrinsic nature of the experiments. These are surface

studies, which require meticulous care for the provision of a clean vacuum

environment. This objective is accomplished through the use of modern vacuum

chambers, pumping systems, and monitoring systems. In addition, the samples studied

often are produced employing molecular beam epitaxy, and characterized with several

standard surface analysis techniques. Since these samples are of acceptable quality for

a limited time, adequate procedures and equipment for the introduction and transfer of

samples into the vacuum chamber are necessary. The main technique used for

analysis, MeV ion scattering, also requires special equipment for the attainment of a

suitable beam, while maintaining good vacuum, and the ability to detect scattered

particles in a variety of geometries. This whole capability is then coupled to an

accelerator (which produces the ion beam) and occupies a footprint consistent with the

limited laboratory space available. An extensive and complete description of this setup

has been given in the Ph.D. thesis of Lyman [Lyma91a]. In the interest of

completeness, this chapter contains a brief discussion of the experimental equipment

and capabilities.

UHV Chambers
Because of the large number of sample preparation and experimental

capabilities required to perform these surface studies, geometrical issues and port

space are of concern. This set of constraints has led to the use of three separate,

interconnected ultrahigh vacuum (UHV) experimental chambers. One chamber is

dedicated to ion scattering and another to sample preparation. A third, smaller

chamber is used as a means of introducing several samples (up to five) into the system

at a time. It provides the ability to change samples quickly (unfortunately, a common

occurrence) without breaking vacuum. Each of these chambers can be isolated via

UHV-compatible valving, making maintenance on any given one possible while the

others are pumped. Figure 2-1 shows an overhead view of the setup. Samples are

clipped to Ta modules that can be picked up and moved under vacuum from one

chamber to the next using rack-and-pinion transfer arms. Vacuum is maintained in the

system by careful attention to the cleanliness and composition of materials inserted and

by baking the preparation and scattering chambers after each break in vacuum. The

baking procedure consists of heating the chambers to -1300C for around 2 days.

Introduction Chamber

The stainless steel sample introduction chamber contains a carousel with

receptacles for five sample modules. Thin Si crystals to be used as substrates are

prepared chemically and cleaned ex situ (see chapter 4), then clipped to modules and

loaded into the introduction chamber five at a time. The introduction chamber is then

evacuated to roughly 2x10-9 Torr, a process which usually takes several hours and

requires no bakeout. The chamber is small (8" diameter) and is pumped by a Varian

150 1/s diode ion pump. A gatevalve separates the pump from the chamber. The

chamber contains an ion gauge for measuring the pressure. This gauge, however, is

seldom used because of a suspicion that it contaminates the samples, which are in

close proximity. There are several viewports to facilitate sample transfer. A module

then can be picked up, translated into the preparation chamber, and attached to the

sample manipulator therein.

Preparation Chamber

The commercial, stainless steel sample preparation chamber (12" diameter) is

pumped by a 300 1/s ion pump, and has a base pressure of 5 x 10-" Torr. The sample

2.5 MeV He+










Figure 2-1. Experimental setup. The three UHV chambers are illustrated schematically
from an overhead view (from the Ph.D. thesis of P. Lyman, p.19) [Lyma91a]. Sample
preparation and surface characterization equipment are contained in the preparation
chamber. The scattering chamber is equipped with 4 solid-state charged particle
detectors. Samples are moved from chamber to chamber with the transfer arms.


manipulator has two angular and three translational degrees of freedom. A sample can

be heated radiatively to >10000C using a Ta strip heater on the sample stage or by a

retractable heat lamp, and can be cooled by contact with a copper braid attached to an
in situ liquid nitrogen reservoir. We have succeeded in cooling a sample to -400C by

suspending a pressurized (-8 psi) 40 1 liquid nitrogen dewer near, and connected with

a short rubber hose to, the nitrogen inlet port and flowing nitrogen through the
reservoir. The heat lamp consists of an in vacuo tungsten-halogen bulb whose
filament is at one focus of an elliptical nickel reflector; the lamp can be translated so

that the other focus of the reflector is near the sample. The lamp has a maximum

power of 250 W and can melt Si. The strip heater is composed of a 1.0cm x 1.5 cm x

0.0127 mm strip of Ta and is shielded from the rear of the sample by a thin sheet of

sapphire. The temperature of a sample is estimated from the current passing through

the Ta strip, using a calibration curve [Lyma91a]. The calibration was performed with
a thermocouple in contact with a thin window. The chamber is outfitted with rear

view low energy electron diffraction (LEED) optics and a cylindrical mirror analyzer

for Auger electron spectroscopy (AES) for surface characterization. The sample

modules are designed so that a given sample can be analyzed with LEED, AES and

transmission channeling without breaking vacuum. Sample preparation instruments

include an ion sputter gun and two effusion cells. A residual gas analyzer (RGA) is

available for the characterization of gas species in the chamber.

Ion Scattering Chamber

The ion scattering chamber has a base pressure of 5x10-" Torr, and is
maintained at pressures below lx10-10 Torr with beam on sample by differential
pumping along the beam line. The chamber is composed of stainless steel, has a 12"

diameter, and is pumped by a 500 1/s Varion diode ion pump. The samples are

mounted on a precision goniometer with two angular and three translational degrees of

freedom (see figure 2-2), typically allowing five well-separated beam spots ( 1 mm2)

per thin window ( 7 mm diameter). In this geometry, with the exception of one point

on the sample, a rotation about either axis may result in a translation of the beam spot.

Near the <100> axial direction, for which this effect is greatest, and for a beam spot

maximally displaced from the axis of rotation, a rotation of the sample about its normal

by 20 (typical for an angular scan) leads to a displacement of approximately 0.05 mm.

This displacement causes only 5% of the beam to irradiate a new part of the surface.

A change in the beam spot on this level is not expected to significantly influence our

results. A Faraday cup is located downstream of the sample, and is fitted with a

viewport in the rear. The beam can be stopped either by a piece of tantalum or by a

quartz disc. The quartz fluoresces when struck by the beam, and the beam profile

thereby can be observed. Four bakeable, passivated, ion-implanted silicon detectors

are mounted at angles of 18, 50, 78, and 1500 to the beam direction. The geometrical

placement of the detectors is illustrated in figure 2-3. The 780 detector is mounted

below the sample, attached to the goniometer (facing up). The detector at 180 is

covered by a thin foil (10 micron Al), and is used to measure elastic recoils [Behr87].


As indicated in figure 2-1, the system of UHV chambers is connected via a

beamline to an ion accelerator. The accelerator is a HVEC 4 MV single-ended Van de

Graaff, that, in practice, is capable of producing a high current (gA) beam of H+ or

He+ ions in an energy range of 0.5 to 3.5 MeV. For a typical experiment, we use 2.5

MeV He+, with a current on sample of 5-20 nA (singly charged, i.e., before passing

through sample). The current is reduced from the nominal ;300 nA produced by the

Van de Graaff by the insertion of a small (1.6mm) aperture in the beamline after

focusing the beam at the end of the beamline to a size of 2x2 mm. This aperture also

serves, along with collimating slits, to limit the beam divergence, which we estimate as


360 /

Figure 2-2. Scattering chamber goniometer. Indicated are the 0 and 4 angular
degrees of freedom. The manipulator to which this is attached also provides 3
translational degrees of freedom.

Detector placement



Figure 2-3. Top view of the detector arrangement in the scattering chamber. After
passing through the sample (center), the beam enters a Faraday cup (not shown) for
current integration.

<0.80. The incident beam energy is determined by measuring the field strength of the
analyzing magnet with a nuclear magnetic resonance gaussmeter.

The strength of the magnetic field has been calibrated against the beam energy
using the 27Al(p,y)Si28 resonance reaction at 991.9 keV [Feld77]. Two points were

obtained for the calibration by counting y's as the energy was varied for both H+ and
H2+ beams. Figure 2-4 shows the data used for the calibration. The vertical axis

displays the number of gammas detected. The horizontal axis displays the frequency

of the NMR probe and is related linearly to the strength of the magnetic field analyzing

the beam. The signal has the form of a step function because the H particles were

incident on a thick Al target and therefore produced gammas at some point in the

target whenever the energy of the beam was above that for the resonance. In each

case (H+ and H2+) the voltage on the Van de Graaff terminal was adjusted to give
992 keV per H particle at the sample when on resonance, yielding the two data points

needed for the calibration.

It should be noted that this procedure was performed after the data presented
in this manuscript were taken. Upon calibration it was discovered that the energy of

the beam was slightly in error using the earlier tabulated frequency settings. This shift

would have caused a slight, systematic underestimation of the absolute coverage

determined for each sample. However, since there was no way to know precisely

when the discrepancy arose, and since any error introduced was within our

experimental uncertainty in the coverage, the data were not corrected.

1 I I I I I I










3110 3120 3130 3140 3150
Frequency (kHz)


3160 3170

6260 6280 6300
Frequency (kHz)



Figure 2-4. Data used for Van de Graaff calibration. The resonance is located at the half-
height of the step. a) Data for H+ beam; b) Data for H2 beam.


10000 -





27 A(p,)2Sie2




b) H2


I I I .

1 i l l


MeV ion beams have enjoyed widespread application for materials analysis for

many years [Feld82; Feld77; Chu78; Van85; Sten92]. For the results presented in this

manuscript, the principle tool was a 2.5 MeV beam of He+ ions. In this chapter, some

of the characteristics of ion beam analysis that make the technique attractive for the

application to surface analysis are outlined. Also included is a discussion of the ways

in which the ions were used in a typical experiment to probe the physical

characteristics of our samples. The discussion begins with a description of a typical

backscattering spectrum (in the absence of channeling) used to study the gross

physical properties such as sample thickness, uniformity, composition and degree of

contamination of the samples. This use represents the most basic and common one for

MeV ion beams in materials analysis. The discussion then progresses to cover the

ideas behind the use of channeling in the transmission geometry to elucidate the atomic

structure of the surfaces of samples. Augmenting this treatment is a brief overview of

the mechanisms governing the channeling process (channeling will be described further

in chapter 5).

Rutherford Backscattering Spectroscopy
One of the most useful and well-established techniques utilizing ion beams is

Rutherford backscattering spectroscopy (RBS) [Feld82]. There are three important

properties of RBS which make it useful for materials analysis, in general, and for this

study in particular. First, the technique is mass dispersive, which allows separation in

energy of the signal from among impurities and from that of the substrate itself

[Van85]. In practice this quality is extremely useful for the identification of unwanted

sample contamination and the confirmation of the chemical makeup of the samples as

prepared. Second, by avoiding beam energies leading to non-Rutherford scattering,
the process is classical, with well-defined and well-described cross sections
(Rutherford), making the technique quantitative [Sten92]. This feature allows the

experimental determination of impurity levels (for coverage above -0.001ML for

heavy nuclei and ~ 1 ML for light). For surface studies this quantitative nature insures

that the absolute coverage of surface adlayers can be determined. In fact, RBS

represents one of the most accurate and trusted methods (in cases where the impurity

has a larger Z than the substrate) for determining the amount of a given foreign

chemical species on a surface or implanted in a near-surface region. Finally, since the

rate of energy loss for protons and alpha particles in most materials has been studied

extensively and is well tabulated [Zieg77], the technique affords some depth

resolution. This makes RBS one of the most accepted techniques for the study of thin
films (see our analysis of the Ge thin windows, chapter 4).

These considerations are summarized graphically in figure 3-1, where is shown
a typical RBS spectrum (scattering angle = 1500) from a Si sample with 0.95 ML of

Sb on the front (beam exit) surface. In the inset the scattering geometry is sketched.

This spectrum also indicates that some Sb (
on the back of the sample. Further illustrated is the widening of the Si signal and the

separation of the Sb peaks on the two sides due to energy loss of the ions through the

sample. The mass dispersive quality of the technique is clear from the separation of

the Si substrate and surface Sb peaks. From such a spectrum (with a calibrated

detector) one can determine accurately the thickness of the Si substrate (5%), the
identity of the surface "contaminant" ( Z = 1), and the amount of the contaminant (













200 300 400 500 600 700 800

Channel Number

Figure 3-1. Sample RBS spectrum. The open circles are the data.
geometry is shown in the inset (the front is the surface studied).

The scattering

For our experiments, RBS has been used mainly to determine the thicknesses
of the thin Si crystals and the amounts of given impurities adsorbed on the surfaces.

For the surface coverage determinations, typically, an accuracy of 3-5% (or even 1-

2%) [Van85] is attainable from RBS in a system optimized for precise quantitative
impurity detection. However, in our experimental setup, designed for thin samples,

implanted standards (common for achieving high accuracy [Cohe83; Watj90]) cannot

be used because we integrate current after the beam passes through the sample. Thus,
our measurement of sample coverage is limited by experimental uncertainties involving

current integration, detector solid angle, and counting statistics. Fortunately,
knowledge of the coverage to within 10% is adequate for the studies presented in this

manuscript (our setup is optimized to measure relative yields). For the thickness

measurement, inaccuracies associated with the empirical, tabulated energy loss

parameters provide the dominant errors. The parameters are known to within

approximately 5% [Zieg77], which translates to an uncertainty of -5% in a

determination of the sample thickness. This level of accuracy is acceptable since the

sample thickness is important in the data analysis only in that it is used in the

calculation of the angular scans (chapter 5). The angular scans are relatively

insensitive to differences in thickness on the present level (hundreds of A).

In addition to providing information on the composition of materials and the
thickness of films, MeV ion beams also can be used to study the crystallinity of

materials. Advantage is taken of the fact that, under the proper conditions, the spatial
distribution of the ion beam can be affected drastically as it penetrates a crystalline

material [Feld82]. This effect occurs because only a small fraction of the total number

of ions suffer a low-impact-parameter collision at a given atomic plane (<<1%). The

vast majority penetrate farther into the crystal, where they are influenced mainly

through large-impact-parameter interactions with the host nuclei (and slightly through
multiple collisions with electrons).

Indeed, if a beam of MeV ions penetrates a crystal while aligned with a major
crystallographic direction, its Coulomb interaction with the atom rows causes the ions
to be gently steered (the interaction is gentle in that the change in the trajectory of a
channeled ion caused by any one crystal atom is small). These interactions lead to

highly a non-uniform flux distribution, which is peaked at the center of the channels
(flux peaking). Eventually (after ~ 1000 A for 2.0 MeV alpha's), the ion beam
assumes the in-plane periodicity of the crystal. As a consequence of this effect
(referred to as "channeling"), the probability that an impurity in (or on) the crystal will

suffer a low-impact-parameter collision (i.e. will scatter an ion significantly) will
depend strongly upon its position with respect to the underlying matrix (substrate).
This process is illustrated schematically in figure 3-2.a. Here one channel is depicted
as viewed from a direction perpendicular to the beam. The channel is defined by the
rows of atoms (open circles). Several ion trajectories are sketched (not to scale), and
the pileup of flux at the center of the channel is shown as the dashed curve. The
motion of any given ion is controlled by the electrostatic potential in the crystal and

the direction and point of entry of the ion into the channel. Interactions with the ion

rows are largely characterized by repeated, correlated collisions which occur over the
distance of hundreds of atoms [Lind65].

The experimental manifestation of the channeling process is a greatly reduced
scattering yield from the substrate for channeled beams. This effect is shown in figure
3-3. Here, spectra are shown for a 2.0 MeV beam of He+ ions incident on Si, aligned
(solid curve) with the <100> axial channeling direction (<100> crystallographic
direction), and in a random direction (dashed curve) near the <100> axis (70 tilt of the

beam with respect to the crystallographic axis). Although taken on the same sample,

beam (MeV He )

flux distr.

000000000000000000 00Q0000

oooooooooooooooooo .oooo'ooooo/
000000000000000000 o 00'0000000o

~ 5000 A
a() ,hin r al adaom

\ VLA y JL U U

on beam-
exit surface

Angular scan





1.2 -0.8 -0.4

0.0 0.4 0.8 1.2

Tilt Angle

Figure 3-2. Schematic illustration of channeling, a) Channeled particles (represented by
solid lines) leading to flux peaking (dashed line); b) Angular scans for those adatom
positions indicated in a.













100 200

300 400 500 600 700 800
Channel Number

Figure 3-3. Sample spectrum from 780 spectrum. Spectra for both random (dashed
curve) and aligned (solid curve) incidence are shown. The scattering geometry is given in
the inset (the surface of interest is toward the detector).

the random spectrum shown here differs in appearance from that in figure 3-1 (RBS

spectrum) because the scattering angle is 780 (as compared to 150). The scattering

angle affects the data in several ways. First, the path length of a scattered ion through

the Si is increased, due to the glancing view the detector has of the sample. This
increased path length causes a corresponding larger energy loss for scattered particles.

For this reason the Si signal appears much thicker (occupies more of the horizontal

axis), and obscures the Sb peak from the back of the sample. Also, the mass

resolution is reduced due to kinematics [Gold80], which moves the Sb and Si signals

closer together. Another effect is an increased cross section, which leads to a greater
scattering yield for a given solid angle. In figure 3-3, however, the reduction in the

yield for the aligned spectrum, as compared to the random, is completely a channeling


In channeling, the flux distribution is also a function of the tilt angle, \, of the

beam with respect to the axial channeling direction. It is the combined dependence of

the scattering yield on position, and this dependence of the flux on N, that is used in

transmission ion channeling for adatom site determination. In practice, this

determination is accomplished by a comparison of the experimentally determined

scattering yield as a function of \, an angular scan, with computer-calculated angular

scans. Also shown in figure 3-2.a, are potential bonding sites for adatoms on the

beam-exit side of the thin crystal (large circles). Figure 3-2.b then shows angular

scans for such positions. Note that, due to the flux peaking, an interstitial site will

exhibit an enhanced yield in the channeling direction (W-0), while a substitutional site

will show a reduced yield (Si occupies, by definition, a substitutional site). At larger

tilts, the yield for each of these sites approaches that for random incidence (defined as

1). Of course, a whole range of adatom sites, and, hence angular scans, is possible.

Qualitatively, much can be inferred about the adatom bonding sites from a

consideration of the shapes and depths of such angular scans. For a quantitative

determination of the bonding site, however, experimental and computer-calculated

scans are compared.

An important consideration concerning the experimental angular scans is the

angle of tilt of the beam with respect to the given crystallographic direction, \. In

principle, it is necessary to know precisely the exact angular position at which each

spectrum is acquired with respect to the crystal (this defines the so-called "crystal

coordinates"). In practice, however, systematic uncertainties introduced by the

physical makeup of the goniometer and sample mounting conspire to make this

knowledge difficult to obtain with precision. These effects can be accounted for and

the crystal coordinates calculated after a lengthy calibration of the experimental setup

[Dygo93]. However, such precision is difficult to maintain reliably for our setup, and
is unnecessary for characterizing small tilts about a known direction. Since we used

angular scans over only ~2, and since we can determine accurately the coordinates of

the axial direction, our main concern then becomes the execution of angular scans

across the axial channel in an appropriate direction. For these studies, care was taken

to conduct the scans in a so-called "random plane". Such scans are meant not to

coincide with nor intersect (except at tilt = 0) any major crystallographic planes of the


In order to avoid problems associated with planar channeling, one has to be

able to determine the positions of the planar directions. Since we use thin crystals,

which allow the beam to pass through the sample, the requisite determination is

possible. After passing through the sample, the beam enters a faraday cup and can be

made to strike a piece of quartz (in front of a 2.75" viewport). Beam alignment is then

accomplished by viewing the distinct patterns on the quartz, which fluoresces when

struck by MeV ions, as a channeling direction is approached [Arms71]. This

phenomenon expedites the aligning of the sample along an axial channeling direction,

because the ion beam forms a doughnut pattern as the axial direction is approached

[Gemm74; Rosn78]. Aligning the sample in this way with any axial or planar

channeling direction can be accomplished with great accuracy in less than 1 min. and

with a beam current of less than 10 particle nA. Angular scans are then planned in

order to avoid the planes. For the <100> axial channeling direction, the < degree of
freedom has a relatively minor affect on the actual tilt angle because the < axis is
nearly parallel to the beam (see figure 2-2 for the definition of the angular degrees of
freedom 0 and <). Therefore, scans are conducted by tilting in 0. Since the angular

positions of the planes will depend upon the orientation in which the sample is

mounted, the planes are avoided by choosing the proper orientation of the sample
module on the goniometer [Lyma91a]. Several orientations often are tried before a
<100> scan is completed. For the <111> direction, the angle between the sample

normal (approximately the J axis) and the beam is 550. Therefore, in this channeling

direction the crystal is oriented reproducibly with respect to the beam from run to run.

Also, no major plane is encountered for tilts in <, so the scans are conducted with >

movement. For the <110> direction, again, the sample is oriented reproducibly.

However, tilts in only 0 or only < both coincide with planes in the crystal. Therefore,

scans are conducted using tilts with both angles. This procedure is illustrated in figure

3-4, which displays a contour plot of the Si yield obtained at tilts about the <110>
axial direction. In this figure, the effect of planes crossing the axial channeling

direction is seen clearly. This contour plot was constructed from a data set composed
of 763 spectra, taken on a grid in 0 and in the RBS chamber (this chamber has a

goniometer which is identical to the one in the UHV scattering chamber). Also shown

as the symbols (*) is a superposition of the cut typically used to acquire data across


this axial direction. This cut is seen to avoid adequately the major planes, satisfying

the conditions defining a random plane.

-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2
0.8 r .--- .









0.4 0.6 0.8

-1.0 1 '1 1i A//A I Y/r I \ i) i 1 i \i i I\\N\ I \ \ \I I I I
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Figure 3-4. Plane of scan for <110> axial channel. The horizontal axis represents the
0 degree of freedom, and the vertical axis 4 (units are degrees). The plane of the
angular scan is shown as the symbols (*). Each contour represents 2000 counts.

1.0 1.2












Transmission ion channeling necessitates the use of thin (from 0.2pm to

2.0im), single crystals that are able to withstand manipulation (for sample loading,

etc.) and thermal cycling (for cleaning, sample preparation, etc.). There exist recipes

for making such windows of a variety of materials. For example, thin Ni windows in

the (100) orientation have been produced by epitaxial growth of metal onto single-

crystal NaCI, followed by flotation of the film from the salt and onto a frame in water

[Jens90]. Similar procedures have been utilized to produce thin crystals of Al, Cu, Pd,

Pt, W, Ag [Jens90], and Ge [Outl84]. The experiments presented in this discussion

were conducted on the Si(100) surface. A suitable technique for the production of

thin Si windows was described in the literature by Cheung [Cheu80]. The procedure,

as carried out in our laboratory, has also been described in detail [Lyma91a], so only a
brief overview will be given here. The basic idea is to employ a selective etchant

which ceases to etch Si when its dopant level is sufficiently high.

At a certain point in time, it became desirable also to have thin Ge windows

available for transmission ion channeling studies. Although the use of self-supporting

Ge crystals several microns thick had been reported in connection with nuclear physics

experiments [Gibs72], no published literature described their fabrication, uniformity or

crystalline quality. Free-standing single-crystal Ge films fabricated by epitaxial growth

onto NaCI and separation from the substrate by differential sheer stress had been

reported [Outl84], but were substantially thicker (10 gm) than desirable and were
therefore unsuitable for use in transmission ion channeling. After considerable effort,

an appropriate procedure was developed for the in-house fabrication of suitable

windows [Gran93d] in collaboration with Dr. F. Namavar at Spire Corporation.

Although the Ge windows have not, as yet, been used for surface studies in this
laboratory, their fabrication and characterization are described below, since a

substantial effort was expended towards their development. Utilization would
constitute a logical extension of the experiments contained herein.
Thin Si Windows

Chemical Preparation.
As received, a Si wafer is 2" in diameter, ~ 200 upm thick, and typically is

doped lightly with Sb. After a brief cleaning step, Boron is deposited on the polished
surface (suspended in an organic solvent) and diffused into the surface region of the
wafer in a tube furnace (at 10500C, under flowing, dry N2). After the diffusion is
accomplished, the wafer is cleaved into 1.3 x 1.3 cm samples and each is etched into a
thin window in two steps. The first step removes Si from a circular region on the back
(unpolished) side of the sample to a thickness of 5 im. The second step selectively
etches the remaining n-type Si until reaching the n+ region of high (5 x 1019 B/cm3)
Boron concentration, at which point etching essentially stops [Bohg71]. These steps

are outlined schematically in figure 4-1. After the thin windows are produced, a

lengthy ex situ cleaning process is employed to remove any metallic and carbonaceous

surface contaminants. That process is very similar to that described by Ishizaka and

Shiraki [Ishi86] (referred to as the "Shiraki oxide" technique), which leaves the Si

protected by a thin oxide (-5-10A). Our ex situ cleaning procedure does differ in a
few details, and therefore is outlined in appendix A. The oxide which results is free
from impurities within the sensitivity of AES (see below), and is desorbed in vacuo by

radiative heating.
Surface Cleanliness
Before surface experiments are performed on a sample, and after the oxide

desorption, it has to be insured that the clean surface is of high crystalline quality and

B doped

Etch iv 1

Etch 8 2

t 1 Nitric:Hydrofluoric:Acetic

Figure 4-1. Sketch of samples at various stages of preparation (cross sectional view, not
to scale). Etch #1 constitutes the gross removal of Si. Etch #2 is selective, removing only
the lightly doped Si. The thick, unetched region surrounding the window is used as a
supporting frame.



is atomically clean. To this end, the surfaces are characterized by AES and LEED.

Both of these techniques are established and extremely useful for the surface scientist.

Both utilize the short escape depth of the exiting electrons to achieve a high surface

sensitivity [Wood86, chapter 3]. To produce the escaping electrons, AES relies on the

excitation of core-levels and their subsequent decay, which can kick out Auger

electrons [Davis78]. This excitation is accomplished in most cases, including ours, by

a focused keV electron beam [Davis78]. While keV electrons are efficient at knocking

out core-level electrons, they also effectively heat the substrate. For thick

semiconductors, the heat is carried away readily. For the samples we use, however,

the heating by the electron beam is drastic enough to bum holes in the thin windows.

Therefore, none of the AES spectra discussed in this work was taken on thin windows.

Instead, spectra typically were acquired on the thick frame next to the window, or

during separate runs on thick crystals. We have no reason to think that these Auger

spectra are not characteristic of the elemental composition of the surface on the thin

windows prepared similarly.

LEED allows an evaluation of the degree of order and the symmetry of the

atomic arrangements on the surface [Zang88, chapter 3; Wood86, chapter 2] and is

accomplished by diffraction of a low-energy electron beam from the surface atoms.

Because of the low energies employed in LEED (10-100 eV), no such heating problem

has been encountered. LEED has been used to monitor the cleanliness and uniformity

of substrates as measured on the thin windows themselves for all samples discussed.

Both LEED and AES indicate that our Si(100) sample surfaces, after the

thermal desorption of the oxide, are clean and well ordered. LEED shows the two-

domain 2x1 pattern characteristic of this surface [Feld86, p. 167; Schl59; Hame86] for

most samples. Although the quality of the patterns varies from sample to sample,

typically they are sharp and the background low (high background or diffuse spots are

indicative of disorder [Wood86, p. 38]). Occasionally, a sample is overheated, leading

to the appearance of satellite spots in the pattern, possibly because of faceting

[Hame86] or B segregation [Lyma91a]. B segregation has been shown to alter the

reconstruction on the Si(l 11) surface [Bens89]. Data from such samples were not

used in the analysis. Auger spectra indicate that the oxide, before thermal desorption,

is free of impurities. Spectra taken after desorption also typically show little

contamination. This trait is illustrated in figure 4-2, which shows Auger spectra before

and after oxide desorption on a Si(100) sample. The positions of common

contaminants, C and O, are indicated. This spectrum is particularly clean, as it is not

uncommon to observe a C peak above background, due mainly to the buildup of C on

the surface as a function of electron bombardment.


Transmission ion channeling studies rely on the crystalline properties of the

substrate of interest. The technique, however, is comparatively insensitive to defects

in a crystal. This insensitivity is evidenced by our channeling studies for highly doped

Si. Although the doping level (5xl0'9 B/cm3) has a tremendous impact on the

electronic properties of the Si, this level represents only one part in 1000 atomic

impurities, and would not be expected to influence channeling measurements

significantly. The indicator of good crystalline quality, as measured by channeling, is

the minimum yield. Table 4-1 shows measured values of the minimum yield for a

number of samples taken over a period of 2 years, along with the standard deviation,

and number of measurements. These values were taken from data for which there was

available an angular scan, to help insure that true alignment of the beam with the axial

direction was obtained. Also shown are calculated values for the different channels

using a formula arrived at from Monte Carlo simulations of the channeling process

[Feld82, p. 44].

200 300

400 500

Electron Energy (eV)

Figure 4-2. Sample Auger spectra from Si(100) surface, a) Before oxide desorption; b)
After oxide desorption.


Si C

Si(100) with protective oxide



Si(100) after oxide


Table 4-1. Si channeling minimum yields

direction Xmin a # calc.t

<100> 0.048 0.013 22 0.033
<110> 0.038 0.012 15 0.023
<111> 0.050 0.013 23 0.028

tThe <110> and <111> values were scaled from the <100> value
[Feld82, p. 45] with d (linear string density).

Our measured values of the minimum yield are seen to agree reasonably well with the

calculated values. For our measurements, the minimum yield is found by integrating

the total counts in the Si peak (500 detector) from scattering throughout the ~ 5000 A

thickness. Due to dechanneling of the beam with thickness, our values are expected to

be higher than the calculated values, which give the minimum yield just after entering

the crystal (see figure 5-7). Also, for our measurements, there exists a finite beam

divergence, which serves to increase the minimum yield slightly [Gemm74]. Overall,

we conclude that the crystalline quality of a typical thin Si window as produced in the

manner described above is adequate for the studies detailed herein.

Thin Ge Windows


Using an Applied Materials 1200 epitaxy reactor at atmospheric pressure,

epitaxial Ge layers with thicknesses of 0.5 4.0 pm were grown at Spire Corporation

on clean Si(100). Prior to Ge growth, to produce clean Si surfaces, the samples were

heated to 1200 oC for 1/2 hour and then etched in situ by passing HCL over the

surface until approximately 1 rm of the Si was removed. Germanium tetrachloride

(GeCl4) was used as a Ge source for growth on these substrates, with the growth rate
ranging from 0.5 to 1.0 gm/min and growth temperatures from 7500C to 8500C.

Transmission Electron Microscopy (TEM) and cross-sectional TEM (XTEM) were

used by Dr. Namavar at Spire to determine defect densities in the layers. A typical

XTEM micrograph of the as-grown Ge film is shown in figure 4-3. The micrograph

shows a defective region of misfit dislocations extending from the Ge/Si interface 2000
- 3000 A toward the surface. A region of high crystalline quality extends about 5000

A from the surface. A plane-view TEM survey of a large area near the surface of the

sample indicates a defect density in the range of 106/cm2 [Nama91]. This defect

density, low for epitaxial Ge layers on Si(100), is attributed to a high dislocation glide
velocity at the relatively elevated growth temperatures employed in CVD [Kvam91;
Nama90]. The defect density is below that detectable by ion channeling.
In order to produce thin windows, several etching steps (outlined below) were

applied to the as-grown samples. The state of the sample after the successive steps is
schematically illustrated in figure 4-4. After the Si(100) wafers with the Ge grown on
them were received from Spire, they were cleaved into 1.3 cm by 1.3 cm squares, and
the front (Ge) and back (Si) of the sample were masked completely in Parafilm
(Parafilm is made by Dixie/Marathon, Greenwich, Conn.) except for a small ( ~2mm)
hole in the center of the back. Then the first etchant, which consisted of

HF:HNO3:CH3CO2 H (2:4:lvolumetric ratio), was applied. It was used to etch the

unprotected region to a thickness ranging from a few to tens of microns. This part of
the sample preparation was similar to that applied in the selective etching of doped Si
for the production of thin Si windows. The purpose of this step was to create a bowl-

shaped cavity in the Si, both to act as a template for the final window, and to expedite

the etching in subsequent steps. For the fabrication of Si windows this step is
terminated when light from a high intensity lamp is visible through the thinned Si,

however Ge is opaque at these thicknesses, preventing the use of this method. The
hope during this step was to come as close to the Si-Ge interface as possible, without
etching the Ge (or etching entirely through the sample), which was accomplished

(roughly 25% of the time) by careful calibration of the etch rate under controlled

Figure 4-3. Plan-view cross-sectional transmission electron micrograph of 1.2 gm Ge film
on Si substrate. The defective region is visible as the dark area.



As grown

Final window

Figure 4-4. Illustration of Ge thin window etching steps. Step one removes large
quantities of Si. Step two employs a selective etch which does not remove Ge. Step three
is not selective, but is brief and removes the defective region from the back of the window.

experimental conditions. The length of time required in this etch bath to reduce a 400

pm wafer to a few microns was ~ 25 min. Next, the Parafilm was removed, and the

sample was submerged in a solution consisting of ethylenediamine:H20:pyrocatechol

(EDP) in the ratio 50 ml:25 ml:10 g, held at 850C, which etches Si at a rate of about

50 pm/hour but etches Ge at a negligible rate. The samples were left in this solution

until the Ge/Si interface was reached, which was indicated by the appearance of a flat,

blue-tinted region in the center of the etch pit. It was found that KOH increased the

window size more quickly and gently than continued etching in the EDP, so the

samples were removed from the EDP just after the interface was reached and were

placed in a 10 Normal solution of KOH at 85C until of suitable diameter. In this

manner, samples of up to -6 mm in diameter were produced (although the larger

samples were not tested for durability under ion irradiation). After these steps, the

sample consisted of a uniformly-thin Ge window backed by a defect-ridden region and

surrounded by a thick Si frame (third panel in figure 4-4). Finally, the defective region

was removed from the back of the window by masking the front of the sample in

Parafilm (without contacting the thin part) and dipping it briefly (~ 6 sec) in a mixture

of HN03:HF (20:1 volumetric ratio).

Characterization of Crystallinity

RBS with channeling was used to further characterize the structure of the

windows. Ion beam analysis of the Ge windows was accomplished in a high vacuum
(~10-6 Torr) ion scattering chamber connected to the UF Van de Graaff accelerator

and equipped with a goniometer having two stepper-motor-controlled angular degrees

of freedom. For the data presented here, the beam was 2.0 MeV He+ and a surface

barrier detector was placed at a scattering angle of 1700. Figure 4-5 shows a

channeling spectrum (solid circles) for a 1.2 pm Ge epilayer on Si exposed to the

etching procedure described above, including the removal of the defective region.

550 1

500 -

450 Channeling

400- ** <100>

350 Ge

S300 Bulk
S250 1
I Si

100 Ge

50 -

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75

Energy (MeV)

Figure 4-5. Channeling spectra from Ge thin window. The open circles represent
scattering from the thick (unetched) portion of the sample. The filled circles represent
scattering from the thin window. The dashed curve is from a channeling spectrum
obtained from a bulk Ge crystal. The peak in the spectrum from the thick region (open
circles) is due to the defect-ridden region, which has been removed from the back of the
thin window (filled circles).

Channeling spectra of the unetched sample (open circles) and of bulk Ge (dashed
curve) are shown for comparison. The two spectra were taken on the same sample,

the former on the thick frame surrounding the thin window and the latter in the center

of the window (see inset). The peak in the channeling spectrum from the unetched

sample near 0.95 MeV is the result of ion beam dechanneling by defects in the Ge

layer near the Si-Ge interface [Feld82, chapter 4], as well as direct scattering of the ion
beam by the distorted channels near defects [Wiel86, Quer68]. A detailed analysis of

the effects leading to this peak is given in appendix B. A visual inspection of figure 4-
5 indicates that the etching process removed the Si substrate and the defective region

of the Ge epilayer. The Ge minimum yield on the window is clearly as low as that in

the unetched region, and as low as that for bulk Ge, indicating that the crystalline

quality of the Ge epilayer is comparable to that of the bulk crystal in the region near

the Ge surface. Channeling spectra on the window have also been taken with the
beam entering through the back (etched side) of the Ge and are indistinguishable from
that shown in figure 4-5. This implies that the crystalline quality of the window at the

back is comparable to that at the front, and further indicates that the defective region

has been completely removed. If the defects had not been removed, immediate

dechanneling as the beam entered the crystal from the back would lead to a dramatic

increase in the yield throughout the sample thickness.

In figure 4-5, the bulk spectrum (dashed line) and the spectra from the thin

window sample were obtained in separate experiments. For the bulk experiment, the

beam was aligned with the Ge <100> axial channel using the standard procedure
[Chu78, p. 228] of determining the polar coordinates of several major planar

channeling directions, plotting these points on polar graph paper, and identifying the
axial direction as the point of intersection of the planes. This procedure was time

consuming (several hours), so, to minimize the influence of beam damage and carbon

buildup, the beam spot was changed before returning the channeling direction (-10
min.) and taking the spectrum shown. In contrast, the procedure used for aligning the

beam on the Ge window was similar to that employed with the Si windows (discussed

in chapter 3). However, for the Ge case (due to the thicker windows and increased

stopping power of Ge over Si [Zieg77]), the pattern on the quartz viewport was

visually less distinct. Although this had the effect of decreasing the precision with
which we were able to align the beam, with care the alignment was accomplished

satisfactorily, as is indicated by the low minimum yields obtained.

Using RBS, it easy to estimate the thickness variation of the thin windows.

Due to statistical fluctuations in the processes governing the amount of energy lost by

energetic ions traversing a crystal, an ion beam having an energy distribution which

can be approximated by a delta function upon entering the crystal will have a different
energy distribution at a thickness t. This effect is termed energy straggling. It has

been shown that this distribution (except at small t) can be well-described by a

Gaussian, with variance given by Bohr theory [Bohrl5] as,

nB2 = 47(Zle2)2NZ2t. (4-1)

Here, Z, and Z2 are the atomic numbers of the incident ion and target, respectively.

The falling edge of a random spectrum taken from a thin film is broadened by

contributions from a finite detector resolution, energy straggling through the sample,

and nonuniform sample thickness. A random spectrum from the Ge window was used

to estimate the thickness variation of the window over an area the size of the beam
spot (1 mm2). Figure 4-6 shows the numerical derivative of the spectrum (dots) in the

regions of the beam entrance and beam exit of the window and Gaussian fits to each

peak (solid lines). The fit to the beam entrance peak yields the energy resolution of

the system at the beam entrance, )D. The quantity OD includes detector resolution

Derivative of random spectrum







500 600 700
x (channel number)

Figure 4-6. Numerical derivative of random spectrum from Ge/Si sample. Portions of the
derivative of the spectrum are shown. Since the rising and falling edges of RBS spectra
take the form of error functions, the derivatives of these regions are Gaussian [Chu78, p.
50]. The functional form used to fit the peaks is given, along with the value obtained for
the fitting parameters.

(dominant), the energy spread of the incident beam, and contributions from the
electronics. Equation (4-1) yields the variance due to straggling at the beam exit

surface. All of these contributions can be treated as Gaussian, so they add

fi2 = CD2+ 2f2B2 + t2. (4-2)

Here, 0 is the width of the Gaussian at the beam exit side in figure 4-6, and represents
the apparent resolution at t. Using this, we place an upper limit on the thickness

variation of the window (D) within the area of the beam spot of 8% for this sample.

Surface Cleaning

Shown in figure 4-7 is a portion of an AES spectrum taken near the edge of

(but not on) a Ge thin window before any in situ cleaning steps. There are clear peaks

from several elements: C, O, Cl, and Ge. These peaks, plus the absence of a LEED

pattern, indicate that the surface is covered by a disordered layer of various impurities,

probably composed of organic molecules and Ge-C, Ge-O species. We have

successfully sputter-cleaned the thin Ge windows using 1.5 keV Ar bombardment
(emission = 22 mA at a pressure of -5x10-5 Torr and for a beam rastered over -2.5

cm2). A spectrum from the sample of figure 4-7 after 2 cycles of sputtering and

annealing is shown in figure 4-8. In contrast to the sample before cleaning, after

cleaning, impurities give no signal above background. From this spectrum the O level

is estimated to be below 0.4%, with C below 0.3%. Sputtered samples have been

annealed to approximately 6000C. These samples typically yielded sharp 2x1 LEED
patterns with low background and with 4th order streaks, as observed by other groups
(for instance [Culb86]), indicative of the of c4x2 domains and hinting at the transition
(from 2x1 to c4x2) which has been observed on this surface just below room

temperature [Culb86, Keva85]. Further, we have found that such a Ge window

exposed to a channeled 2 MeV He ion dose of 42 pCoulombs/mm2 shows no

degradation in crystalline quality (as measured by ion channeling). Since a typical dose

required for an energy spectrum is 6 to 12 iCoulombs, the windows meet all criteria

necessary for use as substrates for transmission ion channeling.









100 200 300 400 500

Electron Energy (eV)


Figure 4-7. Auger electron spectrum from a Ge-window sample before cleaning, a) Raw
counts versus energy, N(E); b) Spectrum as typically displayed, dN/dE (surface impurities
are indicated).


AES spectrum
Ge after cleaning

200 400 600 800 1000 1200

Electron Energy (eV)

Figure 4-8. Auger electron spectrum from Ge-window sample after sputtering and
annealing for two cycles.

The basic motivation behind the use of computer-calculated angular scans for
the interpretation of channeling data is clear. On the one hand, experimental angular

scans provide a very direct glimpse at the registry of the overlayer with the substrate.
This assertion is true because, for channeled beams, the ion flux varies tremendously

over any given axial channel (discussed below), and therefore can give large variations
in the scattering yield from spatially differing bonding sites. On the other hand,

achieving the level of precision demanded by the current state of surface science

requires an intimate knowledge of the flux distribution in any such channel, a

distribution which is not readily measured in a typical experiment. The solution has

been to reproduce the flux using computer calculations of the channeling process.

There are two basic methods employed to find the distribution of channeled
ions in crystals. One method is too simulate the channeling process with some sort of

Mote Carlo technique. In this process, ions are sent into the crystal at random

positions and their trajectories followed. The yield from impurities in (or on) the

crystal is then obtained readily by simply keeping track of how many ions get

sufficiently close to the proposed position of the impurity to scatter. Another method

is to formulate some sort of analytic expression for the flux distribution in the crystal,

and calculate the overlap of the proposed impurity site with the distribution. The code
used in this thesis work is based on the latter idea. It employs a numerical solution to

the analytic equations describing the flux distribution in three axial channeling
directions (<100>, <110> and <111>) of, in this case, Si.

In this chapter the methods employed in the calculation are outlined, and a

discussion is given of the assumptions made. Also discussed is the method used to

determine the projected adatom sites in the off normal (<110> and <111>) channeling
directions. This treatment is included since it is a crucial step in the interpretation of
the calculated scans in relation to the experimental data, and thus the determination of
the bonding site. Several discussions of the reduction of data using simulations of the
channeling process exist, including those by Gemmell [Gemm74] and Morgan


Calculated Yields
Methods involving the analytic solution of the flux distribution in a channel
usually are based on two simplifying assumptions. First, it is assumed that the rows of
atoms can be approximated by lines of charge of the appropriate linear density.

Second, it is assumed that statistical equilibrium for the ion beam is approached in the
crystal. With the first, the problem becomes two-dimensional. The second allows one
to write an expression for the flux distribution in a given channel. The code used for
this study is based upon these two assumptions. It was developed by B. Bech Nielsen

at Aarhus, Denmark and was acquired from him in April, 1991 [Bech88a, Bech88b].
The code was tested, comments added, and modified [Hoog92] by a visiting student,
J. Hoogenraad, before being used for these studies. Modifications concerning the

calculation of angular scans made by him and made afterwards were largely superficial.
The discussion that follows, unless otherwise stated, concerns this code, which is in


Continuum approximation. As an MeV ion beam enters a crystalline solid the
vast majority of collisions with the atoms making up the solid are gentle, high impact
parameter collisions [Lind65]. If the direction of the beam is close to ( 0.50) that

of a low-index axis of the solid, then, to a given ion, the atoms of the solid appear to

be arranged in a periodic array of rows. These effects combine and lead, for most ions

approaching rows, to a series of correlated impacts such that the collision is
characterized by the gentle interaction of the ion with many atoms, rather than a high
angle scattering event with any single atom. That being the case, the row of atoms can
be thought of, to a good approximation, as a line of charge of the same, smeared linear
density. This approximation, referred to as the continuum approximation, was
introduced by Lindhard [Lind65]. In such treatments, it is convenient (and
conventional) to separate the motion into two components, which leads to the
definition of the transverse energy,

E== E p2 + U(r), (5-1)

for a particle traveling at a position r (string at I r 1=0), moving at an angle 9p with
respect to a string, and having potential energy U(r) (see figure 5-1). As the ion is
channeled, this quantity is, to first order, conserved.

The continuum approximation is, of course, only valid as long as the graininess
of the potential is not felt by the ion beam. This condition was expressed by Lindhard

U"(r )d2 / 8E < 1, (5-2)

which leads, using Lindhard's potential for the strings [Lind65], to a restriction on the
distance of closest approach (r.) for which the continuum approximation is valid,

r. > (2dZZ2e2 / 8E)2. (5-3)

Here, d is the atomic separation on the strings (5.43 A for the <100> axis of Si). For
our channeled beam experiments (2.5 MeV He+), the right hand side of equation 5-3
equation is equal to 0.02 A. At first glance, this seems to exclude substitutional sites.
However the crystal atoms have a mean displacement [Gemm74] from their

equilibrium (string) positions (p) due to thermal vibrations of pi % 0.1 A (room

___ .______




Figure 5-1. Sketch of an ion interacting with a string (row of atoms). The effective
charge of ions He ions in Si at these energies (-2MeV) is +1.94. Due to the gentle nature
of the interaction (inset), the atom row can be approximated by a line of charge (from
Feldman [Fe88, p.39]).

n n


temperature). This insures that, under these conditions, the continuum approximation
is valid for use in the calculation of scattering yields for all impurity positions in (or
on) the crystal.
Statistical equilibrium. The assumption of statistical equilibrium for channeling
processes relies on the two-dimensional nature of the space within which the particles
are confined, and was first made by Lindhard. He pointed out in reference to this
system, [Lind65] that the spatial distribution function, F(E1,r), for a system conserving

energy, is proportional to the an integral over the momentum part of the phase space,

81(E-E,)dp, (5-4)

in the microcanonical ensemble (E is conserved), with

dp = 2xdpjpj = 27iMdE1. (5-5)

The integral in equation 5-4 is then constant, implying that the probability density for
particles of given E1 is uniform within the area that their E1 allows them to access. If
the accessible area is termed A(E), then the distribution function becomes,

r I/A(E,) E>2 U(r)
F(E,,r) = J (5-6)
I 0 E<: U(r).

A given channeled particle is confined therefore to move within the area defined by the
value of its transverse energy and the electrostatic potential in the channel. This
concept is illustrated in figure 5-2, where several trajectories are shown for 2.0 MeV
He+ ions in a <100> channel. These trajectories were calculated within the 25-string
Doyle-Turner continuum potential [Doyl68] (equation 5-7, below). The trajectories
shown are for ions entering the channel at various positions (indicated by the arrows)
and aligned with the axial direction. The accessible areas, which are defined by the


-0.1 -0.10 -0.06 0.00 0.05 0.10 0.156


0 0 0L06 0. 1

-0.16 -0.10 -0.06 0.00 0 .-06 0.10 0.1

Figure 5-2. Ion trajectories projected on the <100> plane. The trajectories shown (dots)
were calculated for a 2.0 MeV He ion moving in a continuum potential for a pathlength
(into the page) of 5000 A. Every 10 A the forces on the ion were calculated and its
trajectory adjusted. The arrows indicate the position of entry of the ion into the channel.

Hatched regions indicate inaccessible areas.

0.10. .

0.00 : '..' :.. .

0.00 ". ': .
o. '- .: .' ... ". -. ..^ :


-0.1 -0.10 -0.0 0.00 0.05 0.10 t .o 1





magnitude of EL (equation 5-1, \/ = 0), are shown for each case as the unhatched

regions in which the particles travel.
Their are several problems with the statistical equilibrium hypothesis, which
need to be addressed. The first concerns the ion distribution at shallow depths in the
crystal (< 1000 A). It is clear that just after entering the crystal, the beam does not
satisfy the requirements of statistical equilibrium, because particles oscillate between
the "walls" of their confining potential barrier. This behavior will be manifested by
oscillations in the scattering yield from the substrate atoms as a function of depth
which show a characteristic dampening at positions deeper in the sample due to the

lack of coherence among the individual ion trajectories. This effect has been well
documented [Barr71, Abel75]. The fact that the oscillations die at thicknesses above
2000 A [Barr71, Abel75], supports the assumption of statistical equilibrium as the

beam exits the crystal for our samples, which are much thicker (- 5000 A).

Another concern is related to the influence of crystallographic planes on the
shapes of angular scans. If an ion is incident at a tilt angle such that its momentum
vector pi is in a direction which coincides approximately with that of a major plane of

the crystal, the motion of the particle will be governed by the plane (planar

channeling), and statistical equilibrium will not be realized. The effects of planar

channeling can be minimized with careful attention to the direction chosen for the

angular scan. For our current experimental setup, this problem has been dealt with, as

discussed in chapter 3, by avoiding major planes while tilting about the axial direction.
Also, angular scans that are grossly misshapen and/or are without the appropriate
shoulder level are not accepted. As an example, figure 5-3 shows several angular

scans across the <100> axial channeling direction. The solid circles represent the

normalized yield from a scan conducted in a random plane. Note the symmetry of the

scan and the correct level at larger tilt angles. Also shown are two unacceptable
















Tilt Angle (Degrees)

Figure 5-3. Angular scans across the <100> axis of Si. The angular scan represented by
the filled circles has the expected shape and shoulders for a scan in a random plane. The
scans represented by the dashed curves are affected by planar channeling and are

-* \ normal scan

- : ,


;' affected


minimum yield
............................. I .......................... I ......

angular scans. For these scans, the yield from the overlayer being studied might be

effected in a similar manner, and could yield inaccurate results. For this reason, no

conclusions were drawn from scans showing large and obvious deviation from normal.

This discrimination, in principle, should be sufficient. In practice, the situation is even

better. We have observed that, even when the host angular scan has been affected by

planar channeling, the surface impurity scans are reproducibly unaffected [Lyma9la].

Presumably, this is a benefit of the thickness of our windows. Over the 5000 A

pathlength the ions travel to reach the exit surface, the planar effects are likely washed

out. Therefore, as long as scans are used which are not drastically affected by planar

channeling, the assumption of statistical equilibrium should be valid in this respect.

A final concern with calculations based on the continuum approximation was

studied by Barrett [Barr71, Barr76]. He found that these calculations underestimate

the yield in the channeling direction due to crystal focusing effects and, also, due to a

low energy tail on the transverse energy distribution function. Both of these effects

are not accounted for with the assumption of statistical equilibrium. These omissions,

for substitutional impurities when the beam is aligned in the channeling direction, can

lead to errors in the calculated yield of a factor of 2 to 3. However, for this worst

case, the error with respect to the random yield is only 2%. For the sites displaced

from substitutional, the effect is diminished, since the cause, for both, is a small

component of the beam with high transverse energy. The contribution to the yield

from well-channeled particles increases rapidly as the impurity is displaced from
substitutional, so the high-EL component has a very small influence on the yield for

displaced impurities. Therefore, the assumption of statistical equilibrium is expected

to introduce a negligible error for the calculation of yields for sites displaced (even

slightly) from substitutional, and a small error for exactly-substitutional sites.

With the assumptions of the model in place, the calculation of the scattering
yield as a function of angle is rather straightforward. The basic idea is to calculate the
potential (from lines of charge) in the channel. From this, and with the definition of
transverse energy, the flux distribution can be found. The scattering yield is then the
overlap of the flux distribution with the impurity location (distributed as a Gaussian in
the channel).
Figure 5-4 displays the potential within each channel calculated within the
continuum model. The potential used is the thermally-averaged 25 string Doyle-
Turner potential, given by,

U(r)={ (8r2aoe2Zl)/d} i=,4 {ai/(bi+(27p)2)} exp{-(2xr)2/(bi+(2Cp)2)}. (5-7)

Here ao is the Bohr radius, and ai and bi are the Doyle-Turner coefficients which
describe 4 Gaussians used to fit atomic potentials. These potentials were calculated by
taking the Fourier transform of electronic scattering factors determined with a
relativistic Hartree-Fock method [Doyl68]. Equation 5-7 then represents the
continuum potential derived from those atomic potentials. Note the presence of p, the
two-dimensional RMS thermal vibrational amplitude of the crystal atoms, in the
potential, which arises due to thermal averaging.
It is this potential in the channel that dictates the accessible areas for ions with
a given EL, allowing a calculation of the ion flux distribution. For random incidence,

the flux distribution is uniform, and with total area of the channel being denoted Ao, is
equal to 1/Ao. Then the probability of a close encounter between ion and impurity (TI

in), normalized to that for random incidence, is given by,

nin(E) = AofFa(E,r)Pi(r)dr. (5-8)

<100> axial potential

150 -




<111> axial potential



S 50

Figure 5-4. Electrostatic potentials calculated in the <100>, <110> and <111> axial

directions of Si with a thermally-averaged Doyle-Turner potential. Units are fraction of a

lattice constant.

150 -


a 50


<110> axial potential

Here Fa is as defined above (equation 5-6), and Pi is the Gaussian distribution of the
impurity on the surface,

Pi(r)=l/(xp2)exp {-(jr-ri|/p) }, (5-9)

while, p is the two-dimensional RMS thermal vibration amplitude of the impurity.
Other distribution functions for the impurity are possible. This matter of choice and
considerations involving the magnitude of p are discussed in chapter 9. Equation 5-8
gives the basis for the calculation of scattering yields for channeled beams.
Up to now the discussion has assumed that transverse energy is conserved. In
practice, however, interactions (multiple scattering) with electrons and interactions
with the thermally vibrating strings lead to gradual changes in the transverse energy
distribution of the ions as the beam penetrates deeper into the crystal. Those changes
have the effect of smearing the flux distribution with depth, and lead to changes in the
scattering yield. This problem of dechanneling was studied by Bonderup et al. Their
theoretical treatment [Bond72], to which the reader is referred for details, serves as

the basis for the inclusion of dechanneling in this calculation. The general treatment
relies on the solution of the diffusion equation, which is written,

a g(E,z)/az = 8/8E1 {D(EL)}a/aE{g(Ei,z)}. (5-10)

The quantity D(E), can be thought of as a position dependent diffusion coefficient,

and g(EL,z) is the transverse energy distribution function (z is thickness). The solution

of this equation, and the treatments for the electronic and nuclear contributions to the
dechanneling are discussed in the references [Bond72; Bech88a], and at length in the
PhD. thesis of B. Bech Nielsen [Bech88b]. Shown in figure 5-5 is a channeling
spectrum for 2.0 MeV He+ incident on Si along the <100> axis (dots). For this
spectrum, the horizontal axis is displayed as depth into the crystal, and the rise in yield

due to the increasing cross section with depth (energy) has been extracted, so the


<100> Axial yield/Randonr yield

Calculated Yield
0 1 5 ............................... ..... ................................... ................... ..........
0 .1 5 .- ....... .......... ......... ..

Typical sample
"3 thickness




....... .... ... ............ ... .......^
0 .0 5 ... .... ....... ...... ....... ..... .... ...... ..., .......... 9............................

S. o *
0 .. *-.. *" .

0 2000 4000 6000 8000 10000

Thickness (Angst.)

Figure 5-5. Minimum yield with depth into a Si sample aligned with the <100> axis.
Shown is a channeling spectrum divided by a random spectrum (dots). This roughly
eliminates the increase in yield due to the dependence of the cross section on energy
(1/E2). Also shown is the increase in yield with thickness due to dechanneling predicted
by the channeling calculation.

increasing yield is solely from an increasing dechanneled component of the ion beam.
Also shown is the scattering yield as a function of depth predicted by the calculation
(solid curve). Here, the contribution to dechanneling from electrons [Bech88b] has
been calculated assuming a uniform electron density equal to that given by distributing
the 4 valence electrons per Si atom evenly in the crystal. The nuclear contribution
[Bech88b] is from Si atoms vibrating with a vibrational Si amplitude of psi=0.14 A. It
is clear that the agreement at depths which correspond to the thickness of a typical
window is much better for the calculation including dechanneling, without which, the
predicted yield would show no increase. With the transverse energy distribution
function, the equation used to calculate the scattering yield for a given beam-tilt

X(E,W,z) = J g(E1,z) lin(E)dE. (5-11)

Then, by inputting the (experimentally determined) thickness of the crystal, the beam
energy, and taking a range of tilts, the calculated angular scans are generated for a
given impurity site. Figure 5-6 shows the calculated scattering yield as a function of
position across three low-index channels in Si for a channeled beam (W=0) of 2.5 MeV
He+ ions. The thickness of the crystal was 5000 A, and the vibrational amplitude of
the impurity was taken as 0.2 A. The yield was calculated over this grid by
X_MINN.FOR using equation 5-11. This figure illustrates well that the scattering
yield is very sensitive to the positions of impurities on the surface.

Crystallographic Projections
Required as input for the calculation of the angular scans are the predicted
positions of the impurities of interest in the given axial channel. Our studies were
conducted on Si crystals in the <100> orientation. However, data have been acquired

across three axial channeling directions: the <100>, <110> and <111>. In each of

<100> x (' =0)




1.5 .




0atom rows

atom rows
<111> x (7p =O0)





atom rows

Figure 5-6. Scattering yield across the <100>, <110>, and <111> axial channels for 2.0
MeV He ions channeled in Si (z = 5000A). The figures represent the scattering yield for
an impurity having p=0.2 A at all positions across the channels. These reproduce the
(slightly smeared) flux distributions. The filled circles represent the atom rows defining
the channels.

<110> X (1k =0)

these cases, the scattering yield from a given adatom site is dictated by the position of
the site projected onto the plane perpendicular to the axial direction. For the <100>

direction (perpendicular to the surface), the adatom site coordinates give the position

in the channel. For the off-normal directions (<110> and <111>), however, these

positions must be calculated from the adatom coordinates on the surface. Another
consideration concerns the steps on the surface (typical step densities are 20

steps/micron [Mo90]). Steps on the Si(100) surface separate local crystallographic

domains which appear rotated by 900 [Mo89, Hame86], as is illustrated in figure 5-7.

Here, an STM image [Hame86] of the Si(100) surface is shown. The dimer rows are
visible as the lines running across each terrace, and the elongated bumps making up the

dimer rows are interpreted as representing the individual dimers. Only upon

descending four steps is a position in the crystal equivalent to the original obtained.
This being the case, projections from the <100> adatom surface site must be

considered from each of the possible steps. Because a typical beam spot encompasses
tens of thousands of steps, for nominally flat surfaces, one would expect an equal
contribution from each step. Therefore, each site on the surface yields, in the <110>

and <111> directions, 4 projected sites. Depending on the surface site, some

projected sites will be redundant. However, all are considered such that the yield at a
given W is given by,

Y(y) = Ei=,N (w, yi)/W (5-12)


W = i=lN wi. (5-13)

Here the wi are the weights of the individual sites. Figure 5-8 shows projected
positions for an adatom in a dimer-like site for the three axial directions generated via

a screen dump with the program PROJECT.FOR [Gran92a]. Depicted are the

Figure 5-7. STM images of a clean Si(100) surface (from Hamers [Hame86]). Five steps
are clearly visible. Dimer strings, appearing as lines on the terraces, are either parallel or
perpendicular to the step edges.

Figure 5-8. Projected adatom positions in three axial channels. These plots were
generated via screen dumps from the FORTRAN program PROJECT.FOR. Shown are
projections into the <100>, <110> and <111> channels for a dimer site (only one dimer
atom is projected).

adatom sites (filled circles) on a grid of crystal sites (open circles) defining the

channels. The small closed circle on each represents the origin defined for that

direction. In PROJECT.FOR, the adatoms sites are projected by defining coordinates

on the <100> surface and calculating the positions of the point in the off-normal

directions. For the present case, all the sites shown would then be weighted equally.

The equations governing the transformations were found readily using three-

dimensional rotational matrices. For these, x00o, Yloo, and z1oo are the coordinates of

the site of interest on the surface. Then, if x11o,y11 and x,11,yl 1 are the coordinates of

the projected points, these are given by [Corb50, p 140]:

/ c;r cos6 s;n si;n cosi cos s si6\ / 100

Scos1',cos6 sin, coscose cos sin i

s n sin -sin cosy cose 6 \n 1 00s

The coordinate systems are defined in figure 5-8, and p, 0 and W are the

Euler angles. As an example, table 5-1 gives values for the angles for rotation into the

<110> direction, and the corresponding x and y coordinates in that plane.

Table 5-1. Coordinate transformation for <110> axial direction*.

# Pt Ott Wt X110 Y11o

1 45 -45 90 /2(x+y)-z/(2) (y-x)/(2)
2 -45 -45 -90 '/(y-x)+z/(2)2 -(x+y)/(2)/2
3 135 -45 -90 2(x-y)+z/(2)/2 (x+y)/(2)/2
4 -135 -45 90 -'/2(x+y)-z/(2)'/2 (x-y)/(2)/2

t Angles in degrees.
x, y and z are on the surface, i.e., xlo0, yoo and z0io.


Experimental Angular Scan Extraction

In ion scattering, the data come in the form of energy spectra (see figures 3-1

and 3-3). Each spectrum contains a wealth of information about the sample and

distribution in energy of the scattered ions. For site determinations, however, it is

convenient to have the data in the form of angular scans. In order to display the data

in this manner, the scattering yield for a given region in energy at each tilt angle must

be extracted from each spectrum. For analysis of an overlayer, this involves

determining the integrated sum of counts in the impurity peak of interest. As an

example, figure 6-1.a shows spectra in the aligned and random geometry for an

Sb/Si(100) sample. For data such as these, the extraction is rather straightforward.

One simply has to subtract the approximately linear background from the counts in the

region of the spectrum containing the peak. In practice, the process is slightly

complicated by the fact that the trajectory dependence of the energy loss of ions

traveling in the crystal causes channeled particles to lose less energy than particles

traveling in a random direction through the crystal [Appl67; Mae73; Jin86; Dygo88].

As a consequence, the Sb peak moves on the energy axis when the beam is changed

from random to channeled incidence. This shift is seen clearly in figure 6-1.b, which

shows an expanded view of part of the Sb/Si(100) spectra. However, if one chooses

the region to be summed as the foreground such that this entire range is included (the

peak region), and chooses the regions by which to determine the background to flank

the peak region, an accurate sum can be extracted easily, assuming linear background.

A typical choice of windows for this method of data extraction is indicated on figure













0o /, ----*

0 -front

0 -
'0 Si aligned

100 200 300 400 500 600 700 800
Channel Number


650 700 750 800 850


Channel Number

Figure 6-1. Channeling and random spectra (<100> axis) from an Sb/Si(100) sample.
Dashed curves are for random incidence. Solid curves are for aligned incidence, a)
Expanded view; b) Region of spectra showing Sb peaks. Indicated are the windows
chosen for background subtraction. The total number of channels in the two background
windows is equal to the number in the peak window.


Si random



Another method of data extraction involves fitting. While fitting is not

necessary for spectra such as that in figure 6-1, this method generally lends itself more

easily to automation, and it becomes necessary when peaks from two or more

elements are closely located in energy. This technique is especially preferred (and

necessary) when the peaks overlap. Another concern is that, in general, the lineshapes

of peaks from ML films are Gaussian only for random beam incidence. For channeled

beams, the shapes are complicated [Dygo94]. These difficulties are illustrated in

figure 6-2, where an expanded region of aligned and random incidence spectra from a

sample having on it Ge (0.8 ML) and Sb (1 ML) is shown. The data are represented

by dots. Also shown, as the solid curves, are fits to the data used to extract peak

sums. In order to circumvent the problems involving peak shifts and non-Gaussian

peaks, numerous Gaussians are fit to each peak. The curves shown represent the sum
of 10 Gaussians (5 each for the Ge and Sb peaks) and a line.

For the fitting procedure, the widths of all Gaussians were fixed at a value

determined from a separate fit (of one Gaussian per peak) to a random-incidence

spectrum, and the positions of the Gaussians were fixed at values determined as

follows. The positions of the Gaussians at the extremes of high and low energy for

each element were determined from channeling spectra and shoulder spectra

respectively. This determination proved simple for the Sb peak, which was well

defined for both cases. However, for the Ge peak, which was not well defined in the

channeling direction, the logical assumption was made that the peak shifted, at a

maximum, as much as the Sb peak shifted (i.e., the spacing between the extreme Ge

peaks was set equal to that of the Sb peaks). Then, the remaining 3 Gaussians for

each peak were spaced evenly between these. For determining the low energy bound,

the shoulder spectra were used rather than the random ones, since particles traveling

preferentially near the rows of atoms (as is the case for shoulder data) have, on








1 5 0 <.. ._-] I u >
125 -

100 L-

75 (0.3 p,) detector
25 Ge

2.10 2.15 2.20 2.25 2.30 2.35 2.40

Energy (MeV)

Figure 6-2. Portion of spectra for Sb/Ge/Si sample. Shown are the data (dots) and the
fitted curve used to extract the peak sums. a) Random spectrum; b) Aligned spectrum.
In the inset is the scattering geometry.

average, a slightly higher energy loss than those with random incidence. The heights

of the Gaussians then were varied, along with the slope and intercept of the line for the


A FORTRAN program has been written in order to facilitate the fitting

process. It displays the fits and data graphically for evaluation and allows for the

automatic fitting of multiple spectra. This program, DATAFIT.FOR [Gran93a], uses

a Levenberg-Marquardt method, which constitutes a combination of the inverse-

Hessian and steepest descent methods, and has become common for nonlinear least-
squares fitting [Pres92]. In practice, coefficients for the Gaussians are passed to a

subroutine MRQMIN.FOR [Pres92], along with a list of coefficients to be adjusted

(parameters) after compilation in DATAFIT.FOR. The parameters are adjusted until

the value of chi-squared does not change appreciably for several iterations, after which

the fit is assumed converged, and the data and fitted curve are printed to the computer

terminal for visual evaluation. Represented in figure 6-3 is the fit, as viewed on the

screen, of a spectrum obtained by summing all spectra from a <100> angular scan.

The figure shows the data (dots), the sum of the ten Gaussians and a line (upper

curve), and the individual Gaussians (lower solid curves). Note that, due to the energy

shift with tilt angle, the widths of the Ge and Sb peaks are greater than those obtained

at random-incidence (shown in figure 6-3, and equal in width to the individual

Gaussians, lower curves). Factors which contribute to the width of random-incidence

peaks were discussed in chapter 4 in relation to the Ge thin windows. The widths of
aligned-incidence peaks are influenced by trajectory-dependent energy loss

It has been found that the fitting procedure just described is robust to small

changes in the width of the Gaussians, or even the total number of Gaussians.

However, in order to minimize errors resulting from the fitting, the data were


Chisq= 212.995800


/ 1. / / /
I ,

Figure 6-3. Screen dump from DATAFIT.FOR.


---C~z i

extracted with as little reliance on accurate representations of the lineshapes as

possible. This was done, for the case of Sb/Ge/Si, by subtracting the experimental
counts from the fitted background line, in order to arrive at the peak areas, rather than
integrating the fitted Gaussians to obtain these areas. This necessarily means that one

has to determine which counts belong in which peaks, a task which proves simple
except in the region of overlap. In that region, the fit for the largest and most well-

defined peak, the Sb peak, was integrated, and the integrated value was added to the

total Sb counts outside the region of overlap. The rest of the counts in the interface
region then were added to the Ge counts. As a check, the BS detector (1500)
provides a signal which has Ge and Sb well separated in energy. While the Ge signal
suffers from being in a region of high background (this is one reason the 780 detector

is used instead), the Sb gives a strong, clean signal. The Sb data therefore are easily
extracted by setting windows on the BS spectra, as described above. With this

procedure, the Sb angular scans obtained from the fitting technique described above
can be compared to those obtained directly from the BS detector. Figure 6-4 shows a
normalized <110> Sb angular scan obtained by fitting the Sb/Ge/Si(100) data from the

780 detector (solid circles) along with the normalized scan from the BS detector, taken
simultaneously. These two show good agreement, suggesting further that the fitting

procedure is accurate. This figure also illustrates the difference in the sizes of the

experimental uncertainties from the two detectors, and hence shows another advantage

of the transmission geometry.
In the studies presented here, both methods of data extraction have been used.
For the Sb/Si(100) investigations (chapter 8), the peak sums for the Sb were extracted

by setting windows and using a FORTRAN program, CUT.FOR [Lyma90]. For the
Sb/Ge/Si and Ge/Si studies, peak sums for the Ge and Sb were extracted by fitting the

data with DATAFIT.FOR. For the Ge/Si data (and other test cases), sums were







.4 o BS windows

0.2 i 78 fit

0.0 I I
-0.8 -0.4 0.0 0.4 0.8 1.2

Tilt Angle (Degrees)

Figure 6-4. Comparison of angular scans generated with peak fits and peak sums. The
filled circles were generated using DATAFIT.FOR on data from the 780 detector. The
open circles were generated using CUT.FOR on data from the 1500 detector. The solid
line is to guide the eye. Error bars represent uncertainties due to counting statistics.

checked against those given using CUT.FOR, and in all cases agreement was found.

Our data were then corrected for a small change in the scattering cross section across

an angular scan due to the difference in energy [Dygo94] loss of channeled and
random ions. This task was accomplished by correcting the yield from the channeled

fraction of the beam at a given tilt angle. In order to understand this step, imagine the

yield from the overlayer (y) to be composed of two components, a channeled and
random, such that,

Y = YR + Yc. (6-1)
Here, yc = y (1-Xsi), where Xsi is the normalized yield from the Si host for a given

datum (this gives the random component of the beam). The channeled component (y -

yR) then is corrected by a factor equal to the square of the channeled beam energy
divided by the random beam energy. This factor, for 2.5 MeV He+ incident on a 5000

A crystal, amounts (at maximum) to a 3.2% correction to ye in the <100> channel, a

6.3% correction in the <110> channel, and 5.2% in the <111> channel.
Comparison of Experimental and Calculated Angular Scans

As previously mentioned, for a quantitative determination of an adatom site,

the computer-calculated scans (chapter 5) are compared to the experimental scans. In

this section, the process by which this comparison was accomplished for these studies

will be explained. The data used to illustrate the procedure are from Ge on Si(100)

(0.6 ML) and Sb data from the Sb/Ge/Si sample. The same basic procedure was

applied to both data sets. For the earlier Sb/Si data, a simplified procedure was used,

since the dimers were reported as symmetric in the literature [Richt90] before our

experiments began. The general process is detailed here because a considerable effort

was expended in order to develop a procedure which was accurate, inclusive of all
reasonable possibilities, and unbiased.

1.8 '
1 4
1 2

atom rows

Figure 6-5. Calculated scattering yield across the <100> channel in Si for a 2.0 MeV He
beam with V=0.150. The conditions for the calculation were the same as for figure 5-6
(except for the tilt).

<100> X (P =0.15)

From figure 5-6, it is clear that the position of the adatom on the surface (in

the channel) will have a great influence on the scattering yield from that adatom.
Figure 6-5 illustrates how this yield depends upon the tilt angle, y, across the <100>

channel, given the same conditions as those of figure 5-6. The scattering yield is
directly related to the flux distribution (see chapter 5) in the channel, which is clearly a

strong function of W. Herein lies the utility of the angular scans. These concepts are
further illustrated in figure 6-6, which shows calculated scans across the <100> axial
channel. These scans are for a single adatom on the <100> surface, for a crystal
thickness of 5000 A, and for a range of displacements in x from substitutional (in steps
of 0.1 A). Scans across this direction are very sensitive to such displacements, which
for purposes of analyzing dimerized surfaces, are related to changes in the surface-

projected dimer bond length. Note, however, that this axial direction (<100>) is
parallel to z, making scans across this channel completely insensitive to the degree of
relaxation of the surface. Sensitivity to z displacements can be accomplished in the

off-normal directions (<110> and <111>), which observe both x and z displacements.

This aspect is illustrated in figure 6-7, which shows calculated angular scans across the
<111> axial direction from sites with a fixed x-position (x=1.49 A), but with variable z

positions (again in steps of 0.1 A). Thus, by combining the information from the three
directions, the adatom site can be pinpointed.
Basically, pinpointing the adatom configuration becomes the problem of

finding the coordinates in a multiple dimensional space (parameter space) that give the
calculated scan demonstrating the best agreement with the data. The parameters are
the positions and vibrational amplitudes of the impurities. In general, for a site
determination, the adatoms are assumed to lie in a perfectly formed overlayer, with
each adatom having a well-defined local structure and well-defined local bonding

geometry (see discussion in chapter 11). For our studies of dimerized adlayers, two









0 .0 'I I I I 1
-1.2 -0.8 -0.4 0.0 0.4 0.8 1.2

Tilt Angle (Degrees)

Figure 6-6. Variation in calculated angular scans with x. Shown are calculated scans
across the <100> channel for a symmetric dimer with variable dimer bond length. The
calculation was for 2.0 MeV He ions and for a sample thickness of 5000 A.

1.6 I I i I
<111> angular scans
1.4 (z steps in 0.1 A)
(y=1.92 A)
1.2 (x=1.49 1) z=1.0 A

a 1.0

0 0.8 -

O 0.6 -
Z 0.4

0.2 =0

0.0 1
-1.2 -0.8 -0.4 0.0 0.4 0.8 1.2

Tilt Angle (Degrees)

Figure 6-7. Variation in calculated angular scans with z. Shown are calculated scans
across the <111> channel for a symmetric dimer with a fixed dimer bond length and with a
variable height above the <100> surface. The calculation was for 2.0 MeV He ions and
for a sample thickness of 5000 A.

adatom sites were allowed for each atomic species. Two were allowed because of the

possibility of buckling in the dimers, which leads to two distinct adatom sites. For the

systems we studied, the dimers were not expected to twist out of the plane defined by
the surface normal (z) and the gross direction of dimerization (x), so the y positions

were fixed. The result is a total of 4 positional degrees of freedom to define the dimer
structures. Another potential degree of freedom was p. In practice, when a believable

value was available in the literature for this quantity, or when the value could not be

determined readily from the data, p was not used as a fitting parameter. For instance,

for the Ge/Si data, a calculated [Aler89] value was used (see chapter 9) for the Ge

adatoms. The same is true for the Ge in the Sb/Ge/Si system (see chapter 10),
whereas the Sb vibrational amplitude was varied in all studies. Therefore, in general,

there were five parameters used in the fitting of the data (the vibrational amplitudes for

two dimer atoms were assumed to be equal). The values for the best fit for these were

determined by calculating the x2 goodness-of-fit indicator for each adatom

configuration, considering the data in three crystallographic directions (<100>, <110>

and <111>), and finding the configuration which gave the lowest value. Each

direction was weighted equally. The x2 minimization was carried out over a grid

which spanned all reasonable values of the fitting parameters, and the parameters were

varied independently.

In order to facilitate visualization of the fitting procedure, the regions in

parameter space giving adequate fits were identified by representing the fitting

procedure graphically. Gross estimates of some of the best-fit parameters could be

obtained by observing the behavior of X2 and considering a experimental angular scan

about one of the axial directions. Figure 6-8 shows X2 as a function of two parameters
for the Ge/Si data from a comparison of calculated scans in the <100> direction with

the <100> experimental data. The parameters shown are the x-positions for



xl/a 0.2 0.1

0.3 0.3 x2/a

Figure 6-8. Three dimensional representation of fitting procedure. Depicted is the quality
of the fits (as measured by x2) as a function of the x positions of the two dimer atoms
(asymmetric dimers). The shaded regions represent those regions of parameter space
giving good fits. Units are fraction of a lattice constant.

calculations compared with the Sb data from the Sb/Ge/Si system. In this figure, the

shaded regions represent those values of the two x-coordinates which give fits having

a X2 below a reasonable cut-off, or, in other words, those configurations giving good
fits. By creating similar plots for all three directions, and all parameters, it is assured

that no reasonable geometry is overlooked, and the volume in parameter space

considered can be reduced substantially.

In general, it is not sufficient to consider the three axial directions

independently for a final determination of the bonding geometry. However, when the

x2 is evaluated considering all channeling directions, the possibility of a simple graphic
representation is lost, since the dimensionallity of the space is increased. A procedure

was developed to circumvent this inconvenience. The parameters were used to

calculate quantities of physical significance and these were used to display the data

graphically. For instance, from the x and z positions of the two adatoms making up

the dimer, the dimer bond length and dimer tilt, two quantities of interest for this

study, were calculated for each set of adatom sites giving a good fit. This intermediate

comparison then allowed an inspection in two dimensions of the configurations which

were consistent with the data, inclusive of all parameters. Also, the uncertainties

associated with the determination of these quantities then were determined readily by

measuring the volume in parameter space encompassing the good fits. An example of

such an analysis is shown in figure 6-9. Plotted is the dimer tilt angle versus dimer

bond length represented by the calculated scans that were consistent with the

experimental angular scans across all three axial directions for the Ge/Si system. Since

the parameters giving good fits tend to cluster, it is reasonable to assume that the

correct structure, within error bars, has been discovered. In practice this procedure is

facilitated by sorting the configurations by X2 value, a task which is accomplished

easily with a simple FORTRAN sorting routine (XTRACT.FOR).


0 r


2.2 2.4 2.6 2.8 3.0
Dimer Bond Length (A)


Figure 6-9. Representation of fits using physical quantities. Shown as the filled circles are
the dimer configurations giving acceptable fits to the Ge experimental angular scans. The
dimer bond length and dimer tilt were calculated from the x and z adatom coordinates.







Configurations giving
good fits


I .

" -

The importance of Si in today's society cannot easily be overstated. The

semiconductor-device industry relies almost exclusively on structures grown on the
Si(100) surface [Shur90] for the production of the integrated circuits which power

computers and electronic devices, which, in turn, drive the progress and productivity

of our modern world. For this reason, and because of an interest in surfaces on a
fundamental level, there has been a tremendous effort in the surface science

community to elucidate the structure and dynamics of the Si(100) surface. This effort

began in the 1950's when Schlier and Farnsworth [Schl59] observed with LEED that

the Si(100) surface symmetry could not be explained by a simple bulk termination of

the crystal. This finding led to the conclusion that the surface atoms moved from their

bulk positions [Schl59]. Later studies [East80; M6nc79] supported this early work,

and a picture of the surface as composed of dimers began to emerge. This

reconstruction was eventually understood as a consequence of the high energy cost of

the two dangling bonds per surface atom caused by the cleaving of the crystal at the
<100> plane. By forming bonds in the surface layer the number of dangling bonds is

reduced, and, thus, the surface free energy is reduced.
As more was learned, it was suspected, primarily on the basis of works such as

a medium energy ion scattering (MEIS) study by Tromp [Trom81a; Trom8lb;

Trom83] and calculations such as those by Chadi and others [Ch79; Verw80; Yin81],

that the surface atoms participated in what were described as "asymmetric dimers".
By this it was meant that the dimer bond was tilted with respect to the surface plane.

It was thought that this tilting was a consequence of a partially covalent dimer bond

[Ch79]. This interpretation was also consistent with core-level photoemission data

[Himp79; Hil80; Uh81; HoofO8], in which surface shifted components altering the

photoemission lineshapes suggested a charge transfer between the two atoms
participating in the dimer bond. Symmetric and asymmetric dimers on the <100>
surface are shown schematically in figure 7-1 as viewed along the <011> direction (a)
and from the top (b). See, also, figure 5-7.

The interpretation of the data up to the era of the STM was becoming

consistent. Then, in the mid 1980's STM images (see figure 5-7) [Trom85; Hame86]

of clean Si(100) surfaces showed symmetric-looking dimers, a finding which inspired a
whole bevy of experiments [Jaya93; Uh92; Land92] and calculations [Redo82;

Robe90; Tang92; Dabr92, Stil92] aimed at determining the symmetry of the dimers.

This enthusiasm was fueled primarily by the calculations, most of which showed only a
slight energy difference between asymmetric and symmetric dimers. There have been

numerous interpretations of the available data, leading to a controversy which is yet to

be resolved. Recent work has helped, however. STM images [Wolk92] obtained at

lower temperatures (120K) show asymmetric dimers in greater abundance than for the

room temperature surfaces. This excess has led to the acceptance of a picture (first

proposed by Hamers et al. [Trom85]), which has the dimers flipping at room

temperature [Dabr92, Weak90] between two degenerate states (defined by the two

possible tilting configurations) such that the STM observes an averaged, symmetric

looking dimer, with asymmetric dimers stabilized only at defects or at lower

temperatures. The argument is physical, since the time scale for the passing of the

STM tip over a dimer is such that a dimer will flip -109 times during the measurement

[Dabr92]. This interpretation seemed be consistent with all the available data.

However, until recently, no direct observation of the dimer symmetry had been




side view

b) top


Figure 7-1. Geometric representation of symmetric and asymmetric dimers on the Si(100)
surface, a) Side view (along <011>); b) Top view (along <-100>).

f /

Aft Adak



Over about the past three years, there has been a separate but related

discussion concerning the origin of shifted components in core level spectra taken
from clean Si surfaces and Si(100) surfaces covered with Ge, Sb, or Sb/Ge.

Photoemission has proven an extremely useful tool for clarifying issues on the Si(100)

surface. On the clean surfaces, two shifted components were observed (labeled S and

S') originally [Himp79; Himp80; Wert91]. These were attributed to the up (S) and
down (S') dimer atoms, and, as discussed above, were thought to arise from charge
transfer between the two (therefore, a partially ionic bond). Interest continued on the

problem, and a similar study of Ge on Si(100) provided some intriguing data [Lin91].

These data showed that the S' component from Si lingered until 2 ML of Ge

coverage, after the S component vanished at -1 ML. This behavior indicated that the

two components were not both from the dimer atoms. Additional evidence supporting
this interpretation came from the Ge core-level spectra, which showed the S'

component only after greater than 1 ML of Ge growth. Since Ge at this coverage
should continue to grow as would Si, rendering the Si below bulklike [Lin91], these

data led to the conclusion that the S component was from the top layer (with covalent

bonds), and the S' component from the second layer down. These authors turned out

to be half right.

A recent study [Land92] with excellent resolution on the Si(100) surface has

shown additional components (obscured by the bulk signal), and has identified the S

peaks with the up dimer atom and the S' peak with the second layer atoms. This study
is also supported by a nice work with Sb, Ge, and Si [Cao92]. The latter work

showed that both Si and Ge exhibit the S' peak when directly under a reconstructed

layer, be it Sb, Ge, or Si. The S component, however, was observed only from

reconstructed Si or Ge surfaces. Further, the S' component vanished from buried

layers (for instance Si buried under thick Ge layers). The consensus then seems to be


that the S shifted component comes from the up dimer atom, and is a result of a

partially ionic dimer bond. The physical origin of the S' component, coming from the

second layer, however, is unclear.


Considerable attention has been focused in recent years on the growth of

group V elements on Si. These studies have been motivated partly by the desire to
improve the quality of III-V epitaxy on Si, and by an interest in achieving abrupt

doping profiles. Much progress has been made in clarifying the growth of As and Sb

on Si(100) at temperatures sufficiently high that ordered structures are formed on the

surface[Richt90; Trom92a; Trom92b; Beck88]. It has been shown by STM [K6hl92;
Beck88] that As terminates the Si(100) surface with symmetric-looking (see

discussion in chapter 7 of STM images of dimer structures) As-As dimers. On the

terraces, As-As dimers form a virtually defect-free overlayer and passivate the Si

surface at one ML [K6h192; Beck88; Uh86]. The low defect density is usually

attributed to the small lattice mismatch between Si and As (-1%) and the convenient

coordination, which allows each As atom (5 valence electrons) to satisfy its valence

requirements by bonding to three atoms (two Si and one As). In this configuration, all

dangling bonds on the surface are filled. By comparison of angle resolved

photoemission data and ab initio pseudopotential calculations, Uhrberg et al. [Uh86]

determined the structure of the As dimers.

Similar progress has been made on the Sb deposited Si(100) surface. This

surface would be expected to behave somewhat like the As/Si case, since As and Sb

are both group V elements. In the Sb case, however, the lattice mismatch with the

substrate is larger by -12%. Indeed, for the Sb/Si(100) surfaces grown at elevated

temperatures and annealed at 5500C, the surface as viewed by STM [Richt90;

Nogami91] consists of short strings of symmetric-looking Sb-Sb dimers, interrupted

by numerous antiphase domain boundaries. For these surfaces, photoemission
[Cao92] and ion scattering data [Slij92] suggest that the underlying Si reconstruction
has been lifted by the Sb adlayer. Finally, the structure of the annealed surface has
been determined using SEXAFS [Richt90].
Considerably less progress has been made in our understanding of the

geometry of these interfaces at low processing temperatures, where adatoms

impinging on the surface have insufficient mobility to form an ordered overlayer. The
absence of long range order, however, does not imply the absence of a preferred
bonding site or the lack of well-defined structural features that are possible precursors
to the formation of an ordered epitaxial overlayer. One major barrier to progress has
been that in the presence of considerable disorder, properties such as bonding
geometry are inaccessible to many of the standard surface probes. However, if the
long range goals of a theoretical understanding of and control of interface formation

are to be achieved, it is of utmost importance to probe the microscopic processes that
precede epitaxy.
The -1ML low temperature surface has been investigated by a number of

techniques, such as: core level photoemission [Rich89a; Rich89b], high energy

electron diffraction (HEED) [Rich89a; Rich89b], STM [Richt90, Noga91, Rich89a],

SEXAFS [Richt90; Richt91], medium energy ion scattering (MEIS) [Slij92] and

LEED [Richt90; Richt91; Noga91; Cric93]. Although there is not a true microscopic

characterization of the Sb overlayer among these, some of this work hints at the
structure of the surface. The MEIS study finds that the Sb deposited at room
temperature forms a 2-D film at first, with 3-D Sb islands at higher (>1ML) coverage
[Slij92]. They attribute the formation of the 2-D overlayer to the energy gained by
chemisorption. This is based on their determination of the Sb binding energy on the

Si(100) surface, Eb = 2.6 eV/dimer. They point out that, since the dissociation energy

of Sb tetramers in the gas phase is 1.2 eV/dimer (the deposited species is Sb4), there

exists an incentive of 1.4 eV/dimer for dissociative chemisorption. They also conclude

that upon -1 ML Sb deposition the Si dimers on the surface are broken, and the Si

atoms become bulklike. These authors, however were not able to comment on the

bonding geometry of the Sb overlayer. Consistent results were found by the group of

Rich et al. These authors find that for Sb deposition at 320-3700C, the top-layer Si

becomes bulldike [Rich89b] (although in similar paper they conclude the Si

reconstruction persists [Rich89a]). Again, however the Sb layer is described only as

disordered, and no conclusions are drawn concerning the atomic bonding geometry.

In still another work, it also has been suggested that the Sb forms a disordered

overlayer with three-dimensional Sb clusters upon low temperature deposition


We have used transmission ion channeling to probe the bonding geometry of

the Sb terminated Si(100)-2x1 surface formed at temperatures below 375C. Some

samples were studied for Sb deposition as low as -40C, with ion scattering conducted

at room temperature. In all of these experiments, we have exploited a fundamental

advantage of transmission channeling: It probes the registry of adatoms with respect

to the underlying substrate directly, and, therefore, does not depend on long range

order in the overlayer.


After the cleaning and characterization by LEED of the Si surface, Sb was

deposited from a boron nitride effusion cell. Sb was deposited at a number of

temperatures, however the two most common were 3750C and room temperature. For

Sb depositions at 3750C the surfaces were exposed to several monolayers in order to

insure saturation of the surface with Sb. For the room temperature Sb deposition,

rates were near 0.05 ML/min. After Sb deposition, the samples either were not

annealed, or annealed at 3750C or 5500C for 15 min. All samples having been exposed

to 1 or more ML's had ~1 ML Sb coverage after annealing at 3750C, which is
consistent with thermal desorption studies placing a saturation coverage of Sb at

elevated temperatures (>150*C) of- 1 ML [Barn86]. LEED patterns were noted for
all samples and were (lxl) with weak, diffuse 1/2 order spots for samples having -1
ML of Sb, independent of deposition temperature or annealing. With one sample, the
experimental stage was cooled by thermal contact via a copper braid to an in situ
liquid nitrogen reservoir and Sb deposition occurred at -400C. Our observed LEED

patterns are consistent with the vast majority of LEED results reported by other
groups for the Sb/Si(100) surface [Richt90; Richt91; Noga91].
All coverages were determined by RBS. A beam of 2.5 MeV He+ ions,
collimated to < 0.08* angular divergence, was directed onto each sample after it was
transferred to the scattering chamber. Possible effects of ion beam irradiation on the

sample were monitored carefully. A small loss of Sb during a typical angular scan was

observed due to sputtering of the overlayer by the ion beam. In a separate dosing run,

the Sb loss per unit ion-dose was measured for ions incident in a random direction.

Shown in figure 8-1 is the Sb yield versus dose used to determine the loss rate. The
data are shown as the filled circles, and the dashed line is a linear regression, with the
coefficients indicated in the inset. The Si yield is included, which, as expected, shows

no dependence on dose. The computer calculations used to fit the data have been
corrected to reflect this loss, and the slope in the scans with tilt angle is indicative of
this (see figure 8-2 and 8-3). We have no evidence that ion irradiation alters the site of

the Sb. To the contrary, channeling minimum yields have been measured both before
and after angular scans were completed on a particular beam spot, and have shown no

significant difference.




<100>-random Sb yield
o Si yield

-O- o -o- --~O-q ~ 0 -o0-0- 0 -0-0- 0 -o-0- 0

0.6 -

0.4 -

0.2 -




Dose (/uC)

Figure 8-1. Sb loss rate (random incidence). The filled circles represent scattering from
the Sb overlayer. The open circles represent scattering for the Si substrate. The solid line
is a linear regression used to characterize the Sb loss rate due to sputtering by the ion

Sb yield=-(0.00036)dose + 1.00

Results and Discussion
Annealed Surfaces
Figure 8-2 shows angular scans across three low-index crystallographic

directions for Si(100) samples on which Sb was deposited at 3750C, followed by an
anneal at 5500 C. Coverages for the three samples are 0.9 0.14ML. In this figure
the hollow (filled) squares represent the normalized ion scattering yield at a particular
tilt angle (w) from the Si substrate (Sb overlayer). The solid lines represent calculated
angular scans for the Si host (lower curve) and the adsorbed Sb (upper curve). The
decrease in the experimental and calculated angular scans for the Sb from left to right
is due to the Sb loss with dosing discussed earlier. Three parameters were varied in
the calculated scans to determine the best fit to the Sb data; the lateral displacement in
the (100) plane (related to the dimer bond length) corresponding to modified bridge
site positions, the position of the Sb adatoms perpendicular to the (100) surface
(related to the relaxation), and the thermal vibrational amplitude, pSb, of the adatoms.
Because it is difficult to distinguish a static displacement along the <100> direction
from a change in psb, we fixed Psb (taken to be isotropic) and the lateral displacement
in the (100) plane first using the (100) angular scan, which is not sensitive to static
<100> displacements. The (110) and (111) scans then were fit by adjusting the

position of the Sb perpendicular to the (100) plane.
From fig. 8-2, the calculated scans are seen to be in good agreement with the
experimental data in all three channeling directions. These calculated scans represent a
modified bridge site, with an Sb-Sb bond length of 2.8 0.1 A, within experimental
uncertainty of the bulk Sb bond length (2.90 A), and the value found by Richter et al.
using SEXAFS [Richt90] (2.88 A). Assuming an Sb-Si bond length of 2.63 A (equal
to the sum of the covalent radii of Si and Sb), and the measured Sb-Sb dimer length of

2.8 A, the calculated position of the Sb dimers perpendicular to the (100) plane