Characterization and Control of the Surface of

the Topological Insulator Bi2Se3

by

Avery James Green

A Dissertation

Submitted to the State University of New York

in Partial Fulfillment of

the Requirements for the Degree of

Doctor of Philosophy

SUNY Polytechnic Institute Colleges

of Nanoscale Science and Engineering

2017
Acknowledgments

Any success I’ve achieved in graduate school, and in the field of nanoscience research, must

be attributed to the support of my advisor, Prof. Alain C. Diebold. His guidance and attention gave

me the direction and motivation I needed when I would otherwise be lost in the complex and

difficult field of topological insulators, and the combination of scientific rigor and kindness he

constantly displayed make him all that one could hope for in an academic role model. He helped

me grow from a student of physics to a researcher capable of implementing various laboratory and

analytic skills toward solving a difficult problem. This work, and all of my other efforts in graduate

school, would have been impossible without him, and I am vastly grateful for his mentorship.

I would also like to thank my committee, and all of the other members of the scientific

community who have helped me along the way. Prof. Carl Ventrice (CNSE), Prof. Vince LaBella

(CNSE), Prof. Hassaram Bakhru (CNSE), and Prof. Robert Hull (RPI) all provided essential

feedback on my thesis work, and were instrumental in helping me produce my best possible

dissertation. The samples and discussion provided by Prof. Chris Hinkle (UT-Dallas) and Dr. Lee

Walsh (formerly UT-Dallas) allowed me to address a problem in the topological insulator field

that I would have otherwise had no access to. Dr. George Orji of NIST was a highly valuable

collaborator, enabling me (and all AFM users at CNSE) to make sub-nm topographical

measurements with precision. I would also like to thank Prof. Ken Burch (Boston College) and

Prof. Bradley Thiel (CNSE) for discussions of topological insulators and general materials science

techniques. I must also acknowledge Dr. Brendan O’Brien for his assistance and teaching in TEM,

and Westley Nolting for his STM characterization efforts.

The current and prior members of the Diebold lab group – Dr. Raja Muthinti, Dr. Manasa

Medikonda, Dr. Sam O’Mullane, Dr. Dhairya Dixit, and Dr. Florence Nelson – were both peers

ii
and counselors. They taught me the metrological techniques I needed, ensured that I was never

alone in research, and showed me the path of a successful graduate student.

Finally, I would like to acknowledge my family and friends. I am fortunate be surrounded by

people who are both warmhearted and brilliant. It’s only with the comfort and encouragement they

provided that I had success in Albany after living in California for 22 years. My friends helped me

feel at home away from home, and the solace and inspiration their companionship provided cannot

be overstated. My mother, Donna Alkadri, my father, Richard Green, and my brother, Emerson

Green, have always supported me in my decisions and pursuits. As the youngest of the family, I

have always had three stellar examples to aspire to in life and in work, and have had access to a

wealth of advice and unconditional love. This journey would have been impossible without them.

iii
Abstract

The field of topological insulator (TI) materials is new. The ideal TI contains surface states in

helical Dirac cones that can be used for spintronics or interconnect applications. Of the TI class,

Bi2Se3 is the most promising for applications due to its stoichiometric composition, its relatively

large band gap (0.3 eV), and the central (Γ-point) position of the Dirac cone in its 2D surface band

structure. Although the theoretical solid-state models that the TI field has produced are powerful

and unique, their novel emergent physical properties are not universally observed in every sample.

These materials are difficult to grow and maintain under ambient conditions. Growths tend to

either not be applicable to wafer-scale production or produce high polycrystallinity, and all

samples experience natural oxidation, band bending, and intrinsic n-doping, which generates spin-

degenerate or bulk conduction.

This thesis contains a primer on topologically non-trivial materials, and two studies aimed at

understanding and minimizing defects at the surface of Bi2Se3. In the first, the aging process of

Bi2Se3 when exposed to air at room temperature is investigated. The time scale and topographic

changes of the oxidation process at micromechanically exfoliated surfaces are measured, and an

optical model of the bulk and oxide layers are developed. The surface appears to oxidize starting

at 2 hours after exfoliation, and continuing through 1.5 weeks, by which time, the oxide layer

growth has reached an asymptote of 1.9 nm. New optical characterization methods are developed

to monitor the orientation of the crystal (via second harmonic generation) and to measure the oxide

growth at the surface (using spectroscopic ellipsometry and the derived dielectric functions of the

bulk and oxide layers). The goal of the second study is to assess the use of Se capping and

subsequent thermal decapping to preserve a pristine surface and maintain a constant Fermi level.

This was measured by annealing samples in a UHV environment to successively higher

iv
temperatures until the Bi2Se3 film decomposed, and measuring the surface crystallinity,

topography, surface chemistry, and Fermi level between each anneal. Thermally decapping

samples has no measurable effect on crystallinity, minimal effect on surface topography, reveals

the expected Bi-Se surface bonds, and retains a mid-gap Fermi level. This may serve as a reference

to improve the fabrication process of devices that include Bi2Se3.

v
Table Of Contents

ABSTRACT 2

LIST OF FIGURES 5

CHAPTER 1: Introduction 10

1.1: Moore’s Law 11

1.2: Spintronics 12

1.3: Graphene and High Mobility 2D Materials 16

1.4: Topological Insulators and Bi2Se3 17

1.5 Defects in Bi2Se3 20

CHAPTER 2: Topologically Non-Trivial Systems 23

2.1: The Hall Effect and Anomalous Hall Effect 24

2.2: The Integer Quantum Hall Effect and the Anomalous Quantum Hall Effect 26

2.3: Z2 Topological Insulators 41

2.4: Bulk-Boundary Correspondence 48

CHAPTER 3: Bismuth Selenide, its Topological Properties, and its Defects 53

3.1 The Physical and Electronic Structure of Bi2Se3 55

3.2: Characterizations and Applications od Bi2Se3 60

3.3: Defects in Bi2Se3 and their Effects on Surface States 68

CHAPTER 4: Characterization Method Description 79

4.1: X-Ray Diffraction 79

4.2: Spectroscopic Ellipsometry 89

4.3: X-Ray Photoelectron Spectroscopy 99

4.4:Transmission Electron Microscopy 105

vi
 4.5: Atomic Force Microscopy 109

4.6: Temperature Programmed Desorption 111

4.7: Low Energy Electron Diffraction 118

4.8: Scanning Tunneling Microscopy and Scanning Tunneling Spectroscopy 122

CHAPTER 5: Surface Oxidation in an Ambient Environment 129

5.1: Sample Preparation 129

5.2: Measurement Results 130

5.2: Analysis 154

CHAPTER 6: 156

6.1: Sample Preparation 157

6.2: Measurement Results 158

6.3: Analysis 175

CHAPTER 7: Discussion and Conclusions 178

REFERENCES 186

vii
List of Figures

Ch. 1: Introduction

1-1: Trends in transistors per microprocessor and price per gigabyte of storage.

1-2: Schematic of GMR and spin Hall effects.

1-3: Schematic of a spin-transfer torque device.

1-4: Demonstration of conduction of Dirac fermions in graphene.

1-5: Color map of the spatial variations in charge density of a micromechanically

exfoliated graphene flake.

1-6: A transition between the insulating phase and QSH state, occurring at the predicted

critical thickness.

1-7: Evidence of the first three dimensional topological insulator

Ch. 2: Topologically Non-Trivial Systems

2-1: A survey of topologically trivial and non-trivial states.

2-2: The first measurement of a topological surface state in the quantum hall phase.

2-3: Laughlin’s flux cylinder, used in his derivation of the quantum hall effect.

2-4: A topological view of the simple insulating state and quantum Hall state.

2-5: Hofstadter’s butterfly.

2-6: The physical configuration and final phase diagram of Haldane’s model.

2-7: Band structures and phase diagram of the Kane-Mele model.

2-8: Schematic of a one dimensional, TRI band structure showing Kramers degeneracy

at and between .

2-9: Contour integrations around half the BZ1 in the Kane-Mele model in order to count

the zeros of .

viii
 2-10: Z2 index calculations, performed by multiplying the various di at TRI momenta.

2-11: Topological spin orientations in momentum space, as derived from the modified

Dirac equation.

2-12: Surface conduction and band schematics of topological materials.

Ch. 3: Bismuth Selenide, its Topological Properties, and its Defects

3-1: Schematic of the Bi2Se3 hexagonal unit cell.

3-2: Inversion of bands with opposite parity in DFT calculations of Bi2Se3.

3-3: Band structure of Bi2Se3, including all modifications to the Hamiltonian.

3-4: Spin-polarized Dirac cones, experimentally measured with ARPES.

3-5: Measured and theoretical LDOS distributions on the surface of Bi0.92Sb0.08.

3-6: Weak antilocalization in TRI materials.

3-7: SdH oscillations and analysis.

3-8: Predicted and measured Kerr rotation in thin Bi2Se3.

3-9: MBE growth of Bi2Se3 on SiC (0001) with an epitaxial graphene layer.

3-10: Band bending-induced 2DEG, growing over time in Bi2Se3, as shown by ARPES.

3-11: Generation of the Rashba spin-split 2DEG, as shown by ARPES.

3-12: Some attempts at extrinsically p-doping Bi2Se3 in order to move EF toward ED.

3-13: Bi and Se oxides growing in air at the surface of nanoribbon and bulk Bi2Se3

samples, as measured in XPS at the Bi 4f and Se 3d energies.

3-14: Similar Bi2Se3 surface band response from exposure to different environments, as

shown by ARPES.

3-15: A chemically inert Bi2Se3 surface.

Ch. 4: Characterization Method Description

ix
4-1: The Bragg condition indicates at what angles x-rays may be diffracted from parallel

crystal planes.

4-2: Demonstration of the relationship between real space planes and reciprocal lattice

vectors, using the (231) plane of an orthorhombic lattice.

4-3: Forbidden (002) diffraction in Si from one diamond cubic unit cell.

4-4: Demonstration of omega-2theta experimental configuration and reciprocal space

scan direction for a Si sample with a (001) surface.

4-5: The Ewald sphere and the allowed diffraction points of Si.

4-6: Demonstration of planes with a sample that must be tilted with the sample in order

for diffraction to be measured.

4-7: Band structure and imaginary dielectric function of germanium.

4-8: Polarization of reflected light.

4-9: Schematic of a dual rotating compensator ellipsometer.

4-10: Emission process for photoelectrons and Auger electrons.

4-11: Example of shake-up lines (s) of the Cu 2p observed in CuO and CuSO4, but not

pure Cu.

4-12: Intensity of electrons emerging from depth d01+d12 as a function of their emission

and detection angle.

4-13: Schematic of STEM operation. Images can be produced from the BF, ADF, or

HAADF detectors.

4-14: AFM operation.

4-15: Schematic of TPD system.

4-16: Typical temperature calibration and full array of TPD-measured molecules.

x
 4-17: Universal mean free path curve of electrons traveling through solids.

4-18: Schematic of crystal diffractions in LEED.

4-19: Diagram of the LEED experimental system.

4-20: Schematic of an STM.

4-21: STM Topography measurement of Si(111) 7×7.

Ch. 5: Surface Oxidation in an Ambient Environment

5-1: Aberration-corrected HAADF STEM images of bulk Bi2Se3.

5-2: Omega-2Theta XRD scan of Bi2Se3.

5-3: Pole figure of exfoliated Bi2Se3 sample used in oxidation study.

5-4: Time-dependent surface topography of Bi2Se3.

5-5: Reference topography scan of a contaminated surface.

5-6: XPS and ARXPS results.

5-7: XPS scans of the Bi 4f region taken over various times.

5-8: ARXPS spectra of the Se 3d and Bi 5d region for the sample after ~1.5 weeks of

exfoliation.

5-9: ARXPS analysis.

5-10: Variable angle spectroscopic ellipsometry data and results of optical modeling.

5-11: Contrast maps for Bi2Se3 on a plain Si substrate and on a Si substrate with 300 nm

thermal SiO2.

5-12: Measured rotational-anisotropy SHG (RA-SHG) scans in a dry air environment

immediately, and at 1, 2, 3, 4, 12, and 24 hours after Bi2Se3 sample cleaving.

5-13: XRD orientation of the ( 0115) plane for SHG analysis.

5-14: Schematic of SHG at the surface of Bi2Se3 samples.

xi
 5-15: Schematic of 2-step surface aging process.

Ch. 6: Characterization of Bi2Se3 with Selenium protective layer and after Selenium layer

removal

6-1: SEM images of Se-capped Bi2Se3 films, showing the crystallites on the cap surface.

6-2: Omega-2Theta scan of capped sample SB1.

6-3: Pole figures along the (0115) planes taken on the three rounds of MBE-grown

Bi2Se3 samples prior to any thermal treatment, while the Se cap remained.

6-4: XPS results from a capped Bi2Se3 sample.

6-5: Decomposition and desorption of the Bi2Se3 film.

6-6: Effects of UHV thermal decapping procedure on the surface crystallinity of sample

Rd3_2, as measured via LEED.

6-7: AFM images of the surface of the first set of samples taken over 5×5 μm2 areas.

6-8: STM images of sample Rd3_1, thermally decapped in situ prior to measurement.

6-9: XPS spectra taken on sample Rd3_2 after TPD decapping.

6-10: Typical STS results and analytical fitting used to derive EF - ED .

6-11: Histograms of Fermi level shifts in sample Rd3_1 after annealing temperatures of

150 (a), 180 (b), 210 (c), 240 (d), 265 (e), and 295 °C.

6-12: Statistical trends of Fermi level shifts over various anneals and time.

xii
Chapter 1: Introduction

In order to outline the potential impact that Bi2Se3 and other topological insulators may have

in future electronic devices, it is useful to examine historical trends. These explain why Si devices

are increasingly difficult to improve, and why alternative materials are the subject of intense

research. Many of these alternative materials have electrical properties that are increasingly

divergent from those seen in conventional semiconductors, and may be applied to devices that do

not operate on electrostatic principles. Graphene began a paradigm shift toward materials that host

ballistic electron transport states. Topological insulators are a spiritual successor to graphene

which maintain the linear Dirac cones that lead to ultra-high electron velocities and ‘massless’

carriers while coupling the spin of the electron to its momentum. This helicity causes a wide variety

of physical properties that are not seen in ordinary, topologically trivial (as defined in Ch. 2)

materials, such as forbidden backscattering and negative magnetoresistance. While these materials

have a series of potentially promising device applications, they have proven to be fragile. Many

studies report measurements where the signal comes from defect induced electronic states instead

of from the Dirac cones. While the literature has described some of the defects (which include, but

are not limited to, surface oxidation and atomic vacancy-induced doping), few studies have

focused on understanding the defects themselves. This is a necessary step to deducing the

conditions in which topological insulators can behave as theory predicts. Finding the right

conditions, or the right protective measures, is a key step toward using Bi2Se3 in new devices.

1
1.1: Moore’s Law

The economically driven goal of the semiconductor industry has been to adhere to Moore’s

law1, which states that the number of transistors in a microprocessor will double every two years,

while cost per device will decrease. This was possible for decades that followed this 1965

prediction. During this time, advancements in device design and architecture improved transistor

design, and new memory circuits were coupled with advancements in chip processing such as the

development of chemically amplified resists, advanced UV light sources, and chemical mechanical

polishing. More recently, attempts to improve scaling and performance have included the

implementation of strained silicon, raised source/drain, high-K/metal gates, and a switch from

planar devices to 3D fins with multiple gates. These improvements have driven transistor counts

per microprocessor from the lower thousands in early 1970s (Intel 4004) to upward of 15 billion

in 2016 (NVIDIA GP100 Pascal GPU), with accompanying exponential decreases in price per

transistor (Fig. 1-1).

However, lithographic advancement beyond 193 nm light sources have slowed, and gate and

source-drain leakage in Si devices impose an inherent limit on geometric scaling. Consequently,

advancements beyond the 22-nm node have slowed, and the “double every two years” model has

lengthened to every 2.5 or 3 years. In order to surmount the challenges created by parasitic

tunneling effects in the front end of the line, researchers and companies are looking to alternative

materials and physical phenomena, such as ferroelectricity, functionalized tunneling, and

spintronics. The International Technology Roadmap for Semiconductors introduced a

reorganization of initiatives and research directives in 2014, dubbed ITRS 2.0, which includes

‘Beyond CMOS’ as a major industry topic.

2
 Figure 1-1: Trends in transistors per microprocessor (red) and price per gigabyte of storage
(blue)2,3. The number of transistors per microprocessor increased exponentially for the majority
of time since Moore’s law was predicted, and the price per gigabyte of storage has exponentially
decreased at a similar rate. These trends have changed in recent years, due to inherent limits in
of silicon semiconductor manufacturing and performance. Fit lines are proportional to
1.4196^(year) and 0.5635^(-year) for counts and price, respectively, giving just over a doubling
of counts and a decrease to just under 1/3 price every 2 years during the given years.

1.2: Spintronics

Spintronics is a field of device research that focuses on spin carriers and the magnetization of

materials, rather than charge carriers and electrically polarizable or conductive materials. Some

recent work has focused on giant magnetoresistance (GMR) (Fig. 1-2(a)) and tunnel

magnetoresistance4, the latter of which often takes the form of magnetic tunnel junctions (MTJs).

GMR devices modulate conductivity through magnetic materials based on the alignment of the

carrier spin to the magnetic moment of its host material, i.e. the resistance a carrier experiences

during transport is low if its spin is parallel to the magnetic moment of the material, and high if its

spin is antiparallel to the magnetic moment of the material. MTJs are non-volatile memory

3
 Figure 1-2: Schematic of GMR and spin Hall effects. (a) GMR, demonstrated by the
transmittance of spin - electrons and backscattering of spin ¯ electrons in a material with a
magnetic moment along +ẑ , resulting in a high ratio of resistances for opposite spins. (b) Spin
Hall effect, demonstrated by longitudinal charge injection resulting in transverse spin current.
Adapted from [5].

structures in which two ferromagnetic layers are separated by an insulator, and spin carriers may

tunnel between the magnetic layers depending on their relative alignments. Switching is controlled

by spin-transfer torque-induced magnetization reversal of the ‘free’ ferromagnetic layer, causing

low tunneling amplitudes for the antiparallel configuration, and high tunneling amplitudes for the

parallel configuration. This can be seen in rectangular H-field hysteresis loops, where a critical

field amplitude is necessary to switch the device.6 The tunnel magnetoresistance ratio, defined as

the difference ratio of resistances of parallel and antiparallel magnetization configurations (R AP-

RP)/RP has been reported as high as 600% at room temperature.4 While this does not yet achieve

the on/off ratio required in modern integrated circuits, it demonstrates a promising alternative

paradigm for logic and memory devices.

A significant obstacle to generating MTJ devices is the difficulty in generating high amplitude

directed magnetic switching fields in nanoscale geometries.7 A standard example of this process

is electron conduction through a ferromagnetic or antiferromagnetic material, from which the

gathered output electron has a spin aligned to the magnetic moment of its conducting material.

4
 Figure 1-3: Schematic of a spin-transfer torque device. Spin current from a reference layer with
magnetic moment Mr along + ŷ provides spin torque, which may flip the magnetization of the
free layer if the damping force is overcome. Weaker spin injections may result in magnetization
precession around the effective magnetic field (Heff). By changing the current polarity, the
magnetization can be switched between parallel and antiparallel with respect to the reference
layer. Adapted from [8].

These spin-oriented carriers may then be collected and their collective moments may suffice to

switch the write layer of the MTJ stack (Fig. 1-3). More recently, efforts have been made to harness

the giant spin Hall effect (Fig. 1-2(b)), allowing more efficient conversion from electrical to spin

conduction. Spin Hall angles (ΘSH = Js/Je) have improved from reports of 0.0037 in 2007 to 0.5 in

β-W in 2016. This > 2 orders of magnitude increase indicates that novel physics, e.g. giant spin

Hall effect and quantum spin Hall effect, may provide solutions to ongoing efforts in spintronics

device development.

5
1.3: Graphene and 2D High Mobility Materials

Another relatively recent advent in device physics came in the form of graphene in 2004.9 This

marked the beginning of a shift toward research in purely quantum effects implemented in electric

field effect devices. Importantly, it displayed the first instance of Dirac cones in solid electronic

Figure 1-4: Demonstration of conduction of Dirac fermions in graphene. (a) Dependence of

fundamental Shubnikov-de Haas oscillation (SdHO) field density on the gate voltage-
modulated carrier density of the graphene film. This follows the expected trend for 2D systems
as BF = f f n , where f is the degeneracy (equals 4 in graphene for spin and sublattice), and f 0 is
0

the magnetic flux quantum. (b) Tracking the inverse applied field as a function of SdHO index
for graphene (red) and 5 nm graphite (black). The horizontal shift in graphene can be explained
by its non-trivial Berry phase.9 (c) Relative change in longitudinal conductivity as a function of
temperature shows a steeper drop-off for high gate voltage and high n (lower) than for lower n
(upper), indicating higher mobilities for states near the Dirac point energy. (d) Carrier mass as
a function of doping level or E. Data points conform to mc = E c*2 , and from the function
, where S is the area of Fermi energy orbits, the relationship may
be derived. This proves the existence of a linear cone structure of graphene, and fits to the data
yield c* =1´106 m/s. (e) Schematic of Dirac cone, where the Dirac point is the center of the
linear dispersion, and where n = 0. Adapted from [10].

6
systems. These Dirac cones are local regions in the first Brillouin zone (BZ) where valence and

conduction energy bands are connected at a point, and the energy of the upper and lower bands

depends linearly on the momentum distance from the cone center, i.e. , where E0

is an offset and Ki is the Dirac point location (Fig. 1-4). This semimetallic connection is

nontrivially different from typical materials with parabolic bands, in that the linear dispersion gives

rise to a “massless” electron state, with early Fermi velocities measured at 1×106 m/s10, and

mobilities measured at 230,000 cm2/Vs11.

Despite the exciting physical properties of graphene, its implementation in devices has been

difficult. Perhaps predictably, growth of 2D carbon with controlled doping levels, few structural

defects, and minimal topography has been met with challenges. Though exfoliation has provided

a reliable means to obtaining high quality flakes that are utilizable at the lab scale, efforts to grow

graphene at wafer scale (on hexagonal SiC13, Ru14 and Cu15, (110)-Ge16, hBN17, etc.) have not

Figure 1-5: Color map of the spatial variations in charge density of a micromechanically
exfoliated graphene flake. These may be caused by minor disorder in the graphene topography,
trapped charge in the substrate oxide, doping via exposure to photoresists chemicals,
atmospheric contaminants, etc. Adapted from [12]

7
produced large pristine domains consistently enough for implementation in current integrated

circuits. Further, methods to transfer epitaxial graphene typically result in heterogeneous doping

of the film, and can lead to localized charge ‘puddles’12 (Fig. 1-5), which negatively impact control

of electrons at or near the Dirac point. Research into production is ongoing, but it behooves the

field of device physics to try alternative routes, where true 2D materials may be avoided. This

option presented itself in 2009 in the form of Bi2Se3, a topological insulator (TI) material.18

1.4: Topological Insulators and Bi2Se3

Novel research that occurred soon after the discovery and characterization of graphene came

in 2005, with the theoretical prediction of the quantum spin hall (QSH) state, and a topological

classification to describe it.19,20 The topological nature of this state refers to a Hamiltonian that

generates the same band gap as ordinary dielectric materials, but cannot be smoothly transformed

into any “topologically trivial” states without closing the band gap, which results in exotic interface

states (explained in detail in Ch. 2). These topological surface states (TSSs), like the Ki states of

graphene, occur in Dirac cones, but unlike graphene their spins are locked perpendicular to their

momentum direction, and they are protected against weak perturbations by the topology of the

bulk. The central portions of the theory behind topological classification of solid systems had been

developed decades earlier (and have recently won a Nobel prize)21–23, and some of its

consequences had been demonstrated via the quantum Hall effect.24 The more recent development

of the Z2 topological theory of Kane and Mele contains insights on time reversal invariant (TRI)

systems, which lead to explosive growth in the field of topological insulators (TIs). The first

physical realization of this effect was shown in HgTe quantum wells that displayed the QSH effect

above a predicted critical thickness (Fig. 1-6).25,26 Quantum wells thinner than 6.3 nm contain

8
 Figure 1-6: A transition between the insulating phase and the QSH phase, occurring at the
predicted critical thickness. Curve I shows resistance data from a thin quantum well, below the
critical phase transition thickness of 6.3 nm. The low conductivity level of this film indicates
its insulating phase. Curves III and IV are from measurements of films above the critical
thickness, and show quantized conductivity levels equal to the predicted values of the quantum
hall state. This structure was the first manifestation of the topological QSH effect. Adapted
from [25].

trivial insulator state, but when the thickness increases above that level, the band structure of the

system changes, and the system shifts into a QSH state. Another major advance in the TI field

from Charles Kane came with the development of Z2 theory for 3D systems (another notable

difference from graphene physics) with inversion symmetry.27 This significantly simplified

topological calculations, and facilitated efforts to find and grow 3D TI materials, but initial efforts

were largely focused on non-stoichiometric compounds such as Bi1-xSbx.28 These materials

demonstrated the first examples of Dirac cones on 3D surfaces, as predicted by Z2 theory and

shown in angle resolved photoemission spectroscopy (ARPES) in 200829, but had complex band

structures with small gaps. In 2009, calculations of Bi2Se3 using the same inversion-symmetric Z2

theory revealed that it had a simple band structure and larger gap than any previously discovered

TI. Its surfaces were predicted to contain a single Dirac cone at the gamma point, and a (relatively)

9
large band gap of 0.3 eV. These were quickly realized experimentally30, and research into the

‘second generation’31 of TI materials was born.

Bi2Se3 contains the surface spin-polarized Dirac cones inherent to 3D TI materials. At these

surfaces electron conduction is helical, i.e. the momentum and spin are locked at a perpendicular

angle. These band structures generate multiple interesting properties (see Ch. 3 for more complete

descriptions). As in the case of graphene, these Dirac cone states have ultra high mobility due to

the linear energy dispersion. The spin-polarized nature of these Dirac cones makes backscattering

impossible, as an electron’s spin must to flip in order for it to inhabit a state with the opposite

momentum. Carriers in these systems display an unusual property called weak antilocalization

(WAL) that results in negative magnetoconductivity.32 This emerges from the phase difference

Figure 1-7: Evidence of the first three dimensional topological insulator, Bi1-xSbx. (a) ARPES
measurements show the graphene-like Dirac cone at the surface of this material. (b) A phase
diagram demonstrates the critical Sb concentration that generates the topological insulator
phase. The intersection of bands gives rise to a ‘negative band gap’, a common trait among
topological materials. Adapted from [28].

10
that they acquire during diffusive transport, wherein traveling around a closed loop leads to

destructive interference. These states also generate unusual Shubnikov-de Haas data.33 These arise

in high field magnetoconductivity tests that measure resistance changes while the incident field

shifts Landau level energies through the Fermi level. They provide information regarding the

electrical mobility and density of the system, but also generate a clear sign of the system’s topology

in the phase of the oscillations. The surfaces of Bi2Se3 demonstrate an unusual half-integer

quantized Hall conductivity, although this is difficult to measure as it requires isolating conduction

to a single surface.34 They create an unusual optical behavior called the topological

magnetoelectric effect that results in giant Kerr rotations. Ultimately, they may be used to generate

Majorana fermions for quantum computing purposes, although research in this direction is at a

nascent stage.35

This array of unusual physical behaviors in Bi2Se3 makes for promising potential applications.

Its spin polarized Dirac cones are of particular interest to the spintronics community, where spin-

polarized carriers are needed en masse for many device architectures.

1.5: Defects in Bi2Se3

Although the TSSs of Bi2Se3 are necessary if these materials are going to be applied to

spintronics, crystal growths have so far been plagued by defects. Samples produced for academic

purposes are often grown from liquid melts. These are of high quality and large volume, displaying

flat cleaved surfaces of > 1 mm2.36 However, this growth methodology is not compatible with

modern semiconductor fabrication techniques. More recently, thin film deposition has been

attempted with molecular beam epitaxy (MBE).37 These films are more polycrystalline and have

rougher surfaces than their bulk-grown counterparts. Based on available information, we believe

11
that all MBE films have triangular pyramidal surface structures. These topographic features are

composed of vertically stacked QLs, and usually have heights on the order of > 10 nm above

background. This roughness is a complication in the fabrication process, and may additionally

serve as a detriment to TSS conduction, as these electrons are confined to the surface and may not

be able to smoothly traverse a complex topography. Surfaces are also chemically fragile. The vapor

pressure of Se is high, and Se vacancy formations have been shown to cause n-doping in Bi2Se3

crystals.38 When degenerately doped, Bi2Se3 may conduct metallic carriers in the bulk and on the

surface. These non-TSSs produce Shubnikov-de Haas oscillations at a different phase, and exhibit

spin-degenerate conduction. Moreover, since the density of states of a bulk band is exponentially

higher than that of a 2D system, these may overwhelm the electrical conduction signals created by

TSSs. Perhaps the most detrimental effect of doping emerges from the energy dependence of the

Dirac cone helicity. Carriers that inhabit states below the Dirac point in the surface dispersion have

opposite helicity to the carriers that inhabit states above the Dirac point. Thus, whether EF moves

away from ED (the Dirac point energy) in the negative direction due to p-doping and causes fewer

TSSs to be filled, or moves away from ED in the positive direction and causes the spin polarizations

of occupied states in the upper and lower cone halves to cancel, doping poses a severe threat to the

helicity of TSS conduction. These crystals also naturally oxidize. The first QL of Bi2Se3 and other

similar van der Waals gapped materials39 has been shown to expand to roughly double their initial

thickness during this oxidation process. This oxide exacerbates the surface roughness, correlates

with a surface two dimensional electron gas (2DEG) formation40 (which creates similar spin-

degenerate signals to bulk states), and may create an electrically insulating overlayer on top of the

desired TSSs. While it has been frequently measured in attempts to characterize the Bi2Se3 crystal,

few reports contain information about its electrical or optical properties.

12
 These defects pose a clear challenge to harnessing the unique properties of Bi 2Se3 (and other

structurally similar TIs such as Bi2Te3 and Sb2Te318), and must be studied and mitigated before TI

devices can be realized. However, as it is not practicable to fully comprehend the physical

phenomena that topologically non-trivial systems generate or the impact that defects have on their

unique properties, the following two chapters contain detailed explication of the theoretical and

experimental foundations and recent advances in the field of TIs.

13
Chapter 2: Topologically Non-Trivial Systems

In order to understand the physics of the topologically protected surface states in Bi2Se3, it is

useful to examine the early discoveries and theoretical developments in the field of topologically

nontrivial systems. There are many systems that display topological distinction, each with their

own physical origins, and they each are informative about the physics of modern topological

insulator materials such as Bi2Se3. The 2D electron gas (2DEG), subject to a voltage bias and a

perpendicular magnetic field, creates the classical Hall effect in conductors and, at low

temperature, the quantum Hall (QH) effect in semiconductor quantum wells. Although these

systems both show transverse conduction in response to longitudinal bias, but result from wildly

discrepant physics. The anomalous quantum Hall (AQH) effect was theoretically predicted within

a decade of the development of QH theory, and the quantum spin-Hall (QSH) came two decades

later. These two systems are topologically distinct from vacuum, as well as most dielectric

materials, and require no external potentials to generate topological surface states in linear

electronic bands. These are all useful examples of unusual two-dimensional electronic behaviors

and phases (Fig. 2-1) that result from topological physics, as well as its importance, in modern

solid-state physics. Each theoretical description builds on the theory that explains the previous,

leading to the Z2 theory, which was initially developed to explain the QSH effect in graphene but

underlies the topological nature of many 3D materials such as Bi2Se3 as well. In this way, the

discussion in this chapter will proceed from the theoretical foundations describing the integer QHE

for a 2DEG in a magnetic field at low temperature to topological insulators where the protected

surface states are not a result of an external magnetic field.

14
2.1: The Hall Effect and Anomalous Hall Effect

In the classical Hall effect, the Lorentz force causes the conducting 2DEG carriers to arch in

their paths according to . For a sample with electrons drifting with at volume

density n , subject to a magnetic field , equilibrium is reached where the separated carriers

generate a ŷ -direction bias vx Bz = Ey = VH . Here, w is the sample width and VH is the generated Hall
w

voltage

nqvx ABz I x Bz IB Eq. 2-1

VH = wvx Bz = = = RH x z
nqd nqd d

This has provided a method for measuring net carrier density (where negative RH indicates

majority electron conduction and positive indicates majority hole conduction) for samples of

1
known thickness d and measured longitudinal current Ix, in the form of RH = . Transverse
qn

resistivity in this system is a continuous and linear function of , and can be simply derived from

a Drude model:

Eq. 2-2

where meff is the effective carrier mass and G is the average time between collisions, 1/ t . Using

the relationship , the resulting resistivity is

æ 1 w ct 0 ö
ç ÷
r = r0 ç -w ct 1 0 ÷ Eq. 2-3
çç ÷
è 0 0 1 ÷ø

meff eBz
where r0 = and w c = . This is a useful conceptual precursor to the topologically distinct
nq t
2
meff

IQHE, wherein transverse resistivity in the 2DEG of a semiconductor quantum well has a similar

dependence on , but is quantized step function rather than a continuous one.

15
 Figure 2-1: A survey of topologically trivial and non-trivial states. (a) The insulating state is
characterized by an energy gap separating the occupied and empty electronic states, which is a
consequence of the quantization of the energy of atomic orbitals. (b) In the quantum Hall effect,
at the edge, electrons execute “skipping orbits” as shown, ultimately leading to perfect
conduction in one direction along the edge. (c) The edge of the “quantum spin Hall effect state”
or 2D topological insulator contains left-moving and right- moving modes that have opposite
spin and are related by time-reversal symmetry. (d) The surface of a 3D topological insulator
supports electronic motion in any direction along the surface, but the direction of the electron’s
motion uniquely determines its spin direction and vice versa. Adapted from [41].

The AHE is similar, in that the Lorentz force generates arching carrier paths, and with them,

transverse voltages and resistivities. However, this effect takes place in the absence of an externally

applied magnetic field. It is instead caused by conduction through a material with an inherent

16
magnetic moment , such as a ferromagnet. This is often a much stronger effect than the classical

Hall effect.42

2.2: The Integer Quantum Hall and Anomalous Quantum Hall Effect

The integer QH effect (IQHE) is a physically complex phenomenon, and a simple preemptive

conceptual encapsulation may help the reader. In addition to the localized states found in common

insulators, where the Fermi level, EF, inside the band gap ensures that all valence states below the

gap are filled and non-local scattering effects are prevented, the QH effect displays simple

harmonic oscillator (SHO) states when a magnetic flux is applied to the sample. Electrons

in these states can be understood as having a circular motion with direction given by and the

right-hand rule, circumscribed with radius . Because of the required uniform

directionality of these emergent SHO states, even when EF is in a gap, the effect of edge scattering

is to push an electron into an adjacent circular orbit (Fig. 2-1 (b)), thus creating robust edge

conduction in an otherwise insulating material.

The integer QH effect was first predicted for a 2DEG by Ando et al. in 197543 and

demonstrated in 1981 by von Klitzing et al24. Ando’s theoretical work painted the picture of a

2DEG with robust quantized Hall conductivity in the face of short-ranged scatterers, and von

Klitzing’s experiment in a metal-oxide-semiconductor field-effect transistor at liquid helium

temperature (1.5 K), and with a strong external magnetic field (15 T) confirmed it. In this

experiment, raising the gate voltage while maintaining the strong external magnetic field lead to

discrete changes in Hall voltage (Fig 2-2). This transition occurred at gate bias levels coincident

with the Landau level (LL) energy spacing , and generated quantized

17
 Figure 2-1: The first measurement of a topological surface state in the quantum hall phase.
Recordings of the Hall voltage UH, and the voltage drop between the potential probes, Upp, are
shown as a function of the gate voltage Vg at T=1.5 K and constant magnetic field of 18 T in
the work of von Klitzing et al. The inset shows the experimental setup. The integers on the gate
voltage axis give Landau level energies, between which longitudinal voltage is 0 and Hall
voltage is at a plateau. At the Landau levels, EF is effectively in a small continuous energy band,
thus allowing longitudinal conduction. As EF increases inside LLs, the QHE increases in
strength. Adapted from [24].

N L e e æ neB ö ne2
conductivity changes when EF crossed through LLs s xy = = ç ÷= (LL degeneracy
B Bè h ø h

eB
NL = will be derived later). While von Klitzing’s original work described the conductivity,
h

resistivity, and Hall resistance accurately (and successfully provided an accurate measurement of

m0c
the fine structure constant a = to a new level of precision), a novel quantum mechanical
2nRH

description by Laughlin44 and Thouless et al.23 was required to fully elucidate the quantum

mechanical behavior of the 2DEG, i.e. to explain the quantized plateaus of Hall conductivity and

resistivity, to explain the edge transport generated by dissipationless “skipping” states, and most

importantly to classify the IQHE by a distinct topology of electronic bands in the first Brillouin

18
zone. As this provides a sound basis for TI physics, it is valuable to show some of the details of

Thouless’ and others’ theoretical work.

The integer QH effect occurs in an initially free electron gas in a magnetic field can be quantum

mechanically expressed as:

Eq. 2-4

The ẑ -component is ignored, as it generates a trivial kinetic energy contribution, and additional

scalar potentials are not considered. Here, a Landau gauge choice of (for a vertical

magnetic field ) results in a Hamiltonian that only includes a momentum operator in

the ŷ -direction. Thus, a wave function of the form y = u(x)eiky is valid, and generates

æ p̂ 2 mw c2 2ö
Eu(x) = ç x + ( x + klB2 ) ÷ u(x) , where w c = qB and , which is a familiar quantum harmonic
è 2m 2 ø m

oscillator equation with an oscillation center of x0 = -klB2 . The energy levels of this 2D system are

, which are separated into discrete Landau levels (LLs). Each LL contains states with

degeneracy N that can be calculated from the Gaussian spatial distribution of a SHO with

æ ö
wavefunction u(x) = H n (x - x0 )exp ç - (x - x20 ) ÷ , where the exponential denominator 2lB2 is taken to be
2

è 2lB ø

the oscillation range:

Eq. 2-5

h
where gs = 2 is the electron spin degeneracy and F0 = is called the quantum flux. The third
e

quantity shows that degeneracy can be seen as single state taking up the area of a circle with radius

lB2 , and the last shows that each degenerate state is created by a single flux quantum.

19
 Figure 2-3: Laughlin’s flux cylinder, used in his derivation of the quantum hall effect. (a)
Laughlin’s gedanken experimental geometry involving a 2DEG cylinder penetrated by a
magnetic field Perturbations to H0 result in quantized charge transfer in the n = LDA
f . (b) The 0

IQHE density of states (DOS) is still separated into LLs, which remain quantized and nearly
gapped in the presence of higher temperatures (c). Adapted from [44].

The addition of an electric field in the x̂ -direction to the Hamiltonian,

px2 mw c
( x + klB2 ) - eEx ,
2
H= + causes carriers to drift. Completing the square with the external term,
2m 2

px2 mw c æ æ 2 E öö
2
mE2
H= + çç x + ç klB - ÷÷÷ + eEklB - 2 , a new SHO center can be defined as x1 = -klB + wEcB . In
2 2

2m 2 è è w c B øø 2B

addition to this shift from the SHO center, the energy spectrum results

in a drift velocity of

Eq. 2-6

Thus, analogously to the transverse arcing trajectories in the classical case caused by the Lorentz

force, the quantum Hall effect results in transverse drift when a longitudinal field is applied.

The question of the origin of quantized conductivity was answered by Laughlin in a physically

valid thought experiment. He imagined a Hall configuration in which a cylinder that contained a

2DEG was penetrated by a perpendicular magnetic field (Fig. 2-3). For this geometry, when the

20
applied magnetic field is perturbed by an additional magnetic vector potential DAŷ , the SHO center

E DA DA
shifts linearly as x2 = -klB2 + - = x1 - . This carries through to the energy dispersion,
wc B B B

, thus generating a linear relationship to the perturbation. Due to

the geometry of this experiment, however, there is a gauge restriction on values of DA , which

through the Aharonov-Bohm effect, must return the phase of the conducting electrons to their

original values upon completion of a loop around the cylinder, i.e. , or

nh f
DA = = n 0 . Thus, the perturbations added must be integer multiples of the magnetic flux
eL L

quantum divided by the circumference of the loop, L. An addition of f 0 maps the system to its

original state, and results in the transfer of n electrons from one edge to the other. This n is a valid

quantum variable for defining states of particles looping the cylinder, and mirrors the degeneracy

neVH
found in LLs. The current can therefore be related to the induced transverse voltage as I = ,
f0

I e
resulting in the quantized hall conductivity, s xy = = n . Comparing to the resistivity of classical
VH f0

B F N
Hall effect r xy = = (temporarily using h = as areal carrier density for clarity) this resultant
he Ne A

f0
quantized resistivity r xy = s xy-1 = is a very reasonable and apparent quantum mechanical analogy.
ne

The most important aspect of the IQHE to note, however, is the topological nature of its

electronic states. This can be shown through the Berry phase of the system, which is the phase

generated by adiabatic transformations in a parameterized Hamiltonian system , where the

wave function and phase changes are given by . The time-dependent

Schrödinger equation, (where ¶t is the partial time differential) produces:

21
 , left-multiplying and integrating,

, undergoing substitution to remove time dependence,

, over a closed path, using ,

, using Stokes’ theorem,

, over a closed surface, using ,

Eq. 2-7

where is the Berry connection, and is not gauge independent (as can be checked by

computing for y1 = eiqy0 ), where as the Berry curvature is gauge independent, and thus

physically measurable. This Berry phase is a purely geometric parameter that describes the phase

acquired by a state that is translated around the entire surface extended by Hamiltonian parameters

, such as the momentum generated by Bloch waves, . This value is only calculable for a surface

that’s closed over the first Brillouin zone for a given EF, as is the case for insulators, but is not the

case for metals, wherein the Fermi level intersects band boundaries. The Berry curvature can also

be expressed as

Eq. 2-8

where Einstein summation notation is implied for all elements of , and from which its identity

as a surface curvature is more apparent. The integral of this value over a compact surface, as seen

from the Gauss-Bonnet formula (where K is the Gaussian curvature of the surface,

K ^ dA gives the magnitude of the curvature at a surface point, and c is an even integer called the

Euler character of the surface), is a quantized value that depends on the topology of the surface.

22
Using this Gauss-Bonnet formulation, one can classify smooth shapes distinctly, as in the 2-sphere

and all of its smooth deformations, which have an Euler character of c = 2 , whereas the torus and

all of its smooth deformations have an Euler character of c = 0 . This integrated curvature, which

results in quantized Euler characters, can be physically visualized as differentiating the surfaces

by the number of different ‘holes’ or ‘handles’ they contain (Fig. 2-4), or by the required number

of points where vector fields on the surface of the shape must converge (singularities, as

demonstrated at the poles of a sphere predicted by the “Hairy Sphere” theorem, or by the smooth

field lines that cover a torus). The interaction of spaces that have differing topologies, and the

resultant interface states, is the essence of the conductive states that are generated in topological

insulators.

Figure 2-4: A topological view of the simple insulating state and quantum Hall state ((a) and
(d)). Although both are separated by gaps ((b) and (e)), the curvatures of their entire magnetic
Brillouin zones are categorically different. In the same way that the orange ( c = 2 ) and donut (
c = 0 ) ((c) and (f)) cannot be smoothly morphed into one another without closing the donut
hole, which is a discontinuous process that creates undefined curvature, the Hilbert spaces that
define the insulating state and quantum hall state cannot be smoothly connected without a
sudden band gap closure. Because BZ1 can’t be as easily shown to have holes or handles, it is
sometimes easier to note the singularities (or lack there of for the torus (h)) in the wave function,
as shown at the poles of the sphere (d). This is the essence of topological differentiation and
the Chern number. Adapted from [34] and [45].

23
 Returning to the IQHE, while origin of the quantization of the Hall conductivity has been

answered, the quantum Hall state’s behavior in true periodic lattice conditions and general

topological character have still not been accounted for. For these properties, we turn to the work

of Thouless et al.23, who used a Bloch wave in both x- and y- dimensions, and rather rather than

perturbing the vector potential, used the Kubo formula to derive the quantum Hall state.

Accordingly, Thouless’ Bloch Hamiltonian is

Eq. 2-9

where U(x, y) is the periodic potential of a lattice, and the conductivity of a sample can be

computed as follows:

Eq. 2-106

Here, the summation of excited states can be taken to represent the bands in a typical solid-state

system. The velocity operators can be substituted by , which produce expected value

terms in the numerator that are in the form equal of the derivative of a first-order perturbation

theory state: . The inner product of these k perturbation terms results in

, and this is exactly the form of the Berry curvature generated by the

quantum variables kx and ky,

Eq. 2-11

While this is the correct form for all bands, it is useful to separate contributions from individual

bands n, as in a real system, only some bands are filled (additionally, if all bands were filled, the

24
system would necessarily be topologically trivial46). Thus, the Hall conductivity in a periodic

lattice is directly proportional to the sum of Berry curvatures for filled bands,

Eq. 2-12

This is the Hall conductivity for a given state k. This readily becomes topologically significant

when we look at the expected Hall conductivity, which requires averaging over all k values:

Eq. 2-13

This is more commonly written in terms of the Chern number, Cn = 2pc n , named after Shiing-Shen

Chern for his work on winding fiber bundles47, giving the exact expected form for the IQHE,

e2
s xy =
h n
åCn Eq. 2-14

where each Cn is itself an integer (although it may be 0 in many cases) and the sum is consequently

quantized. This is the result found by von Klitzing. Completing the calculation of the Chern

number for a periodic potential is difficult and complex, and can be found in the work of Thouless

et al.23

The distinct phases given by these topological distinctions may be elegantly plotted with the

Hofstadter butterfly48 (Fig. 2-5).49 In an insulator, quantum hall phase emerges from any non-zero

magnetic flux and EF.

The next logical step in describing topological systems is to look at the AQHE through the

work of Duncan Haldane22. His is a slightly generalized50 graphitic model51, containing distinct

atomic species on the typical diatomic graphene lattice, threaded by periodic antiparallel magnetic

fluxes, such that the net flux in any unit cell equals zero (Fig. 2-6(a)). The 2D unit cell is hexagonal,

25
 Figure 2-5: Hofstadter’s butterfly. This is a phase diagram, showing topologically distinct
phases as a function of EF and the magnetic flux through the system. Each color represents a
sum of Chern integers, where positive regions are red and negative regions are blue. If this plot
were to continue vertically, the butterfly would be repeated. Thus, the topology of phases is
periodic, and one period is given by an incident magnetic flux quantum. A closer look evinces
a fractal pattern, although the causes and consequences of that are beyond the scope of this
work. Adapted from [52].

given by the lattice parameters and , with the reciprocal lattice

. This results in important first Brillouin zone (BZ1) corners given by

and , as will be seen in subsequent calculations.

This system, modeled in a the second quantized scheme, receives contributions from

differential on-site energies ( M ), nearest neighbor (NN) hops ( t1 ), and next nearest neighbor

(NNN) hops ( t2 ) that receive a directionally dependent phase contribution from the Aharonov-

Bohm effect:

Eq. 2-15

where and refer to NN and NNN vectors, and are atomic sites in the unit cell, and

26
 Figure 2-6: The physical configuration and final phase diagram of Haldane’s model. (a) The
graphitic lattice used has 0 net flux due to its periodic distribution in the hexagonal Wigner-
Seitz unit cell. (b) Phase diagram of the 2D system, in which topologically trivial phases occur
M M
for > 3 3sin f or < 3 3sin f , and AQHE occurs for the central regions labeled n = ±1 .
t2 t2
Adapted from [22].

may refer to either lattice site (for sites given by i). The creation and annihilation operators ci+ and

ci may be Fourier transformed so that the Hamiltonian may be written in terms of :

Eq. 2-16

The mathematics simplifies with a substitution of Pauli matrices . The Hamiltonian and

energy may be written as 20

, where e and are Hamiltonian matrix energy

coefficients:

Eq. 2-17

The two bands may only touch at the and points in BZ1, as in graphene (which is

equivalent to this model for t2 = M = 0 ), where and (for the three

hops). These are the only topologically significant points in BZ1, as the rest of the dispersion

27
resembles a trivially gapped insulator. A Taylor expansion near the points results in a

Hamiltonian of

Eq. 2-18

where e = -3t2 cos f can largely be ignored, , and . is a

Dirac Hamiltonian with linear dependence, similar to graphene, as shown by the simplified

equation E± = e ± dx2 + dy2 + dz2 , where . However, the d z term can open gaps at ,

and the signs of these gaps (a bit of a misnomer given that either sign generates separation between

bands) will be shown to determine the topology of the system. These signs are then completely

determined by the term.

Among the ways to determine topology and the quantized Hall conductivity are to utilize the

Kubo formula and integrate the Berry curvature over the whole BZ1 as Thouless did for the QHE,

to use the Streda formula53 for quantized Hall conductivity as Haldane did, to integrate the solid

angle generated by the Pauli coefficient unit vector as shown by Berry54

34
, or to monitor whether a single well defined Bloch wave function may

or may not be defined over the whole of BZ1 in order to perceive topological distinction. The latter

method, which can indicate the presence of singularities in and thus topological distinction and

non-zero Hall conductivity46, is useful here. Because singularities do not manifest physically, they

cannot be observed in the energy dispersion of a system. Rather, they are only directly calculable

through complex, non-Hermitian operations such as the Berry connection, or indirectly via

integrated, similarly non-physical parameters over regions of Hilbert space.

28
 In order to show the singularities of the Haldane system, it is sufficient to show that any viable

wave function for the system has 0 amplitude at either of the points, or that it must travel

through 0 between them. The Schrodinger equation may be solved for the conduction and valence

bands as

, , , and Eq. 2-19

Because , these simplify significantly, and can be represented by the coefficients

and . Taking the first wave function in the valence

band, for example, the sign of b-I is negative for dz > 0 and b-I = 0 for dz < 0 . Thus, if d z switches

signs between the points, then b-I must be 0 at one of the points. If we temporarily ignore

the differential onsite energy contributions, then dz = -3 3t2m sin f , where m = ±1 indicates the

point, and d z switches signs between points. Thus, y-I is not definable over BZ1. This

calculation can be repeated to show that y-II , y+I , and y+II , are also not definable over BZ1. There

is no single wave function that can cover the entirety of either the valence or conduction band, and

thus at least one singularity emerges. This guarantees that the system is topologically non-trivial.

The more general requirement for d z to switch signs between points is M < 3 3t2 sin f , and the

equality of these factors is what separates the AQHE phase from the trivial insulator (Fig. 2-6(b)).

Because the two wave functions of the valence band are only differentiated by a gauge shift (

), it is easy to calculate the Chern number with a contour integration of the Berry

connection:

29
 Eq. 2-20

Here, the integration over the whole BZ1 is separated into contours surrounding the domains which

are nonzero for , ¶DI and for , ¶DII . The gauge shift of the Berry connection

was used for simplification. In this case, the first domain can be taken to be a

small circle surrounding a point, thus allowing an angular substitution for :

2p
2p C- = - ò ¶zq dz = q (z = 0) - q (z = 2p ) = 2p n Eq. 2-21
0

The actual values of q (z = 0) and q (z = 2p ) depend on the specific gauge shift. In this case,

. Near a point, the Dirac

Hamiltonian parameters suffice, and produce q = ±z , and thus C- = ±1 . In the case that

M < 3 3t2 sin f and sin f > 0 , C- =1, in the case that M < 3 3t2 sin f and sin f < 0 , C- = -1 , and in the

case that M > 3 3t2 sin f , C- = 0 , as shown in the Fig. 2-7 phase diagram.

The final theory step toward generating a topological model for 3D crystals, including Bi2Se3,

is to look at the QSHE and Z2 theory developed by Charles Kane, staring in 2004.19,27,55 The QSHE

model for graphene is remarkably similar to Haldane’s AQHE model, wherein spin orbit coupling

(SOC) in Kane’s model takes the place of the periodic flux in Haldane’s model.

H = t åci+c j + ilSO å vij ci+s z c j + lv åxi ci+ci Eq. 2-22

ij ij i

In this case, the system is topologically classified by lv > 3 3lSO , which generates the insulating

phase, or lv < 3 3lSO , which generates the QSH phase (Fig. 2-7). A full derivation is omitted here.

30
A significant difference of note is that because there are no interactions between the spin states,

the Hamiltonian is separable by spins, and we get a decoupled pair of Hamiltonians with opposite

topological indices:

æ H ö
H+ 0
H = H­ + H¯ = y + ç ÷y Eq. 2-23
ç 0 H -H ÷
è ø

Here, the creation and annihilation vectors have a 4-component basis for both spins and sublattices,

( )
y = a- b- a¯ b¯ , and the H +H is a Haldane-like model for a positive contribution from SOC.

The sign of the SOC contribution only changes the sign of the d z term, which has the effect of

negating the Chern number of the system. Because these Hamiltonians do not interact, the Kane-

Mele model for SOC in a graphitic geometry contains two spin-polarized copies of the AQHE

system.20 Each of these contribute equal charge conductivity in opposite directions at the edges,

Figure 2-7: Band structures and phase diagram of the Kane-Mele model. This model takes into
account differential on site energies, NN and NNN hopping with SOC, and a Rashba term.
Along the axes lv / lSO and lR / lSO , which are ratios of on site and Rashba energies to the SOC
energy and along which axes only one is considered, the system is in a QSH state for
lv / lSO < 3 3 and lR / lSO < 2 3 , but is a trivial insulator otherwise (inset). (a) and (b) show the
band structure in the bulk, along with surface dispersions for the QSH phase and insulator
phase. The surface bands are chiral, leading to spin current in the QSH phase, but the insulator
phase does not conduct because the surface bands are filled. Adapted from [19].

31
but because they are spin polarized, electrons with spin up moving in the opposite direction as

electrons with spin down generates a net spin current. This is what gives the QSHE its name.

There are two complications to this model that are worthy of consideration, and are integral to

the understanding of topological insulators as a general class. The first is an additional term in the

complete Kane-Mele model that involves interactions between spin states. This is the Rashba term,

, and its addition requires a model that goes beyond chiral Haldane pair

model. The second complication is that for time reversal invariant (TRI) systems, each band has a

partner with the opposite Chern number (as can be seen from ), resulting in a net

Chern number of zero.56 Thus, while it is possible to use a similar method to previous one to

calculate wave function singularities and individual Chern numbers for the general system, this is

a useful situation in which to introduce the more general Z2 theory, which may characterize the

system as a whole. Further, the Z2 theory is generalizable to 3D systems with both TRI and

inversion symmetry, and is thus extremely useful to Bi2Se3, which is a system that contains those

symmetries.

2.3: Z2 Topological Insulators

The Z2 theory of topological materials is more general than the Chern classification of bulk

topologies, although the price for this additional usefulness is paid in abstractness of the theory.

However, while the Chern classification is rendered moot by TRI, the Z2 theory simplifies

(somewhat). This is largely due to the emergence of Kramers degeneracy. This concept refers to

spin ½ systems, in which TRI guarantees the existence of an isoenergetic state at ± p̂ . In Bloch

ic
a = -e Q uk,IIa
I
systems, this refers to momenta , and thus Bloch functions are related by u-k, k,a

32
 Figure 2-8: Schematic of a one dimensional, TRI band structure showing Kramers degeneracy
at and between . At the points, labeled ±p , the states are doubly degenerate in
momentum and energy because the momenta are equal. Between the points, a state is only
guaranteed to exist at . In general, these states belong to different bands. Adapted from [57].

ic
II
and u-k,a =e Q uk,I a , where a indicates the Kramers degenerate band pair, and I or II indicates
- k ,a

the band within the pair (Fig. 2-8).57 In general, these states are separated only by the direction of

momenta and a gauge shift, but they belong to different bands of a pair. There are various points

in Bloch systems at which e.g. for any reciprocal lattice

vectors in 3D, where n Î {0,1} . Thus, the system is required to be doubly degenerate at these

momenta.

The band doubling effect of TRI is what nullifies the net Chern number. Instead, the Z2 theory

looks at individual bands of each pair, and uses TRI implications in order to assess topology. The

main tool in this assessment is the interaction matrix between the span of eigenstates and the time-

reversed eigenstates,

wmn (k) = um (-k) Q un (k) Eq. 2-24

Here, there is only overlap between the Kramers band pairs. Thus, wmn (k) is a matrix composed of

33
æ i c k ,a ö
ç 0 e ÷ blocks along the diagonal, and wmn (Gi ) becomes antisymmetric at the Kramers
ç -ei c - k ,a ÷
è 0 ø

momenta. For antisymmetric matrices M , the Pfaffian Pf(M)2 = det(M) is an apt descriptor of the

matrix. For this specific TRI overlap matrix, Pf(wmn (k)) = ±Õa eic , where the sign of the product is
k,a

positive for even numbers of band pairs, and ambiguous for odd numbers of band pairs. For one
i c Gi
2 ´ 2 matrix, the Pfaffian is simply Pf(w(Gi )) = ±e , and Pf(w(Gi )) =1 .

The Kane-Mele QSH model was initially topologically characterized with a slightly different

overlap matrix, . The heart of this topological argument is what while in

general, time reversed pairs are only differentiated by a momentum and gauge shift, there exist

some pairs of points in BZ1 where , and that these cannot reconnect with one another

because their opposite positions ensure that they would have to do so at a Gi point where, much

like in the case of the prior overlap matrix, P(Gi ) =1 . Thus, an individual pair is unable to

reconnect, and these are permanent singularities in the Bloch space, and serve to

topologically characterize the space. These singularities will later be shown to be connected to the

Berry parameters previously discussed. Because the Pfaffian is a product of wave functions, the

Pfaffian zero is sensitive to a singularity in any band.

TRI ensures that a search over half of BZ1 is sufficient to find the Pfaffian zeroes. Around

each, the phase of propagates in a given direction, giving it each a certain vorticity (Fig. 2-

9(a,d)). It is possible for multiple pairs of zeros to exist. In this case, singularities of

either pair that have opposite vorticity may rejoin and annihilate each other. In that case, the

topology of the system is trivial. If the system has an odd number of singularity pairs, however, it

34
 Figure 2-9: Contour integrations around half the BZ1 in the Kane-Mele model in order to count
the zeros of . This process evinces the topological order of the system. (a) The familiar
QSH system in graphene, which contains singularities in either half of BZ1. The vortcitiy of
generates a topological index for the system. (b) The same system as (a) for equal onsite
energies, where the Pfaffian zero is smeared into a line and the system is topologically
undefined. (c) A plot of along the a 2 vector (d) for insulating (dashed) and QSH phases
(solid). (d) A more generalized unit cell with TRI points labeled p , at which P(p ) =1 act as a
barrier to topological shifts for individual pairs of singularities. Adapted from [19].

is inherently topological regardless of some potential annihilations. The calculation of the number

of Pfaffian singularities takes the form of a contour integration

Eq. 2-25

This tracks the total phase change along the contour, and results in an integer. In the Kane-Mele

model, this encloses a singularity at ‘M’ point of the graphitic BZ1 (Fig. 2-9 (a,b)). The QSH

phase contains singularities, while the insulating phase does not. The Z2 index is I mod2 . Because

I is the number of singularities, the Z2 index can also be written n = 12 (C- - C¯)mod 2 ,58 where

opposite Chern numbers for opposite spins in the TRI system result in a nonzero Z2 index.

35
 Another important tool is the time reversal polarization parameter, Pq .57 While the Chern

number is related to the charge polarization, which is a sum of partial polarizations

(for one dimension) over bands s Î {I, II} , the Z2 index is related to

Pq = P I - P II Eq. 2-26

Because of the gauge dependence of the Berry connection discussed previously, and the gauge

transformations generated by TRI, a partial polarization may be expressed as

. The gauge portion of the partial polarization can be

rewritten with the matrix wmn (k) tool as , and the time

reversal polarization is . This can be recompacted

through properties of wmn (Gi ) , ,

into . This easily integrates into

1 æç æ det [ w(p )] ö æ Pf [ w(p )] öö

Pq = log ç ÷ - 2 log çç ÷÷÷÷ , which can be rewritten to show
2p i çè çè det [ w(0)] ÷ø è Pf [ w(0)] øø

det [ w(0)] det [ w(p )] Eq. 2-27

(-1)Pq =
Pf [ w(0)] Pf [ w(p )]

Thus, the only tool necessary to show topologically significant time-reversal polarization is the

antisymmetric overlap matrix wmn (Gi ) . For two dimensions, this takes the form

4 det éëw ( Gi )ùû 4

(-1)n = Õ = Õdi Eq. 2-28
i=1 Pf éëw ( Gi )ùû i=1

where n is the Z2 index, which is 0 for trivial insulators and 1 for topological insulators.

36
 There are only a few extra considerations for 3D and inversion symmetric systems, and the

latter simplifies calculations considerably. The calculation for the Z2 index naturally evolves into
8
(-1)n 0 = Õdi . Here, n 0 is the ‘strong’ Z2 index. There are three other ‘weak’ Z2 indices, n 1 , n 2 , and
i=1

n 3 , for each dimensional axis. These are calculated by taking the product of four di at

Gi = 12 ( n1,i b1 + n2,i b2 + n3,i b3 ) for one n =1 and the other two are 0 or 1 (Fig. 2-10). The topological

systems in which n 0 = 0 but at least one weak index equals one are referred to as ‘weak’ TIs, and

can be interpreted as stacked 2D QSH states, which each have n =1 but sum to n mod2 = 0 . These

are not robust in the face of weak disorder, and thus weak TIs are undesirable for applications.

This stems from the fact that weak TI Fermi surfaces enclose more than one TRI momentum, thus

allowing singularities to reconnect between them. The strong TI is the only system that requires

Dirac cones at each surface. Weak TIs have surface bands that reach the boundary of the BZ1, and

include sides with no surface bands at all.27

Figure 2-10: Z2 index calculations, performed by multiplying the various di at TRI momenta.
The first ‘strong’ index is the product of all di , and the latter three are the ‘weak’ indices,
calculated by the product of di at the (100) face, the (010) face, and the (001) face, respectively.
The bottom row shows the (001) surface projections, which clarify the calculation of the third
weak index, and show possible Fermi surface intersections for each distinct topology. Adapted
from [27].

37
 Inversion symmetry simplifies Z2 calculations. The inversion operator is defined as

for Bloch Hamiltonian functions. The spin conservation distinguishes this from the

time reversal operator. Again, an important tool for these simplified calculations is the overlap

matrix

Eq. 2-29

Because the Z2 indices only depend on the product of time reversal polarizations at Gi , the previous

overlap matrix can be expressed as wmn (Gi ) = ym,G P(PQ) yn,G = xm (Gi )umn (Gi ) , where it should be
i i

noted that the wave vectors are complete Bloch functions, and not their periodic components

. These are still inversion eigenstates, with eigenvalues x n (Gi ) = ±1 (as is required for P 2 y = y ).

2N

Then the relevant Pfaffian at the TRI momenta is Pf [ w] = det [ w] = det[u ]Õx n , where 2N is the
2

n=1

number of occupied bands. Kramers degeneracy guarantees that each parity eigenvalue for the

band of a Kramers pair is equal which generates a eigenvalue product of 1. The Pfaffian, however,
M

may be rewritten Pf[w] = Pf[u ]Õx 2m , where M is the number of Kramers degenerate band pairs. In
m=1

order to address Pf [u ] , it is useful to note that because the Berry curvature is odd under inversion

and even under TRI, it must be zero56, and thus there must be a gauge in which the Berry

connection . Again, utilizing some properties of the overlap matrix produces

, and so the Pfaffian must be constant, i.e. . Thus, the time

reversal polarization for at each TRI point for inversion symmetric systems can easily be calculated

as

38
 det éëw ( Gi )ùû M
di = = Õx 2m ( Gi ) Eq. 2-30
Pf éëw ( Gi )ùû m=1

and the strong topological index for 3D, TRI, inversion symmetric systems is
8 M
(-1)n 0 = ÕÕx 2m ( Gi ) Eq. 2-31
i=1 m=1

Interestingly, all that is necessary to calculate the topology of such a system is knowledge of the

parity eigenvalue of occupied bands. This simplification is an essential facilitator for computations

of topology in real systems, including Bi2Se3.

2.4: Bulk-Boundary Correspondence, Surface States

While it is valuable to understand the physics that define topological insulators in their bulks,

the potential application value of these materials lies in the states that emerge at their surfaces.

This is where topological transitions take place, where exotic states emerge (barring Weyl

semimetal physics59–63 or other systems that contain topological nodes or lines in the bulk64,65),

and where the promising conductivity properties of topological insulators reside. The only way to

transition between systems containing differing topological variants is to close the band gap (as

can be seen in topological transitions within the Haldane model), and likewise, the only way to

transition between two media with different topologies is to close the band gap at the interface.

This generates linear dispersions in 2D (Fig. 2-1(c)), and Dirac cones in 3D (Fig. 2-1(d)), as will

now be shown.

Solutions involving a topological transition at an interface resemble the solution of the Dirac

equation generated by Jackiw and Rebbi66 for a system involving mass inversion at an interface.

The Dirac equation is a relativistic quantum mechanical equation containing linear momentum

39
terms. In solid systems, parabolic bands depend on quadratic momentum, and thus the Dirac

equation applied to topological solids can be written

Eq. 2-3232

where and b are Pauli (or Dirac for 3+ dimensions) matrices and B has 1
m units. This is already

similar to the Hamiltonian of the Haldane model for , b =sz , , and

other similar momentum terms. This system has a band gap of mv 2 , but without the quadratic

momentum term, the energy spectrum of this system is symmetric under m ® -m , and topological

information is lost. The quadratic term thus behaves as a vacuum “reference”, giving topological

significance to the trivial case, and to the non-trivial case by distinction. With the quadratic term,

topological information is encoded in the sign of mB , i.e., mB > 0 is a topologically non-trivial

system, and mB < 0 is a topologically trivial system (like vacuum), as shown in Fig. 2-11.

In one dimension, the Dirac-like system can be written as h(x) = vpxs x + (mv2 - Bpx2 )s z . The only

viable edge solutions to this Hamiltonian have zero energy. For the Dirilecht condition

Figure 2-11: Topological spin orientations in momentum space, as derived from the modified
Dirac equation. (a) For mB < 0 , spin is the same for p = 0 and p = ¥ , and tracking the spin
orientation along px , for example, shows a rotation toward x̂ and then back to ẑ (twice). (b)
Whereas for mB > 0 , spin makes a complete rotation from corner to corner. The topological
distinction is seen in the relative spin orientations at p = 0 and p = ¥ . Adapted from [67].

40
 C æ sgn(B) ö - x/x+ - x/x-
y (0) = y (¥) = 0 , topological solutions come in the form ysgn(B) (x) = ç ÷(e -e ) , where
2 çè i ÷
ø

, and C is a normalization constant. This solution extends readily to higher

dimensional systems with four-component wave functions:

Eq. 2-33

In the case of 2D at the x-boundary, . The Hamiltonian for this

system is NG = Dn mod2 , and the energy dispersion is Ep ,± = ±vpy sgn(B) where the sign indicates both
y

the spin and velocity direction. This energy dispersion shows helical conduction. Dirac cones at

3D system boundaries have the same solution, but with a gauge shift can be written as

Eq. 2-33

py
where tan q = , and conduction remains spin-momentum locked. Predictably, the energy
pz

dispersion here is E p ,± = ±vp^ = ±v px2 + py2 in the case of a z-boundary. These edge solutions are all
^

products of mass inversion, which can be seen in all of the topological models, which have at least

one negative mass singularity. As previously discussed, these inverted mass singularities are

topologically protected, thus resulting in topologically protected (and TRI protected in the Z2 case)

surface states.

All of these topological systems display linear surface states (Fig 2-1(b-d)). Further, the

topology of these systems can be seen directly from the surface bands, which cross the Fermi

41
energy between Kramers degenerate momenta at an odd number of points in the bulk band gap,

whereas topologically trivial states contain even numbers of EF band crossings at the surface (Fig.

2-12). Topological transformations can be directly assessed from these surface band crossings as

NG = Dn mod2 , where N G is the number of band crossings between Kramers momenta and Dn is

the difference between Z2 invariants at the interface. In the case of the QH state, there is only a

single linear surface band. In the case of the QSH state, the bands are required to cross at a Kramers

momentum point, thus only allowing one Fermi level intersection in the gap. This applies to 3D

Z2 TI materials, such as Bi2Se3, as well.

42
Figure 2-12: Surface conduction and band schematics of topological materials. (a,b) Generic
band structures of topologically trivial and non-trivial TRI materials, showing surface bands
extending between Kramers degenerate momenta. (a) Topologically trivial systems host pairs
of surface bands between the Gi and Gj points, and a mid-gap EF intersects leads to an even
number of surface states. This pairwise connection can be eliminated by pushing the bound
states out of the gap. (b) Topological materials host single surface bands between Gi and Gj ,
resulting in odd numbered surface states at EF, which are topologically protected. (c)
“Skipping” conduction states at the QH state interface, and (d) its linear, spin degenerate,
unidirectional surface band. (e) Charge neutral spin conduction in the QSH state stemming from
(f) its linear, spin-oriented surface bands. Adapted from [34].

43
Chapter 3: Bismuth Selenide, its Topological Properties, and its Defects

Bismuth selenide has been described as the “ideal topological insulator” due to its large band

gap (0.3 eV), the central location of its single surface Dirac cone, and its stoichiometric chemistry.
18,30
The large band gap diminishes the effects that doping levels have on the generation of metallic

states in the surface or on the bulk. For normal doping conditions, this band gap allows Bi2Se3 to

display its unique topological surface state behaviors at room temperature, which is not the case

for all other TI materials. The Dirac cone at the center of the momentum dispersion allows TSS

conduction in any direction on the surface. Other TI materials have Dirac cones on the sides of

their BZ1, such as Bi1-xSbx in which the linear surface states are projected from the bulk L point.29

In such systems, isotropic behaviors are not expected, and attempts to utilize the surface states

must account for the orientation of the lattice. The Bi2Se3 Dirac cone shows minimal deviation

from a circular shape (at any given mid-gap energy), allowing precisely isotropic surface

conduction, or equal conductivity at arbitrary in-plane surface directions. In general,

stoichiometric compounds are easier to fabricate than alloys, which require significant control and

tuning efforts to achieve the desired composition. These factors make Bi2Se3 a promising TI

candidate for applications as they lead to minimized metallic conductivity and facilitate

fabrication.

Its uses extend beyond typical implementations of conductive layers in transistor or memory

devices. Its topology and helical surface states create some highly unusual phenomena that can

have potential uses in spintronics or optical phenomena. Most of these make use of the p Berry

phase or surface helicity. They are discussed in detail in Ch. 3.3.

Bi2Se3 is still at the nascent stage of its development toward practical purpose, however. There

are a variety of issues that plague the crystal, and inhibit measurements of the TSSs. The main

44
barrier toward implementation of this material is the metallic states that tend to emerge on the

surface and in the bulk of the crystal. There are many examples in the literature that show

degenerate n-doping.28,30,38,68,69 When EF rises to the conduction band in the bulk, spin-unpolarized

carriers dominate transport measurements due to the higher density of states of bulk bands, and

the TSSs are unobservable. At the surface, band bending causes 2D electron gas formation. This

has a similar effect to degenerate n-doping, as it creates conductive metallic states at the surface

with higher DOS levels than the TSSs. While TSSs still exist, their behaviors are not detectable

above the higher densities of other filled states. Therefore, none of the phenomena discussed in

Ch. 3.3 are observable under these conditions. Surface oxidation is another distinct defect of

interest. This has been shown to correlate with 2DEG formation40 as well as decreased TSS signals

in Shubnikov-de Haas oscillations70. Further, the oxide layer of Bi2Se3 creates an insulating barrier

to electrical contacts, as will be discussed in Ch. 5. Before any progress can be made toward

producing devices that contain Bi2Se3 for the purpose of using its topological properties, these

defects need to be addressed in further detail than is currently available in the literature.

3.1: The Physical and Electronic Structure of Bi2Se3

Bi2Se3 is a three-dimensional crystal with a hexagonal unit cell containing 6 bismuth and 9

selenium atoms (Fig. 3-1). As such, this unit cell is three times the volume of the primitive cell,

which belongs to the trigonal space group R3m . Its lattice parameters are 4.14 Å in the in-plane

directions and 28.64 Å perpendicular to the plane. Each unit cell is composed of three vertically

(along the out-of-plane c-axis) stacked quintuple layers (QLs). These QLs each contain five

vertically stacked atomic layers, SeA2 -BiB -Se1C -BiA -SeB2 , which are covalently bonded in an octahedral

configuration and rhombohedrally stacked at (A, B, C atomic sites. Each QL is terminated by a

45
selenium atom, separated by a van der Waals (vdW) gap of 2.42 Å71, and also rhombohedrally

stacked. This continuity of this rhombohedral stacking results in a screw axis-like structure of

atomic layers. These vdW-gapped QLs lead to a natural cleavage plane at the (0001) hexagonal

face, terminated by a Se atomic layer. The Se atoms are relatively light, and play the role of electron

acceptors (as will be addressed in Ch. 2.7), while the Bi atoms are heavy, and are electron donors.

The heaviness of the Bi atoms has another major effect in this system. It generates strong spin-

orbit coupling, which is the genesis of the topological distinction of Bi2Se3.

Spin-orbit coupling (SOC) is a relativistic atomic effect, in which the electron in its own rest

frame experiences a dynamic electric field produced by the moving nucleus, as well as a magnetic

Figure 3-1: Schematic of the Bi2Se3 hexagonal unit cell. Each unit cell, which contains six Bi
atoms and nine Se atoms, is three times the size of a primitive cell. Its lattice parameters are
4.14 Å in the in-plane directions and 28.64 Å perpendicular to the plane. Each unit cell is
composed of three QLs

46
field . Substituting momentum for velocity, using the radial dependence of a central

electric field, and expressing that field as a potential results in . Using the angular

momentum provides , and the first order perturbation Hamiltonian contribution

, where is the Pauli matrix spin indicator in a 2-band system and l

is the spin order parameter, which describes the strength of the SOC. In solid systems, this has the

effect of splitting some bands with opposing spins. For bands with orbital angular momentum

quantum number L > 0 (e.g. p-, d-, f-, etc.), this effect produces significant splitting, but has no

effect on the s-like bands. The magnitude of the energy split is proportional to the atom’s atomic

number to the fourth power, Z 4 . Looking at some relevant atoms, and comparing the Z 4 ratios of

their splitting strength shows that before finer considerations, D Bi

D Se = 36 , D Bi
DSi =1, 235 , and

D Bi
DC = 36, 619 . This large SOC is the basis for the broad presence of Bi in prospective TIs (BiSb,

Bi2Se3, Bi2Te3, Bi2Se2Te, etc.) This interaction maintains TRI ( QL ® -L and Qs ® -s ), and is

symmetric under inversion ( PL ® L and Ps ® s ), which means that the Z2 topological

categorization with the inversion consideration holds (as was the case for the Kane-Mele QSH

model). Thus, systems with TRI, inversion symmetry, and SOC can still be categorized according

to their band parities.

The bands of Bi2Se3 can be calculated from density functional theory by initially organizing

the atoms in their sites, and subsequently turning on chemical bonding (stage I), crystal-field

splitting (II), and finally applying SOC (III) (Fig. 3-2). The focus of Zhang et al.18 was to do these

calculations near the G point, where Eg is at a minimum. This optimizes a search for SOC-induced

band inversion, in which bands of opposite parity may exchange positions around EF, thus resulting

47
 Figure 3-2: Inversion of bands with opposite parity in DFT calculations of Bi2Se3. (a) Steps (I-
III), showing the energy effects of interatomic bonds, crystal field splitting, and SOC on the
opposite parity orbital states. The energy of the states with opposite parity crosses during the
SOC step, specifically at x » 0.58 (b). This results in net negative Kramers band parity at the G
point, driving the system into a topological insulator phase. Adapted from [18].

in a sign switch in the parity of occupied bands and a topologically nontrivial band structure. As

the net parity of core bands is zero, only the states near the Fermi surface need to be considered.

Prior to chemical interactions, these are Bi 6s26p3 and Se 4s24p4, and the s-orbitals are ignored.

Organizing the orbitals of a primitive cell (2 Bi and 3 Se) by parity results in two odd and one even

state from each Se p orbital, and one odd and one even state from each Bi p orbital. Chemical

bonding (I) hybridizes these states, pushing down ‘acceptor’ Se and lifting up ‘donor’ Bi orbitals.

The applied crystal field (II) splits the in-plane from out-of-plane states according to Bi2Se3’s point

group symmetries, and the pz states are left closest to EF. Finally, the energy is turned

on (III). Because the SOC mixes spin and orbital angular momenta while preserving the total

angular momentum, there is a repelling effect between P1+z ,- and P1+x+iy ,¯ . Opposite repulsion

effects on neighboring P1+z ,- (¯) , P2-z ,- (¯) states results in band inversion at the G point, and the

M
opposite parities of these bands results in d ( G) = Õx 2m (G) = -1. However, because there is no band
m=1

48
 M
inversion at any other Kramers momentum d ( Gi ¹ G) = Õx 2m (Gi ) = +1 , and the total product of time-
m=1

8 M
reversal symmetries at is negative (-1)n = ÕÕx 2m (Gi ) = -1 , and thus n 0 =1 and Bi2Se3 is a strong
0

i=1 m=1

topological insulator. The magnitude of the SOC that generates band inversion can be seen in

Figure 3-2(b), where the SOC is increased from zero to its empirical magnitude as l (Bi) = xl0 (Bi)

and l (Se) = xl0 (Se) , where l0 (Bi) =1.25 eV and l0 (Se) = 0.22 eV. The level crossing occurs where

x » 0.58 , and through the process the gap energy changes from Eg » 175 meV with non-inverted

bands to Eg » 300 meV with inverted bands. The strong SOC, therefore, produces a mean band

deviation of D » 300+175
2 = 237.5 meV, and generates the largest band gap of the 3D TI material class

discovered so far.

The surface states of Bi2Se3 can be calculated from a low-energy model, taking the

Hamiltonian up to k2. Considering a four-state system for two spins and two bands, centered near

the G point,

Eq. 3-1

where k± = kx ± iky , , and . All parameters except are fit

to the appropriate energies via an ab initio calculation. This system is representable by the

appropriate choice of Dirac matrices in the as-mentioned modified Dirac equation,

Eq. 3-2

Given ( M, B1, B2 ) > 0 , one can see from the equation for that the sign of the Hamiltonian of

this system inverts from k = 0 to k = ¥ , as expected from a TI. In this formulation, the required

49
 Figure 3-3: Band structure of Bi2Se3, including all modifications to the Hamiltonian. (a) Band
structure in BZ1 between high symmetry points of Bi2Se3. SOC is included. For an undoped
system, EF is in the middle of the gap at , and the band gap is 300 meV. (b) Surface band
structure centered at . The topological nature of Bi2Se3 produces a spin-polarized (not shown)
surface Dirac cone. Adapted from [18].

time ( Q = K ×isy Ä s 0 , where K is the complex conjugate operator), inversion ( P = s0 Ä s 3 ), and

three-fold rotation ( C3 = exp( i3p sz Ä s 0 ) ) symmetries are apparent, and the Hamiltonian has uniaxial

anisotropy along kz . The surface states can be calculated by keeping the quantum numbers k x and

k y , but substituting momentum for kz . The solution here is to the 1D ( ẑ ) modified Dirac equation

projected to the surface boundary, with the solutions given in Sec. 2.4, and the previously given

power expansion of k x and k y around the G point. The low energy surface Hamiltonian becomes

æ 0 A2 k- ö
H surf = ç ÷ Eq. 3-3
ç A2 k+ 0 ÷ø
è

where the two states are for opposite spins, and each comprised of a superposition of P1+z ,- (¯) and

P2-z ,- (¯) . The energy dispersion is E = ±A2 kx2 + ky2 , and the states are spin-momentum locked (Fig.

3-3). The velocity at these Dirac cones is ms-1.

There is an apparent issue with this surface dispersion. This comes from the fermion doubling

theorem, which states that having Hall conductivity in integers of the magnetic flux quantum

50
requires an even number of Dirac points in TRI systems.72 When a magnetic field is applied to a

strong TI such as Bi2Se3, the surface Hall conductivity should be quantized in half integers,

. However, for experiments with physical leads on the top and bottom of a slab sample,

both surface Dirac cones are inherently measured, and the conductivity ‘doubles’, and appears as

an integer again. This coupling can only be surmounted with an experiment that can decouple these

surface measurements. This ability to generate a single Dirac fermion without violating TRI is a

unique property of the strong TI transition.

3.2: Characterizations and Applications of Bi2Se3

The topology and surface states of Bi2Se3 create some very interesting physical phenomena,

although they can be difficult to measure. Among these are high mobility and backscatter

protection in surface transport, weak antilocalization, unusual Shubnikov-de Haas oscillations, a

surface quantum Hall effect, topological magnetoelectric effects, and Majorana fermion

emergence at superconductor interfaces. These lend to a variety of applications, some of which

were discussed in Ch. 1. However, before these can be harnessed and manipulated, experimental

confirmation of topology and its accompanying surface states is necessary.

Various research groups have measured the surface Dirac cones of Bi2Se3 with angle resolved

photoemission spectroscopy (ARPES) (Fig. 3-4). This has become the standard method of

observing topological surface states (TSS) in physical samples, due to its surface sensitivity and

the wealth of band structure information it provides. In this experiment, (further explained in Ch.

4) photoemitted electrons are detected from their respective energy bands at the surface as a

function of momentum and energy, thus effectively displaying the surface band structure of the

51
 Figure 3-4: Spin-polarized Dirac cones, experimentally measured with ARPES. (a, b) Dirac
cones with at 300 and 10 K, indicating the robustness of these surface states caused by
SOC and band inversion up to room temperature. (c) An n-doped surface dispersion, where the
Fermi level ( ) is in the conduction band. (d) A cartoon schematic of the spin-polarized
Dirac cone. Spin polarization was measured to 90° momentum locking ± 5°. Adapted from [28].

material. In Bi2Se3, while the Dirac point is not necessarily located at the intrinsic EF level, it is in

the band gap, and doping (intentional or not) can shift EF to the Dirac point. The more common

Fermi level shift is toward the conduction band, as will be explained in Ch. 3.3. In general, the

bulk bands, which contain high electron densities, emit a large number of photoelectrons, and are

seen at the high- and low-energy limits of most published ARPES spectra. The surface bands,

which have a significantly lower DOS (although the ratio of detected densities is a function of the

surface sensitivity of the measurement), emerge from the bulk bands, connecting the bulk valence

and conduction bands through the Dirac cone.

The band structure measured in ARPES lends significant credibility to the theory that predicts

the rest of the emergent physical phenomena. Among the most experimentally challenging of the

remaining characteristics is the lack of direct observation of the absence of backscattering. In

theory, there is no backscattering in conducting Dirac cone states because the state traveling

in the opposite direction, but with the same spin, , does not exist, and as primary scattering

effects do not flip carrier spin, the transition between these states is impossible.34 In an effort to

52
 Figure 3-5: Measured and theoretical LDOS distributions on the surface of Bi0.92Sb0.08. (a) The
measured FT-STS data clearly resemble the spin-dependent SSP prediction for scattering,
rather than the spin-independent JDOS prediction, indicating the validity of the model that
includes the spin transfer matrix, and indicating the lack of backscattered carriers. (b)
Schematic showing possible and impossible scatter events as a function of spin and momentum.
The integration of these probabilistic, spin-dependent scattering events over BZ1 generates the
SSP. Adapted from [73].

infer the existence of this effect, an experiment from the Cava group showed that the measured

spatial distribution of carriers on the surface of Bi0.92Sb.08 matched with the distribution predicted

by the spin-dependent scattering probability.73 As the Bi0.92Sb0.08 system is subject to the same

band inversion physics as Bi2Se3, this indicates that the physics that drives both of these

topological phases can cause Bi2Se3 to exhibit backscatterless transport. This experiment

investigated the periodicity of the spatial distribution of the local local density of states (LDOS)

on the Bi0.92Sb0.08 sample surface with dI/dV scanning tunneling spectroscopy (STS) (Fig. 3-5).

Fourier transformed LDOS maps were then compared to spin-dependent scattering probability

predictions,

Eq. 3-4

where the transfer matrix T is just the overlap matrix of spin states with different momenta. This

decreases as spins variance increases. When integrated over the whole BZ1, the SSP indicates

which states should be filled with given probability, in a manner that’s highly sensitive to spin

53
polarized modes. The control for this experiment is the spin-independent joint density of states,

, which is the same as SSP for a transfer matrix of unity. The FT-STS

spectra match the SSP prediction and not the JDOS, allowing the control to be rejected, and

providing strong evidence that backscattering in these 3D topological systems is suppressed.

Weak antilocalization (WAL) is a purely quantum effect in the diffusive transport regime in

which magnetic field-dependent scattering leads to decreased conductivity (Fig. 3-6). In most

topologically trivial TRI materials without significant SOC, carriers that are scattered in opposite

direction along a closed contour are subject to opposite phase acquisitions over the loop. However,

the overlap of these TRI contours creates constructive interference of the oppositely traveling

states, and thus these scattered carriers have increased wave function amplitudes in localized areas,

and conductivity is lowered. When a magnetic field is applied, this phase overlap is complicated

by the TRI-breaking magnetic field phase factor, constructive interference is lowered, and the

localization effect is negated. However, the presence of a Berry phase in topological materials

Figure 3-6: Weak antilocalization in TRI materials. (a) In the diffusive transport regime,
carriers may be scattered around a closed loop. (b) In the case of most materials, the overlap of
the contours that propagate oppositely creates constructive interference, weak localization, and
low conductivity. This effect is mitigated by an applied magnetic field. (c) For systems with a
p Berry phase such as the surface of Bi2Se3, this phase change is destructive rather than
constructive, leading to weak antilocalization and high conductivity absent a magnetic field.
(d) Experimental measurements of the weak antilocalization signature in the
magnetoconductivity of Bi2Se3 as a function of applied field angle. Adapted from [67] and [74].

54
causes an additional phase change of p along the scatter contour for oppositely directed carriers.

Thus, these same TRI scattering contours generate destructive interference in the absence of a

magnetic field. This is the weak antilocalization effect. In the local area of the scattering contour,

the deconstructively interfering states have lowered amplitudes, and conductivity increases

accordingly. The magnitude of this effect can be expressed as ,

where Y is the digamma function and Lf is the phase coherence length. This effect is dependent

on the angle of the applied magnetic field relative to the TI surface, as the phase change of

traversing carriers is dependent on the relative angle between the contour normal and the applied

field. In real systems, in which bulk is often a factor, thicker samples and higher applied fields can

often result in positive magnetoconductivity.

Among the more useful measurements of carrier transport in topological insulators is high field

magnetoconductivity and Shubnikov-de Haas (SdH) oscillation analysis. Incident high magnetic

fields on these systems induce Landau levels around the Dirac point, and varying the field strength

spreads the levels as , where the 0th level is pinned at the Dirac point,75 barring

very high field strengths that can open a band gap. When EN crosses EF, the system’s carrier

concentration changes, and its transverse ( s xy ) and longitudinal ( s xx ) conductivities oscillate, as

seen in the integer QHE. The difference between the energies in this quantization scheme (and the

Landau levels in the integer QHE), , may give early indication of a measured system’s

topology. Additionally, when B increases to push EN through EF, SdH oscillations occur (Fig. 3-

é æ öù
7). These can be modeled with the Lifshitz-Kosevich theory77 as Ds xx = A0 RT RD RS cosê2p ç F - 1 + b ÷ú
ë èB 2 øû

, where F is the frequency of oscillations in 1

B , b is the phase offset ,

55
 Figure 3-7: SdH oscillations and analysis. (a) Oscillations in Rxy (transverse resistance) show
the energetic shift and angle dependence of Landau levels. (b) Plotting the LL trend against 1 B
shows the frequency F as a linear trend, and the x-intercept b gives g = 0.8p , close to the
expected Berry phase. The insets show Ds xx and its Fourier transform for F. (c) Dingle analysis
of SdH oscillation amplitude gives TD, from which the surface mobility is calculated. The inset
shows the temperature dependence of the oscillation amplitude and calculations for the
cyclotron parameters. Adapted from [76].

, and RS = cos( 12 p gme mc ) are the temperature, Dingle, and spin damping

factors.78 g , w c , and mc are the electron g-factor, cyclotron frequency, and cyclotron mass, and

the Dingle temperature gives the scattering energy from which the carrier lifetime can

be inferred. The frequency of SdH oscillations can be used to measure the

surface carrier concentration and Fermi wave number, the phase offset 2pb = g can be used to

calculate the Berry phase, the temperature trend of SdH oscillation amplitude can be used to

calculate for comparison against theoretical and ARPES band structures, and the

magnetic field strength trend can be used to calculate the Dingle temperature to

extract the surface mobility. These experiments have been conducted by multiple groups in order

to show well-behaved Dirac fermions, and to measure the insulating quality of the bulk. Because

56
the magnetic field may only induce states perpendicular to its direction, varying the angle between

the magnetic field and the surface can decouple the contributions from 2D surface and 3D bulk

carriers.

Bi2Se3, perhaps unsurprisingly, displays an unusual surface QHE. In relatively low magnetic

fields, surface Hamiltonian H = vF ( pxs y - pys x ) produces a half integer quantized Hall conductivity,

e2
s xy = - ( N + 12 ) .79 The half-integer portion can be understood to be a result of the p Berry phase.78
h

Experimentally, it is very difficult to measure this half quantized property because each surface

contributes equally to Hall measurements, which doubles the effect and results in whole integer

conductivities. In a high enough magnetic field, Zeeman splitting opens a band gap and the

Hamiltonian gets a massive contribution, H = vF ( pxs y - pys x ) + Ds z . The Kubo formula can again be

sgn(D) e 2
used to give half quantized hall conductivity at zero temperature s xy = - at the surface.
2 h

This is, of course, subject to the same experimental difficulties, as the top and bottom surfaces of

a slab geometry often combine to produce integer results.

From the topological field theory,80 an unusual magnetoelectric term appears in the TI

æ ö æ ö
Legrangian, s xy = 1 ç eE 2 - 1 B 2 ÷ + ç a 2 ÷q E × B . In ordinary materials q is zero, and all of the electric
8p è m ø è 4p ø

and magnetic behaviors can be described by the common Maxwell eqautions. However for Z2 TIs,

the term q = p , and the equations for electric polarization and magnetization become

Eq. 3-5

and thus an electric field can induce magnetization and a magnetic field can induce

electric polarization in an otherwise unaffected system. Unfortunately the ballistic

57
 Figure 3-8: Predicted and measured Kerr rotation in thin Bi2Se3. (a) q K as a function of ratios
of the surface band gap D and the incident frequency w to the bulk band gap e c . A step
function is expected at low energies, under which the Kerr rotation is exactly - p 2 . (b)
Measured Kerr rotation from a 16 QL sample at 5 K a gap-opening field of 10 T.
shows good results, but may more closely match the predictions for EF nearer to the Dirac point.
Adapted from [81] and [82].

surface states tend to foil any attempts at static polarization, although using Zeeman splitting to

open a gap in the surface states may surmount this issue. This magnetoelectric signature can also

be measured optically, via giant Faraday or Kerr rotation. These effects cause the polarization of

transmitted (Faraday) or reflected (Kerr) light to rotate from its incident polarization direction to

a significant degree. Defining, for example, the Kerr rotation as qK = ( arg E+r - arg E-r ) 2 , where + (-)

indicates the right-handed (left-handed) polarized light component, and using b = 4p Res xy and the

reflected field expected from a thin TI film such as Bi2Se3 , the Faraday

rotation is preditcted to become qK = -tan-1 b -1 » - p 2 at low frequencies and given a magnetic field-

induced surface band gap D > 0 .81 This has been measured to as high as 65° (Fig. 3-8), which is

larger by an order of magnitude than rotations typically found on high mobility GaAs

heterostructures.83

One of the strangest effects predicted to occur in Z2 TI materials is the emergence of Majorana

fermions at a TI/superconductor interface.35 Majorana fermions are their own antiparticles, and

58
their existence has never been confirmed, despite being predicted in 1937. They may only exist in

particle-hole symmetric, spin non-degenerate systems, and the former criterion is readily produced

in superconductors, but the latter has proven difficult to achieve simultaneously. The rigorous

physics of these states, as well as their applications, are far beyond the scope of this work, but

researchers have considered promising for quantum computation for more than a decade.84

3.3: Defects in Bi2Se3 and their Effects on Surface States

Despite all of these promising emergent phenomena the surface of Bi2Se3 and the potential

applications they should allow, the measurement and manipulation of topological surface states

(TSSs) has proven very difficult at nearly every step, even at the lab scale. Growth and

maintenance conditions of Bi2Se3 need to be strictly controlled in order to minimize structural and

chemical defects. Even in the most tightly controlled environments, there have been no reports of

films devoid of polycrystallinity or doping, and they are all subject to oxidation and band bending,

as discussed in Ch. 1.

The main methods of growing Bi2Se3 are by the Bridgman method, chemical vapor transport

(CVT), and molecular beam epitaxy (MBE). The former two produce large, high quality chunks,

but are not applicable to wafer deposition, and are therefore only useful for research purposes. The

latter growth method can be applied to wafers (given an appropriate substrate), but shows

polycrystallinity with large grain sizes being on the order of 10 μm grains, and more often < 500

59
nm in length.85–87 The grain boundaries may act as scattering walls, accumulating bulk charge88,

changing the bulk gap,89 and imposing a limit on the phase coherence length of topological carriers,

limiting the desired ballistic transport and high mobility properties of Bi 2Se3. These grain

boundaries may appear as twins such that the angle between them given by their normalized

hexagonal in-plane lattice vectors f12 = cos-1 ( â1 × â2 ) = p 3 , generate ‘zipper’ boundaries for the

angles in between 0 and p 3 , or may cause a vertical edge dislocation and laterally truncated QLs.

MBE also shows screw dislocations37 caused by spiral growths, and can induce sufficient strain

alter the surface states, depending on the substrate and growth quality (Fig. 3-9). These strain

Figure 3-9: MBE growth of Bi2Se3 on SiC (0001) with an epitaxial graphene layer. QLs are
intact, but grains are grown below 100 nm, have rounded edges, and exhibit clear screw
dislocations. (a) A SiC miscut step causes grain structure irregularities. (b) Step edges appear
to generate spiraling vertical growth. (c) These screw dislocations can occur at significant
densities. (d) Domains are slightly rotated between neighboring domains of a spiral growth.
Adapted from [37].

60
effects may raise or lower the Dirac point relative to the Fermi level, and could thus either help or

hinder attempts to lower the carrier density.90

Band bending is another major complication in attempts to measure the TSSs. The bulk bands

tend to lower toward Bi2Se3 surfaces and interfaces. This causes the conduction band to intersect

EF, generating an electron gas within 10 nm91 of the surface which is thus referred to as the two-

dimensional electron gas (2DEG). In micromechanically exfoliated samples, the bands decrease

over time after cleavage, and the 2DEG moves to lower energy and appears to increase

Figure 3-10: Band bending-induced 2DEG, growing over time in Bi2Se3, as shown by ARPES.
(a) After cleavage of a degenerately n-doped sample, the Dirac point is measured at 0.3 eV
below EF, and the diffuse ARPSE signal above the Dirac cone is due to the filled bottom of the
conduction band. (b) The sample is remeasured after three hours. During the interim, the bands
have bend downward, lowering the Dirac point to 0.5 eV, and creating a lower rim of significant
intensity below the conduction band. This is growth of the 2DEG near the surface, existing
simultaneously with the TSSs. Adapted from [91].

in occupation density (Fig. 3-10). Rashba effects may subsequently remove the spin degeneracy

of the 2DEG92, potentially generating confusion between the spin-polarized TSSs and 2DEG

61
 Figure 3-11: Generation of the Rashba spin-split 2DEG, as shown by ARPES. (a) The
degenerately n-doped Bi2Se3 spectrum is initially devoid of the 2DEG. (b) With time, the
surface bands shift down and the 2DEG forms. (c) After further doping, Rashba splitting moves
2DEG with opposite spin orientations into different bands. (d) Plan view diagram of Rashba-
induced spin polarization. Adapted from [92].

states. Low energy electron diffraction shows that in these samples, there is no change in

crystallography at the surface, implying that chemical changes in the bulk, defect formation, or

surface impurity adhesion drives the 2DEG formation.91 Intentional band bending has been

conducted in these materials by exposing the surface to carbon monoxide at a constant pressure of

8´10-9 mbar, resulting in 2DEG emergence state after 20 minutes.93 A similar experiment was

performed with applied water vapor, producing a 2DEG with between 18 and 54 Langmuir of

exposure.40 In thin films, this can break the inversion symmetry of the bulk and destroy the TSSs.94

In 2D-sensitive conductivity measurements (usually for TSSs) such as SdH oscillations, it provides

a source of confusion as the additional source of 2D surface states makes it difficult to filter

between the TSS from the bulk signals.70 Because one of the especially promising uses of Bi2Se3

is in spintronics, wherein spin conduction of surface states is desired, the 2DEG, spin degenerate

or not, poses a major threat to the spintronics performance of Bi2Se3. This is because even in

Rashba split surface states, bands of spin up and spin down occur on either side of the gamma

point (Fig. 3-11), and cannot produce spin-filtered carriers for spintronics applications. In the case

62
of the spin degenerate 2DEG, any attempts to gather spin-filtered TSS carriers are overwhelmed

by the higher current densities of the 2DEG. While in specific samples the 2DEG can be

counteracted an electric field effect,95 the inconsistency of the usually incidental 2DEG emergence

threatens the viability of Bi2Se3 usage in integrated circuits.

Somewhat connected to the issue of the 2DEG is natural bulk n-doping of the crystal. This is

largely caused by Se vacancies ( VSe•• )96,97, which act as double electron donors, and to a lesser extent

by Se anti-site defects ( Se•Bi )69,98, which act as single electron donors. Both of these formations are

thermodynamically favorable, and doping is difficult to prevent unless growth is done with excess

Se under appropriate conditions. Degenerate n-doping (or p-doping, which is the worst case

scenario as no Dirac cone states are filled, although this is rarely an issue) in Bi2Se3 can move the

bulk EF to move into the conduction band. When this happens, the entire material behaves

metallically. Quantum Hall effects are overwhelmed by other contributions, and SdH oscillations

are almost entirely generated by 3D states causing Berry phase offsets to be zero. High

concentration spin-degenerate metallic states dominate TSS spin conduction, and WAL is replaced

by weak localization from the bulk, etc. Once the bulk behaves as a metal, Bi 2Se3 is rendered

useless as a host to TSSs. n-doping is also problematic in slighter, but nondegenerate levels, as this

can enhance the surface 2DEG and shift the EF upward along the Dirac cone. When the entire

material’s EF is shifted up, less band-bending is required to produce the quantum well for the

2DEG states. Further, slight n-doping can eliminate the spin-polarized nature of Dirac cone

conduction, as the top and bottom halves of the Dirac cone contain inverted helicities. The surface

spin conduction is likely to drop off as EF moves away from from ED proportionally to the Dirac

cone spin-polarized DOS,

63
 Eq. 3-6

where E0 represents the surface area of the lower half of the cone, and EF = 0 at the Dirac point.

Because TSSs are the only helical and ballistic states, bulk conduction hinders the potential of

Bi2Se3 for spintronics use. In order to mitigate n-doping and its deleterious effects, epitaxy of

Bi2Se3 often involves significant overpressure of Se vapor in order to minimize vacancies, but bulk

carrier densities are still typically measured to 1019 cm-3. Other attempts to counteract n-doping

have included the introduction of Ca38,99 and Cd100 dopants to the bulk, and growing doped, non-

stoichiometric alloys such as Pb-doped (Bi2-xSbx)Se3 (Fig. 3-12). These extrinsic dopants may

introduce extra issues in the form of point defects that act as scattering centers, reducing the

electron mobility.99

Surface oxidation in Bi2Se3 is another complicating factor. While some of the proof-of-concept

TSS measurements, such as ARPES, allow in situ cleavage and measurement to be completed in

vacuum and at low temperatures, many others, such as electrical conductivity measurements that

require leads to be attached after sample deposition or cleavage, require or are facilitated by

measurement at room temperature and in air. It beehoves researchers to consider the effects of

these ambient conditions, as the specific materials science involved is of interest to academia, and

the potential consequences to device performance as a result of ambient exposure are of interest to

industry. Surface oxidation has is an expected consequence of exposure to air, as it is a natural and

thermodynamically favorable process. It has been called “the likely material origin of surface

degradation and environmental doping.”70 Kong et al. proposed that on both nanoribbons and bulk

crystals, moderate doping ( n £ 5´1018 cm-3) should cause SdH oscillations to come from the bulk

64
 Figure 3-13: Bi (a,c) and Se (b,d) oxides growing in air at the surface of nanoribbon (a,b) and
bulk (c,d) Bi2Se3 samples, as measured in XPS at the Bi 4f and Se 3d energies. This report
shows Bi oxide growing immediately after cleavage, although this specific process is debated
by other reports. This oxidation process is suspected to be the cause of extrinsic doping in
Bi2Se3. Adapted from [70].

Figure 3-12: Some attempts at extrinsically p-doping Bi2Se3 in order to move EF toward ED.
(a) Plot of some recorded dopants and their effect on the bulk resistivity of Bi2Se3, measured at
4 K. The most insulating film had a Bi replacement level of 0.125% Ca. (b) Hall-measured
carrier level in variously doped Bi2Se3 samples. The lowest carrier concentration was measured
in the same Ca-doped sample. (c-e) ARPES measurements of variously Ca-doped samples,
taken 20 minutes after cleavage. Higher dopants cause an n- to p- shift, and EF drops below ED.
(f-h) Momentum distribution curves of the ARPES measurements. (i-j) Diagram of the Ca
doping effect. (l, m) Typical ARPES spectrum and momentum distribution taken 18 hours after
cleavage. (n) Band-bending diagram responsible for the changes shown in (l, m). Adapted
from [28].

contributions as n3D
23
. Their SdH measurements resulted in oscillations that indicated in electron

SdH
densities of n3D = 4.7´1018 cm-3, indicating only moderate doping. However, their Hall

Hall
measurements indicated bulk concentrations of n3D =1.7´1019 . The discrepancy was explained by

the inability of low-mobility carriers to react to and participate in SdH measurements, and that

these slow carriers were a result of environmental doping. Subsequent x-ray photoelectron

spectroscopy (XPS) investigations into the chemical changes that occur on the surface reveal

oxidation to be the primary source alterations in Be2Se3 surface chemistry (Fig. 3-13). The reported

growth of BiOx and SeOx, as seen in the Bi 4f and Se 3d XPS spectra, respectively, was observed

65
in both bulk and nanoribbon samples. In both systems, BiOx was shown to form immediately after

cleavage, and SeOx was shown after 2 days of exposure to air. The universal applicability of the

oxidation behavior reported in reference [70] is the subject of debate. Other measurements,

including those included in this thesis, taken on the Bi 5d XPS peaks, show oxidation occurring at

a different time scale. These reported BiOx peaks are more likely to be Se 3p peaks, which overlap

with Bi 4f. Regardless, because oxidation has been shown to contribute to environmental doping

in semiconductors such as Ge and GaN, it is expected that oxide formation may have the same

effect in Bi2Se3. The mechanisms here may include surface charge trapping in the oxide or at the

Bi2Se3/oxide interface, or that oxide formation causes interfacial defect states that pin the Fermi

level, and induce band bending. Oxide formation may be generated by a variety of deposited

oxygen-containing species. Pure O oxidation may occur, and is likely to be a p-doping process.101

Following the controlled deposition of water vapor, Bi2Se3 may react to form H2Se gas and a

Bismuth hydroxide surface:

Bi2Se3 +6H2O ® 3H2Se­+2Bi(OH)3 Eq. 3-7

This reaction was presented to explain Se vacancy formation, the resultant n-doping, and surface

band bending.40 However, the same band-bending (and Rashba splitting) seem to occur in samples

66
 Figure 3-14: Similar Bi2Se3 surface band response from exposure to different environments,
as shown by ARPES. (a) When exposed to residual UHV gases, (b) when cleaved in air, or (c)
when subjected to 324 Langmuir of water vapor, the surface bands seem to bend and form a
Rashba-split 2DEG in similar fashions. Thus, while surface exposure to water vapor may, in
part, be a source this band bending, it is likely not the only cause. Adapted from [40].

that were left in residual vacuum gas at 3´10-10 mbar, and samples that were cleaved in air (Fig.

3-14). If oxidation is cause of Se vacancy formation and band bending, it is essential to that the

oxidation process and the oxide layer be well understood. Some studies report that cleaved surfaces

are highly resistive to oxide growth.36,102,103 These studies show differences in growth, cleavage

area, and surface stability. The growth method in these reports was to melt a Bi2Se3.01 mixture in

an evacuated fused quartz ampule, as to avoid a sequence of peritectic reactions in Bi-heavy

phases. Subsequently, they were annealed and recrystallized with the Bridgman method, resulting

in ~4 cm3 sized Bi2Se3 rods. These samples were left in an air environment over two weeks and

subjected to a variety of measurements that showed the surface to be inert. XPS showed correct

stoichiometry and no oxidation with XPS at the Bi 5d, Se 3d, and Bi 4f energy ranges (Fig. 3-15),

AFM measured near atomic flatness (barring few QL steps), STM topography showed the

expected interatomic spacing for a Se-terminated surface, and ellipsometry showed strong

67
 Figure 3-15: A chemically inert Bi2Se3 surface. (a) A cleaved surface shows expected peaks in
the Bi 4f (also Se 3p) range, the Se 3d range, and nothing at the O 1s range. This spectrum is
consistent for a surface that has been exposed to air for 1 month (b), as well as a surface that
has been dosed with NO2 and O2 (c). The trend changes when the surface is etched with argon
and exposed to NO2 and O2 (d). The resultant XPS spectra show oxidized Bi as well as a O 1s
peak. Adapted from [102].
absorption near 2.14 eV (as will be discussed later).36 Subsequent experiments showed inertness

after cleavage in the face of air exposure as well as NO2 and O2 dosing, and were only oxidized by

a combination of Ar+ etching and NO2 + O2 dosing. This was explained with DFT calculations that

show a minimal enthalpy change by oxidation reactions that arent initiated at Se vacancies.102 The

differences in reported oxidation behavior are surprising. One reference attributes this to

insufficient mixing during growth of bulk samples.36 Some studies also claim that cleavage

techniques can affect the surface quality, and possibly induce defects.102 It is clear that oxidation

has not yet been sufficiently studied, that conflict remains in the literature, and that the dynamics

of the oxidation process are not well understood.

68
Chapter 4: Characterization Method Description

In order to get a complete picture of the changes at the surface of a material (Bi2Se3 in this

case), it is almost always necessary to use a variety of experimental techniques. A combination of

optical, chemical, electronic, and topographically sensitive measurements are used in this study in

order to characterize the oxidation and Se decapping process of Bi2Se3.

4.1: X-Ray Diffraction

X-ray diffraction (XRD) is a well-developed method of gathering information about periodic

materials, such as crystals. It has been an essential tool in materials science since its discovery and

development by von Laue in 1912, and is used as a primary tool in measuring the lattice

parameters, point group, and chemical properties of unknown crystals, nanostructures, and

composite materials. Because absorption of x-ray wavelengths is relatively low, with typical

attenuation lengths of ~65 μm in Si and ~4 μm in lead, XRD is largely a bulk measurement method,

and provides structural data for large volumes (typical probe size ≫ crystal unit cells) and density

of defects. Compton scattering, an inelastic process, provides information about the attenuation

length of materials, but the main use of XRD is in measuring elastically scattered rays, which exit

the material at the same wavelength at which they were emitted. The periodicity of crystalline

samples allows XRD analysis to provide information about the chemical composition, as the

intensity of reflections is a function of the electron density and distribution in a material, and

geometry of a unit cell. Likewise, deviations from expected single crystal XRD patterns may be

used to measure perturbations in crystallinity. These deviations may include extra diffraction spots

or unexpected spot intensities, and may be caused by polycrystallinity or atomic defects such as

interstitials, vacancies, contamination, or other effects.

69
4.1.1: Theory of XRD

XRD is caused by coherent elastic scattering from periodic structures. The most essential

concept underlying XRD is Bragg’s law. Bragg’s law describes the conditions under which parallel

atomic planes can reflect x-rays while maintaining phase coherence. As the phase acquired by an

x-ray traveling between planes is dependent on its angle relative to the planes, as well as the

distance between the planes, this diffraction requirement also determines the angles at which x-

rays at given wavelengths may be diffracted by crystals with given interplanar distances (Fig. 4-

1). For adjacent planes, a ray must be able to travel some distance to the top plane for reflection,

and also be able to travel that distance plus a distance of

nl = 2d sinq Eq. 4-1

to be reflected from the bottom plane and return to its original separation from the top reflection

while having its phase changed by . This is the required condition for coherent scattering, and

is the starting place for many XRD measurements.

Figure 4-1: The Bragg condition indicates at what angles x-rays may be diffracted from parallel
crystal planes. d is the interplanar distance, and θ is the angle that the incident beam makes with
the planes. In this figure, in order to maintain coherent rays, the phase of the top beam must
match the phase of the bottom beam upon reflection. The extra distance travelled by the bottom
beam, 2dsinθ, must result in an integer number of oscillations n of wavelength λ. Thus,
diffractions occur for nl = 2d sinq .

70
 Figure 4-2: Demonstration of the relationship between real space planes and reciprocal lattice
vectors, using the (231) plane of an orthorhombic lattice. The plane is given in real space by
connecting , , and . Its reciprocal vector is . is an essential tool to
calculating sample orientations for diffraction measurements.

In order to determine the complete structure, including the space group, of a crystal, it is

essential to consider the three-dimensional periodicity of the sample, and not just simple surface

oriented planes. This is simplified with considerations of the unit cell and its symmetries, which

may be used to generate a reciprocal lattice, which is an essential map to determining sample

orientations that correctly orient crystal planes for diffraction. The unit cell is described in real

space by its lattice parameters, a1 , a2 , and a3 , where the sum of arbitrary integers of lattice

parameter results in a symmetric translation, and leaves the system unchanged. A reciprocal lattice

is a representation of crystal planes in reciprocal space, each of which may be represented by a

vector (Fig. 4-2). Plane (hkl) has coordinates in reciprocal space

Eq. 4-2

where is the reciprocal vector, hkl are the miller indices denoting that this plane has real space

nodes at , , and , and . Each plane in the unit cell has a reciprocal lattice

vector that gives diffraction direction and Bragg angle, but further considerations must be made to

71
calculate diffraction intensities and forbidden diffraction planes. This is largely accomplished with

the structure factor.

The structure factor gives the amplitude of diffraction that a crystal plane produces. It is a

weighted Fourier transform of the unit cell, calculated as:

Eq. 4-3

where atom i is at real space coordinates , is the normalized position

of an atom in the unit cell, and is the plane being considered. is the atomic

scattering factor of atom i that scatters light with wavelength l at angle q , and can be calculated

as , where is the electron probability density function and

is the scattering vector, pointing along the bisection of the incident and reflected

beams. The intensity of a scattered wave can be calculated from the structure factor of the plane

as . Therefore, the structure factor is essential in predicting and analyzing diffraction

patterns. It contains all information about a crystal lattice, but because only intensity is measured,

phase information is lost in conventional diffraction measurements. This causes confusion in back-

calculating lattices from diffraction patterns, because while distances between atoms may be

accurately measured, atomic positional offsets are lost with phase information. For crystals with

known lattices, the structure factor provides essential information about expected diffraction

intensities.

Because the structure factor is zero for many planes, caused by destructive interference of

diffracted waves, diffraction is forbidden. This is only the case for non-primitive lattices. Primitive

lattices may diffract for a plane with any Miller indices. However, for example, the structure factor

predicts that the diamond cubic lattice will not produce diffractions at planes where the Miller

72
 Figure 4-3: Forbidden (002) diffraction in Si from one diamond cubic unit cell. Each plane
may reflect the incident X-ray, and every other plane may reflect coherently, but the
interspersed planes reflect at a π phase offset, creating destructive interference and eliminating
the diffraction signal.

indices have a mix of even and odd numbers or where sum of the miller indices h + k + l = 2(2n +1)

for any integer n. In the (002) diffraction, x-rays do reflect coherently from every other plane, but

the ones between also coherently reflect with a phase offset of exactly p , thus causing destructive

interference and eliminating the diffraction (Fig. 4-3). Rules like these greatly can simplify

structural measurements. Further, measurements involving forbidden diffraction planes are

convenient because they don’t require any calculations for electron density-dependent intensities,

and can still be highly informative about the symmetry of a crystal sample. They may obviate full

Patterson analysis in tests for single crystallinity. If significant diffraction is measured where a

family of forbidden planes is predicted, it is likely that point defects are not the cause.

Polycrystallinity is a common cause of supposedly forbidden plane measurement, due to the

potential variety of orientations of grains in the sample.

73
4.1.1.1: XRD in Practice

XRD data was acquired with a Bede Metrix-L diffractometer for this work. It uses a stationary

copper K-α source, rotating a point detector, three goniometers controlling sample Euler angles,

and has 3-axis high-resolution XRD (HRXRD) capabilities. This tool is designed to measure 300

mm wafers, but more irregular samples may be mounted to a 300 mm wafer and measured.

Because the tool is effectively ex situ, few other sample requirements exist. The main

measurements used in this work are -2θ and pole figure XRD scans. Omega-2theta scans are

performed by changing the detector angle at twice the rate of source angle such that the reflection

Figure 4-4: Demonstration of -2θ experimental configuration and reciprocal space scan
direction for a Si sample with a (001) surface. (a) The detector is scanned at twice the rate of
the source, 2 dtd w = dtd 2q , such that the bisectrix of the reflection is maintained. If they are at the
same position, as is shown, then the scan runs perpendicular to the sample surface. (b) -2θ
reciprocal scans are representable in reciprocal space by linear vectors anchored at the origin.
Therefore, -2θ scans of sufficient range measure plane (hkl) and planes that are linearly scaled
from its reciprocal vector, .

74
bisectrix is maintined (Fig. 4-4(a)). They measure a straight path in reciprocal space, and thus can

gather signal from many sample planes with the same orientation (Fig. 4-4(b)).

-2θ scans are used to measure planes and reciprocal vectors with the same orientation. For

scans of sufficient range, the interplanar distance can be calculated with the Bragg condition, and

the AOI that produces diffraction peaks. However, because of forbidden planes, Bragg angle

information can be incomplete, and thus extra information in the form of the sample space group

or other diffraction data may be required to correctly index the measured planes and to calculate

interatomic distances. However, because angular information is often accurate to within 100

nl
arcsec, measurements of interplanar distance are as accurate as dhkl = , which for copper
2sin(q B ± d )

K-α measurements the Si (004) peak, (Bragg angle q B = 34.56° ), results in accuracy of

Si
d004 =1.357± 0.001 Angstroms, which gives the lattice parameter of Si to within 0.5 picometers. This

accuracy makes XRD an invaluable tool for the measurement of crystal structures.

The second important scan type used in this work is the XRD pole figure. This is a natural

complement to -2θ scans, because while the data it produces does not contain information about

interplanar distances (although a proper experimental configuration where interplanar distances

are taken into account is required to produce useful pole figure data), it is completely sensitive to

the direction of the reciprocal space vectors producing diffraction. It can simultaneously show

planes in any direction given that they have the same interplanar distance. This is an invaluable

tool for scans of polycrystalline samples, in which certain grain orientations may be preferred, and

each of which produce diffraction spot patterns in different directions. This scan is easily

demonstrated with the Ewald sphere (Fig. 4-5).

In order to measure a diffraction spot with the reciprocal space vector , the x-ray

configuration must satisfy

75
 Figure 4-5: The Ewald sphere and the allowed diffraction points of Si. The radius of the sphere
is taken to coincide with the 224 family of planes, which are plotted red. (a.) Three-
dimensional representation, showing that all 224 reciprocal vectors intersect the sphere.
Because is constant for the family, any XRD scan that maintains a constant angle between
the source and detector while rotating the sample can, in theory, measure all 224 diffraction
spots. (b.) Tomograph of (a.), showing the x-ray configuration required to measure the (224)
plane of Si.

Eq. 4-4

(Fig. 4-5(b)), where and are the ingoing and outgoing x-ray momenta respectively, and is

the reciprocal vector of the measured plane. This equation is valid for all elastic scattering events.

A pole figure is a scan at constant values of , only varying the Euler angles of the sample

with respect to the fixed source and detector. In Fig. 4-5 this could be seen by plotting a vector

from the origin to any point on the Ewald sphere, and rastering it both on azimuthal and elevation

axes to generate a solid angle of data. It thus produces 2D maps of diffraction spots with the same

magnitudes of reciprocal vector, and can measure all planes of any symmetric family given enough

scanning range.

For samples with expected plane magnitudes and directions, it is possible to calculate the

orientation of the sample, as well as the x-ray source and detector angles, to detect any diffraction

76
plane(s). The formula for interplanar distance for a completely generic (trigonal) crystal is

somewhat complex:

1
d2
= V12 (S11h2 + S22k 2 + S33k 2 + 2(S12 hk + S23kl + S13hl))

S11 = b2c 2 sin 2 a S12 = abc2 (cosa cos b - cosg )

S22 = a2c2 sin2 b S23 = a2bc(cos b cosg - cosa )

S33 = a2 b2 sin 2 g S13 = ab2c(cosa cosg - cos b ) Eq. 4-5

Here a, b, and c are the crystal lattice magnitudes, V is the unit cell volume , and α, β, γ,

are the angles , , and . However, this is significantly simplified for crystals

with more symmetry. For cubic and hexagonal crystals, it simplifies to the following:

1 h2 + k 2 + l 2
=
d2 a2

1 4 æ h 2 + hk + k 2 ö l 2 Eq. 4-6
= ç ÷+ 2
d2 3 è a2 ø c

This interplanar distance informs the user of the x-ray angular separation, via the Bragg

condition. The only other step is to predict the necessary orientation of the sample relative to the

surface normal, in order to measure the desired diffraction spot. This requires knowledge of the

azimuth and elevation (tilt) angles necessary to orient the diffraction planes normal to the desired

x-ray reflection bisectrix (Fig. 4-6(a)) so that the Ewald sphere condition is satisfied.

In actual XRD experiments, goniometer axes are fixed, and so to align a reciprocal vector with

, two separate rotations are often needed (Fig. 4-6(b))). Fortunately, because pole figures

are often performed with 360° range, only the tilt of the desired diffraction plane is required. Given

the common surface plane direction s = g001 , the tilt angle required is easily calculated as

. With the Bede Metrix-L tool, the x-ray angle given by

77
 Figure 4-6: Demonstration of planes with a sample that must be tilted with the sample in order
for diffraction to be measured. (a.) A simplified diagram showing x-rays impinging on a
sample, as well as some planes in the sample. These planes are tilted at angle θ relative to the
x-ray bisectrix, which is given by K and is initially normal to the sample surface. The sample
must only be tilted by θ for successful XRD measurement. (b.) For experimental measurements,
samples must be tilted azimuthally so that is in the n-s plane (where s signifies sample
surface direction, and n signifies sample in-plane notch direction, and is immaterial for most
scans), and then along the elevational axis to make //K for successful XRD measurement.

the Bragg condition and the tilt angle given by the Ewald sphere suffice to produce hemispherical

pole figures, as will be shown later for Bi2Se3.

4.2: Spectroscopic Ellipsometry

Spectroscopic ellipsometry (SE) is an optical technique measures the changes in light

polarization that occur upon reflection from (or transmission through) a sample surface as a

function of the energy of the light. Changes in polarization that occur at interfaces are caused by

interactions between the incident EM wave and the electrically polarizable sample surfaces, and

therefore, ellipsometry is sensitive to the dielectric properties of the media at the interface(s).

Further modifications to the incident light polarization arise in experiments that involve

transmission through materials, where light absorption and phase procession need to be accounted

for as the polarized light propagates through the sample. But with sufficient supplementary

78
information (usually coming in the form of at least a partial knowledge of the optical properties of

the sample materials), these experiments can give fast (< 10 seconds), accurate (often within 1

nm), and non-destructive information about dielectric or semiconducting layers, as well as thin

metallic films, including their thicknesses, roughness, electronic, and optical properties.

Ellipsometry has become an invaluable tool in the semiconductor industry for measuring deposited

thin films in rapid production environments.

4.2.1: Theory of Spectroscopic Ellipsometry:

In order to understand changes in polarized light that is reflected from a sample surface, it is

useful to understand the electronic response of the reflecting material, and to explore the physics

from which the dielectric function of the sample is derived. In optical experiments, the most

consequential physical process is light absorption, and accompanying direct excitation of electrons

to higher energy states. The band structure provides information about intra- and inter-band

transitions. For gapped materials, the number of available direct transitions as a function of

transition energy can be expressed via the joint density of states (JDOS)

dSk
ò Ñ (E
1 Eq. 4-7
D j (Ecv ) =
8p 3 k cv )

where Ecv = Ec - Ev is the energy difference between the valence and conduction bands. Fermi’s

Golden rule for a continuous state distribution (as in band theory),

Eq. 4-8

(where E(w ) is the incident electric field, Pcv is a quantum mechanical transfer matrix constant,

79
and ), uses the band structure to give the transition rate at a given light energy. In the first

formulation, the Kroneker delta implies that the incident light must have the same energy as the

band difference, but as k is treated as continuous in band structures, the discrete sum is replaced

by the continuous JDOS integral in the second formulation. Multiplying that rate by energy

to get power loss, and comparing to power loss in a simple Beer-Lambert rule,

dI d ac æ eiw öæ n 2 2ö ew
- = - ( I 0 e-a x(t ) ) = I = ç 2 ÷ç E(w ) ÷ = i E(w ) Eq. 4-9
2

dt dt n è n øè 8p ø 8 p

one gets a formulation for the imaginary portion of the dielectric function versus JDOS:

æ 2p e ö 2 Eq. 4-10
ei = ç ÷ Pcv D j (Ecv )
è mw ø

This formulation demonstrates that the imaginary portion of the dielectric function is a largely a

function of the number of available direct electronic transitions between valence and conduction

bands. Further, because ei (w ) ³ 0 and the dielectric function does not diverge, the Kramers-Kronig

(KK) relations may be used to compute the real portion of the dielectric function,
¥
w ¢ei (w ¢)
Pò
2 Eq. 4-11
er = 1+ dw ¢
p 0 w ¢2 - w 2

where P is the Cauchy principle value. Thus, a complete reconstruction of the dielectric function

is possible through some knowledge of the band structure. Further, these dielectric properties are

directly connected to the complex refractive index of the material as

Eq. 4-12

and therefore the band structure analysis provides a prediction methology for electronic behavior,

and light behavior, during optical/solid interactions.

80
 There are places in the band structure where many momentum-adjacent transitions may occur

at the same energy. These are called critical points, or van Hove singularities, because they occur

where D j diverges, which is where

Ñk (Ecv ) = 0 Eq. 4-13

At these critical points, ei reaches a local maximum, and appears as peak in the ei (w ) spectrum (or

d 2ei (w )
more rigorously, a local minimum in due to other potential contributions to ei (w ) ). Because
dw 2

these are highly visible features in ellipsometric analysis, they are very useful for comparison to

theory. In systems with parabolic band transitions, critical points are categorized according to the

curvature along their axes. The transition energy can be written

Eq. 4-14

where the signs of the alpha coefficients determine the curvature directions of the critical point.

All positive curvatures are classified as M0, mixed sign curvatures are classified as M1 or M2, and

all negative curvatures are classified as M3, for van Hove singularities in three dimensions. Near

M0 points Ecv increases in all directions, near M1 and M2 Ecv may increase or decrease depending

on the direction, and near M3 Ecv decreases in all directions. Typically, and as in Ge, there is a

M0 critical point at the center of the band structure ( G point). This is often the lowest direct

transition. Due to the cubic symmetries of Ge and alike crystals, the maxima in e 2 are often

generated by M2 critical points along the momentum-degenerate 100 and 110 directions (Fig.

4-7). The critical points transitions in these directions are called E1, distinct from G -point E0

transitions and the 111 -direction E2 transitions. Knowledge of these trends is helpful in analyzing

band structures based on dielectric functions extracted from ellipsometric measurements.

81
 Figure 4-7: Band structure and imaginary dielectric function of germanium. (a) Band structure,
showing critical points where the band k derivatives are equal, and the types of singularities are
labeled. (b) Dielectric function of Ge from theory (solid) and experiment (dotted). The E2
critical point is type M2, along the 6-fold symmetric (100) axis, and produces the maximum
value in the visible range. Adapted from [104].

However, because information about electron momentum is lost in linear optical measurements

(in that for optical processes that don’t include a 2-step phonon-mediated process, which are rare,

are independent of electron momentum), any feature in the measured dielectric function can only

be taken to represent an integration of JDOS contributions at a given transition energy over all

momenta. For complex band structures with multiple critical points at similar energies, assignment

of ei to one or more of these particular critical points is often impossible, as is the case in Bi2Se3.

Thus, while physical theory may support measurements of ei , complete confirmation of the band

structure based on is ei impossible, and changes to the band structure can’t feasibly be parsed

based purely on ellipsometric measurements of ei .

Another important parameter to consider is the optical conductivity of a material, s . This

gives the AC conductivity of the impinged material at the light frequency, J(w ) = s (w )E(w ) . This

parameter is above nonzero even for electrically insulating materials, which have DC

82
conductivities of 0 below band gap biases. The relationship between the optical conductivity and

the dielectric function is

s (w ) = -iw (e (w )- e0 ) Eq. 4-15

Polarization changes measured in ellipsometry are given by

rp Eq. 4-16
r= = tan YeiD
rs

Here, rp and rs are the complex reflectivities of p- and s- polarized light,

rp æ rp rp ö
Y = tan -1 D = d p - ds = arg çç ÷ Eq. 4-17
rs ÷
è rs rs ø

are the angle given by the ratio of reflected amplitudes, and the difference between phase changes

incurred by p- and s-polarized light upon reflection from a sample surface, respectively (Fig. 4-

Figure 4-8: Polarization of reflected light. This light, traveling in the +ẑ direction (toward the
rp
reader), has a left-handed ellipticity. Y = tan -1 is the angle given by the ratio of reflected p
rs
æ rp rp ö
and s amplitudes, and D = d p - ds = arg çç ÷ is the difference of phase changes incurred by p-
÷
è rs rs ø
and s- polarized light. For a linearly polarized incidence (in which Ei,max p and Ei,smax occur
simultaneously), D can be computed from the time between E max
f , p and E f ,s as D = w (t f , p - t f ,s ) .
max max max

83
8). The reflection and transmission that occur at a sample surface can be determined by the Fresnel

equations (as derived in the Appendix):

N1 N 22 - N12 sin 2 j - N 22 cos j 2N1 N 2 cos j

rp = tp =
N1 N - N sin j + N cosj
2
2
2
1
2 2
2 N1 N - N12 sin 2 j + N 22 cos j
2
2

N1 cos j - N 22 - N12 sin 2 j 2N1 cos j Eq. 4-18

rs = ts =
N1 cos j + N - N sin j
2
2
2
1
2
N1 cosj + N 22 - N12 sin 2 j

Here, N1 and N 2 are the complex refractive indices of the interface materials, and j is the angle

of incidence. For an experiment involving a flat substrate in air, N1 » 1 , and the Fresnel reflection

equations may be directly solved for N 2 . The dielectric function may then be analyzed for

information about the critical points of the substrate’s band structure.

4.2.2: Spectroscopic Ellipsometry Experiments

The measurements for this work were performed with a J.A. Woollam RC2 dual rotating

compensator ellipsometer (Fig. 4-9). It is a spectroscopic (210 to 1690 nm acquisition range),

variable angle (25 to 90° AOI) system, in which complete changes in polarization may be

measured with four optical components. Incident light may be set to any polarization with a

rotating polarizer and first rotating compensator, and after reflection from the sample, a second

rotating compensator and rotating analyzer select specific polarization intensities to be measured.

The dual rotating compensator design provides significantly higher precision in the determination

of D than rotating polarizer models. Further, depolarization may be assessed, thus providing

information about sample roughness or thickness inhomogeneity.105,106

In standard SE, reflection from a sample is represented by a Jones matrix S:

84
 Figure 4-9: Schematic of a dual rotating compensator ellipsometer. Light from a bright Xe
source is polarized at a given angle and ellipticity by the rotating polarizer and first rotating
compensator components. After reflection from the sample, polarization direction and
ellipticity are filtered by the second rotating compensator and analyzer before the intensity is
measured at all wavelengths by the detector.

Eq. 4-19

In standard ellipsometry, it is assumed that no cross polarization (i.e. transferrance between p- and

s- polarized light) takes place, which is valid for most unpatterned samples. SE is generally

performed near the Brewster’s angle of the sample, which is

n2 Eq. 4-20
q B = tan -1
n1

for transparent samples. At this angle, p-polarized light is completely transmitted through the

interface, and all reflected light is s-polarized, resulting in high precision ellipsometric data.

Once Y and D data are collected, a model may be generated to match the optical response of

the sample. This process was done with J.A. Woollam CompleteEASE for all of the ellipsometric

analysis in this thesis. CompleteEASE allows the user to construct a model sample composed of

epitaxial layers on a substrate, and to fit the model layers’ physical parameters (e.g. thickness,

refractive index or dielectric function, interlayer mixing or surface roughness, etc.) to match the

collected data. Because physical materials have KK-consistent optical responses, it is useful to use

85
KK-consistent oscillators to model the electronic response of materials. Two common physical

models used in ellipsometric analysis are the Drude model and the Lorentz oscillator. The Drude

function takes into account metallic behavior, wherein electrons are not localized, but experience

resistance in their medium. It may be derived with simple Newtonian considerations, taking into

account the impinging EM wave, and a drag coefficient: . Here, we use an

oscillatory , and take stationary solutions of x to match its frequency. The amplitude of x

is easily solved, and using the classical definition of the dielectric function,

D = e E = e0 E + P = e0 E + np = e0 E - nex0 , The Drude dielectric function may be written as

æ ne2 ö 1 w2
e (w ) =1+ ç ÷ 2 =1+ 2 p Eq. 4-21
è e0 m ø w - iwG w - iwG

where w p is the plasma frequency. As seen in the denominator, the dielectric function is complex,

and results in light absorption. The Lorenz oscillator has similar. It is used for locally bound states,

and thus has a “spring” center contribution: . This results in a dielectric

function

æ ne2 ö 1 w p2
e (w ) =1+ ç ÷ 2 = 1+ Eq. 4-22
è e0 m ø (w 0 - w ) - iwG (w02 - w 2 ) - iwG
2

where w 0 is the oscillation frequency of the bound state. Real samples have a variety of interband

and intraband transitions, each of which often produce their own contribution to the dielectric

function, and are therefore modeled with sums of oscillators that contribute to the complete

complex dielectric function as e (w ) = å fDm + gLn , where m and n index individual band transitions,
m,n

and fD and gL are the Drude and Lorentz dielectric functions, respectively. These two oscillator

types are often all that’s needed to accurately represent the electronic and optical behavior of a

sample, and are used to model various transitions in Bi2Se3.

86
 Once a model is made, in order to attempt a forward solution in which there are more model

parameters (e.g. m, Γ, ω0, and others for a Lorentz oscillator) than there are data points to fit it (Ψ

and Δ), it is necessary to perform error-minimizing procedures to match the parameterized model

to the acquired data. This is done with the Levenberg-Marquardt algorithm in CompleteEASE:

b̂ = min b S(b ) = min å(yi - f (xi , b ))2 Eq. 4-23

i

where b and xi are the set of parameters (as a vector) that define the model f and the independent

variable that produces a signal to match the measured output yi, and b̂ is the optimal parameters

that minimize the error S. This is solved by perturbing the model with a Jacobian Ji = ¶f (¶xb,b ) , which i

differentiates the model with respect to the various parameters, and setting the resultant differential

error to zero:

S(b + d ) » å(yi - f (xi , b ) - Jid )2 = (y - f (b ))T (y - f (b )) - 2(y - f (b ))T Jd + d T J T Jd ,

i

(J T J)d = J T (y - f (b )) Eq. 4-24

In order to vary parameters to match data without changing them excessively, a damping function

is required. With it, a solution for an error-minimized function is reached efficiently:

(J T J + ldiag(J T J))d = J T (y - f (b )) Eq. 4-25

4.3: X-ray Photoelectron Spectroscopy

X-ray photoelectron spectroscopy (XPS) is a highly surface sensitive technique that uses

x-rays to excite and eject core electrons in order to measure surface chemistry. Because sets of

binding energies of core electrons are unique to elements (Table 4-1), and bonding states produce

predictable core energy shifts, the detected electron energy spectra can completely identify all

surface species and their molecular states.

87
 Excitation Line 2s 2p1/2 2p1/2 L23M23M23
Energy (eV) 151 100 99 1394

Table 4-1: Binding energies in Si for Al K-α x-rays. Orbitals give photoelectron lines and x-
ray line transition notation gives an auger line.

Because the energy of emitted electrons is low, their extinction distance is quite low as well

(the inelastic mean free path of Si 2p lSi2 p » 1 nm). Thus, the majority of XPS signal is produced

from the top few nanometers of the sample surface, and signal intensity decays exponentially with

depth. This surface sensitivity is advantageous for samples with epitaxial layers, as composition

as a function of depth is measured fairly easily. This is facilitated with angle-resolved XPS

measurements (ARXPS), which make use of detector angle, and signal intensity to determine path

length and signal origin depth. Spectra are fit with appropriate Bremsstrahlung background models

and oscillator-like XPS and auger peaks, and fit to minimize error between the complete model

and acquired data. Quantitative analysis is facilitated with libraries of information about XPS

energies, chemical states, extinction distances, etc. from NIST.

4.3.1: Theory of XPS

XPS uses high-energy photons to excite and eject, and subsequently measure, electrons from

materials. The absorption of x-rays provides more than enough energy to allow core electrons to

escape from their nucleus, and sometimes, to travel sufficient distances through the sample to

escape the surface. The amount of energy required to eject a core electron from a sample (non-

bonding s, p, d…) is equal to the energy difference between its core state and the Fermi level,

called the binding energy, plus the energy difference between EF and vacuum, called the work

function. If any kinetic energy from a state i excited by an x-ray remains after it escapes from the

surface, it is measured by the XPS system.

88
 Figure 4-10: Emission process for photoelectrons and Auger electrons. (a) Photoelectrons
come from core states that directly absorb an incoming x-ray and are subsequently ejected from
the atom. (b) Auger electrons are produced by a secondary process, in which after a
photoelectron is ejected, a higher state relaxes to fill its place, and releases energy for nearby
electrons to absorb and excite out of their core state. Adapted from [107].

Eq. 4-26

This effect was originally described as the photoelectric effect by Albert Einstein, which won him

the Nobel Prize in Physics in 1921.

For a given element, the energy of a state at a given principle quantum number n can be

calculated to first approximation in a Bohr model form, but significant tuning is required to

calculate the correct, empirically measured energies. Sources of shifts in state energy include the

effects of the orbital and spin quantum numbers, j = l + s , as well as finer terms such as Zeeman

splitting, etc. Hydrogenic levels up through fine structure considerations are given by the

Sommerfield expression,

mc 2a 2 é a2 æ n 3 öù Eq. 4-27
Enj = - ê1+ 2 ç 1 - ÷ú
2n 2 ë n è j + 2 4 øû

where m is the reduced mass of an electron traveling around its nucleus, and a is the fine structure

constant. Fortunately, typical XPS peaks are only split by n, l, and s, producing individual peaks

such as Au 4f7/2, Si 2p3/2, C 1s, etc. Further, a well-documented history of XPS usage has provided

89
reliable maps of emitted photoelectron energies in a multitude of materials, so quantum mechanical

calculations are not needed to identify peaks. Because core states in given elements are not

drastically changed by the bonding states of the host material, and because core states in different

elements occur at different energies, the binding energies detected are highly indicative of the

elements present in the material.

Core electron energies are not completely independent of bond states, however. The relative

electronegativities of bound atoms either acts to attract or repel nearby electrons. When a valence

electron is attracted (repelled) to another atom, the core electrons below it experience less (more)

screening, which causes an increase (decrease) in their binding energies. This effect can shift the

energy levels of core electrons of up to a few eV. For example, the binding energy of Si in a pure

crystal is often recorded at ~99.5 eV, but the binding energy of Si in SiO 2 is ~103.5 eV. These

“chemical shifts” are too little to confuse different photoelectron peaks, but are easily large enough

to be measured, and thus can be used to identify the chemical state of each constituent element.

Core emissions are not the only ones that take place in XPS experiments (Fig. 4-10). Signals

from Auger electrons, x-ray satellites, x-ray ghost lines, shake-up peaks (Fig. 4-11), multiplet

splitting, energy loss lines, and valence states may also be used to identify the chemistry of a

sample.107 On the other hand, secondary peaks such as these may also be a source of confusion in

the identification of elements in a sample, and can greatly hinder efforts to quantify the

concentrations in a given material, due to peak overlapping that can be difficult to decouple.

90
 Figure 4-11: Example of shake-up lines (s) of the Cu 2p observed in CuO and CuSO4, but not
pure Cu. These can cause misidentification in peaks and elements, as they may overlap with
other photoelectron, Auger, etc. peaks. Adapted from [107].

The energies of collected electrons are not the only information gathered in XPS. The intensity

of electron emission at each measured energy peak can also be used to calculate the concentration

of the elements in the material. These intensities are a function of the x-ray brilliance, the

absorptive properties of the given electron orbitals, the ability of an electron to travel to the sample

surface, and the sensitivity of the detector. For a homogeneous sample, the signal intensity is a

substantial, but relatively simple function

I = nf sq yl AT Eq. 4-28

Here, I is the measured intensity, n is the volumetric density of atoms of the element, f is the x-ray

photon flux, s is the photoelectric cross section for the orbital of interest, q is an angular

efficiency factor based on the experimental configuration, y is the efficiency of the photoelectric

91
process for formation of photoelectrons of the normal photoelectron energy, l is the mean free

path of the electron in its host material, A is the area of the sample illuminated, and T is the

detection efficiency for electrons emitted from the sample. The majority of these factors can be

grouped together such that I = nS , where S is called the atomic sensitivity factor. In experiments,

these sensitivities are a required part of the quantitative calibration process. They may change from

system to system and from material to material. However, well-known sensitivity factors allow a

relatively simple calculation to be made for the atomic fractions within a sample

Ci = ni ån = ( I S ) å( I
j i i j Sj ) Eq. 4-29
j j

There is, however, an additional problem of separating a photoelectron (or Auger, etc.) peak

from the background spectrum. This superimposed continuous signal is from Bermsstrahlung, or

braking radiation, and contributes to all XPS signals. It is caused by electrons that are being slowed

down and emitting photons as their kinetic energy is lost. Power loss density of traveling electrons

is modeled with the quantum electrodynamic Thomas-Fermi function, but the total background

spectrum in XPS can simply, generally, and empirically be described by Kramers’ law:

æ l ö1
I(l )d l = K ç -1÷ 2 d l Eq. 4-30
è lmin ø l

hc
The low wavelength (high energy) cutoff is given by lmin = , where V is the x-ray tube
eV

accelerating voltage. There are a variety of methods to decouple this background spectrum, so that

true peak intensities and areas may be measured, and quantitative analysis may be performed

accurately.

Many XPS systems are equipped with angle-resolved detectors. This enables angle-resolved

XPS (ARXPS), which provides information about samples with compositions that change with

depth. Electrons are able travel through their host material approximately according to a Beer-

92
 Figure 4-12: Intensity of electrons emerging from depth d01+d12 as a function of their emission
and detection angle. The exponential decline with increasing angles allows depth analysis in
XPS measurements.

Lambert trend, I = I 0e-x l . Here, l = lr (E) is the mean free path of the electron, which is specific to

its energy and the medium through which it’s traveling. Because an electron has to travel up to the

surface in order to be detected, the intensity of electrons that escape is dependent on the angle they

æ d ö
take toward the surface as I = I 0 exp ç- ÷ , where d is the depth from which the electron is
è l cosq ø

initially emitted, and q is the angle that the electron’s path makes with the surface normal (Fig.

4-12). It is important to examine this effect on multilayered systems. The intensity that results from

a buried layer n is

æ 1 n d ö
I = I 0 exp ç -
q
å l (E)
i
÷ Eq. 4-31
è cos i=1 i ø

Here, di is the vertical distance to the layer (or sample surface) above (Fig. 4-12). This is a bit of

an oversimplification, as the initial intensity itself can be a function of emission angle. However,

for known mean free paths, these relationships can be used to calculate the depth from which an

elemental signal originates, and can thus be used to non-destructively find the thicknesses of

93
distinct thin layers. For substrates with single epitaxial layers, the low angles often have far higher

bulk intensities, and the high angles have far higher epitaxial intensities.

4.3.2: XPS Equipment and Measurement Description

XPS and ARXPS data were collected using a Thermo Scientific Theta ProbeTM system using

monochromatic Al Kα (1486.6 eV) x-rays of 400 μm spot-size. All samples were loaded

simultaneously into the UHV chamber, and the measurements were made at a base pressure of

8´10-9 Torr. The measurements were taken with an energy step of 0.05 eV and at pass energy

settings of 100 and 50eV for Bi 5d and Se 3d peak regions, respectively. The energy resolution is

estimated to be 1.0 eV from the full width at half maximum of the Au 4f7/2 core level. In the

absence of a well-defined C 1s peak after exfoliation, the energy scale of the XPS spectra was

calibrated with respect to the Se 3d5/2 core level (53.7 eV) according to the value in the National

Institute of Standards and Technology (NIST) database. We estimate the uncertainty with all peak

binding energies to be 0.05 eV from peak fitting. ARXPS data were measured by collecting

photoelectrons in 16 channels over an angular range of 24.875 –81.125 using a spherical sector

analyzer fitted with an electrostatic lens having 60 angular acceptance and an angular dispersive

multichannel detector.

4.4: Transmission Electron Microscopy

Transmission Electron Microscopy (TEM) is a multifunctional technique that, while time

intensive, can provide a wealth of information from a localized area regarding crystallinity,

structural defects, chemistry, and electrical behavior. Many of these can be done simultaneously.

As an imaging technique, its combination of fast acquisitions and sub-angstrom resolution are

94
unique, although some scanning tunneling microscopes may image a surface with similar

resolution. Recently, advanced ‘intelligent’ software has enabled automated sample preparation

and data acquisition, drastically decreasing the human time and effort requirements to take these

data. While these experiments are still time intensive and destructive, and cannot be used in-line

to assess samples at a manufacturing scale, this new automation is facilitating the expansion of

TEM systems into production-level quality assurance and research and development, making them

ever more relevant and in demand.

4.4.1: Theory of TEM

For an electron to transmit through a sample, it must be traveling with extremely high kinetic

energy, and the sample must be very thin. In order to project an electron at such rates, a TEM uses

accelerating voltage in the range of 100 kV – 1 MV. This results in a relativistic velocity and

wavelength

é æ ö ù
-1 1/2 -1/2
eV é æ eV öù
v = c ê1- ç +1÷ ú l = h ê2m0 eV ç1+ ÷ú Eq. 4-32
êë è m0 c 2 ø úû ë è 2m0c 2 øû

For an accelerating voltage of 200 keV, the de Broglie wavelength is 2.5 pm, which gives a

0.61l 0.61l
Rayleigh resolution limit of dR = » = 2.2 pm. However, no electron source or lens
m sin b b

systems behave ideally, and the aberrations in real TEM systems each effectively increase the

probe size. These may originate from the source, d0 = 2 I 1 = C0 (where I is the current, b is the
p ba a

brightness, and a is the convergence angle), from spherical aberrations, dsph = Csa 3 , and from

DE
chromatic aberrations, dchr = Cc a . The final resolution of a TEM can thus be approximated as
E

d = d02 + dsph
2
+ dchr
2
+ dR2 . While picometer resolution may not yet be attainable in modern TEM, sub-

95
angstrom features may be resolved with aberration correction. This process involves the use of

extra lenses and monochromators which each must be tuned individually.

A TEM can be operated in two modes. Conventional TEM is performed with a beam that is

parallel upon incidence with the sample. The exiting beam is scattered (at varying intensities) into

a diffraction pattern, which may be measured, or some diffraction spots may be selected with

apertures to produce images. Propagated diffractions create images that may contain varying

information depending on the choice of spot(s), and are generated by different forms of contrast.

Many TEMs are also capable of operating in scanning TEM (STEM) mode, where the beam

converges to a small point in the sample in a conical shape (Fig. 4-13). In some ways, this mode

is analogous to an ultra high-energy scanning electron microscope. Like TEM, this mode produces

both a direct beam and scattered electrons that are still able to transmit through the sample, but

apertures are replaced by detectors to generate different types of images.

Amplitude contrast imaging in TEM and STEM can be taken in bright field (BF) or dark field

(DF) mode. The former is a measurement of electrons that pass directly through the sample without

being scattered, and the latter involves selecting one or more scattered electron beams to propagate

toward the detector to create an image. BF and DF mode are each capable of displaying mass-

thickness contrast. This is a mode in which elastic Rutherford scattering events decrease the current

DI
transmitted through the sample as =1- e-Qt » Qt , where Q is the total elastic scattering cross
I

section (which takes into account atomic weight, number density, and the single atom cross

section) and t is the sample thickness.108 Additionally, DF STEM imaging itself can be performed

in two different ways, depending on the exit angle of the scattered transmitted beam. In addition

to elastic scattering events that deflect the beam at at relatively low angles, inelastic scattering

events generate highly deflected beams that may be detected to generate images. These are often

96
called ‘Z-contrast’ images, as the intensity of inelastically scattered electrons is proportional to Z2,

making this technique highly sensitive to heavier elements. This may be performed in annular dark

field (ADF) or high angle annular dark field (HAADF) mode. Both of these modes utilize disc-

shape detectors around the direct beam to collect electrons scattered above ~10 mrad. ADF was

developed earlier, and functions by collecting electrons scattered at an intermediate angle (

10 < q < 50 mrad), and may be susceptible to signals from diffracted electrons that can deflect at

lower angles. HAADF collects electrons inelastically scattered at a high angle ( > 50 mrad), and

is a commonly used tool in high resolution STEM imaging. To produce these images, the STEM

beam rasters a (aberration corrected, in the high resolution case) sub-angstrom sized probe across

the sample, and electrons scattered from individual atomic columns are detected, thus producing

an atomic-resolution direct image of the crystal lattice, wherein brightness correlates with heavier

elements.

4.4.2: TEM Experiments

The TEM used in this work FEI Titan3 G2 microscope, operating at 300 keV in HAADF STEM

mode with spherical aberration correction. The source is a field emission gun (FEG) equipped with

a monochromator, which in this case, is used to decrease the beam intensity at all energies rather

than to filter for a narrow energy band. This lowered current level facilitates high resolution

measurements. Emitted electrons are accelerated into the column, where 3 condenser lenses

generate a highly focused probe. Before the beam impacts the sample, a spherical aberration

corrector further decreases the probe size to sub-Angstrom levels needed for high resolution STEM

imaging. The beam then interacts with the sample, deflecting off of atomic columns, and

97
inelastically scattered electrons are collected by the HAADF detector to produce atomic-resolution

images.

4.5: Atomic Force Microscopy

Atomic force microscopy (AFM) is a high-resolution, potentially non-destructive

topographical measurement technique. It can produce maps of surface features that contain sub-

nm resolution height information over > 1000 μm2 surface areas, generate 3D profiles of nanoscale

features such as FinFETs, identify biological formations with unique structures, etc. It is an

essential tool for any topographical studies, and can be used to measure surface feature dimensions

with higher throughput (although lower resolution) than TEM, and higher resolution (although

lower throughput) than optical methods. Its ability to measure surface textures makes it highly

valuable for fabrication processes that critically depend on line edge roughness or line width

roughness.

4.5.1: Theory of AFM

In topographic scanning modes, AFM uses a vibrating, tipped cantilever to detect a sample

surface level with high accuracy. The sharp, pyramidal tip is located under one end of the

cantilever, and the other end is mounted to the microscope. The most common type of

measurement is called ‘tapping mode’ AFM, in which the cantilever is vibrated at its resonant

frequency and placed near the surface. Meanwhile, a laser is reflected off the top surface of the

vibrating cantilever, and a photodetector detects the frequency and amplitude of these vibrations.

(Fig. 4-14) Highly sensitive piezo motors in the z-direction move the cantilever down until its

vibrations cause they tip to lightly tap the sample surface. The force that the surface generates on

98
 Figure 4-14: AFM operation. (a) The tip interacts with the sample while a laser reflects off its
back and into a detector. (b) The amplitude and frequency of the tip’s vibrations are measured
during the scan. As the height of the surface increases, the force on the tip increases, and its
vibration amplitude decreases. As the height of the surface decreases, the force on the tip
decreases, and its amplitude increases toward the resonant frequency peak.

the tip causes the cantilever to deviate slightly from its maximum amplitude and its resonant

frequency, f0 = 21p k
m0 , where m0 is the effective mass of the cantilever and k is its spring constant,

k = Ewt
3

4l 3
(E is the cantilever’s Young’s modulus, w is its width, t is its thickness, and l is its length).

These deviations are corrected by a feedback loop that uses a PID error minimization algorithm to

adjust the tip height and keep the amplitude of oscillation at its set point. This set point is not

directly at the center of the resonant frequency peak during scanning, but (typically) just below it,

which allows the microscope to discriminate between amplitude changes due to higher tip force

levels (which lower the amplitude from the resonant peak) and lower tip force levels (which raise

the amplitude toward the resonant peak). The force that the surface generates on the tip is caused

by changes in z-height at topographical features. The high sensitivity of this cantilever amplitude

modulation detection mode and the PID correction algorithm help generate highly sensitive

topographical surface maps.

99
4.5.2: AFM Measurements

AFM measurements in this thesis were performed with a Dimension Icon AFM in tapping

mode. The probes used in topographical measurements were OTESPA aluminum back-coated,

etched silicon tips from Bruker. These have resonant frequencies near 300 kHz, and the aluminum

back coating helps create stronger reflections. Scans were performed at 512 or 1024 px2 resolutions

at rastering frequencies below 1 Hz. No damage was measured on any of the samples during

acquisitions. Some images contained highly regular corrugations that were assumed to be artifacts

due to their coexistence with more defined features of expected sizes and depths, such as QL steps.

These may have been caused by the slight surface tilt of exfoliated bulk samples. These were

removed with Fourier filters in analysis.

4.6: Temperature-Programmed Desorption

Temperature-programmed desorption (TPD) is a method of tracking surface desorption rates

as a function of temperature. Because rates can be tracked at UHV pressures, this technique is

capable of measuring the vapor pressure of thin epitaxial layers. Analysis with TPD can be

informative about the kinetics of a desorption event, the coverage of adsorbates on a surface, and

with a sufficiently accurate system, it can provide the activation energies of individual desorptions.

This may be employed to find the energy of binding sites at the surface of samples.

4.6.1: Theory of TPD

Desorption is a thermodynamic process in which surface molecules evaporate into the

surrounding gaseous environment. The rate at which molecules desorb is dependent on the energy

100
of their surface bond and the temperature of the system, and is given by an Arrhenius relationship,

n des = n 0 exp -EkT . For homogeneous materials simply evaporating into the surrounding environment,
a

the surface evaporation rate (in molecules cm-2 sec-1) is given by the surface molecular density s

and their desorption frequency,

E = n dess Eq. 4-33

For systems in which a surface has a specified fractional coverage of evaporating adsorbates, a

model may created in which evaporation rate is dependent on fractional surface coverage. The

basis for this is given by the Polanyi-Wigner (P-W) equation,

r(q ) = - ddtq = -q nn = -q nn 0 exp kT

Ea Eq. 4-34

Here, q is the fractional surface coverage and n is the kinetic order of the desorption event.

Different n values correspond to different kinetics of desorption e.g. n = 0 gives the equation for

bulk or multilayer desorption in which the desorption rate is independent of coverage, n =1

describes direct desorption from a surface in which the desorption rate is linearly dependent on

coverage, and n = 2 gives the equation for a multi-step surface desorption process (for example,

recombination of two H atoms to desorb as the gaseous H2 molecule). The n = 0 case results in

evaporation rates that are purely dependent on the atomic surface density and the Arrhenius

desorption frequency, and the evaporation rate reverts to Eq. 4-33. The P-W equation (Eq. 4-34)

describes the rate of surface desorption as a function of time.

TPD is generally performed by linearly increasing the temperature of the sample i.e.

T(t) = T0 + bt . This allows substitution of the time variable, dq

dt = ¶T
¶q ¶T
¶t = b ¶T
¶q
, and the P-W may be

rewritten

101
 ¶q
= qb n 0 exp -E
n
a Eq. 4-35
¶T kT

In this study, the desorbing material is a bulk Se cap, which experiences a zero order kinetics, and

simplifies the equation to ¶q

¶T = nb0 exp -E
kT
a
. This describes what fraction of the surface will desorb per

second per increase in degree Kelvin. Desorption rates are directly proportional to the partial

pressure they create, ¶q

¶T µ ¶T
¶P
, which is an easily measurable thermodynamic variable. The kinetics

of desorption can then be derived with pressure measurements,

¶P
µ nb0 exp -E a Eq. 4-36
¶T kT

In ideal conditions, TPD can be used to measure this pressure change during a linear temperature

¶P
ramp, and the activation energy can be extracted from an Arrhenius plot ln ¶T = ln nb - kT
E
+ C by taking 0 a

fitting the slope of the curve as 1/T, -k ¶(1¶T ) ln ¶T

¶P
= Ea .

Another method to predict desorption rates in UHV conditions makes use of the vapor pressure

at the surface. This is the pressure required to create an equilibrium such that the rate at which

molecules are emitted is the same as the rate at which they are adsorbed. The rate of adsorption is

calculated with the flux, Fg = 14 nv , where n is the volumetric molecular density of the gas, and v

is the average thermal velocity. This is the rate at which gas molecules impinge on an area of the

surface. The average thermal velocity is computed with Boltzmann statistics,

ò ò (
¥ ¥
)
2
- mv
v= vf (v)dv = 4p v 3e dv =
3/2
m
2 p kT
2 kT 8kT
pm
Eq. 4-37
0 0

Using the ideal gas law allows substitution of n, and the flux is

Fg = P Eq. 4-38
2 p mkT

The units for this are molecules cm-2 sec-1. The equilibrium is described by the case when the net

102
flux is zero, as given by the difference between the vapor pressure caused by desorption from a

surface and the ambient partial pressure of the surrounding gas,

F= P-Pv Eq. 4-39

2 p mkT

In a UHV environment, the partial pressure is very low, but a material’s vapor pressure may be

high. The surface and surrounding gas systems are not in equilibrium when the vapor pressure

dominates the pressure of the ambient gas. In this case, the net flux (desorption) is given by the

vapor pressure. Further, if material density is known, Eq. 4-2 can be modified to calculate changes

in thickness due to desorption. This can be expressed as

¶d
= Pv n Eq. 4-40
¶t 2 p mkT

where n is the molecular density of the material.

4.6.2: TPD Experiments

TPD was performed in a chamber custom built by Carl Ventrice and his research group (4-15).

It contains a UHV chamber for measurements and a high vacuum chamber for pre-pumping

samples, thus allowing successive measurements without exposing the UVH chamber to air, and

obviating the need for bake-outs between samples. The TPD measurements rely on temperature

regulation and mass spectrometry systems that run simultaneously.

Temperature is constantly measured by a type-K thermocouple in physical contact with the

sample surface. This thermocouple is thermally isolated from the rest of the Cu column by ceramic

mounting washers. As the voltages produced by thermocouples are low, this signal is increased by

a preamplifier before being fed into a data acquisition board which is able to communicate with a

nearby computer. Behind the sample, there is a tungsten filament that radiates when at a

sufficiently high current. An annular aperture on the sample mounting plate allows this to directly

103
 Figure 4-15: Schematic of TPD system. (a) After insertion through a load-lock, the sample is
placed in a large UHV chamber. Signals from the thermocouple are fed into a nearby computer,
where a PID program regulates heating to a linear ramp by signaling a power source to increase
(or decrease) the current running through a heating filament behind the sample. Meanwhile, a
mass spectrometer tracks the partial pressure of desorbing gases. (b) The sample is mounted on
a plate with an annular hole that lets the filament radiation through to directly impact the back
of the sample. A thermocouple in physical contact with the surface sends temperature readings
to the computer.

reach the back of the sample. On the computer, a program written in LabVIEW monitors the

temperature of the sample surface, and sends an outgoing signal to a power source to supply current

to the W filament, and is therefore able to adjust the heating rate based on the recent temperature

trend. This program relies on a PID error minimization model in order to keep a steady, linear

ramp rate.

The mass spectrometry is conducted with a Hiden Analytical HAL 201 RC quadrupole mass

spectrometer (QMS), which is a type of a residual gas analyzer (RGA). Prior to entering the QMS,

molecules are ionized with 70 eV electrons. The QMS is composed of a four parallel metal rods,

which when observed along their long axis appear as corners of a square. Rods in opposite corners

104
are paired, and given an equal bias. The two pairs have opposite biases, which are fluctuated at

radio frequencies. These biases are able to filter out singly ionized molecules according to their

masses, and the QMS is able to quickly sweep through relevant mass ranges in order to quickly

detect desorbing molecules. This model is has a mass range of 200 amu, which disallows the direct

measurement of Bi desorption. However, all isotopes of Se atoms are measurable in this range.

Desorption rates, as estimated from vapor pressure, can be calculated in two ways. One is to

use a lookup table, such as from Honig and Kramer109, in order to find the vapor pressure. Another

way is to use perform a direct calculation with the semi-empirical simplified Antoine equation,

log10 Pv = A + TB , and use reported literature values for the A and B parameters. For known solids, the

vapor pressure can be used to calculate temperature. In the system used for these experiments, the

temperature could not be accurately calibrated for Al2O3/Bi2Se3/Se samples. The thermocouple

read accurately near room temperature, but high temperature tests involving electron beam heating

of a Si coupon sample showed significant (consistently >200 °C) discrepancies between the

thermocouple and a visible wavelength pyrometer. As the pyrometer could not read below 600 °C,

these trials were conducted between 600 and 1100 °C. As this region is far outside the range of

interest (Se expected to desorb near 130 °C and Bi2Se3 expected to desorb near 325 °C), and the

measured values between the pyrometer and thermocouple were significantly different, the initial

Se desorption event was used to calibrate temperature. The thickness change of the Se cap due to

evaporation is given by Eq. 4-40. The vapor pressure of Se reaches 5×10-5 torr near 130 °C109,110,

Se has three atoms per unit cell, and the volume of a unit cell is V = a2csin60 = 81.78 Å3, giving a

thickness evaporation rate of 2.77 nm sec-1. This is approximately the average rate needed to

105
evaporate the Se cap in 2 minutes, as shown in Fig. 4-16(a), and thus is assigned to the middle of

the Se desorption increase ramp. The desorption, as indicated by high partial pressures of residual

gases, is then plotted as a function of temperature in the region where temperature is a linear

function of time.

Unfortunately, this method of temperature calibration is not accurate beyond a short range, and

not reliable between samples. Se cap desorption events were measured between 60 and 110 °C, so

no single TPD spectrum could be used as a general calibration. Rather, TPD measurements were

performed at a nominally linear temperature ramp, and the temperature data were recalculated

based on the Se peak position in each measurement. TPD measurements that were calibrated this

way indicated higher temperatures than possible beyond the Se desorption peak. Figure 4-16(b)

shows all of the gas species measured in TPD, and shows that a temperature of 375 °C is not

sufficient to produce a second Se peak due to Bi2Se3 desorption. The partial pressure of Bi2Se3 is

Figure 4-16: Typical temperature calibration and full array of TPD-measured molecules. (a)
Temperature calibrations were performed near the midpoint of Se desorption rate, and taken to
indicate a vapor pressure of 5´10-5 Torr. (b) Even this direct calibration is not valid, as it results
in ‘calibrated’ temperatures of up to 375 °C in the linear ramp range that do not show Bi2Se3
desorption. TPD is therefore is only used as an indicator of desorbed layers, but not as a valid
measurement for desorption temperatures.

106
known to reach 1´10-4 Torr at 375 °C, and we therefore conclude that no attempted method of

TPD temperature calibration is sufficiently reliable to produce the correct temperature readings.

TPD is therefore used as an indicator of desorption events, and is taken to be a reasonable

qualitative guideline to temperatures near Se desorption peak, but is not sufficiently accurate for

calculations regarding the kinetics of Se or Bi2Se3 desorption.

4.7: Low Energy Electron Diffraction

Low energy electron diffraction (LEED) is a method of probing the crystalline structure of

samples very close to the surface. It relies on the wave nature of the electrons to diffract with the

periodicity of the lattice and the low kinetic energy of the electrons to only produce signal within

the top ~1 nm of the surface. It was one of the early confirmations of the de Broglie wavelength

of electrons, as it created lattice patterns that be compared to reference x-ray diffraction

experiments (a 100 eV electron has l =1.2264 Å, similar to the Cu Kα wavelength of 1.5406 Å)

where the wavelength dependence of light diffraction was known. It is a standard surface science

tool, operating in UHV conditions to indicate the 2D crystallinity of the growths at surfaces.

Further, LEED can be used in a quantitative mode where the intensity of diffraction points as a

function of the electron accelerating voltage can be informative about specific atomic positions (as

opposed to lattice parameters) by comparison to a dynamical scattering model, or used to calculate

the work function at the surface by detection of emitted secondary electrons as a function of

incident electron accelerating voltage and sample bias. As only the verification of intact crystal

structure was needed in this thesis, LEED was only used for lattice assessment. It should be noted

that observation of a LEED pattern does not guarantee that the whole surface is ordered, as LEED

107
spot sizes are often on the order of 1 mm2, and a detailed assessment of surface crystallinity

requires a supplemental microscopic characterization technique.

4.7.1: Theory of LEED

LEED is a highly surface sensitive measurement. This is due to the low kinetic energy of the

electrons traveling through the solid sample, and the low mean free path that such a low kinetic

energy generates. Unlike in particle travel through gas or vacuum, which may be modeled with a

Beer-Lambert function, electrons traveling through a solid are subject to many different scattering

events, such as interactions with phonons, plasmons, individual valence electrons, etc. These

interaction events happen at wildly varying energy levels, and thus a complex model is required

to calculate the mean free path of an electron as a function of its kinetic energy. In lieu

Figure 4-17: Universal mean free path curve of electrons traveling through solids. For kinetic
energies between 20 and 200 eV, this is less than 1 nm, which indicates that electrons traveling
at these energies scatter rapidly. This produces a high degree of surface sensitivity in LEED, as
few electrons are backscattered after the top 1 nm of the surface. Adapted from [5].

108
111
of such a calculation, a reference to a “universal curve” for the mean free path of electrons will

suffice (Fig. 4-17). Because surface sensitivity is desired in LEED, these experiments are usually

carried out with accelerating voltages between 20 and 200 V. The mean free path in this energy

region is below 1 nm, which ensures that electrons that impinge on the sample surface are rapidly

scattered, and that any information gathered from scattered electrons originates near the surface.

LEED detects electrons that are elastically and coherently backscattered. The coherence

maintained during this process leads to a collective phase acquisition during scattering, which

Figure 4-18: Schematic of crystal diffractions in LEED. The incoming beam, labeled k0, is
diffracted with reciprocal lattice vector ghkl, and leaves the crystal with momentum k. Because
the interaction volume is so short, the reciprocal lattice vecors are blurred in the surface
direction, thus allowing many reciprocal lattice vectors with g // gsurf to produced the same
diffraction. Even though this k-k0 does not exactly coincide with a reciprocal lattice vector, the
overlap with the truncation rods allows diffraction.

109
creates a diffraction pattern. This can be described by the change in momentum (similar to XRD)

that an electron receives during the scattering process,

Eq. 4-41

k (k0) is the outgoing (incoming) electron momentum (from the de Broglie equation

), and ghkl is a reciprocal lattice vector. Because the interaction depth is so low, the effective

reciprocal lattice points tend to spread out into truncation rods in the surface direction, thus

allowing some diffractions that coincide with changes in k along the surface plane, but do not

exactly coincide with the surface direction of the ghkl vector (Fig. 4-18). This allows diffractions

in many Laue zones to be measured simultaneously.

4.7.2: LEED Experiments

LEED is conducted with an electron gun pointed normal to the sample surface (Fig. 4-19) and

a hemispherical detector grid surrounding it. The gun filament is negatively biased to extract

electrons, which are then accelerated at voltages below 100 eV toward the sample. Upon reflection,

these electrons leave the surface with angles acquired by their diffraction vectors. The detector

system is composed of four concentric hemispherical grids and a phosphor screen. The middle

grids are negatively biased, and serve to filter out inelastically scattered electrons from the surface

that would otherwise generate noisy signals. The outer two grids are attached to ground and prevent

stray fields. Diffracted electrons have sufficient energy to surpass the filters, and impact the biased

(2-5 keV) phosphor screen, which creates a visible LEED pattern.

110
 Figure 4-19: Diagram of the LEED experimental system. The electrons extracted the cathode
are accelerated and focused by a negatively biased Wehnelt cylinder. These pass through
focusing rings and impact the grounded sample at normal incidence. The detector is composed
of four grids that filter out inelastically scattered low energy electrons and a positively biased
phosphor screen on which the LEED pattern can be seen.

4.8: Scanning Tunneling Microscopy and Scanning Tunneling Spectroscopy

Scanning tunneling microscopy (STM) is a method of imaging a surface by rastering a

conductive tip close enough to it that electrons are able to tunnel through the vacuum that separates

the tip from the surface (Fig. 4-20). Its resolution (sub-angstrom) and surface sensitivity make it a

uniquely useful tool (Fig. 4-21). Its applications to basic science earned its inventors a Nobel Prize

in 1986. STM has been used to observe surface diffusion and the resultant growth and expansion

of terraces, to image crystalline surface reconstructions, and to manipulate the positions of atoms

by attracting them with a bias ramp, dragging them with the STM tip, and applying another bias

ramp to release them112. Scanning tunneling spectroscopy is an extension of STM, in which the tip

bias is varied and the current generated between the tip and sample is measured. This technique

111
gives electrical information at the surface with high spatial resolution, thus allowing some

electronic properties to be mapped as a function of position.

4.8.1: Theory of STM/STS

STM measurements are conducted such that the probe tip never touches the sample surface,

although it approaches very closely. There is no direct line by which electrons may classically

conduct in these experiments. Instead, the fundamental physics of STM/STS involves quantum

tunneling. Electrons on the surface of the sample do not have sufficient energy to make it over the

energetic barrier of vacuum, but with sufficient bias and minimal distance from tip to sample, the

tunneling current rises to measurable levels. Calculations of tunneling current in a typical STM

environment require a brief derivation.

An electron traveling with energy E can be described by the Schrodinger equation

. The wave function has the solution

Figure 4-20: Schematic of an STM. An atomically sharp tip is positioned within 1 nm of a

surface and biased. The resulting tunneling current is monitored, and usually kept constant with
a feedback loop that is sensitive to the sample-tip distance.

112
 Figure 4-21: STM Topography measurement of Si(111) 7×7. (a) The surface is resolved well
beyond Angstrom resolution in the x- and y- dimensions, and individual atoms can be seen as
slight surface protrusions. (b) The topographic contour line from (a) shows the height of atoms
in reference to the background to within picometers.

y (x) = y0 exp(ikx) Eq. 4-42

For regions where U(z) < E , k is real, and electrons oscillate at with wave number k. This is a

classically viable energy regime. However, for regions where U(z) > E , classical particles are

forbidden. In the quantum regime, though, this creates an imaginary k value, and the wave function

has exponentially decreasing amplitude. For an abrupt transition into the high potential, this

amplitude is dependent on the distance traveled into the classically forbidden region, x - xU ,

y (x) = y0 exp(-k(x - xU )) , and the probability that an electron may have a given position in this

potential is P(x) = y0 exp(-2k(x - xU )) . In the case of STM, the distance between the tip and surface
2

is on the order of 1 nm or less, and will be given here by d, and the difference between the high

vacuum potential between the surface and tip, and the average electron energy (EF), is given by

the work function, ϕ. The probability of an electron tunneling to the tip in such a system is

Eq. 4-43

This probability is extremely small for d > 1 nm, as in most systems, the work function is

113
significantly greater than (the product for the electron rest mass, d = 1 nm, and the

work function of Si, 4.75 eV, resulting in a negligible probability). However, the tip distance can

be modified with sub Angstrom-level precision in an STM, allowing the surface-tip distance to

decrease to a few angstroms or less, greatly increasing tunneling probability and resulting in

measurable current levels. Further, a bias between the sample and tip increases the tunneling

probability by changing the Fermi level of the tip or sample. Instead of a constant high vacuum

potential, this results in a ‘slide’ down the potential barrier from sample to tip (or from tip to

sample given the opposite bias).

In order to calculate real current, a model that takes into account temperature and density of

states DOS is required. The rate at which an electron can transfer between states of different

energies, e.g. surface and tip states, is given by Fermi’s golden rule, , where

M. For varying surface and tip DOS, many such transfer events may occur, thus requiring an

integration

Eq. 4-44

Here, Nt,s are the DOS of the tip and sample, and f is the Fermi-Dirac distribution. The multiple

uses of f indicate that there must be both a filled state to tunnel from (in the sample, in this case),

and an empty state to tunnel into (in the tip). In real systems, although bias pushes current in one

direction, there’s a real probability of back current, and thus the net current requires a sum of

current directionalities,

Eq. 4-45

Further simplifications can be made for low temperatures and biases. At T = 0, f acts as a step

function, which allows the integration to be made between EF and EF + eV. For tips that have near

114
constant densities of states in these energy ranges, that term can be treated as a constant. Finally,

a consideration by Bardeen showed that for most realistic quantum tunneling situations, the matrix

element is virtually independent of the difference in energies between the two sides. The tunneling

current may finally be written

Eq. 4-46

and the sample’s local density of states (LDOS) at the tunneling site can be represented by the

derivative

Eq. 4-47

This is the essential equation that governs STS. It allows a surface LDOS to be investigated to

high precision.

Further STM physics considerations can be made for the often-spherical geometry of the tip,

the transfer matrix value, the real DOS for both the tip and sample, and the effect of orbital shape

on current levels, but those are left to the reader.

STM experiments are usually conducted such that the tip bias is held constant during scans,

and the tip distance is modified such that the current is maintained at a specified level. As shown

in Eq. 4-32, higher currents require the tip to be closer to the surface, and vice versa. This

relationship makes the STM sensitive to surface topography. As the tip moves to a feature

protruding out of the surface, the current is briefly increased as the tip distance decreases, and the

system raises the tip in order to keep the current at the specified level. Likewise, dips in topography

briefly decrease the current, and the system responds by pushing the tip downward toward the

surface, maintaining the current level.

115
4.8.2: STM/STS Experiments

Before inserting each Bi2Se3 sample into the STM, the sample is placed on a molybdenum

plate and loaded into a UHV chamber for annealing, in order to degas the surface. A radiative

heater is utilized to remove the Se cap in samples that had not been previously annealed. For

samples that were thermally decapped in a previous chamber (TPD), the STM anneals had

maximum temperatures that were 20 °C below the initial anneal, in order to minimize further

modifications to the Bi2Se3. This heating chamber is located in situ with the STM, connected by

UHV transfer tubes, allowing surface measurements to be performed without breaking vacuum,

thus minimizing surface contamination. Samples were photographed in order to ensure that the

anneals caused complete Se cap removal.

All STM and STS measurements are performed with an Omicron LT STM. Measurements are

done at room temperature (300K) and under UHV. The probe is a Pt-Ir wire that has been

mechanically cut to form an atomically sharp tip. Each STM measurement is done with a tip bias

of 0.487V to ensure stable tunneling above the band gap of Bi2Se3. A feedback loop is monitored

with a tunneling current of 2.1 nA in order to maintain a constant current across the 1000×1000

nm image area. Each STM image is taken with 2.5 nm step resolution.

In STS measurements, a 10×10 grid area is used over the 1 um2 imaging area, producing

spectral maps with 100 nm node spacing. The tip bias in each spectrum is swept from +0.487 V to

-0.487 V with a resolution of 2.5 meV, and data is acquired at a rate of 320 μs per point. The

feedback loop is disabled during STS acquisitions. Forward and reverse sweeps are performed at

each grid location for a total of 200 spectra taken per image. Because we assume minimal

hysteresis, these are then averaged to increase the quality of the data. A 50us delay is used between

each sweep of STS to account for any change in the tip location due to drift.

116
117
Chapter 5: Surface Oxidation in an Ambient Environment

Aging and surface oxidation are highly disruptive processes at the surface of Bi 2Se3. The

alterations they cause in the chemistry, structure, and electronic behavior near the surface can have

detrimental impacts on attempts to utilize TSSs40,70, but are not well understood mechanistically.

It is critical to the TI behavior of Bi2Se3 that the Fermi level remains in the gap, even when the

crystal is exposed to air, as is a likely scenario during fabrication. Exposure to various air gases

(CO, H2O, O2…) effects have been correlated with band bending and the emergence of a Rashba-

split 2DEG92, which exhibits metallic conduction with a sufficiently high DOS to drown out

conductivity measurements of the TSSs.40 Various properties of the oxide itself are also not well

documented. Bi and Se oxidation states have been shown to grow on the surface,70,113 but overall

composition of the oxide, the rate of its growth, the surface morphology it creates, and its electrical

and optical properties must be studied in order to assess the impact that oxidation has on the Bi2Se3

crystal and its surface states. Therefore, attempts to characterize these processes are critical to a

comprehensive physical picture of the overall electrical behavior of Bi2Se3 and its TSSs. This is a

necessary preliminary step to implementing Bi2Se3 in devices, as fabrication processes involving

Bi2Se3 must be sensitive to its defect vulnerabilities to circumvent them.

5.1: Sample Preparation

Ex-situ cleaved Bi2Se3 crystals from Sigma Aldrich (CAS 12068-69-8, SKU 401080) were

used for all of the measurements. Cleavage was performed with Scotch tape. This method has been

shown to produce high quality flake samples,9 and can be used in a similar fashion to produce

large, flat, shiny measurement areas on the bulk surface. Samples were exposed to air at 21 ± 2 °C

118
with humidity levels at 45 ± 5 percent. These conditions were held constant during the aging

periods of each sample, which ranged from 0 minutes to 10.5 days.

5.2: Measurement Results

This study makes use of a hybrid approach to characterizing the aging process. The goals are

to understand the rate at which the surface oxidizes in air, to observe the surface morphology that

it creates, and to develop optical models for both the bulk Bi2Se3 and its surface oxide. Multiple

experimental methods are necessary to accomplish this. The methodology of our study involves

characterizing the bulk of the samples with XRD and HAADF STEM to assess their crystallinity,

exposing them to air in a controlled environment for specified time periods, and characterizing the

time-dependent oxidation process with a variety of instruments. These include XPS for near-

surface chemical analysis, AFM for topography, and both spectroscopic ellipsometry and second

harmonic generation for optical properties. This generates a complete characterization of the

oxidation process and its physical properties, and is informative about the time scale of oxidation

caused by air exposure as well as the electronic properties of the oxide, both of which impact the

device development process. As we are looking to create a reference for this aging process, it is

important to ensure that our samples are of high quality before they are modified.

5.2.1: Verification of Single Crystal Sample Quality

Before conducting a variety of measurements on Bi2Se3 to test its oxidation dynamics, the

crystalline quality of the samples was assessed. Previous microscopy studies of Bi2Se3 and Bi2Te3

show a variety of lattice defects including Bi-Bi layers in the bulk of Bi2Te3114 and deviations from

the ideal lattice (Fig. 3-1) could have significant effects on the oxidation process. The QL

119
Figure 5-1: Aberration-corrected HAADF STEM images of bulk Bi2Se3. (a) A top-down image
of the crystal along the [0001] zone axis, showing its hexagonal structure, along with a
simulated atomic model in the upper left, and the power spectrum of the image in the upper
right. Though each QL has a honeycomb structure (two lateral sites occupied by both Bi and
Se, and one singly occupied only by Se, generating one relatively weak site per hexagonal cell,
and thus a honeycomb structure when viewed in the [0001] axis), their rhombohedral stacking
leads to a 2D hexagonal close-packed image. A hexagonal unit cell is outlined in red.
Differences between the hexagonal shape and the observed image are attributed to sample drift.
(b) A cross-sectional image of the crystal. In hexagonal notation, the zone axis is normal to the
plane (0 1 1 0). The QL structure of the crystal is apparent from the vdW gaps separating
sections of five atomic layers. One can also see the Se-Bi-Se-Bi-Se order of QLs from the
relatively high brightness of the Bi atoms. A QL is outlined in green, with the red dotted
extension outlining a unit cell. Each figure has an ideal atomic schematic and a power spectrum
in the insets. The lattice parameters measured from these images are a = 0.429 nm and c = 2.958
nm. Adapted from [115].
120
structure is integral to understanding the rates of oxidation and the presence of major structural

defects may change this time scale, or disrupt the expected equilibrium, drastically. To assess the

crystallinity of these samples, local measurements are made with TEM to show pristine QL

structure by direct imaging methods, and bulk measurements are made with XRD to show that the

expected QL structure found in TEM is present across the sample. Further, XRD pole figures show

the lack of twinning or major rotational stacking defects between QLs in the crystals we studied.

HAADF STEM was conducted in cross-sectional and top-down geometries in order to obtain

images of the Bi2Se3 crystal structure, as shown in Fig. 5-1. These images were taken with the

assistance of Brendan O’Brien, formally of CNSE. The cross-sectional intensity seen in each QL

indicates the expected Se-Bi-Se-Bi-Se structure, and each QL is vertically stacked with smooth,

vdW-gapped interfaces. Because the ( 01 10) normal zone axis is used in this cross section, it does

not indicate the presence or absence of distinct twinned grains, but a lack of visible grain

boundaries (and Bi bilayers) indicate a high quality single crystal. The STEM image and electron

diffraction pattern of the samples used in this study show that these crystals have the expected

lattice structure and lattice constant values, as is readily seen by comparison of atoms in the bulk

STEM images with a schematic of the ideal bulk structure in the inset. Hexagonal lattice

parameters were measured at a ~ 0.429 nm and c ~ 2.958 nm respectively, which are both within

3% of values found in literature.116 Aberrations from regular hexagonal shape in the top down

images are attributed to drift-induced stretching effects. However, because these are highly

localized measurements, bulk crystallinity measurements are also conducted.

XRD characterization also shows a high quality QL structure (Fig. 5-2). XRD -2θ

measurements yielded strong diffraction peaks at the (0 0 0 3n) indices, and produced a c value of

2.833 nm, which is within 1% of literature values.116 The diffraction peak intensities follow those

121
 Figure 5-2: Omega-2Theta XRD scan of Bi2Se3. This measurement was performed on a large
granule, of approximately 2x2 cm in area and 0.5 mm in thickness. The (0 0 0 3n) indices are
the only c-axis planes that are not forbidden. Thus, we have labeled each peak with its l index.
These scans give values of 28.427 Å per unit cell, or 9.476 Å per QL. Adapted from [115].

previously reported117. Pole figures were measured in order to in order to search for grains in the

samples over a larger volume than possible with STEM. The ( 0115) peaks were chosen for their

high intensity (highest predicted intensity in powder diffraction experiments118) and their three-

fold symmetry. They are highly visible and indicative of the presence or absence of twinned grains.

This can be seen in the azimuthal symmetry of the pole figure, which becomes 6-fold for samples

containing roughly equal amounts of grains with the two predominant orientations being rotated

by 60°. This was observed in MBE-grown samples, as discussed in Ch. 6. For flat samples, the

plane normals that give the position of diffraction spots have expected tilt angles of

°. These spots may stray from the expected value for tilted samples

such as the bulk growths involved in this study. The acquired pole figure has no 60°-separated

twin diffraction, and thus indicates that only one grain orientation is present in these samples (Fig.

122
 Figure 5-3: Pole figure of exfoliated Bi2Se3 sample used in oxidation study. Only two
diffraction points are present. These are caused by the Bi2Se3 ( 0115) plane and the Si (111)
substrate plane. Beyond this range, the Si spots appear with four fold symmetry, facilitating
their identification. In the 120° azimuthal range shown, only one diffraction point appears from
Bi2Se3. This indicates that there is only one major grain orientation. Because it is slightly
smeared toward the negative direction, it is likely that there are multiple grains with slight
angular mismatches. However, because there is no twin diffraction point at 152 + 60 = 212 °, this
sample is devoid of twinning.

5-3). We thus conclude that since local measurements definitively show a vdW-gapped structure,

and XRD shows a well-layered structure in the bulk as well as a single azimuthal orientation, these

may be considered to represent high quality Bi2Se3 systems, devoid of twinned grains. Since the

measurements of the lattice constants of bulk Bi2Se3 produced values close to the literature, and

top down and cross-sectional lattice images show no signs of defects, the origin of time dependent

changes to this crystal sample of Bi2Se3 presented in the next section are considered to arise from

the basic properties of the exfoliated surface of crystalline Bi2Se3. The quintuple layer (QL)

structure and weak vdW inter-QL bonds of the Bi2Se3 crystal allow for samples to be prepared via

micro-mechanical exfoliation, as has been done with graphene since 2004.91 Scotch tape exposes

a fresh QL as the sample surface, enabling measurements on pristine surfaces, and further

measurements as the samples age in air.

123
5.2.2: AFM Characterization of Time-Dependent Surface Morphology

The time dependence of the surface morphology, structure, and chemical state were studies by

AFM, ARXPS, XPS, and RA-SHG. First, the observation of islands on the surface by AFM is

presented. AFM analysis was done with the open-source scanning probe microscopy software

Figure 5-4: Time-dependent surface topography of Bi2Se3. All scans were taken with Bruker
TESPA AFM tips. Until 33 minutes after exfoliation, the crystal is atomically smooth. After 33
minutes, surface patches arise, but stop growing at ~100 minutes. They are very uniform in
height (3.6 ± 0.2 Å). We hypothesize that these patches are Bi bilayers. The background
roughens through the full time scale, increasing from 0.4 Å RMS at 11 minutes to 1.1 Å RMS
at 1251 minutes. Adapted from [115].

124
Gwyddion119. AFM scans completed within 25 minutes after exfoliation show atomically flat

surface topographies. Initial surface roughness in the sample shown in Fig. 5-4 is 0.43 Å RMS.

Subsequent scans show the nucleation and growth of patches that within measurement error are

uniform in height (3.6 ± 0.2 Å). These patches and their growth process are also noticeably

different from samples that experience buildup of contaminating particles (Fig. 5-5). Their height

above background is consistent with the height of Bi bilayers in the bulk of Bi2Te3114 which is

structurally similar to Bi2Se3. Based on this consideration, we hypothesize that the patches are

bilayers of Bi. The existence of Bi bilayers on the surface of Bi2Se3 samples is well documented

in literature.117,120 This patchy growth provides an explanation for why surface Bi bilayers are

observed by some methods and not by others.71,120 Some attempts to identify these patches with

Figure 5-5: Reference topography scan of a contaminated surface. Time-resolved AFM images
of a clean surface (a), followed by patch growth (b-c), and then dirt adhesion (c-d). These
patches, as in other measurements, have a uniform thickness of 3.6 ± 0.2 Å. The bright spots in
(d) have thicknesses roughly equivalent to their diameters (>100 nm). Whereas the patches are
quite large and flat, the dirt is roughly spherical, and contributes significantly to image
aberrations. Adapted from [115].

125
Auger electron spectroscopy were conducted, but returned inconclusive based on the spatial

resolution of the tool. If they are Bi bilayers, these scans are also indicative of a vertical diffusion

process, in which Bi rises to the surface. We expect that this diffusion may alter the band structure

of Bi2Se3 at the surface, changes the conductance of the topologically protected surface states, and

could contribute to 2D electron gas formation. The majority of patch growth occurs before 2 hours.

After 2 hours, the sample appears to roughen ubiquitously, and we attribute this change to surface

oxidation in agreement with XPS measurements to follow. This roughening process slows over

time, leaving the background with RMS roughness of 1.1 Å after 22 hours. It is possible that oxide

growth up to a certain length of time after the exfoliation provides a barrier for further surface

diffusion, thus distinguishing the growth of Bi bilayers on top of the oxidized surface from the

initial oxide growth.

5.2.3: Characterization of Chemical Changes during Oxidation with XPS and ARXPS

XPS and ARXPS provide insight into the growth of the oxide overlayer over a period of ~ 1.5

weeks after exfoliating the samples. Fig. 5 contains angle-integrated, time dependent XPS

Binding energy (eV) ± 0.05 eV Thickness

Sample ID Bi 5d5/2 in Se 3d5/2 in O 1s in (nm) ± 0.2
(time since Oxide Oxide Oxide
exfoliation) Bi2Se3 over- Bi2Se3 over- over-
layer layer layer
t ~ 0 min 25.2 -- 53.7 -- -- --
t ~ 119 min 25.2 -- 53.7 -- -- --
t ~ 189 min 25.2 26.4 53.7 58.7 531.5 1.0
t ~ 309 min 25.2 26.4 53.7 58.7 531.5 1.4
t ~ 479 min 25.2 26.4 53.7 58.7 531.5 1.6
t ~ 1.5 week 25.2 26.4 53.7 58.7 531.5 1.9

Table 5-1: Results of time dependent XPS and ARXPS investigation on Bi2Se3 after exfoliation
along (0001). Adapted from [115].

126
spectra showing aging of the Bi2Se3 surface, and provides information about the constituent

elements by means of the Se 3d, Bi 5d, and O 1s level peaks. Table 5-1 lists the values of binding

energy of the samples after exfoliation. Spin orbital splitting leads to a pair of peaks for the Se 3d

(3d5/2 and 3d3/2; separated by ~ 0.9 eV) and Bi 5d (5d5/2 and 5d3/2; separated by ~ 3.0 eV) levels.

The freshly exfoliated sample (t = 0 minutes) shows no signs of oxidation as evident by a flat O

1s peal and the absence of higher binding energy peaks in the Se 3d and Bi 5d regions. We estimate

the binding energy reference for Bi metal 5d5/2 peak as 24.0 ± 0.3 eV and Se chalcogenide 3d5/2 as

55.4 ± 0.6 eV after calculating mean and average of all the reported values present at the NIST

XPS database.121 Thus, for the freshly exfoliated sample (t = 0 minutes), the Bi 5d5/2 and Se 3d5/2

in Bi2Se3 show a charge shift of +1.2 eV and –1.7 eV respectively which agrees well with the

previously reported values for Bi2Se3.122,123 The more electronegative Se (2.55 in Pauling scale)

pulls a greater portion of the electron cloud towards itself as compared to the less electronegative

Figure 5-6: XPS and ARXPS results. Angle summed XPS spectra of (a) Se 3d, (b) Bi 5d, (c)
O 1s region for various times after exfoliation. The oxide peaks corresponding to SeO2 (marked
by a dotted box in (a)) and Bi2O3 (marked by vertical arrows in (b)) are observed after ~119
minutes from exfoliation along with the observation of a strong O 1s peak. The results also
indicate that oxygen adhesion and subsurface oxidation proceed simultaneously. Adapted from
[115].

127
Bi (2.02 in Pauling scale), thus giving some ionic nature to the bonds in the Bi 2Se3 crystal.124 No

evidence of oxidation is visible in the XPS spectra of samples up to t ~ 119 minutes from

exfoliation (Fig. 5-6). AFM measurements show the growth of small patches (3.6 ± 0.2 Å in height,

less than 3 μm in diameter) for t < ~119 minutes, and subsequent roughening of the whole surface.

The relative concentration of those patches with respect to the analyzing volume is below the

detection limit of our XPS measurement. For samples exposed to ambient air longer than

t ~ 189 minutes, a sharp O 1s peak appears implying adsorption of oxygen on the (0001) surface

of the crystal with time. At the same time, new features show up near the Se 3d (broad peak at ~

59 eV) and Bi 5d (shoulders towards higher binding energy side) regions at higher binding energy

values consistent with oxidized Se and Bi confirming the oxidation of both Bi and Se at the surface

as discussed below. Top layer adsorption would produce O 1s peaks with no corresponding oxide

peaks in the case that oxygen did not chemically bond to the surface, or O 1s and Se oxide peaks

but no Bi oxide peaks in the case that oxygen only bonded to the topmost layer, as indicated by

low patch coverage in AFM. The observed values at the higher binding energy region of the Se 3d

and Bi 5d (Table 5-1) are close to the values for Bi2O3 and SeO2, the most stable oxides of Bi and

Se under ambient conditions. As the O 1s peak show only one binding energy value (531.5 eV),

we conclude that the average bonding environment around the O atom stays similar over time and

the O atom remains surrounded by both Bi and Se atoms, i.e., the chemical environment in the

oxide overlayer is similar to a mixture of Bi2O3 and SeO2. At the beginning of the oxidation

process, it is possible that some O-vacancy related defects will remain in the oxide overlayer.

Adsorption of more oxygen atoms over time will heal some of these defects and also increase the

thickness of the oxide overlayer as more Bi and Se from the QL layer reacts with the atmospheric

O. The O 1s, Se 3d oxide, and Bi 5d oxide peaks all continue to grow as evidenced by spectra

128
collected at subsequent times (Fig. 5-6) although the rate of growth of the oxide overlayer appears

to decrease after ~8 hours. This may be due to the vdW gapped structure of Bi2Se3, which could

act as a temporary barrier to oxidation beyond the first QL.39 We expect that over longer time scales

(months), oxidation may continue through further QLs via defect-mediated movement of O in the

intercalation layer. The time dependence of the oxidation process is quite unusual, in that it

displays a significant delay between surface exposure and oxygen incorporation. It has been

reported that a delay to oxidation of up to 20 hours is normal for the (0001) surface of certain

Figure 5-7: XPS scans of the Bi 4f region taken over various times. Before t ~ 119 minutes, no
oxidation is observed in the XPS data (see Fig. 5). As such, the presence of the shoulders to the
left of the main Bi 4f peaks before t ~ 119 minutes are attributed to overlapping Se 3p peaks.19,20
For samples exposed to air for longer times, the Bi 4f oxide peaks also show up in the same
region as that of the shoulders and grow in intensity over time. Adapted from [115].

129
Bi2Te3 crystals.39 Further study of the factors relating the surface behavior of Bi 2Se3 and Bi2Te3

(e.g., Bi presence, topological phase, QL structure, etc.) is required.

Another research group performed an analysis of the oxidation of Bi2Se3 and found from the

presence of a shoulder in the Bi 4f peak that an oxide is present immediately after cleavage in all

samples (Fig 2-24).70 Fig. 5-7 shows the time evolution of the Bi 4f XPS peaks. At t = 0 minutes,

the main Bi 4f7/2 and 4f5/2 XPS peaks are accompanied by shoulders to the higher binding energy

side. Although the Se 3p 3/2, 1/2 doublet alter the higher binding energy spectrum of the Bi 4f

peaks, the effect of oxidation can be observed over time in the continued growth of higher binding

energy shoulders. However, since there is no evidence of oxygen adsorption at the (0001) surface

for 0 < t < ~ 119 minutes (see the O 1s spectra in Fig. 5-6(c)), the presence of these Bi 4f shoulders

may not be used for interpreting the early presence oxidation of the (0001) surface of Bi2Se3. These

shoulder peaks are attributed to potential shake-up peaks21 and overlap with peaks from Se, as

shown in the literature19,20. For samples exposed to the ambient for longer time periods, the

shoulder peaks grow in intensity as the contribution from the Bi 4f oxide peaks also appear in the

same region. The Bi 5d peak region (Fig. 5-6(b)) is devoid of any such complications arising from

peak overlap. Thus, we used the Bi 5d XPS peak region, along with the Se 3d and O 1s regions, to

understand the growth mechanism of surface oxide on the Bi2Se3 (0001) surface.

Analysis of angle-resolved XPS data provides a profile of the growth of the oxygen overlayer

with time. Among other factors, photoelectron counts are a function of collection angle from

normal incidence (θ), and electron origin depth. This is largely due to inelastic scattering events in

the sample125, which can be approximately modeled with a Beer-Lambert function: I = Io

exp[−αd/cos(θ)], where Io is the intensity of emitted electrons, d is the vertical depth from the

sample surface to the origin of the emitted electron, and α is a material-dependent density

130
 Figure 5-9: ARXPS analysis. (a) Relative intensity depth profile for the oxide overlayer for
various times after exfoliation (t ~ 189 minutes to ~ 1.5 week). The thickness of the oxide layer
is estimated from these plots where the relative intensity profile comes down to 50% of the
maximum value. The oxide layer grows over time, and appears to reach a metastable
equilibrium at ~ 1.9 nm after ~ 1.5 weeks of exfoliation (values mentioned in the plot legends).
The figure in the inset shows that the information depth varies as a function of the angle (θ)
with
Figurerespect to the surface
5-8: ARXPS spectranormal.
of the A (a)smaller
Se 3d value
and (b) θ corresponds
of Bi 5d region fortothe signals
samplecoming
after from
~1.5
layers that lie deep in the material (bulk) and a higher value is more sensitive
weeks of exfoliation. The angle in ARXPS profile is measured from the surface normal. Spectra to the features
close to theangles
at smaller surface.
are The
fromRelative intensity
layers deeper intodepth profile (bulk)
the material are obtained
and theafter modeling
spectra theangles
at higher angle
resolved
are coming normalized
from layerspeak areatoastheshown
closer in (b)offorthethe
top surface t ~ 1.5The
sample. week sample.
relative The ARXPS
intensities of the
modeling
signals dueistodone according
oxidized to Se
Bi and theincreases
maximum forentropy
spectraapproach
collected37,38 . Thelayers
from peakcloser
areas are calculated
to the surface
from 115
(24.9°the plots in to
compared Fig. 6 and
69.9°) normalized
indicating the with the instrument
overlayer transmission
is rich in oxygen. function,
Adapted from [sensitivity
].
factors and energy compensation factor. The calculations are limited between 25° to 60° to
minimize the effect of elastic scattering at higher angles37. Adapted from [115].

parameter. Fig. 5-8 shows the representative plots for Se 3d and Bi 5d regions’ ARXPS data

collected for the sample ~1.5 week old. The relative intensities of the oxide signals increase in

going from a smaller angle (more sensitive to signals coming from the bulk) to angles close to the

surface implying the depth of the oxide overlayer is limited to few QLs during our measurement

time. Applying a maximum entropy approach126,127 to the normalized ARXPS data as implemented

in the ARProcess software in Thermo Avantage, growth of the oxide overlayer was quantified as

shown in Fig. 5-9. Oxide depth appears to increase from 0 nm before 119 minutes to 1 nm at 189

minutes, to ~1.9 nm at the end of 1.5 weeks. This matches previously reported values very closely,

which describe the oxidation process of the top-most QL in Bi2Se3.70 It also indicates that the

131
expansion of a QL due to oxidation increases its volume by ~2x from 0.96 nm in an oxidized QL.

This expansion rate is similar to that seen in Si128.

5.2.4: Time-Sensitive Optical Characterization with Ellipsometry

Spectroscopic ellipsometry was used to characterize the changes in freshly exfoliated Bi2Se3

continuously over a 27-hour period. Three repetitions of the experiment, each at different angles,

gave consistent results. Standard ellipsometry data Ψ, the relative polarization amplitude change,

132
 Figure 5-10: Variable angle spectroscopic ellipsometry data and results of optical modeling.
(a) and (b) show Ψ(E) and Δ(E) at 10 minutes and ~5 hours after exfoliation, at angles of
incidence between 55 and 75 degrees from the surface normal. (c) shows Δ(t) for various
energies, and its inset highlights the rapid initial decline of Δ, especially at higher energies. (d)
shows the oxide and other surface growths on Bi2Se3 over time. The rapid initial change of the
surface layer corresponds to the initial decline of Δ. Figures (e) and (f) show the real and
imaginary dielectric functions of bulk Bi2Se3 and Bi2Se3 ‘oxide’, respectively. Adapted from
[115].

and Δ, the relative phase shift of polarizations, are shown in Fig. 5-10(a,b). During the first three

hours after exfoliation, Δ initially decreases rapidly followed by a period of slow decrease, while

133
Ψ shows little change over time (Fig. 5-10(c)). Changes in Ψ are small at most energies

(wavelengths). The well-known Drude approximation indicates that at an appropriately selected

wavelength Δ is a linear function of film thickness for thin dielectric films on metals or

semiconductors. Sub monolayer changes in Δ can easily be measured.40 That is only one possible

explanation for the observed changes in Δ during the first 3 hours. Δ shows significant change at

higher energies (smaller wavelengths) over the time frame when surface patch growth is observed

by AFM. Because XPS does not observe an oxide layer prior to 3 hours, the changes in Δ may be

due to the formation and growth of surface patches. Bi2Se3 thicknesses of > 1 mm lead to complete

absorption of light above the 0.3 eV bandgap. Thus, bulk Bi2Se3 behaves as an infinitely thick

substrate for our entire measurement range of 1.24 to 4.0 eV, and is treated accordingly in our

models. Initial measurements after exfoliation are used to extract a dielectric function of bulk

Bi2Se3. As can be seen in Fig. 5-10(e), the main peak in the imaginary portion of the dielectric

function, ε2, is located at 2.00 eV, and the real portion of the dielectric function, ε1, crosses zero at

2.27 eV. This matches well with FP-LAPW band structure calculations129, for which transitions

F1v ® F1c , G3v ® G1c , L1v ® L2c , and Z1v ® Z1c near 1.96 eV contribute most strongly to the optical

response of Bi2Se3, and with previous experiments, in which pseudodielectric functions were

calculated from ellipsometric data130. Utilizing these substrate properties and the observed changes

in Ψ and Δ over time, we are able to model the changes to the (0001) surface that we observe with

our other methods. Because of the variety of growing species and topographical changes to the

surface, e.g. sparse unidentified patches, possible 2D electron gas formation, bismuth and selenium

oxide growth, surface roughening, as well as the thinness of these new layers (sub-nanometer in

most cases), monitoring the individual contributions of these changes is currently beyond the

analytical capabilities of our system. Thus, for the first 3 hours the surface ‘oxide’ layer used in

134
this analysis models the total optical response of a combination of changes in the surface. After 3

hours, the model tracts the observed growth in the oxide layer. It may be possible to track each of

source of change individually in the future, but without access to reference data on the exact Se or

Bi oxides that grow on Bi2Se3’s (0001) surface, or any more information about the identity of

surface patches, we have not yet been able to extract meaningful analysis from each affect

simultaneously. Thus, we have generated a single dielectric function that most strongly resembles

that of previously reported bismuth and selenium oxides131–133, which we consider to be the major

Figure 5-11: Contrast maps for Bi2Se3 on a plain Si substrate (a) and on a Si substrate with 300
nm thermal SiO2 (b). Light regions indicate higher reflectivity of a flake at a given wavelength
and flake thickness relative to the reflectivity of the substrate. Likewise, dark regions indicate
lower reflectivity of a flake relative to the reflectivity of the substrate. For regions at 0 contrast,
there is no difference between the intensity of light reflected from the flake and from the
substrate. It is clear that having a 300 nm thermal SiO2 layer between the substrate and the flake
causes higher levels of contrast, especially in the 375 nm and 630 nm regions. Adapted from
[115].

135
contributor to ellipsometric signal generated by our observed surface alterations, after the first 3

hours of ambient exposure. The ‘oxide’ dielectric model is composed of Tauc-Lorenz oscillators,

and has a band gap of 2.7 eV (Fig. 8f), which matches previously reported values for Bi oxide.131,132

Se oxide may also contribute to ε2 at energies above its band gap of 2.8 eV133, but we cannot

currently parse the contributions of each oxide species.

In addition to providing information about electronic behavior, these dielectric functions

(refractive indices) can be used to model reflectivity and the contrast of Bi2Se3 flakes on variety of

substrates as seen in an optical microscope with the Fresnel equations at normal incidence:

r +r exp(-i2 b ) Eq. 5-1

r0,1 = nn00 -n1
+n1 bk = 2 pln d k
r0-k+1 = 1+r0-k0-krk,k+1 k
k ,k+1 exp(-i2 bk )

Here, r0,1 represents the electric field reflectance at the interface between layers 0 (taken as

ambient) and 1, nk is layer k’s complex refractive index, and βk gives the absorption of and phase

acquired by light of wavelength λ propagating through layer k of thickness d, and r0-k+1 is the field

reflectance produced by layers 0 through k+1 The reflection intensity is given by |r|2. Fig. 5-11

shows maps of the contrast created by unoxidized flakes on a Si wafer, as well as on a Si wafer

with 300 nm of thermal oxide, as a function of flake thickness and incident wavelength. The
2 2
rflake - rsub
equation for contrast is C(l, d) = rsub
2 . For flakes between 0 and 20 nm thick, an increase in

thickness causes an increase in contrast with the background. However, since the trend across

wavelengths changes minimally with changes in thickness, the change in color of a flake may not

be a reliable metric for flake thickness.

5.2.5: Non-Linear Optical Characterization with Second Harmonic Generation

The nonlinear optical technique of second harmonic generation (SHG) is particularly suited

for the study of Bi2Se3, because SHG arises from symmetry breaking at the surface, and thus should

136
detect changes at the surface preferentially over the bulk. The generated second harmonic light of

frequency 2ω using the incident fundamental light at frequency ω is forbidden in centrosymmetric

systems such as Bi2Se3. However, at the surface, there is a break in the symmetry due to the

transition to ambient, allowing SHG from the crystal surface. Therefore, SHG is a powerful probe

for surface-specific properties, such as surface electronic, spin, and lattice structures.134,135 In our

rotational-anisotropy SHG (RA-SHG) experiment, the sample was rotated azimuthally about the

surface normal (0001), and azimuthal angle dependent SHG intensity was measured for p-in/p-out

polarization configuration. In Fig. 5-12, the 3-fold symmetric RA-SHG signal appears at the

beginning stage after sample cleaving, corresponding to the 3-fold symmetry of the Bi2Se3 crystal,

and it develops over time. Initially, there are three major lobes at 60°, 180°, and 300° and three

minor lobes at 0°, 120°, and 240°, of sample azimuthal angles. The minor lobes appear in a fresh

sample when the outgoing beam is nearly parallel to the ( 01 15) plane normal, and the major lobes

appear in a fresh sample when the incident laser is nearly antiparallel to the ( 0115) plane normal,

as verified with XRD (Fig. 5-13). As the ( 0115) plane normal is parallel to the Bi-Se(2) bond at the

top QL surface, alignment between the top Bi-Se(2) bond and the outgoing SHG beam produces

minor lobes, and alignment between the top Bi-Se(2) bond and the incident laser produces major

lobes. The major lobes decrease in amplitude while the minor lobes increase, eventually becoming

equal in amplitude for a 6-fold symmetric RA-SHG scan at ~3.5 hours. Subsequently, the lobes

continue to grow at 0° (and other equivalent angles) and shrink at 180°, thus switching their major

and minor characters. When the sample has not experienced any prior laser exposure history, this

major-to-minor lobe switching process is complete at 12 hours with a majority of the change

137
 Figure 5-12: Measured rotational-anisotropy SHG (RA-SHG) scans in a dry air environment
immediately, and at 1, 2, 3, 4, 12, and 24 hours after Bi2Se3 sample cleaving. Each scan was
taken at a different sample spot so that the measurement spot had not experienced any prior
laser exposure before each measurement. Thin lines are experimental data, and thick smooth
curves are fits to Eq. (2). The empirical fit parameters a0 and a3 are plotted as a function of
time. While a3 only undergoes minor changes in the early stages of measurement, a0 increases
significantly and monotonically throughout the full oxidation process. Adapted from [115].

occurring by 6 hours. Analysis was done with the following RA-SHG equation for light that was

p-polarized before and after reflection:

I 2w (f ) = a0 + a3 cos(3f )
2 Eq. 5-2

Here a0 and a3 are respectively the isotropic and anisotropic complex coefficients, both

determined by Fresnel factors of the beam configuration and electric susceptibility of Bi2Se3, and

𝜙 is the sample azimuthal angle between the incident plane and the mirror symmetry plane, with

the directionality defined as in Fig. 5-14. Both coefficients are time-dependent when the surface

138
 Figure 5-13: XRD orientation of the ( 01 15) plane for SHG analysis. (a) XRD setup, showing
the sample orientation when a ( 01 15) diffraction peak is measured. The green arrows show the
incident and diffracted x-ray beams. (b) Zoomed image at the XRD stage, clearly showing the
Figure 5-14: Schematic
direction down and of outSHG at page.
of the the surface
(c) SHGof Bi 2Se3 showing
setup, samples.theIn aorientation
freshly cleaved
of thesample,
sample
each QL surface
and beam directioncontains
when aaminor
Bi-Se(2)
lobe bond
of athat points
fresh samplein the (0 1 1 5) or
is measured, plane
whennormal direction.
a major lobe of
When
an oxidized sample is measured. The red ray indicates the incident w beam, and the (0
the incoming ω light’s propagation vector is nearly aligned antiparallel to the 1 1ray
blue 5)
plane normal (a), a major SHG lobe is generated (60°, 180°, 300° in Fig.
indicates the reflected 2w beam. (d) Clarified image of the SHG setup, rotated 90° for10). As the sample is
rotated around
comparison to the
XRD.c axis,
Thethe outgoing
threefold 2ω light’soutgoing
symmetric propagation vector becomes nearly parallel to
2w beam orientation indicates that a
the (0 1
minor lobe 5) plane normal (b), and a minor SHG is
1 for a fresh sample follows the same direction as generated (0°,
. 120°, 240° in Fig. 10). These
SHG orientations are shown in alternate views in (c) (side view along the a axis) and (d) (top-
down view along the c axis). Adapted from [115].

is modified, and can be described as linear functions of the second-order susceptibility of Bi2Se3

as follows136:

a0 = a0r +ia0i = c01 c zzz + c02 c zxx + c03 c xxz a3 = c3 c xxx Eq. 5-3

Here  zzz ,  zxx ,  xxz ,  xxx are the four non-zero components of the second-order susceptibility

tensor  ( 2 ) of the 3-fold symmetric Bi2Se3 surface,137 and c01, c02, c03, c3 are constants that are

139
determined by Fresnel factors of the beam configuration and dielectric functions of the Bi 2Se3.

Although both a0 and a3 are complex in general, in our experiment, a3 can be chosen to be real,

and a0 requires only a minor imaginary contribution. The significant changes in a0 can be attributed

to surface charge accumulation137,138 and the trapping of photoinjected electrons at the

Bi2Se3/oxide interface, which produces an electric field-induced SHG effect. Figs. 5-12(a-c) show

the measured RA-SHG scans in a dry air environment immediately, and at 1, 2, 3, 4, 12, and 24

hours after Bi2Se3 sample cleaving. To avoid the effects of prior laser exposure, the sample was

shifted to a different measurement spot before each RA-SHG measurement. The thin lines are

experimental data of SHG, and the thick smooth curves are fits to Eq. (2). The a0 term contains a

contribution from the surface dipole induced SHG and electric field induced SHG, as described by

Eq. (3). The impact of the a0 term changes as the surface oxidizes. Fig. 5-12(d) shows empirical

fits to a0 and a3 over time. The major change to a3 to occurs within 1 hour. As other measurement

methods detect no oxide in that time, we attribute this change in a3 to be either caused by 2D

electron gas formation from band bending137,138 or Se vacancy formation. After the first hour, the

value of a3 remains near 1.44. The effects of charge accumulation, however, change the isotropic

parameter a0 from -0.19 to +0.19 while the first QL is being oxidized. The a0 trend closely

resembles the trend of oxide thickness observed in the previous sections of this work. We thus

consider the surface dipole SHG contribution, as represented by a0, to be a valid metric for

measuring oxide thickness via RA-SHG.

We can reasonably assume that a 6-fold RA-SHG pattern corresponds to the time when the top

two layers and part of the middle layer of the top QL is oxidized, and a full major-to-minor lobe

switching corresponds to the time when a full QL is oxidized. In comparison to the XPS results

140
with regard to the time scales of surface oxidation, RA-SHG shows a gradual evolution of

oxidation within a QL during the 12 hours period after sample cleaving.

5.3: Discussion

The surface aging process occurs in sequential stages (Fig. 12). First, surface islands of height

3.6 ± 0.2 Å form over the first 25 minutes as measured by AFM. The ellipsometry parameter Δ

also changes during the same time period. Since no O 1s peak is observed by XPS, the

observations are interpreted as being due to segregation of Bi to the surface over the first 25

minutes. The AFM data shows that patch growth continues for the first two hours, followed by

surface roughening. The presence of a surface oxide layer composed of both Bi oxide and Se oxide

is clearly present after 189 minutes (~3 hours) in the XPS data. Calculated values of the oxide

thickness from both XPS and SE find that the oxide film present at ~3 hours is ~1 nm thick. Using

the estimated expansion factor of 2 (as discussed in section 3), the first ½ QL is oxidized at ~3

hours. Here we refer to the oxidation of the top two layers plus part of the middle layer as the top

½ QL. The RA-SHG data show that half of the change in the top QL has occurred by ~3 hours,

based on the six-fold symmetric RA-SHG pattern and the associated data from other

methods. Thus the oxidation starts soon after 2 hours and continue until the first QL is oxidized

at ~1.5 weeks. The RA-SHG for the p-in/p-out polarization combination shows that most of the

Figure 5-15: Schematic of 2-step surface aging process. (a) Initially, a freshly cleaved sample
(0001) surface is atomically smooth. (b) Between 25 and 120 minutes after exfoliation, thin
(~3.6 Å) patches nucleate and grow on the crystal surface. (c) After 120 minutes, the surface
begins to oxidize and roughen (not shown), obscuring the patches and expanding the surface
QL, eventually reaching an oxide thickness of ~1.9 nm. Adapted from [115].

141
oxidation induced change is done after ~12 hours. There is only a minimal change in the RA-SHG

data between 12 and 24 hours. The single oxide film model for the SE data also shows a small

change in oxide thickness between 12 and 24 hours. The dielectric function generated shows that

this oxide layer has a sufficient band gap to electrically insulate TSSs from conduction to layers

ostensibly in contact with the Bi2Se3 surface. This may act as a unwanted barrier to transmission

in device applications of Bi2Se3 if not accounted for.

142
Chapter 6: Characterization of Bi2Se3 with a Selenium Protective Layer and

after Selenium Layer Removal

The surface of Bi2Se3 is fragile, and may easily be altered by any of the various defects

mentioned in Ch. 3. As previously described, defects such as surface oxidation and Se or Bi

segregation or desorption alter the surface band structure. In order to utilize a Bi 2Se3 in an

electronic device, the material must be kept in pristine condition. This may be a particularly

difficult task during the fabrication process, where the variety of chemical environments that a

material may be exposed to (including ion etching, chemical and physical depositions, polishing,

etc.) are capable of inducing defects at the Bi2Se3 surface. Although some evidence supports the

viability of surface oxide removal process as well as an electrostatic doping correction, the capacity

to repair a Bi2Se3 surface once it is integrated into a device is miniscule. Some promising means

of producing these devices involve protecting the Bi2Se3 surface during fabrication with an inert

surface layer, or cap. This may be formed in a variety of ways. Some oxides that are familiar to

the semiconductor industry, such as ZnO, are inert. These types of dielectric caps may both protect

and maintain a pristine crystal, as well as isolate the conduction at the Bi2Se3 surface. This is in

contrast to an inert metal deposition, such as gold, which may not interact with the Bi2Se3, but

provides conductive states for TSSs to transfer into, thereby losing their helical character. In either

of these cases, the cap may not be applied in the same chamber that the Bi2Se3 is grown (given that

a dedicated chamber is a likely requirement for high quality Bi2Se3 growths), and there may be no

way to selectively remove the cap without damaging the Bi2Se3 surface, as may be required in

order to make contact to a device. One solution to these problems is to use bulk Se as a cap. It is

trivial to introduce this deposition into a MBE Bi2Se3 growth process, as it is only necessary to

curtail the Bi flux and allow Se to accumulate at the surface once the desired Bi2Se3 layer has been

143
grown. A sufficiently thick Se cap serves as a barrier to surface damage and contamination.

Moreover, it is conceivable that this cap may improve the chemistry of Bi2Se3 films that contain

Se vacancies, as the cap may act as a quasi-infinite Se source from which Se may diffuse toward

vacancies present in the adjacent Bi2Se3 layer. However, the process of removing the cap for the

deposition of contacts to the Bi2Se3 has not yet been sufficiently examined. One way that his may

be accomplished makes use of the high vapor pressure of Se. At relatively low temperatures ( <

150 °C), Se desorbs in a vacuum environment at the rate of > 1 nm/s. Thus, a wide range of

annealing temperatures exist at which Se can be removed in bulk (over the course of minutes for

sub-micron caps), while the underlying Bi2Se3 phase remains. An investigation of this process,

and its impact on the crystallinity, chemistry, and electronic properties of the Bi2Se3 follows.

6.1: Sample Preparation

Se-capped Bi2Se3 samples were grown by Dr. Lee Walsh and Prof. Chris Hinkle’s research

group at the University of Texas at Dallas. Bi2Se3 growth was performed in a V80H MBE system

that is part of a three-chamber MBE cluster system with each of the growth chambers

interconnected with UHV transfer tubes (base pressure = 10-11 mbar). The TMD growth chamber

is equipped with Knudsen effusion cells for the evaporation of Bi and Se and in-situ reflection

high-energy electron diffraction (RHEED) for characterization of the grown films. The sapphire

substrates were sequentially cleaned in acetone, methanol and isopropyl alcohol for 10 minutes

each. The sapphire samples were then loaded into the system and annealed at 600 °C for 90 minutes

and 750 °C for 10 minutes to degas and complete the surface clean. Before each growth, the Bi

and Se sources were outgassed for two hours. The Se:Bi flux ratio was maintained at 20:1 in order

to minimize n-doping. The Se and Bi shutters were opened and closed simultaneously to initiate

144
growth. Growth was performed using a two-step process where a nucleation step was performed

at 110 °C and then the Bi shutter was closed while the sample was annealed to 320 °C under a Se

flux to improve the crystallinity, the second step was then performed at 320 °C until the desired

cap thickness of 200 nm was achieved. The growth rate was determined to be ~0.4 nm/min.

Three rounds of growths were conducted and shipped to Albany for experiments. Samples

from the first round are named SBn, where n indicates the sample in the set. This round also

includes one sample that was never capped, Bi1, which is used as a reference as no annealing was

necessary to reveal its surface. The second set of samples contained severe abnormalities, and is

generally not included in analysis. Samples from round three are labeled Rd3_n, where n again

indicates the sample in the set. All three rounds of samples were sealed in protective packaging

before transportation and arrived in nominally ideal condition.

6.2: Results

The purpose of these experiments is to test whether an optimized annealing process may

remove the Se cap, while maintaining an intact, high quality Bi2Se3 film. The criteria for assessing

the quality of these films are their crystallinity, surface chemistry, and key electronic properties

such as the Fermi level. The films must retain a highly crystalline structure, and they must maintain

the expected, appropriate properties throughout the annealing process. Therefore, these criteria are

measured before and after anneals. The preliminary reference measurements are performed with

XPS and SEM to assess the Bi2Se3 crystal structure and cap coverage. It is necessary for the usage

of Bi2Se3 TSSs in devices that the Fermi level resides in the band gap. Thus, the chemical analysis

of the film surface is paramount in diagnosing the validity of the annealing decap technique, and

the utility of the Se cap in general. To this end, particular care was taken in examining the surface

145
chemistry and the EF after anneals. STS electrical analysis is performed in addition to XPS

chemical analysis, so that direct measurements of the Fermi levels of these samples could be made.

6.2.1: Crystallinity and Chemistry of a Capped Sample

Because the goal of this study is to measure the crystalline and chemical changes in the MBE

Bi2Se3 film that occur with Se-decapping anneals, it is necessary to make preliminary structural

and chemical measurements that serve as a baseline. A survey consisting of SEM imaging, XRD,

and XPS spectra is performed in order to generate a complete picture of the initial samples.

Initially, AFM measurements were attempted, as the Se cap was expected to be amorphous, and

only rough on the order or < 10 nm. However, it was immediately apparent that the surface had

much larger topographical features. This motivated SEM imaging, which provided more complete

morphological information.

SEM imaging is performed with a Zeiss LEO 1550 at 5 kV accelerating voltage and with the

system’s InLens detector. A standard aperture of 30 μm is used, which provides enough signal

with minimum sample damage to clearly see surface features. Images are taken normal to the

surface and up to 75° from normal incidence in order to allow vertical measurements of the

features. All images of the cap clearly show crystallites laying on the surface (Fig. 6-1). In many

areas, these are rods, lying flat and bundled together. In areas with well-aligned rods, patches of

rods are rotated in 30° intervals, making right angles and triangular structures. These are measured

at up to 5 μm in length. There are also areas where crystallites have a less defined

146
Figure 6-1: SEM images of Se-capped Bi2Se3 films, showing the crystallites on the cap surface.
(a) Large domains exist on the surface of Se cap (bright and dark). In some, there are large
crystallite formation areas that are oriented in symmetric directions. In others, the orientation
of crystallites appears more random. (b) An image taken at 75° from normal shows the heights
of crystallites. Some are up to 200 nm.
147
 Figure 6-2: -2θ scan of capped sample SB1. This sample shows the expected peaks for the Si
carrier wafer, the Al2O3 substrate, the Bi2Se3 film, and the crystalline Se cap. Peaks that are
labeled only with number indicate that the peak comes from the (000n) plane of the Bi2Se3 film.

patches of crystallites with well-defined angular separations. However, they still form bordered

structures on the order of 0.5 μm that appear crystalline. In both areas, crystallites of up to 200 nm

in height are measured. There are no large areas that are smooth, which indicates that while the

cap structure is not as expected, it appears to completely cover and protect the film.

XRD is again performed with a Bruker Bede Metrix-L Diffractometer, with Cu Kα x-rays at

1.5406 Å. Because the cap is less than 1 μm thick, and the Bi2Se3 film is less than 50 nm thick,

these measurements are able to acquire diffractions from the Se cap, the Bi2Se3 film, and the Al2O3

substrate. An -2θ scan along the surface normal direction shows the (004) peak of the Si carrier

wafer, the (0 0 6n) peaks of the Al2O3 substrate, the (0 0 0 3n) peaks of the Bi2Se3 film, and some

peaks from the crystalline Se cap (Fig. 6-2). Again, the Bi2Se3 intensities are as

148
 Figure 6-3: Pole figures along the (0115) planes taken on the three rounds of MBE-grown
Bi2Se3 samples prior to any thermal treatment, while the Se cap remained. The scale in the
upper left of all figures gives the tilt of the plane from the (0001) surface normal. (a) Round 1
shows six diffraction points, each azimuthally separated by 60°. Their intensities are not 3-fold
symmetric. It is likely that this pole figure should show 6-fold symmetry, but some unexpected
sample tilt during measurement caused deviation in the form of weaker diffraction points near
120° and 300°. (b) Samples from round 2 still had six diffraction spots. However, the intensity
they produced was nearly an order of magnitude smaller than the other two rounds (see scale
bar), and the scan was noisier. These samples had much poorer crystalline quality. (c) The third
round of samples produced outstanding pole figures. These were 3-fold symmetric, containing
major and minor diffraction spots. This indicates that while there are twinned grains in this
sample, one orientation is dominant.

expected, showing strong diffractions from the (0006) and (0 0 0 15) peaks. This scan also contains

strong evidence of good epitaxial MBE growth, as the l planes of the hexagonal Al2O3 substrate

and Bi2Se3 films are in alignment. Pole figures along the (0115) planes are also performed in order

to assess the polycrystallinity of the samples. As in the case of the oxidation study, these are chosen

for their high intensity, as well as for the grain information contained in their 3-fold symmetry, or

deviance from it. Three sets of film growths were used in this study, and each produces a different

quality pole figure, indicating differing crystallinity between the growth rounds (Fig. 6-3). In all

sample sets, the 4-fold symmetric Si (111) peaks from the carrier wafer are also present. In the

first round, pole figures are generally 6-fold symmetric, barring tilt effects. In the second round,

pole figures were still six-fold symmetric, but the peaks produced less intensity, and the images

149
were noisier. In the third round, the pole figures are three fold symmetric, with one 3-fold

symmetric set of points producing significantly higher intensity than the other. This indicates that

there was a dominant grain direction in the third round of samples, but not in the first two. The

poor diffraction quality of the second set of samples also indicates that the crystalline structures in

that round were not well formed. This data in all measurements of the second round is similarly

poor, and complete crystallographic and chemical analyses were not performed on these samples.

However, from these XRD measurements, it is concluded that crystallinity seen in rounds 1 and 3

were suitable for study of the effects of thermal decap anneals.

XPS measurements are also conducted prior to annealing. These provide a chemical identity

of the capped surface, and may indicate gaps in the cap that could expose the Bi 2Se3 film to the

surrounding environment. Fig. 6-4 shows the resultant Se 3p, Bi 5d, and O 1s spectra. There is a

significant Se 3d peak in these capped samples, as expected. The Se 3d5/2 peak is centered at 54.73

eV. This is a 1 eV shift from the peak center previously seen in cleaved Bi2Se3 samples (53.7).

This shift is due to the change in Se chemical bonding states. Whereas the Se cap is unary, the Se

atoms in Ch. 5 were influenced by surrounding donor Bi atoms, which had the effect of driving

electrons toward the Se and lowering its core binding energies. The Bi 5d and O 1s spectra are not

present. This indicates that no Bi was near enough to the surface to eject photoelectrons, i.e. the

film is completely capped. It also indicates that only trace amounts of oxygen may be on the

surfaces of these capped samples. This is interesting to note, as it indicates that the Se caps do not

oxidize. The Se oxide seen in Bi2Se3 is likely facilitated (or perhaps

150
 Figure 6-4: XPS results from a capped Bi2Se3 sample. (a) Se is the only element producing any
major features in these scans (as well as others such as the survey scan). Pictured is the Se 3d
photoelectron energy range, showing the Se 3d5/2 (54.73 eV) and Se 3d3/2 peaks. (b) No peaks
are apparent in the Bi 5d range. Comparison to Table 4-1 shows that any Bi 5d signal from
Bi2Se3 should occur at 25.2 eV. (c) No peaks are apparent in the O 1s range. This is interesting,
as we were unaware that the Se cap does not oxidize.

enabled) by the larger chemical or structural environment of the Bi2Se3 surface. From these data,

we gather that the thermodynamic process of breaking Se-Se bonds to form Se-O in these

crystallites is less energetically favorable than the process of breaking Se-Bi bonds to form Se-O

at the Bi2Se3 surface, or that the kinetics pose a large, inhibitive barrier to the reaction.

6.2.2: Cap Desorption and Thermal Effects on Surface Crystallinity

The first experiment performed on these MBE films was aimed at tracking the surface

desorption caused by various annealing temperatures, and subsequently measuring the crystallinity

of the surface of the Bi2Se3 film. Annealing and desorption measurements were conducted with

TPD, as described in Ch. 4.7. Samples are loaded into the high vacuum pre-pump chamber and

left to outgas for a minimum of 12 hours before anneals were performed. Because the system was

not capable of measuring atomic masses above 200 amu, the cap desorption and Bi2Se3 desorption
78 80
events were measured with Se spectra. Both of the major isotopes, Se and Se (which have

151
 Figure 6-5: Decomposition and desorption of the Bi2Se3 film. (a) TPD measurements show a
second, minor peak of both Se isotopes at high temperature. This is measured at a much higher
temperature than the expected Bi2Se3 desorption temperature of 325 °C, as discussed earlier.
The inaccuracy in this early measurement is exacerbated by the thermocouple’s placement near
the edge of the sample, a thicker thermocouple wire, and metallic mount washers, which were
replaced for later measurements.

expected ratios of 23.77%:49.61%, or 1:2.09) are measured during these processes. Anneals from

room temperature ranging between 90 °C and 500 °C showed multiple Se desorption events. Those

measured near 130 °C are ascribed to the Se cap desorption, and used for temperature calibration.

As discussed in Ch 4.7, while this was the best method found to calibrate TPD anneals, it is not

accurate at higher temperatures, and thus only used qualitatively, as an indicator of desorption

events. Measurements that contained a second Se desorption peak, resulted in a visibly damaged

Bi2Se3 film. These were usually close to 400 °C, which is known to be due to bad temperature

measurement based on reports of Bi2Se3 decomposition in literature. This indicates that the second
78
Se peak is due to Bi2Se3 decomposition and desorption (Fig. 6-5). All peaks showed Se:80Se

ratios that were in agreement with the expected values.

152
 Most samples are subjected to TPD anneals between the Se and Bi2Se3 desorption

temperatures, as this is the region of interest. Crystallinity is measured with three different surface-

sensitive techniques: LEED, AFM, and STM. Not all samples were measured by all techniques,

as LEED is the only measurement that could be performed in situ with TPD, and time spent out of

vacuum is known to affect the Bi2Se3 surface. The time of this exposure was minimized in order

to keep surface contamination low. For exposure times on the order of hours, oxidation is likely

take place (see Ch. 4). As oxidation penetrates a full quintuple layer, expanding it to 2 nm thick,

the surface crystallinity at the surface may be severely altered if insufficient care is taken during

transfer between measurement stations. Transfers to AFM and STM were completed within 10

minutes, during which time the sample was enclosed in a dry plastic case. Many ex situ

measurements showed no signs of degradation due to transfer.

LEED is performed in situ immediately after the decapping anneal. All samples from rounds

1 and 3 heated between the Se and Bi2Se3 temperatures produced LEED patterns that indicated a

high degree of surface crystallinity. The acceleration energies of these experiments are set to 30,

72.7, and 96.3 eV in order to see the reciprocal lattice of the surface at different scales (Fig. 6-6).

As LEED is only sensitive to the in-plane crystal structure, diffraction spots in these measurements

are denoted by the first three indices of Miller hexagonal 4-notation, e.g. (101n) ® (101) . The 30

eV images show three {101} diffraction spots and their {101} twins. In the first set of samples,

these are approximately equal in intensity. This indicates that there are twinned grains at the

surface of near equal density. In the third set of samples (Fig. 6-6(b)), there is visible 3-fold

symmetry in the {101} diffractions, indicating that the twin domain dominance seen in the

preliminary XRD pole figures appears both in the bulk and on the surface. 72.7 eV LEED images

capture both the ±{10 1} and ± {112} families of diffractions, and the 96.3 eV images capture out to

153
 Figure 6-6: Effects of UHV thermal decapping procedure on the surface crystallinity of sample
Rd3_2, as measured via LEED. (a) A TPD anneal of shows a clear Se cap desorption peak in
both isotopes, and no other Se desorption activity below the maximum temperature of 200 °C.
The warping seen effect is assumed to be due to magnetic deflections from the type-K chrome-
alumel thermocouple. (b) LEED at 30 eV shows a slightly deformed 3-fold symmetric pattern.

the ±{202} points. 3-fold symmetry can still be seen in these higher energy images, but the

intensity difference between twins is smaller. This slight intensity change requires a further

quantitative analysis, which is beyond the scope of this work. These LEED measurements show a

similar level of crystallinity to that shown in the preliminary XRD measurements, albeit at the

154
surface level and not throughout the bulk of the film. This demonstrates that there are areas of

surface where the crystal lattice remains intact during the decapping anneal, although as LEED is

a macroscopic measurement, these patterns do not rule out the presence of defects in the surface

structure. As we expect the crystallinity in the bulk of these films to be as good or better than that

seen at the surface (which may be subject to atmospheric effects), these LEED experiments allow

us to conclude that significant portions of the Bi2Se3 films maintain their crystallinity during the

thermal decapping process at all attempted temperatures.

Topographic measurements were performed with a Bruker Dimension Icon AFM (Fig. 6-7).

Scans were taken at 1, 5, and 10 μm2 in order to see the changes in topography induced by

annealing over a large area, as well as to see the details of grains. However, STM topographic

measurements are limited to a maximum of 1 μm2, and are better suited for observations of detailed

grain structure changes. Samples from round 1 were subjected to AFM measurements after TPD

decapping and LEED measurement. These films have pyramidal topographies, composed of QL

layered triangular grains, which shrink in area with their vertical stacking (Fig. 6-7). These features

are similar to the MBE-grown films seen in literature37, and create surfaces with ~20 nm

roughnesses. While these films are not flat enough to be suitable for most applications, their > 1

μm2 grains make them useful for intragranular measurements of chemical alterations, and for

observations of the changes in crystallinity of visible features. Sample Bi1 was unique in that it

was never capped. This served as a reference for the effect of annealing on grain structures. All

round 1 samples were topographically similar, and contained grains of up to ~1.5 μm2. Decapping

anneals did not appear to significantly alter the triangular structure of the grains, although no

samples displayed the sharp 60° corners seen in sample Bi1. While LEED indicated little change

of the lattice at the surface, AFM shows that the geometry of grains is likely slightly altered by

155
 Figure 6-8: STM images of sample Rd3_1, thermally decapped in situ prior to measurement.
Figure 6-7: AFM images of the surface of the first set of samples taken over 5×5 μm2 areas.
(a) Image taken after a 165 °C anneal, showing the flat surfaces and rough edges of the Bi2Se3
Although these scans show different scales of roughness, results not shown indicate that these
grains. (b) Image taken after a 270 °C anneal, showing the spiral growth structure seen in many
levels are very similar, and that all films have pyramidal structures rising ~20 nm above the
grains. There is little difference in the structure of layered (or spiraled) QL grains seen between
background. (a) Sample Bi1, which was never capped or decapped, may act as a reference for
these anneals.
changes induced by annealing. (b) Sample SB5 was annealed at 225 °C. Beyond somewhat
more rounded corners in SB5, the structure remains.

annealing. It should be noted that because all annealing temperatures produced similar

topographies, there is some ambiguity in whether the capping or decapping process created these

modifications to grain structure. However, as the film growth process was identical between Bi1

and SB5, and Se accumulation is not expected to damage the surface, we conclude that the changes

in grain seen in AFM are likely due to thermal annealing.

STM imaging was performed with an Omicron LT STM. Samples Rd3_1, Rd3_5, and Rd3_6

were all annealed in situ prior to STM measurement. These samples were initially annealed at

different temperatures, e.g. sample Rd3_1 was annealed at 165 °C, sample Rd3_6 was annealed at

200 °C, etc. Each was also subsequently ramped to higher temperatures at approximately 35 °C

intervals until the Bi2Se3 film decomposed, and measurements were taken between each heating

step. The Se and Bi2Se3 desorption events (as could be seen by visible changes in the surface) were

used to calibrate the anneal temperatures. These measurements were difficult and inconsistent (in

156
that many images contained too many artifacts to be of use, and image quality was difficult to

correlate with sample preparation conditions), but some scans provided topographic information

at small scales. No attempts were made at atomic resolution imaging. Most of the scans in this

study were 1 um2, which is on the order of the size of grains seen in round 1 and 3 samples. Samples

subjected to this annealing and measurement process displayed generally triangular pyramidal

stacked QL grain structures. Some grains showed a spiral vertical growth pattern (Fig. 6-8(b)),

and the corners and edges of these round 3 samples were more irregular than the annealed round

1 samples. Because the first set of samples (which were annealed to different temperatures in TPD)

all had straighter edges, and the topography of the 3rd set of samples were generally unaffected by

successively hotter anneal temperatures, it is likely that the difference in edge shape seen in these

images is due to slightly different growth conditions, and not due to the annealing process. Similar

to AFM images, STM topographic scans show that annealing between the Se and Bi2Se3

temperatures had minimal effect on the surface topography of these films, and that these films

remain crystalline during the annealing process.

6.2.3: Thermal Effects on the Surface Chemistry and Fermi Level

The surface chemistry and Fermi level of these samples is of critical importance, as it is the

determining factor to whether the TSS are overwhelmed by metallic bulk conduction. Some

samples that were subjected to TPD and LEED measurements were then taken ex situ to the XPS

system in order to characterize the surface chemistry of the Bi2Se3 film. Others were annealed in

157
situ in the STM and immediately measured with STS to find the Fermi level. The second set of

samples produced linear current levels near 0 bias. This behavior is indicative of metallic states,

and is not an expected quality in ideal TIs. The first and third set of samples each contained a

region near 0 bias where there was very little current, due to the band gap.

XPS was performed under the same conditions as in previous measurements. Spectra were

acquired at the Se 3d, Bi 5d, and O 1s peaks so that chemical changes between the capped and

decapped samples could be measured (Fig. 6-9). The annealed films demonstrated the expected

surface composition. The Bi 5d signals are strong. Because XPS extinction distances are low, this

excludes the possibility of any Se cap greater than a few nm. Additionally the Se 3d 5/2 peak was

measured at 53.63 eV, which is within normal error of the Se binding energy in Bi 2Se3, and

significantly different from the cap Se 3d5/2 binding energy of 54.73 eV. The higher energy portion

of this spectrum is entirely attributable to the Se 3d3/2 peak. There was some oxygen measured in

most post-decap XPS experiments. This was to be expected from ex situ measurements, as even

Figure 6-9: XPS spectra taken on sample Rd3_2 after TPD decapping. This sample was
annealed to 165 °C, and shows a complete decap, as well as some oxidation. (a) The lack of Se
cap can be seen from the energy of the Se 3d peak, which has shifted 1 eV to its expected Bi2Se3
chemical state position of 52.63. (b) Because XPS is sensitive to within a few surface nm, the
high Bi signal also indicates that there is no Se cap. However, there is some Bi oxidation, as
can be seen by the left-skewed Bi 5d doublets. (c) Oxygen is more present at this surface than
it was in the capped sample surface. This may originate from the oxidized film or from adsorbed
hydrocarbons.

158
rapid transfers between tools result in air exposure, and once adsorbed, may result in oxidation

within the time period that XPS is performed.

The main results of this study are seen in STS measurements. The critical criterion for

annealing is that it does not shift the EF into the bulk or conduction band. STS is used to directly

calculate the EF by measuring shifts in the LDOS at the surface (see Ch. 3.9). This is possible

because not only is the DOS function of Bi2Se3 known, the Dirac point provides a clear reference.

STM data is used to calculate how far the Fermi energy has shifted from the Dirac point, EF - ED .

STS measurements are performed on samples from round 3 after in situ anneals. Each sample is

heated to an initial temperature, measured, and heated again at a higher temperature for a

subsequent STS measurement. This is repeated until the Bi2Se3 film decomposes. In all

measurements, the samples are held at ground while the tip was biased. A lock-in amplifier was to

perform dI/dV measurements on some of the first set of samples, but equipment issues prevented

its further use. Instead, IV measurements are performed and fit with the analytical function,

f (x) = eb1 (x-x1 ) - e-b2 (x+x2 ) +Cx + D (Eq. 6-1)

in proximity to inflection point, which is taken to represent the Dirac point, as can be seen in Fig.

159
 Figure 6-10: Typical STS results and analytical fitting used to derive EF - ED . These spectra
were measured on sample Rd1_1 after it was annealed to 265 °C. (a) IV spectrum, with data in
circles and the fit function in red. (b) Direct derivation of the data in (a), calculated with
significant smoothing. As most spectra had more noise, which was amplified by derivation,
fitting the IV curves was found to be more reliable. The dI/dV spectrum shows the Dirac point
(minimum dI/dV) at higher energy than the Fermi level (0 bias), indicating that this sample is
slightly p-doped.

6-10. Many spectra were too noisy for evaluation due to either the sample quality after Se cap layer

desorption or equipment issues, and the spatial distribution of Fermi levels was found to be

inconsistent, and thus a statistical approach is taken to assess the doping levels of these films. Over

200,000 IV spectra were analyzed between samples Rd3_1 and Rd3_6. Spectra with a high noise

level or were taken when the system was not functioning correctly were not used in the analysis.

Selection of the spectra was done using the spectral shape as well as the error of the fit curve.

Fermi shifts are extracted from the remaining spectra, as shown in Fig. 6-11. The

160
Figure 6-11: Histograms of Fermi level shifts in sample Rd3_1 after annealing temperatures
of 150 (a), 180 (b), 210 (c), 240 (d), 265 (e), and 295 °C (f). Each inset shows the number of
fit spectra used to construct the histogram, n, the median Fermi level shift, E*, and the standard
deviation of the set of measurements. While no median Fermi level is shifted more than 35 meV
from the Dirac point, there was significant variance in these measurements.

161
resolution of each of these measurements is determined by the Fermi-Dirac distribution. At room

temperature, the FWHM of its derivative is 2kT ln (3+ 2 2 ) = 0.092 eV. Thus, each individual

measurement may not be as precise as desired for this experiment, but the acquisition of many

ensures the accuracy of the EF study. The histograms constructed from filtered EF - ED show low

median shifts (maximum EF - ED median is 35 meV), but high standard deviations (maximum is

125 meV). The trend of these medians and standard deviations is more easily seen in Fig. 6-12. In

sample Rd3_1, the standard deviation of Fermi level shifts is lowest below anneals of 200 °C. For

anneals of 210 °C and above, the standard deviations rise above 100 meV. This indicates that lower

temperature anneals produce higher spatial uniformity of Fermi levels in these films. Increasing

anneal temperatures do not lead to uniform increases in the Fermi level. In both samples Rd3_1

and Rd3_6, the highest median Fermi level occurs between anneals of 210 and 255 °C. These

median shifts were 35 and 37 meV, respectively. Hotter anneals appear to correlate with a return

to the Dirac point. This effect is not well understood, and will be discussed in the analysis section.

Gaps of time between measurements are also significant factors in the Fermi level shifts, as clearly

shown in Fig. 6-12(b). After sample Rd3_6 was annealed to 225 °C, it was left in vacuum for more

than three weeks while scans were intermittently performed. This dwell time appears to cause a

rise in the Fermi level, which is reset by the next anneal. During STS measurements, at no time

did the median Fermi level rise or fall to levels near the valence or conduction bands, as the band

gap of 300 meV is an order of magnitude larger than measured EF shifts.

162
Figure 6-12: Statistical trends of Fermi level shifts over various anneals and time. (a) Sample
Rd3_1 was annealed at 150 °C and higher until the Bi2Se3 film decomposed. The numbers
above each line indicate how many days had passed since the film was decapped. (b) Data from
sample Rd3_1 plotted as a function of time rather than temperature. (c) Sample Rd3_6 was
annealed from 225 °C until it decomposed. The effects of dwell time are emphasized in this
sample
163
6.3: Discussion

A variety of experiments were performed on Se-capped Bi2Se3 films to test whether they may

remain structurally and chemically intact during a Se decapping anneal process. Many

temperatures were tested between the Se desorption temperature and the Bi2Se3 decomposition

temperature. Comparison between XRD measurements conducted before decapping and LEED,

AFM, and STM conducted after decapping shows that the anneals have little affect on the structure

of the film. LEED shows that the lattice is intact at the surface, and AFM and STM topographical

scans show that triangular grains and pyramidal structures, as often seen in literature, remain after

anneals. Chemical changes at the surface are measured with XPS and STS. The former shows that

the Se bonds detected at the surface contain the expected chemical shifts for Bi2Se3. The latter

shows a variety of annealing effects on the Fermi levels of the films. Between annealing

temperatures of 150 and 255 °C, the Fermi level tends to rise. However, subsequent hotter anneals

correlate with a return to the Dirac point. The only proposed p-doping mechanism for these films

is adsorption of O2. Oxygen may have been adsorbed during measurement or between anneals, as

some samples were left exposed to 10-10 Torr for as much as three weeks. However, no p-doping

trends are seen during in situ dwell times. Because there are multiple factors that may potentially

affect the electrical environment at the surface of these films, including oxidation and Se

evaporation, it is difficult to determine the cause of some changes. However, it may be concluded

that thermal annealing is a suitable method to decap these Bi2Se3 films, as they maintain the

structure of their growths, and exhibit no significant Fermi level shifting during the process.

164
Chapter 7: Discussion and Conclusions

This thesis illustrates the potential utility of Bi2Se3 as a TI material, some of the current

obstacles barring applications including a detailed study of oxide growth, and shows that a Se cap

may be used to protect the Bi2Se3 surface.

The physics that govern the TI class was discussed in detail. This provides the reader

information about how the time reversal polarization of Z2 topological insulators may be

calculated, and how it acts as a topological invariant. This is central to understanding the

emergence of TSS and their protection. After the lengthy explication of TI physics, the emergent

phenomena of TSSs and their potential applications were discussed. Finally, a variety of defects

in Bi2Se3 seen in the literature were introduced, and an assessment of their potential impacts on

surface conduction was presented. The experimental work in this paper is aimed at creating a better

understanding of the generation of oxide formation, a common defect, and how it and other defects

may be avoided. A chapter regarding the experiments used in this thesis followed. It included

information about vacuum and surface science, as well as the theory and practical applications of

XRD, spectroscopic ellipsometry, XPS, TEM, AFM TPD, LEED, and STM/STS. These tools were

all essential in the hybrid approach taken to studying Bi2Se3 defects.

There were two major experimental studies presented in this thesis. The first investigated the

oxidation process at the surface of exfoliated Bi2Se3 samples. This has been the subject of some

debate in the academic literature, with some research groups reporting immediate oxidation of all

samples upon exposure to air, and others reporting no oxidation up to a time period of years. This

study addresses the issue with a hybrid approach, using multiple techniques to assess the dynamics

of oxidation at the Bi2Se3 surface and generating multiple optical methods to non-destructively

track its formation. The samples used in this study were initially measured with XRD and TEM to

165
show that they were of high quality. TEM demonstrated a pristine lattice along the (0001) and

(1010) zone axes, showing the hexagonal lattice with expected lattice parameters in plane view

and intact and vdW separated QLs in the cross section. As TEM may only be conducted over small

areas that do not necessarily represent the macroscopic properties of a sample, XRD was

performed to supplement measurements of the bulk crystals. Omega-2theta and pole figures

showed the expected vertical separation of layers in each sample, and pole figures showed that all

grains were oriented in the same direction. This showed that these samples while there may have

been some structural imperfections in these crystals, they were nearly single crystalline and

suitable for studies of morphological and chemical changes. AFM measurements showed that the

exfoliation technique produced atomically flat surfaces, but when exposed to air, showed the

growth of unidentified surface patches and began to roughen over the time scale of hours. XPS

performed on samples exposed to air over the same time period showed the nucleation and growth

of a surface oxide. Chemical shifts in the Se and Bi spectra show that this layer contains both Bi

and Se oxides. Samples that were exposed for a variety of time periods before XPS measurement

show that no oxide forms within 2 hours of exposure, but that a surface oxide layer is universally

present when crystals are exposed to air for 3 or more hours. Subsequent spectroscopic

ellipsometry measurements confirm this time scale. Optical models created from 245 to 1700 nm

wavelengths facilitate the measurement of this oxide growth, and provide a means to monitor the

surface quality in real time. The dielectric model of this ‘oxide layer’ shows a significant band gap

of 2.7 eV. Additional SHG measurements were performed and calibrated with XRD. This

experiment showed the impact of surface bond directions oxide formation on SHG. Second

harmonic waves are generated away from the surface bond direction when the Bi2Se3 surface is

fresh, and are generated toward the surface bond direction when it is oxidized.

166
 The second experimental study in this thesis is aimed at solving the issue of defect state

formation at the surface of MBE-grown Bi2Se3 films. The proposed technique is Se surface

capping, which may protect the film from physical damage, environmental contamination, and Se

evaporation. This may mitigate or eliminate the effects of 2DEG formation, n-doping, and surface

oxidation. In order for this solution to be viable for most experiments or device fabrications, the

Se cap must be removable with minimal damage or alterations to the underlying Bi 2Se3 film. The

high vapor pressure of Se (5×10-5 Torr at 130 °C) motivates a series of low temperature anneals

(150 - 295 °C) to thermally desorb the cap. The criteria laid out to assess the quality of the film

are that it maintains its initial crystallinity and that the Fermi level at the surface remains inside

the band gap. This is a strict requirement for any implementations of Bi2Se3 that involve

measurements or manipulations of the TSSs, as metallic behavior overwhelms the TSS signals

when EF is in the bulk or conduction band. As in the oxidation study, initial measurements were

performed to assess the crystallinity of MBE-grown Bi2Se3 films. Multiple growth rounds created

differing crystallinities, as assessed with omega-2theta patterns and pole figures, and showed

strong 6-fold symmetric diffractions in the first round of samples, poor though still 6-fold

symmetric diffractions in the second, and strong 3-fold symmetric diffractions from the third round

of samples. Anneals were performed with TPD measurements or in-situ in the STM chamber.

Samples annealed with TPD were subjected to LEED measurements that showed surface

crystallinity that was highly correlated with pre-anneal XRD measurements. Samples retained their

6-fold or 3-fold symmetries, and the intensities of LEED patterns after annealing matched the

intensities of XRD pole figures. Subsequent AFM and STM measurements were performed on

annealed samples and compared to a sample that had never been capped or annealed. These

displayed similar surface QL pyramid features. Annealed samples maintained triangular surface

167
geometries, although corners and edges of these grains were more rounded, and in some cases,

more irregular than seen in the never-annealed sample. XPS showed the effect of anneals on the

surface chemistry of the films. All anneals resulted in ~1 eV binding energy shifts of the Se 3d

peak. This indicates that no Se-Se bonds remain on the surface, and that the Se cap is removed to

within detection limits. Shifts of the Fermi energy were measured with STS. IV spectra show the

LDOS at the sample surface, which is indicative of the Fermi level relative to the Dirac point

energy. Anneals of up to 250 °C showed increasing n-doping due to EF shifts above the Dirac

point, with maximum EF - ED reaching 38 meV. This was expected from Se vaporization trends.

However, anneals beyond 250 °C resulted regressions of EF back to ED. This behavior is still not

well understood, and is left to future directions in the field. Higher anneal temperatures also

appeared to cause a larger spatial distribution of EF shifts, as is visible in the statistical analysis of

the STS data. Dwell time also appeared to be a factor in Fermi level shifts. A sample left in UHV

for three weeks showed a steady increase in the Fermi level over time. This effect was negated by

the next anneal. It is possible that this trend was due to surface contamination in the chamber that

was removed by subsequent heating.

These studies show that Bi2Se3, while delicate, may be placed in optimized conditions that

allow its reliable usage in devices and experiments. The surface oxidation study presented rectifies

some disagreements in literature about the time scale of oxidation. A previous study that showed

immediate oxidation incorrectly attributed this to the presence of a Se 3p peak in XPS

measurements that occurs at the same energy as the Bi 4f oxide peak. While immediate oxidation

is not the cause of degradation in TSS signals, exposure to air reliably causes oxidation to occur

in Bi2Se3, even at atomically flat surfaces. The surface patches in AFM may also serve to bridge a

gap in literature regarding the existence of Bi bilayers in these systems. While unconfirmed, we

168
propose that these may be caused by surface Bi segregation. They are of uniform thickness similar

to previously reported Bi bilayers, they appear immobile during measurement, they display a

nucleation, growth, and stagnation process that appears less random than adsorption of surface

contaminants, and differ categorically from other scans showing significant surface contamination.

Additionally, ellipsometric and SHG methods developed to monitor the oxidation process may be

used in the future to help understand the mechanisms of oxidation, and could be used to help parse

the different defects that emerge at the Bi2Se3 surface. While ‘oxide’ layer of the ellipsometric

model may actually be a representation of multiple surface modifications at some time scales, the

insulating properties it exhibits pose an additional challenge to fabricating a Bi 2Se3 devices, as

making electrical contact through this oxide layer may be a non-trivial task. SHG could also be

used as a method of non-destructively determining the orientation of Bi2Se3 crystals during the

growth process to supplement or supplant RHEED. The analysis of Se decapping anneals shows

that this method can be used to protect the surface of Bi2Se3. This may be paramount to fabrication

procedures, where the surrounding environment may differ significantly from MBE growth

chamber conditions. As the oxidation study discusses and the literature reports, these surfaces are

highly fragile in the face of oxidating species, and may be detrimentally doped by exposure to

other elements in fabrication chambers such as metals. The Se capping procedure allows the Bi2Se3

film to remain in good condition until surface exposure is desired, and the thermal decapping

procedure shows minimal effects on the quality of the Bi2Se3 films. Ideal decapping temperatures

were found under 200 °C, where the Se cap was completely removed, high surface crystallinity of

the films was confirmed, and EF was measured very near ED. Excessive annealing temperatures

result in some n-doping and higher degrees of spatial EF variance on the surface, although the

median EF was never shown to shift to the valence or conduction band for any attempted anneal.

169
Prior to such alterations, the Bi2Se3 samples decomposed. We can therefore conclude that this is a

promising method to help the implementation of Bi2Se3 in devices.

Further experiments are needed to understand the behavior of the Bi2Se3 surface. We were

unable to definitively answer questions regarding Bi bilayer formations or what the microscopic

behavior is that causes anneals above 250 °C to show ED » EF . Dynamic SIMS is a likely technique

to help address these issues, as it is precise to orders of magnitude lower than the surface coverage

of observed patches requires. As the only tool available during this thesis work utilizes a Bi beam,

such experiments were not possible. Additionally, SIMS may help determine the Se concentration

trends for Bi2Se3 samples under various thermodynamic conditions. For known Se concentration

and EF levels, the ionization energy of Bi donors near Se vacancies may be assessed, and used to

help determine what surface doping levels are purely attributed to Se evaporation. Many

characterization steps are also necessary for the fabrication process. Interactions between various

metal contacts have been assessed, and the basic properties of TSSs are well known, but few

devices have been fabricated. Decoupling the surfaces is a necessary step toward spin filtering.

This may involve multiple vertical contact layers to each terminal of the sample so that the net

spin conduction of the top and bottom surfaces do not cancel. This may also be accomplishable by

doping the surfaces to significantly different levels, as net spin conduction of any completely filled

surface Dirac cone is zero, due to the opposite helicities of the top and bottom halves of each cone.

It may also be useful to decap parts of these films while leaving other parts unaffected. This may

be accomplishable with precise, targeted heating methods such as rapid laser annealing.

Implementations of Bi2Se3 for more exotic purposes such as quantum computing may soon be

possible, and outside the realm of TI, new topological systems such as Weyl and Dirac semimetals

170
are recent subjects of academic interest. Each prospect will require rigorous scientific study and

due diligence, but may result in a dynamic new regime of solid state physics and its applications.

171
Chapter 8: Appendix

8.1: Quantum Hall Calculations from Linear Response Theory

The Kubo formula used by Thouless et al. to calculate the quantum Hall effect can be derived

from linear response theory. This derivation involves calculating the linear response of a system

(given by the Hamiltonian) subject to a time-dependent perturbation. In order to do so, it is

common to use the interaction picture, which is distinct from the Schrodinger and Heisenberg

pictures in that both wavefunctions and operators are functions of space and time.

Take a base 2D Hamiltonian and eigenstates H 0 n = En n , where n = 0 denotes the ground

state. To get the conductivity of this system, we perturb the system with an electric field,

, which gives the vector potential (seen in Maxwell’s equation

for f = 0 ). Later, we’ll take the limit w ® 0 , but it’s easier to have an oscillator for now. The

addition of the electric field brings the Hamiltonian to H (t ) = H 0 + DH(t) , where

, and Ĵ = mL-ep̂L is the current density operator. For now, the goal is to calculate
x y

J . In the interaction picture, operators evolve via a unitary change of basis

Ô(t) = V -1 (t - t0 )Ô(t0 )V(t - t0 ) , where , and states evolve as y (t - t0 ) =U(t - t0 )y (t0 ) ,

where (where T stands for time ordering). Thus, the

current density at time t of state 0 is:

172
Taking only up to the linear terms of the matrix exponential leaves

Taking only up to the linear terms of the perturbation leaves

The first term is ignored as it represents the current in the absence of an electric field. Plugging

in for the perturbation, we get:

For clarity from here, we’ll take Lx Ly to be the total sample area, A , and use Einstein

summation notation over perturbation components j.

We may use u-substitution t¢¢ = t - t¢ . We also take t0 ® -¥ , and for simplicity, drop t0 from 0

To further simplify, we use Ô(t) = V -1 (t)Ô(0)V(t) :

0 éë Ĵ j (t - t¢¢), Ĵi (t)ùû 0 = 0 éëV † (t - t¢¢)Ĵ j (0)V(t - t¢¢),V † (t - t¢¢)Ĵi (t¢¢)V(t - t¢¢)ùû 0

= 0 V † (t - t¢¢)Ĵ j (0)V(t - t¢¢)V † (t - t¢¢)Ĵi (t¢¢)V(t - t¢¢) 0 - 0 V † (t - t¢¢)Ĵi (t¢¢)V(t - t¢¢)V † (t - t¢¢)Ĵ j (0)V(t - t¢¢) 0

173
 = 0 éë Ĵ j (0), Ĵi (t¢¢)ùû 0

Plugging back into get the current density:

Since the only t-dependence of the expected current density comes from e-iwt , the system

responds to an applied electric field at frequency w with current at the same frequency. This is the

linear response that we’re interested in. To get the Hall conductivity from this point is trivial:

This is the Kubo formula applied to conductivity. There is a more general Kubo formula that

expresses the reaction of any system in linear response to a time-dependent perturbation, but that

is beyond the scope of this section. To see the topological implications of this formula, we further

modify by expanding the commutator and multiplying by ån n:

n

The next step is to plug in the evolution of Ĵ x (t) . And again, 0 Ĵ(t) 0 = 0 , so we may take the

sum for n ¹ 0 , and integrate.

From here, we expand the fractions in as . The first term vanishes,

and we can plug in w = 0 to get the response for a DC field:

174
8.2: Derivation of the Fresnel Equations

In optical experiments, the reflection plane is defined by the propagation vectors (ki, kr) of the

incident and reflected waves, which act as basis vectors for the plane, barring a normal reflection

geometry. Light polarization is given by electric field components of the wave that are aligned in

the reflection plane (p polarized, for parallel) and perpendicular to the reflection plane (s

polarization, from the German ‘senkrecht’ for perpendicular). Given any sample, one can make a

classical approximation wherein an incident polarized wave traveling in the x̂ - ẑ reflection plane

meets a locally flat surface (with surface normal ẑ ), as given by Fig. 8-1. Because fields generated

by the electrically polarized surface are oriented vertically, all EM fields in the plane are

conserved:

Eix + Erx = Etx Hix + H rx = H tx

Eiy + Ery = Ety H iy + H ry = H ty (Eq. 8-1)

Figure 8-1: Schematic of an EM wave incident upon an interface between media. Amplitudes
Erp, Ers, Etp, and Ets, as well as their accompanying phase considerations, are all functions of the
incident complex wave given by Eip Eis, as well as the angle of incidence φ1 and the refractive
indices of both media, N1 and N2. These relations are given by the Fresnel equations, Eqs. 8-4
and 8-5. Figure adapted from [24].

175
 Note that in the given geometry, s-polarization coincides with the y-axis, and p-polarization
105
is in the x̂ - ẑ plane. These equations constitute the basis from which Fresnel coefficients, which

give the complex reflection and transmission of an interface, are derived. The Fresnel coefficients

are functions of angle of incidence and the complex refractive indices of the interface materials.

Thus, in order to derive the Fresnel coefficients, we must incorporate equations containing

information about the angle of incidence and complex refractive indices of at the interface into the

continuity equations.

Toward deriving reflection and transmission coefficients for s-polarized light, we make use of

the electric field equation in the ŷ -direction, and create another from the magnetic continuity in

the x̂ -direction. This is done through the relationship between the electric field, magnetic field,

and propagation vector, (where e = e (w ) is the complex dielectric function of the

media that the light is traveling through). Further, utilizing the definition of the complex refractive

index, k = wc N (where N = N(w ) is the complex refractive index), one can make a complete

substitution of the desired variables to produce N1 cosj1 (Eis - Ers ) = N2 cosj2 Ets . Combining this with

the initial ŷ -direction electric field equation gives the Fresnel coefficients for s-polarized light:

j1-N2 cosj 2 cosj1

rs = NN11 cos
cosj1+N2 cosj 2
and ts = N1 cos2 Nj11+N 2 cos j 2
. These can be further reduced with Snell’s law to remove the

transmission angle:

N1 cos j1 - N 22 -N12 sin 2 j1

rs = (Eq. 8-2)
N1 cosj1+ N 22 -N12 sin 2 j1

2 N1 cosj1
ts = (Eq. 8-3)
N1 cosj1 + N22 -N12 sin 2 j1

Fresnel coefficients for p-polarized light are calculated with the remaining continuity equations

for electric field in the x̂ -direction and magnetic field in the ŷ -direction, as well as the previously

176
given electric/magnetic relation and definition of the refractive index. The p-polarized Fresnel

coefficients, after Snell’s law substitution, are:

N22 -( N1 sin j1 )2 -N22 cosj1

rp = N1 (Eq. 8-4)
N1 N22 -( N1 sin j1 )2 +N22 cosj1

2 N1N2 cosj1
tp = (Eq. 8-5)
N1 N22 -(N1 sin j1 )2 +N22 cosj1

Because the Fresnel coefficients are complex, they completely calculate the change of

amplitude and phase of s- and p- polarized light at an interface. Thus, barring cross-polarization,

which is not expected to occur in optically isotropic samples, they provide complete information

for ellipsometric measurements.

177
 References
1
G.E. Moore, IEEE Solid-State Circuits Newsl. 20, 33 (2006).
2
K. Matthew, (2009).
3
Wikipedia (2017).
4
A. Brataas, A.D. Kent, and H. Ohno, Nat. Mater. 11, 372 (2012).
5
D. Pesin and A.H. MacDonald, Nat. Mater. 11, 409 (2012).
6
E. Hirota, H. Sakakima, and K. Inomata, Giant Magneto-Resistance Devices (Springer Berlin
Heidelberg, Berlin, Heidelberg, 2002).
7
L. Liu, C.-F. Pai, Y. Li, H.W. Tseng, D.C. Ralph, and R.A. Buhrman, Science 336, 555 (2012).
8
A. Makarov, T. Windbacher, V. Sverdlov, and S. Selberherr, Semicond. Sci. Technol. 31, 113006
(2016).
9
K.S. Novoselov, Science 306, 666 (2004).
10
K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, M.I. Katsnelson, I.V. Grigorieva, S.V.
Dubonos, and A.A. Firsov, Nature 438, 197 (2005).
11
K.I. Bolotin, K.J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim, and H.L. Stormer,
Solid State Commun. 146, 351 (2008).
12
J. Martin, N. Akerman, G. Ulbricht, T. Lohmann, J.H. Smet, K. von Klitzing, and A. Yacoby,
Nat. Phys. 4, 144 (2008).
13
W.A. de Heer, C. Berger, X. Wu, P.N. First, E.H. Conrad, X. Li, T. Li, M. Sprinkle, J. Hass,
M.L. Sadowski, M. Potemski, and G. Martinez, Solid State Commun. 143, 92 (2007).
14
P.W. Sutter, J.-I. Flege, and E.A. Sutter, Nat. Mater. 7, 406 (2008).
15
X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, S.K.
Banerjee, L. Colombo, and R.S. Ruoff, Science 324, 1312 (2009).
16
J.-H. Lee, E.K. Lee, W.-J. Joo, Y. Jang, B.-S. Kim, J.Y. Lim, S.-H. Choi, S.J. Ahn, J.R. Ahn,
M.-H. Park, C.-W. Yang, B.L. Choi, S.-W. Hwang, and D. Whang, Science 344, 286 (2014).
17
M. Wang, S.K. Jang, W.-J. Jang, M. Kim, S.-Y. Park, S.-W. Kim, S.-J. Kahng, J.-Y. Choi, R.S.
Ruoff, Y.J. Song, and S. Lee, Adv. Mater. 25, 2746 (2013).
18
H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang, Nat. Phys. 5, 438 (2009).
19
C.L. Kane and E.J. Mele, Phys. Rev. Lett. 95, 146802 (2005).
20
C.L. Kane and E.J. Mele, Phys. Rev. Lett. 95, (2005).
21
(n.d.).
22
F.D.M. Haldane, Phys. Rev. Lett. 61, 2015 (1988).
23
D.J. Thouless, M. Kohmoto, M.P. Nightingale, and M. Den Nijs, Phys. Rev. Lett. 49, 405 (1982).
24
K. v Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980).
25
M. König, S. Wiedmann, C. Brüne, A. Roth, H. Buhmann, L.W. Molenkamp, X.-L. Qi, and S.-
C. Zhang, Science 318, 766 (2007).
26
B.A. Bernevig, T.L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006).
27
L. Fu and C.L. Kane, Phys. Rev. B 76, (2007).
28
D. Hsieh, Y. Xia, D. Qian, L. Wray, J.H. Dil, F. Meier, J. Osterwalder, L. Patthey, J.G.
Checkelsky, N.P. Ong, A.V. Fedorov, H. Lin, A. Bansil, D. Grauer, Y.S. Hor, R.J. Cava, and
M.Z. Hasan, Nature 460, 1101 (2009).
29
D. Hsieh, D. Qian, L. Wray, Y. Xia, Y.S. Hor, R.J. Cava, and M.Z. Hasan, Nature 452, 970
(2008).
30
Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y.S. Hor, R.J. Cava,
and M.Z. Hasan, Nat. Phys. 5, 398 (2009).

178
31
J. Moore, Nat. Phys. 5, 378 (2009).
32
Y.S. Kim, M. Brahlek, N. Bansal, E. Edrey, G.A. Kapilevich, K. Iida, M. Tanimura, Y. Horibe,
S.-W. Cheong, and S. Oh, Phys. Rev. B 84, (2011).
33
J.G. Analytis, R.D. McDonald, S.C. Riggs, J.-H. Chu, G.S. Boebinger, and I.R. Fisher, Nat.
Phys. 6, 960 (2010).
34
M.Z. Hasan and C.L. Kane, Rev. Mod. Phys. 82, 3045 (2010).
35
L. Fu and C.L. Kane, Phys. Rev. Lett. 100, (2008).
36
V.V. Atuchin, V.A. Golyashov, K.A. Kokh, I.V. Korolkov, A.S. Kozhukhov, V.N. Kruchinin,
S.V. Makarenko, L.D. Pokrovsky, I.P. Prosvirin, K.N. Romanyuk, and O.E. Tereshchenko,
Cryst. Growth Des. 11, 5507 (2011).
37
Y. Liu, M. Weinert, and L. Li, Phys. Rev. Lett. 108, 115501 (2012).
38
Y.S. Hor, A. Richardella, P. Roushan, Y. Xia, J.G. Checkelsky, A. Yazdani, M.Z. Hasan, N.P.
Ong, and R.J. Cava, Phys. Rev. B 79, 195208 (2009).
39
H. Bando, K. Koizumi, Y. Oikawa, K. Daikohara, V.A. Kulbachinskii, and H. Ozaki, J. Phys.
Condens. Matter 12, 5607 (2000).
40
H.M. Benia, C. Lin, K. Kern, and C.R. Ast, Phys. Rev. Lett. 107, 177602 (2011).
41
C. Kane and J. Moore, Phys. World 24, 32 (2011).
42
R. Karplus and J.M. Luttinger, Phys. Rev. 95, 1154 (1954).
43
T. Ando, Y. Matsumoto, and Y. Uemura, J. Phys. Soc. Jpn. 39, 279 (1975).
44
R.B. Laughlin, Phys. Rev. B 23, 5632 (1981).
45
A. Pressley, Elementary Differential Geometry (Springer, London ; New York, 2001).
46
K. Sun, (2012).
47
J.E. Avron, R. Seiler, and B. Simon, Phys. Rev. Lett. 51, 51 (1983).
48
D.R. Hofstadter, Phys. Rev. B 14, 2239 (1976).
49
D. Osadchy and J.E. Avron, J. Math. Phys. 42, 5665 (2001).
50
G.W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984).
51
P.R. Wallace, Phys. Rev. 71, 622 (1947).
52
J.E. Avron, D. Osadchy, and R. Seiler, Phys. Today 56, 38 (2003).
53
P. Streda, J. Phys. C Solid State Phys. 15, L717 (1982).
54
M.V. Berry, Proc. R. Soc. Lond. Ser. Math. Phys. Sci. 392, 45 (1984).
55
L. Fu, C.L. Kane, and E.J. Mele, Phys. Rev. Lett. 98, (2007).
56
F.D.M. Haldane, Phys. Rev. Lett. 93, 206602 (2004).
57
L. Fu and C.L. Kane, Phys. Rev. B 74, 195312 (2006).
58
D.N. Sheng, Z.Y. Weng, L. Sheng, and F.D.M. Haldane, Phys. Rev. Lett. 97, 36808 (2006).
59
X. Wan, A.M. Turner, A. Vishwanath, and S.Y. Savrasov, Phys. Rev. B 83, 205101 (2011).
60
A.A. Burkov and L. Balents, Phys. Rev. Lett. 107, (2011).
61
S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, B. Wang, N. Alidoust, G. Bian, M.
Neupane, C. Zhang, S. Jia, A. Bansil, H. Lin, and M.Z. Hasan, Nat. Commun. 6, 7373 (2015).
62
B.Q. Lv, H.M. Weng, B.B. Fu, X.P. Wang, H. Miao, J. Ma, P. Richard, X.C. Huang, L.X. Zhao,
G.F. Chen, Z. Fang, X. Dai, T. Qian, and H. Ding, Phys. Rev. X 5, (2015).
63
S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z.
Yuan, C.-C. Lee, S.-M. Huang, H. Zheng, J. Ma, D.S. Sanchez, B. Wang, A. Bansil, F. Chou,
P.P. Shibayev, H. Lin, S. Jia, and M.Z. Hasan, Science 349, 613 (2015).
64
G. Bian, T.-R. Chang, R. Sankar, S.-Y. Xu, H. Zheng, T. Neupert, C.-K. Chiu, S.-M. Huang, G.
Chang, I. Belopolski, D.S. Sanchez, M. Neupane, N. Alidoust, C. Liu, B. Wang, C.-C. Lee,

179
 H.-T. Jeng, C. Zhang, Z. Yuan, S. Jia, A. Bansil, F. Chou, H. Lin, and M.Z. Hasan, Nat.
Commun. 7, 10556 (2016).
65
S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, T.-R. Chang, B. Wang, N. Alidoust,
G. Bian, M. Neupane, D. Sanchez, H. Zheng, H.-T. Jeng, A. Bansil, T. Neupert, H. Lin, and
M.Z. Hasan, Proc. Natl. Acad. Sci. 113, 1180 (2016).
66
R. Jackiw and C. Rebbi, Phys. Rev. D 13, 3398 (1976).
67
S.-Q. Shen, Topological Insulators: Dirac Equation in Condensed Matters (Springer,
Heidelberg ; New York, 2012).
68
G.R. Hyde, R.O. Dillon, H.A. Beale, I.L. Spain, J.A. Woollam, and D.J. Sellmyer, Solid State
Commun. 13, 257 (1973).
69
J. Horák, Z. Stary, P. Lošťák, and J. Pancíř, J. Phys. Chem. Solids 51, 1353 (1990).
70
D. Kong, J.J. Cha, K. Lai, H. Peng, J.G. Analytis, S. Meister, Y. Chen, H.-J. Zhang, I.R. Fisher,
Z.-X. Shen, and Y. Cui, ACS Nano 5, 4698 (2011).
71
D.D. dos Reis, L. Barreto, M. Bianchi, G.A.S. Ribeiro, E.A. Soares, W.S. e Silva, V.E. de
Carvalho, J. Rawle, M. Hoesch, C. Nicklin, W.P. Fernandes, J. Mi, B.B. Iversen, and P.
Hofmann, Phys. Rev. B 88, 041404(R) (2013).
72
H.B. Nielsen and M. Ninomiya, Phys. Lett. B 130, 389 (1983).
73
P. Roushan, J. Seo, C.V. Parker, Y.S. Hor, D. Hsieh, D. Qian, A. Richardella, M.Z. Hasan, R.J.
Cava, and A. Yazdani, Nature 460, 1106 (2009).
74
J. Chen, X.Y. He, K.H. Wu, Z.Q. Ji, L. Lu, J.R. Shi, J.H. Smet, and Y.Q. Li, Phys. Rev. B 83,
241304 (2011).
75
R. Jackiw, Phys. Rev. D 29, 2375 (1984).
76
A.A. Taskin, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. Lett. 109, 66803 (2012).
77
D. Shoenberg, Magnetic Oscillations in Metals (Cambridge University Press, Cambridge, 1984).
78
Y. Ando, J. Phys. Soc. Jpn. 82, 102001 (2013).
79
Y. Zheng and T. Ando, Phys. Rev. B 65, 245420 (2002).
80
X.-L. Qi, T.L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, (2008).
81
W.-K. Tse and A.H. MacDonald, Phys. Rev. Lett. 105, 57401 (2010).
82
R. Valdés Aguilar, A.V. Stier, W. Liu, L.S. Bilbro, D.K. George, N. Bansal, L. Wu, J. Cerne,
A.G. Markelz, S. Oh, and N.P. Armitage, Phys. Rev. Lett. 108, (2012).
83
D.K. George, A.V. Stier, C.T. Ellis, B.D. McCombe, J. Černe, and A.G. Markelz, J. Opt. Soc.
Am. B 29, 1406 (2012).
84
A.Y. Kitaev, Ann. Phys. 303, 2 (2003).
85
N.V. Tarakina, S. Schreyeck, T. Borzenko, C. Schumacher, G. Karczewski, K. Brunner, C.
Gould, H. Buhmann, and L.W. Molenkamp, Cryst. Growth Des. 12, 1913 (2012).
86
N.V. Tarakina, S. Schreyeck, T. Borzenko, S. Grauer, C. Schumacher, G. Karczewski, C. Gould,
K. Brunner, H. Buhmann, and L.W. Molenkamp, J. Phys. Conf. Ser. 471, 12043 (2013).
87
S. Schreyeck, N.V. Tarakina, G. Karczewski, C. Schumacher, T. Borzenko, C. Brüne, H.
Buhmann, C. Gould, K. Brunner, and L.W. Molenkamp, Appl. Phys. Lett. 102, 41914 (2013).
88
Y. Liu, Y.Y. Li, D. Gilks, V.K. Lazarov, M. Weinert, and L. Li, Phys. Rev. Lett. 110, 186804
(2013).
89
K. Song, D. Soriano, R. Robles, P. Ordejon, and S. Roche, 2D Mater. 3, 45007 (2016).
90
Y. Liu, Y.Y. Li, S. Rajput, D. Gilks, L. Lari, P.L. Galindo, M. Weinert, V.K. Lazarov, and L.
Li, Nat. Phys. 10, 294 (2014).
91
M. Bianchi, D. Guan, S. Bao, J. Mi, B.B. Iversen, P.D. King, and P. Hofmann, Nat. Commun.
1, 128 (2010).

180
92
P.D.C. King, R.C. Hatch, M. Bianchi, R. Ovsyannikov, C. Lupulescu, G. Landolt, B. Slomski,
J.H. Dil, D. Guan, J.L. Mi, E.D.L. Rienks, J. Fink, A. Lindblad, S. Svensson, S. Bao, G.
Balakrishnan, B.B. Iversen, J. Osterwalder, W. Eberhardt, F. Baumberger, and P. Hofmann,
Phys. Rev. Lett. 107, 96802 (2011).
93
M. Bianchi, R.C. Hatch, J. Mi, B.B. Iversen, and P. Hofmann, Phys. Rev. Lett. 107, 86802
(2011).
94
Y. Zhang, K. He, C.-Z. Chang, C.-L. Song, L.-L. Wang, X. Chen, J.-F. Jia, Z. Fang, X. Dai, W.-
Y. Shan, S.-Q. Shen, Q. Niu, X.-L. Qi, S.-C. Zhang, X.-C. Ma, and Q.-K. Xue, Nat. Phys. 6,
584 (2010).
95
H. Steinberg, J.-B. Laloë, V. Fatemi, J.S. Moodera, and P. Jarillo-Herrero, Phys. Rev. B 84,
(2011).
96
G.R. Hyde, H.A. Beale, I.L. Spain, and J.A. Woollam, J. Phys. Chem. Solids 35, 1719 (1974).
97
J. Navrátil, J. Horák, T. Plecháček, S. Kamba, P. Lošt’ák, J.S. Dyck, W. Chen, and C. Uher, J.
Solid State Chem. 177, 1704 (2004).
98
D.O. Scanlon, P.D.C. King, R.P. Singh, A. de la Torre, S.M. Walker, G. Balakrishnan, F.
Baumberger, and C.R.A. Catlow, Adv. Mater. 24, 2154 (2012).
99
J.G. Checkelsky, Y.S. Hor, M.-H. Liu, D.-X. Qu, R.J. Cava, and N.P. Ong, Phys. Rev. Lett. 103,
246601 (2009).
100
Z. Ren, A.A. Taskin, S. Sasaki, K. Segawa, and Y. Ando, Phys. Rev. B 82, 241306(R) (2010).
101
Y.L. Chen, J.-H. Chu, J.G. Analytis, Z.K. Liu, K. Igarashi, H.-H. Kuo, X.L. Qi, S.K. Mo, R.G.
Moore, D.H. Lu, M. Hashimoto, T. Sasagawa, S.C. Zhang, I.R. Fisher, Z. Hussain, and Z.X.
Shen, Science 329, 659 (2010).
102
V.A. Golyashov, K.A. Kokh, S.V. Makarenko, K.N. Romanyuk, I.P. Prosvirin, A.V. Kalinkin,
O.E. Tereshchenko, A.S. Kozhukhov, D.V. Sheglov, S.V. Eremeev, S.D. Borisova, and E.V.
Chulkov, J. Appl. Phys. 112, 113702 (2012).
103
L.V. Yashina, J. Sánchez-Barriga, M.R. Scholz, A.A. Volykhov, A.P. Sirotina, V. Neudachina
S., M.E. Tamm, A. Varykhalov, D. Marchenko, G. Springholz, G. Bauer, A. Knop-Gericke,
and O. Rader, ACS Nano 7, 5181 (2013).
104
P.Y. Yu and M. Cardona, Fundamentals of Semiconductors (Springer Berlin Heidelberg,
Berlin, Heidelberg, 2010).
105
H.G. Tompkins and E.A. Irene, editors , Handbook of Ellipsometry (William Andrew Pub. ;
Springer, Norwich, NY : Heidelberg, Germany, 2005).
106
H. Fujiwara, Spectroscopic Ellipsometry: Principles and Applications (2003).
107
J.F. Moulder, W.F. Stickle, P.E. Sobol, K.D. Bomben, J. Chastain, R.C. King Jr., and Physical
Electronics, Incorporation, editors , Handbook of X-Ray Photoelectron Spectroscopy: A
Reference Book of Standard Spectra for Identification and Interpretation of XPS Data
(Physical Electronics, Eden Prairie, Minn., 1995).
108
D.B. Williams and C.B. Carter, Transmission Electron Microscopy: A Textbook for Materials
Science, 2nd edition (Springer, New York, 2009).
109
R.E. Honig and D.A. Kramer, Vapor Pressure Data for the Solid and Liquid Elements (RCA
Laboratories, David Sarnoff Research Center, 1969).
110
L.S. Brooks, J. Am. Chem. Soc. 74, 227 (1952).
111
M.P. Seah and W.A. Dench, Surf. Interface Anal. 1, 2 (1979).
112
M.W. Browne, N. Y. Times (1990).
113
C.R. Thomas, M.K. Vallon, M.G. Frith, H. Sezen, S.K. Kushwaha, R.J. Cava, J. Schwartz, and
S.L. Bernasek, Chem. Mater. 28, 35 (2016).

181
114
D.L. Medlin and G.J. Snyder, JOM 65, 390 (2013).
115
A.J. Green, S. Dey, Y.Q. An, B. O’Brien, S. O’Mullane, B. Thiel, and A.C. Diebold, J. Vac.
Sci. Technol. A 34, 61403 (2016).
116
C.A. and editors of the volumes Iii/17e-17f-41c, in Non-Tetrahedrally Bond. Elem. Bin. Compd.
I, edited by O. Madelung, U. Rössler, and M. Schulz (Springer Berlin Heidelberg, 1998), pp.
1–4.
117
J. Suh, D. Fu, X. Liu, J.K. Furdyna, K.M. Yu, W. Walukiewicz, and J. Wu, Phys. Rev. B 89,
115307 (2014).
118
K. Momma and F. Izumi, J. Appl. Crystallogr. 44, 1272 (2011).
119
D. Nečas and P. Klapetek, Open Phys. 10, 181 (2012).
120
X. He, W. Zhou, Z.Y. Wang, Y.N. Zhang, J. Shi, R.Q. Wu, and J.A. Yarmoff, Phys. Rev. Lett.
110, 156101 (2013).
121
A.V. Naumkin, A. Kraut-Vass, S.W. Gaarenstroom, and C.J. Powell, NIST X-Ray
Photoelectron Spectroscopy Database, 4.1 (Measurement Services Division of the National
Institute of Standards and Technology (NIST), Gaithersburg, MD, 2012).
122
V.B. Nascimento, V.E. De Carvalho, R. Paniago, E.A. Soares, L.O. Ladeira, and H.D. Pfannes,
J. Electron Spectrosc. Relat. Phenom. 104, 99 (1999).
123
M.R. Thuler, R.L. Benbow, and Z. Hurych, Chem. Phys. 71, 265 (1982).
124
S K Mishra and S Satpathy and O Jepsen, J. Phys. Condens. Matter 9, 461 (1997).
125
E. Bersch, M. Di, S. Consiglio, R.D. Clark, G.J. Leusink, and A.C. Diebold, J. Appl. Phys. 107,
43702 (2010).
126
A.K. Livesey and G.C. Smith, J. Electron Spectrosc. Relat. Phenom. 67, 439 (1994).
127
R. Opila, Prog. Surf. Sci. 69, 125 (2002).
128
H. Kageshima and K. Shiraishi, Phys. Rev. Lett. 81, 5936 (1998).
129
Y. Sharma, P. Srivastava, A. Dashora, L. Vadkhiya, M.K. Bhayani, R. Jain, A.R. Jani, and B.L.
Ahuja, Solid State Sci. 14, 241 (2012).
130
J. Humlíček, D. Hemzal, A. Dubroka, O. Caha, H. Steiner, G. Bauer, and G. Springholz, Phys.
Scr. T162, 14007 (2014).
131
K. Jayachandran, Electrical, Optical, and Structural Studies in Bismuth, Antimony, Bismuth
Oxide and Antimony Oxide Thin Films, Mahatma Gandhi University, 1997.
132
L. Leontie, J. Optoelectron. Adv. Mater. 8, 1221 (2006).
133
K. Tripathi, M. Husain, and M. Zulfequar, Chalcogenide Lett. 6, 517 (2009).
134
Y.R. Shen, Nature 337, 519 (1989).
135
R.M. Corn and D.A. Higgins, Chem. Rev. 94, 107 (1994).
136
Y.Q. An, A.J. Green, and Diebold, A.C., (n.d.).
137
J.W. McIver, D. Hsieh, S.G. Drapcho, D.H. Torchinsky, D.R. Gardner, Y.S. Lee, and N. Gedik,
Phys. Rev. B 86, 35327 (2012).
138
D. Hsieh, F. Mahmood, J.W. McIver, D.R. Gardner, Y.S. Lee, and N. Gedik, Phys. Rev. Lett.
107, 77401 (2011).

182