direct heteroepitaxial growth of iii-vs on si by hvpe1380220/... · 2019-12-18 · semiconductors...
TRANSCRIPT
Direct Heteroepitaxial Growth of III-Vs on Si by HVPE
ZHEHAN XU
KTH ROYAL INSTITUTE OF TECHNOLOGY
S C H O O L O F E N G I N E E R I N G S C I E N C E S
DEGREE PROJECT IN SCHOOL OF ENGINEERING SCIENCES, SECOND LEVEL, SK202X STOCKHOLM, SWEDEN 2019
i
Direct Heteroepitaxial Growth of III-Vs on Si by HVPE
Zhehan Xu
2019-11-21
Master’s Thesis
Examiner Sebastian Lourdudoss Supervisor Yanting Sun, Axel Strömberg
KTH Royal Institute of Technology
School of Engineering Sciences
Department of Applied Physics
SE-100 44 Stockholm, Sweden
ii
Abstract
III-V compound semiconductors are widely used in photonic devices such as LEDs and
lasers. Since 1980s, III-V compound on Si heterostructure has shown great potential in improving
device performance and reducing the cost. At the same time, many challenges such as defects arising
in its monolithic integration by heteroepitaxial growth still need further investigation. In this thesis
work, we focus mainly on the growth and characterization of three III-V materials, GaAs, GaP and
GaAsP, on (001) and (111) Si substrates with various surface morphology by hydride vapor phase
epitaxy (HVPE) in order to prove the feasibility of this growth approach, improve its technique, and
investigate the properties of the heterostructure for its application in photonic devices in the future.
In this work, the Si substrates sliced from 4’’ n-type Si wafers were used. Three kinds of Si
substrates are prepared before layer growth: the planar, patterned and textured substrates. The planar
substrate only requires cleaning of its surface and removal of native oxides, while the patterned
substrate requires to form a patterned SiO2 layer by plasma enhanced chemical vapor deposition
(PECVD) and photolithography, which is used for selective area growth of III-V nanopillars by
HVPE. The textured Si is widely used in solar cells and is processed by wet etching using the
mixture of KOH and isopropyl alcohol (IPA). Afterwards, different III-V materials, GaAs, GaP and
GaAsP are grown on the Si substrates by a two-step HVPE method, which involves a low-
temperature 2 min buffer layer growth and high-temperature 15min overgrowth. The prepared
samples are characterized by high resolution x-ray diffraction (HR-XRD), scanning electron
microscopy (SEM), atomic force microscopy (AFM), photoluminescence spectroscopy (PL),
Raman spectroscopy, and ellipsometry to analyze the properties of heterostructure such as
thickness, morphology, defects, resistivity, emission wavelength, doping concentration, etc.
The analysis to the patterned samples proves selective area growth (SAG) of nanopillars
and patches, which is a merit of HVPE. It also proves the influence of surface kinetics,
especially crystal orientation, to epitaxial growth in HVPE. This is a main difference to
diffusion-controlled metal-organic vapor phase epitaxy (MOVPE). In addition, temperature is
also observed to affect the growth result, though more researches are required to investigate the
influence on each material. The planar samples show presence of defects like anti-phase domain
(APD) and misfit dislocation, as well as facets of other orientations on the surface. The
crystalline quality measured by XRD and PL spectroscopy varies in different samples and still
need to be improved in the future work. Besides, the growth condition also needs to be further
optimized to improve the growth rate.
In summary, we have demonstrated the direct heteroepitaxial growth of III-V
semiconductors on (001) and (111) Si substrates by cost effective HVPE technology. The
outcomes of this work will facilitate the realization of III-V/Si based photonic integrated
circuits and high efficiency tandem solar cells at a competitive cost in near future.
Keywords
III-V/Si, hydride vapor phase epitaxy, selective-area growth, heteroepitaxy
iii
iv
Acknowledgement
First of all, I would like to thank Dr. Yanting Sun and Axel Strömberg being my supervisor who
always guide me and help me in my thesis work. Their kindness, hardwork, and a great knowledge
ensures the smooth process of the project. I also would like to thank Prof. Sebastian Lourdudoss
being my examiner and admitting me to Photonics Group, where I met the most talented and
knowledgeable colleagues in the world. Thank Giriprasanth Omanakutten, Yinjun Liu and Tangjie
Mu, you are the consummate colleagues and are always ready to help me in my project. I shall never
forget the days working with you in the cleanroom. I also appreciate for the help and equipment
training from other professors, researcher and PhD students in Electrum laboratory, they are Prof.
Anand Srinivasan, Olof Öberg, Reza Nikpars, Per Wehlin, Cecilia Aronsson, Carl Reuterskiöld-
Hedlund, Fei Ye and Corrado Capriata, At last, I would like to give my gratitude to my parents and
my girlfriend Minchao An, who supports me during my study in KTH and shares with me each
progress and each moment of happiness.
v
Table of Contents
Abstract…………………………………………………………………………………….…ii
Acknowledgement………………………………………………………………….………….iii
Table of Contents……………………………………………………………………………iv
List of Abbreviations……………………………………………………………………………...v
1. Introduction……………………………………………………………………………………1
1.1 III-V/Si heterostructure and photonic devices……………………………………………1
1.2 Thesis outline………………………………………………………………………….…1
2. Theoretical background……………………………………………………………………….2
2.1 III-V materials……………………………………………………………………………..2
2.1.1 GaAs…………………………………………………………………………….2
2.1.2 GaP………………………………………………………………………………2
2.1.3 GaAsP……………………………………………………………………………2
2.2 Processing technologies……………………………………………………………………3
2.2.1 PECVD…………………………………………………………………………3
2.2.2 Photolithography………………………………………………………………….4
2.2.3 RIE………………………………………………………………………………..4
2.2.4 HVPE……………………………………………………………………………..5
2.3 Defects……………………………………………………………………………………7
2.3.1 Misfit dislocation………………………………………………………………..7
2.3.2 Anti-phase domain………………………………………………………………7
3. Sample preparation……………………………………………………………………………9
3.1 Preparation of substrates………………………………………………………………….9
3.1.1 Patterned substrates………………………………………………………………9
3.1.2 Textured substrates………………………………………………………………11
3.2 HVPE for epitaxial growth…………………………………………………………….13
4. Characterization techniques…………………………………………………………………15
4.1 SEM and EDS…………………………………………………………………………….15
4.2 AFM……………………………………………………………………………………16
4.3 XRD and RLM………...……….…………………………………………………………16
4.4 PL spectroscopy………………………………………………………………………19
4.5 Raman spectroscopy…………………...…………………………………………………20
4.6 Ellipsometry…………………..………………………………………………………….22
5. Result analysis………………………………………………………………………………26
5.1 GaAs on Si………………………………..………………………………………………27
5.2 GaP on Si…………………………………..……………………………………………..33
5.3 GaAsP on Si…………………………………..…………………………………………..37
5.4 GaP buffer layer growth………………………..…………………………………………41
6. Conclusion and future work………………………...…………………………………………45
7. References…………………………………………………………………………………47
vi
List of Abbreviations
AFM: Atomic Force Microscopy
AlGaAs: Aluminum Gallium Arsenide
APD: Anti-phase Domain
AsH3: Arsine
BSE: Backscattering Electrons
CCD: Charge-coupled Device
DC: Direct Current
EBL: Electron-beam Lithography
EDS: Energy Dispersive Spectroscopy
FWHM: Full Width at Half Maximum
GaAs: Gallium Arsenide
GaAsP: Gallium Arsenide Phosphide
GaCl: Gallium Chloride
GaN: Gallium Nitride
GaP: Gallium Phosphide
HCl: Hydrochloric Acid
HF: hydrogen fluoride
HRXRD: High Resolution X-ray Diffraction
HVPE: Hydride Vapor Phase Epitaxy
InCl: Indium Chloride
InP: Indium Phosphide
InGaAsP: Indium Gallium Arsenide Phosphide
IPA: Isopropanol
IR: Infrared
KOH: Potassium Hydroxide
LED: Light-emitting Diode
LO: Longitudinal Optical Mode
MBE: Molecular Beam Epitaxy
MOVPE: Metalorganic Vapor Phase Epitaxy
NW: Nanowire
PECVD: Plasma Enhanced Chemical Vapor Deposition
PH3: Phosphine
PL: Photoluminescence
RIE: Reactive ion etching
RLM: Reciprocal Lattice Mapping
SAG: Selective Area Growth
SE: Secondary Electrons
SEM: Scanning Electron Microscopy
TEM: Transmission Electron Microscopy
TO: Transverse Optical Mode
XRD: X-ray Diffraction
ZB: Zinc Blende
1
1. Introduction
1.1 III-V/Si heterostructure and photonic devices
Si and III-V compounds such as GaAs, InP and GaP are two kinds of semiconductor
materials used to work in different fields. Si is widely used in electronics and solar cells because
of its good electric and thermal properties and low cost. While III-V compounds are more often
used for photonic devices such as LEDs and lasers due to their various emission wavelength and
direct bandgap. The monolithic integration of III-V materials on Si substrate to combine the
advantages of two materials is the theme of this project, which is not only expected to reduce
the cost of photonic devices while retaining or improving their property, but also to make it
easier for integration with Si-based electronic chips in future.
Since the idea of III-V/Si heterostructure has been proposed in 1980s, several problems in
heteroepitaxial growth always need to be faced, including the control of growth rate and doping,
and defects like anti-phase domain (APD) and misfit dislocation which have deep influence on
the device performance.[1][2][3][4][10][11][12] Recently, GaAs, GaP and InP nanowire (NW)
structures on Si with good uniformity and optical properties have been produced by Tomioka et
al. using metal organic vapor phase epitaxy (MOVPE),[1] though misfit dislocation and doping
issues still need further investigation. Other research groups focused on the growth of planar
GaP/Si structure using either MOVPE or molecular-beam epitaxy (MBE) face the problem of
APD which influence the flatness and electron mobility.[2][3][4]
Another problem lies on the scaling of devices. With their dimensions scaled down,
fabrication and characterization of these micro- or nano-structures becomes more challenging
because of the limit of precision, which requires to improve current techniques or a trial of new
methods.
In this project, we focus on the growth of III-V/Si heterostructure using HVPE. The
problems mentioned above will be investigated, and different characterization methods will be
explored to improve the quality of heteroepitaxial growth of III-V/Si.
1.2 Thesis outline
The following chapters are focused on the fabrication and characterization of III-V/Si
heterostructures. In Chapter 2, III-V materials used in epitaxial growth in this project, including
GaAs, GaP and GaAsP will be briefly introduced, followed by their growth mechanism in
HVPE. Two types of crystallographic defects that appeared in the growth will also be introduced
at the end of this chapter. Chapter 3 shows how these samples are processed, from substrate
preparation to layer growth by HVPE. Chapter 4 introduces the characterization devices such
as high resolution x-ray diffraction (HR-XRD), scanning electron microscopy (SEM), atomic
force microscopy (AFM), photoluminescence spectroscopy (PL) and Raman spectroscopy, and
in Chapter 5 we analyze the growth results including layer roughness, defects, strain, crystalline
quality, bandgap, resistivity, etc. using characterization methods. The conclusion and future
work are drawn in Chapter 6.
2
2. Theoretical background
2.1 III-V materials
III-V materials are compound semiconductor materials composed of group III elements
such as aluminum (Al), gallium (Ga), indium (In) and group V materials like arsenic (As),
phosphorus (P) and nitrogen (N) in the periodic table. This combination yields compounds like
GaAs, GaP, InAs and InP. Further mixing these compounds at a certain ratio gives ternary and
quaternary compounds like AlGaAs and InGaAsP. Each of the material has its unique features
including bandgap, lattice constant, electron mobility, refractive index, etc. But in general, III-
V compounds have bandgap energy within visible light and infrared (IR) regime and are mainly
used for making photonic devices.
In this chapter, three III-V materials involved in this project will be briefly introduced.
2.1.1 GaAs
GaAs is a material which is widely used in photonic devices such as LEDs, solar cells,
laser diodes and photodetectors. GaAs and Si has 4.09% lattice mismatch. Comparing with Si,
GaAs has a direct bandgap and higher electron mobility, which allows some better electric and
optical properties in the devices. GaAs typically has zinc blende (ZB) structure in the bulk.
However, when the size is down to 40nm, wurtzite structure would be more stable. This needs
to be taken into account when growing nano-sized structures such as nanowires and quantum
dots [5]. In this project, GaAs is used in different functions: it can be a substrate for the growth
of other epilayers including GaP and GaAsP, or it can be an epilayer or nanopillar array grown
on Si substrate.
2.1.2 GaP
The lattice constant of GaP is very close to Si, which is 5.45Å, having only 0.37%
mismatch with Si. Therefore, the heteroepitaxial growth of GaP on Si is expected to have less
crystalline defects and better quality. GaP has a large, indirect bandgap of 2.24eV in the bulk.
The pure GaP LED emits 550nm green light, while the doped LED emits red light or yellow
light depending on the doping type .[30] The low absorption over a large range from visible light
to infrared also makes GaP an excellent material for nonlinear optical devices.
2.1.3 GaAsP
Gallium arsenide phosphide (GaAs1-xPx) is a ternary alloy of GaAs and GaP mixed at a
certain molar fraction x. The composition of As and P can be tuned, thus changing its bandgap
and lattice constants. Figure 2.1 shows the increase of bandgap and decrease of lattice constant
when molar fraction x varies from 0 to 1. It is worth noting that, when x is below 0.45, it has a
direct bandgap and when above it, the bandgap becomes indirect. The bandgap engineering of
GaAsP can be applied in LEDs for emitting visible light with different wavelength from red to
yellow. GaAsP is often grown on GaP substrates[28][29], while in this project, its growth on Si
substrates with GaP buffer layer will also be studied.
3
Fig. 2.1 Lattice constant and bandgap of GaAsP vary with x[23]
2.2 Processing technologies
2.2.1 PECVD
Plasma-enhanced chemical vapor deposition (PECVD) is used for deposition of thin films
like SiO2, Si3N4, and SiC, etc. on the substrate. The schematic diagram of PECVD equipment
is shown in Figure 2.2. It has two parallel, conductive electrodes: the top electrode driven by
RF and the bottom electrode where the substrate is placed. The substrate is heated up to 250-
350°C. The gaseous reactants like SiH4 (carried by N2), NH3, O2, and N2 are introduced to the
space above the substrate between two electrodes. The capacitive coupling of two electrodes
causes ionization of reactant and generates plasma. Finally, the chemical reaction of the plasma
results in its product deposited on the substrate and forms a thin film, and the byproducts are
pumped away.
Fig 2.2 Schematic diagram of PECVD[24]
In this project, a Plasmalab 80Plus (Oxford PECVD System) is used to deposit the
dielectric SiO2 layer on Si substrates. The deposition temperature is 300°C. It has a dual RF
source on top electrode producing low (10-400 kHz) and high (13.56 MHz) frequencies, and its
plasma power ranges 0-500 W. This PECVD system is capable of producing dielectric layers
with high quality and uniformity.
4
2.2.2 Photolithography
Photolithography is used to transfer a desired pattern from a mask to photoresist and finally
to wafer by light exposure. It firstly needs to cover the substrate surface with a photoresist.
There are in general 2 types of resists: positive ones, where the part exposed to light induces
change in chemical properties and can be removed by a developer, while for negative ones, the
part unexposed to light is soluble to developers. After covering the surface with photoresist, a
mask aligner is used to expose the resist to UV light. The schematic diagram of mask aligner is
shown in Figure 2.3. It has a light source, an optical system to adjust beam path, and a
photomask allowing light beam to pass in some areas, while to be blocked in others, so that
patterns are transferred to photoresist. Then by using developers, the exposed or unexposed part
will be removed. Finally, by etching and removal of residual photoresist, the pattern can be
transferred to wafer.
Fig 2.3 Schematic diagram of mask aligner
The Mask aligner MA6/BA6 Karl Suss is used in this project. The light source is a 350W
Hg lamp. The resolution of the equipment reaches 0.6µm and allows maximally 6’’ wafers.
2.2.3 RIE
Reactive-ion etching (RIE) is a dry etching method which uses reactive plasma to etch the
wafer surface. A wide range of materials can be etched by RIE, including dielectric materials,
Si-based materials and III-V compounds. Like PECVD, it also has a top electrode driven by
RF to generate plasma, a bottom electrode where the wafer is placed, and a pumping system
(Figure 2.4), but the gases used are different. Commonly used gases are CF4, O2 and Ar, etc.
depending on the type of etched materials. The positive ions in the plasma are accelerated by
strong electrical field and hit to wafer surface. These ions with large kinetic energy can knock
off surface atoms or react with them, thus etching the wafer. Since the motion of most ions are
directional, the etch profile of RIE is anisotropic.
Photoresist
Optical system
Light source
Photomask
Wafer
5
Fig 2.4 Schematic diagram of RIE[25]
The RIE equipment used in this project is Plasmalab80Plus (Oxford RIE System) which
is capable of etching SiO2 and Si3N4. The frequency is 13.56 MHz and plasma power ranges
0-500 W, which are basically the same as PECVD system.
2.2.4 HVPE
HVPE (Hydride vapour phase epitaxy) was initially used in the growth of GaN in 1960s,[27]
and until now, the most common application of HVPE is still to obtain GaN substrates with
high quality. Beside this, HVPE is also applied for epitaxial growth of other III-V
semiconductors such as GaAs, GaP, and InP because of its merits.
As a method growing III-V compounds at near-equilibrium condition, the advantages of
HVPE include high growth rate and selectivity of epitaxial growth. The growth rate, commonly
reaching as high as 100 μm/h, is realized by varying mass input rate of group III/V precursors.
This is perfectly suitable for the growth of structures with high aspect ratio such as nanowires
and nanopillars. Besides, the selectivity allows it to grow into different patterns as desired. [6]
This is obtained by firstly depositing a dielectric mask on the substrate. In the process of
epitaxial growth, the group III precursors is unable to be adsorbed on the masked areas, so that
selective area growth (SAG) can be obtained.[7]
Fig.2.5 Schematic diagram of LP-HVPE reactor
Figure 2.5 shows the layout of a low-pressure hydride vapour phase epitaxy (LP-HVPE)
reactor. It consists of several heating zones, whose temperature increase in gradient from
loading zone on the left to source zone on the right in order to avoid parasitic nucleation. [8] The
Source Zone Loading
Zone
Ga source
Deposition
Zone
Mix
Zone
AsH3, PH3
H2, N2, H2S
AsH3, PH3
H2, N2
HCl
H2, N2 GaCl
Gate
6
source zone contains two chambers, which separately introduce group III and V precursors, and
allow them to mix and flow to deposition zone. The sample is loaded on another side of the
equipment at loading zone. There is also a part of group V-hydrides and carrier gas introduced
from stabilization line near loading zone for the low-temperature buffer layer growth. The
sample is sent via preheat zone to deposition zone, where epitaxial growth takes place. After
growth, the sample is sent back to the loading zone and cool down for a while before taken out.
HVPE utilizes the reaction of group III and V precursors inside its quartz reactor, which
are III-chloride (GaCl, InCl, etc.) and V-hydride (AsH3, PH3, etc. ), respectively. Take the
reaction process of GaAs and GaP for example, the gaseous III-chloride GaCl can be
synthesized inside reactor by the reaction of HCl and III-source such as liquid Ga,
2Ga(l) + 2HCl(g) ⇋ 2GaCl(g) + H2(g) . (1)
The liquid Ga is stored in a boat in the source zone. The gaseous HCl flows through the boat,
reacting with Ga at high temperature, and carrying away the gaseous products into the growth
zone. GaCl is stable at high temperature but will be decomposed into GaCl3 below a critical
point. Hence the temperature should be well-manipulated during their generation and
transportation.[31] The V-hydrides such as AsH3 and PH3 undergo pyrolysis process in
deposition zone, where for AsH3, it is fully decomposed into As2/As4 gaseous molecules at
high temperature,
2AsH3(g) ⇋ As2(g) + 3H2(g) , (2)
2As2(g) ⇋ As4(g) , (3)
and for PH3, partially into P2,
2PH3(g) ⇋ P2(g) + 3H2(g) . (4)
These products react with III-chlorides, and finally form GaP or GaAs on the substrate. This can
be described as
2GaCl(g) + P2(g) + H2(g) ⇋ 2GaP(c) + 2HCl(g) (5)
and
4GaCl(g) + As4(g) + 2H2(g) ⇋ 4GaAs(c) + 4HCl(g) (6)
In this crystallization process of GaAs or GaP, supersaturation plays a critical role. When
a vapor has a partial pressure that exceeds its vapor pressure, then it comes to the supersaturation
state. This state is a metastable state and is at non-equilibrium, which means a slight fluctuation
in condition will bring its pressure back to vapor pressure to retain equilibrium and
simultaneously crystal nucleation occurs. It has been proved that supersaturation has great
influence on crystal size, nucleation rate and growth rate.[9] In HVPE, kinetics and diffusion are
very fast to allow it always work at near-equilibrium.
Supersaturation of GaAs in HVPE growth can be described as the following formula[6] (for
GaP and InP, they also have similar formulas),
γ + 1 =[GaCl][As4]1/4[H2]1/2
[HCl]Keq(T) (7)
Keq(T) =[GaCl]eq[As4]eq
1/4[H2]eq1/2
[HCl]eq (8)
γ is the supersaturation parameter which is defined as the ratio of partial pressure of III-
and V precursors above the substrate divided by the ratio of their equilibrium partial pressure,
𝐾𝑒𝑞(𝑇). Obviously, γ is 0 at equilibrium. 𝐾𝑒𝑞(𝑇) is called reaction equilibrium constant and it
varies with temperature.
7
HVPE growth is temperature-dependent. At lower temperature, it is governed by surface
kinetics, so crystal orientation, desorption, or temperature variation can strongly affect the
growth. At higher temperature, it is on the regime of thermodynamics, and the growth rate is
mainly tuned by input mass flow.[10] In this project, HVPE is a critical step for growing the
sample. These growth conditions must be well-tuned to improve growth rate and sample quality.
2.3 Defects
2.3.1 Misfit dislocation
Misfit dislocation in heterostructure is due to lattice mismatch of two materials. The lattice
mismatch between GaP and Si is 0.37%, and for GaAs and Si, it reaches 4.09%. In
heteroepitaxial growth, the lattice mismatch is accommodated by either strain or misfit
dislocation at the interface. There is a critical thickness for each heterostructure, below which
the lattice mismatch is completely accommodated by strain, and epilayer can coherently grow
on the substrate. This is known as pseudomorphic growth. When the layer thickness exceeds
the critical thickness, strain relaxation of the layer gets greater by the formation of misfit
dislocations until completely relaxed. The critical thickness for GaP on Si is reported to be about
90 nm and for GaAs on Si, it is estimated to be less than 1nm.[11]
Misfit dislocation can be characterized by either transmission electron microscopy (TEM)
or X-ray diffraction (XRD). In this project, a high-resolution X-ray diffraction (HRXRD) is
used to make reciprocal lattice mapping (RLM) to evaluate the misfit dislocations in the samples.
This will be discussed in the following chapters.
2.3.2 Anti-phase domain
Anti-phase domain (APD) is a planar defect in the epitaxial growth. It is essentially formed
by the difference of material polarity in heterostructures, i.e. a polar compound material such
as GaAs, GaP on a non-polar material such as Si and Ge. APD appeared on the samples in a
region configured in an opposite order of elements from normal regions. Take GaP on Si
substrate for example, Si has a diamond structure and GaP typically has a zinc blend (ZB)
structure. The only difference between them is that diamond structure consists of one element,
while zinc blend has two. Therefore, it can be either Ga or P atoms occupying the first atomic
layer on the surface of Si, which makes two adjacent regions have opposite configurations. This
is known as APD.
APD can propagate through the epilayer along its growth direction and form terraces with
large thickness variation on the layer surface (Figure 2.6a), or self-annihilated inside the layer
and form pits on the surface (Figure 2.6d). The self-annihilation of APD typically happens in
the directions of <111>. Thus using (111) substrates is estimated to be able to reduce APD.[12]
8
Fig.2.6 Anti-phase domain in the epitaxial growth of GaP on Si[3]
Another method for APD annihilation in the growth on other substrates like Si(001) is by
processing the substrate surface into a surface with several stages, each stage contains two
atomic layers (Figure.2.6b). Then due to energetic preference, one element can uniformly grow
on the surface. This kind of surface is typically processed by 4° or 6° off-cut and annealing at a
temperature above 1000 °C. One problem of this method is that it is not practical in designing
the devices which need a planar substrate. Other methods relying on the growth condition are
quite dependent on which epitaxy equipment is selected. For example, in MOVPE, the density
of APD are reported to increase at high growth temperature,[4] while in MBE, a low growth rate
and low V/III ratio is preferable.[3] The investigation of APD in HVPE shall be done in this
project.
9
3. Sample preparation
In this project, the III-V/Si heterostructures we expected to grow are GaAs, GaP and
GaAsP epilayers on planar and textured Si substrates, as well as nanopillars on patterned (001)
and (111) Si substrates. The planar growth only needs a clean on sample surface and removal
of native SiO2 using hydrogen fluoride (HF) before epitaxial growth by HVPE, while the
preparation of textured and patterned substrates involves more steps like wet etching and
photolithography.
3.1 Preparation of substrates
3.1.1 Patterned substrates
The fabrication of nanopillars on Si substrates in general involves in 2 steps: 1. preparation
of patterned substrates, and 2. selective area growth of nanopillars by HVPE.
In the first step, PECVD, photolithography and dry etching are used to grow SiO2 layer
with circular openings on Si substrate. The procedure is shown in Figure 3.1
Fig.3.1 Preparation of patterned substrates
Both 4’’ n-type Si(001) and Si(111) wafers are selected. After cleaning with 50% HF for
1min to remove native SiO2 on the sample surface and rinsed with water, the wafer is treated
with a Plasmalab 80Plus (Oxford PECVD System) for 2 minutes to deposit a SiO2 dielectric
layer on the front side (Figure 3.1b). The deposition rate of SiO2 for this equipment is about 70
Si
(a) HF cleaned substrate
Si SiO2
(b) Oxide deposition
Si SiO2
(c) Spin coating of photoresist
Photoresist
Si
Photoresist Mask
SiO2
(d) Light exposure
Si
Photoresist SiO2
(e)Patterns formed on the photoresist
Si SiO2
Photoresist
(f) Reactive-ion etching
Si
(g) Photoresist stripping
SiO2
10
nm/min and the time is set to be 2min in order to control to the layer thickness about 140nm.
After deposition, a Leitz interferometer is used to measure the layer thickness, which turns
out to be 140nm and 144nm for Si(001) and Si(111), respectively.
Next, the wafer is loaded into a Star 2000 HMDS/Vapor prime oven to improve adhesion
to the photoresist and then stands still for 2min.
A photoresist is selected. Since it is required to create openings on SiO2 layer in this project,
and considering available masks, the positive photoresists is preferable. The S1818 G2 (SP16)
resist is selected and spin-coated onto the wafer surface.
Then in order to let the photoresist evenly spread on the surface (Figure 3.1c), an OPTIspin
SST20 spin coater is used. The wafer is positioned at the center of its chuck, and spins at the
speed of 4000rpm for 30s. After that, the wafer is baked at 100℃ for 90s so that the resist
becomes a solid film covering wafer surface.
Next step is exposure (Figure 3.1d,e) using a Mask aligner MA6/BA6 Karl Suss equipment.
Since the size of openings is several microns, which does not reach the resolution limit
determined by wavelength of the light used, which is typically at the order of 100nm,
photolithography can be used in this case. Otherwise, electron-beam lithography (EBL) would
be preferable.
A suitable mask is selected. The wafer is aligned to the pattern on the mask. Then it is
exposed to light for 5.5s. After exposure, the CD-26 developer is specially used for the removal
of S1818 resist. The wafer is developed for 45s and then rinsed with water.
The pattern on wafer is observed by a microscope. The circular openings can clearly be
seen, along with some residues of photoresist on the surface. These are stripped by a TePla
Model 300 Plasma System for surface cleaning.
Now the pattern is transferred from the mask to the photoresist of the sample. The next
step is to etch SiO2 inside the openings (Figure 3.1f). A Plasmalab80Plus (Oxford RIE System)
is used, which emits plasma to react with sample surface, thus etching thin films of SiO2 in one
direction. The process lasts for 17min, until it etches deep down to the Si substrate.
The last step is photoresist stripping (Figure 3.1g) by acetone and IPA (isopropanol) to
wash away the resist and clean the sample surface. The residues are cleaned by the plasma
system as before.
Now the patterned substrates are prepared, they are observed under the microscope (Figure
3.2). The blue areas are SiO2 layers and white dots are opening with diameters ranging from 0.9
to 4.5μm. These openings have the same arrangement and uniformly spread on the sample
surface, which proves that the substrate preparation is successful.
(a) (b)
11
Fig.3.2 Patterned (a) n-type Si(111), (b) n-type Si(001) substrates under the microscope
3.1.2 Textured substrates
As a monocrystalline material, Si is anisotropic in the wet etching, which means that it has
different etch rates on different crystal orientations. On <111> directions, for example, the etch
rate is much lower than on <001>. This creates facets along all the directions of <111> during
the etching, which finally forms pyramids on wafer surface. This process, known as texturing,
improves the performance of optoelectronic devices. For example, texturing of Si substrate in
solar cells can effectively reduce the reflection of incident light and improve light trapping, thus
increasing the conversion efficiency. Moreover, texturing of substrate has other advantages in
performance, which will be studied in this project.
IPA and alkalis such as KOH are used for texturing on Si(001) wafers. The concentration
of IPA should be high enough to increase the wettability, allowing enough contact of KOH and
wafer surface. The concentration of KOH is relevant to the size of pyramids and should also be
controlled. In order to select the best recipe, different concentrations of KOH and IPA, and
different etching time has been tried, which turned out that 3% KOH+10% IPA mixture is the
best recipe to form pyramidal structures uniformly covered on wafer surface.[19]
The wafers used in the project are n-type Si(001). The KOH-IPA mixture is heated in a
water bath to 80℃ and keeps this temperature during wet etching. Then the wafers are immersed
in the solution for 30min. If the etching time is less than 30min, it will be insufficiently etched
and the pyramids will not fully cover the wafer surface. The same problem appears when the
concentration of IPA is not high enough.
Figure 3.3 is the result of using different recipes during the project, where (a) shows that
by using 3% KOH and 10% IPA mixture for 30min, the surface is totally covered by pyramids
with length less than 10μm. (b) is the comparison of two recipes(firstly the wafer is processed
14% KOH+3% IPA for 30min, and then 3% KOH+10% IPA for 30min). The larger pyramids
are formed by using 14% KOH and 3% IPA. It turns out that higher concentration of KOH
results in higher etching rate, forming pyramids with length around 30μm. However, since the
concentration of IPA is too low, these pyramids only partly cover the surface, and the most parts
are flat. These flat areas form pyramids by using 3% KOH and 10% IPA in the second step.
(a) (b)
Fig.3.3 Etching result of n-type Si(001) wafer using (a) 3% KOH and 10% IPA for 30min
(b) firstly 14% KOH+3% IPA for 30min, then 3% KOH+10% IPA for 30min
12
13
3.2 HVPE for epitaxial growth
The HVPE equipment used for the III-V heteroepitaxial growth on the patterned, textured
and planar Si substrates is a commercially available AIXTRON low pressure HVPE (LP-HVPE)
reactor.
A two-step growth is involved in order to improve crystalline quality of epilayers and
nanopillars. It includes a low-temperature buffer layer growth to facilitate the nucleation of III-
V semiconductors on Si and reduce defects like dislocation and residual stress in the sample,
and a high-temperature overgrowth with high grow rate to achieve a thick III-V layer. During
the buffer layer growth, group V-precursors were injected from the stabilization line and mixed
with the III-precursors in the deposition zone, while during overgrowth, mixing happens in the
mixing zone where the temperature is relatively higher.
The growth conditions and growth rates of samples grown in this project are listed in Table
1. Runs of 3750 and 3770 are focused on the growth of GaAs and GaP nanopillars and planar
layers on Si substrate, run 3800 is GaAsP layer growth on different Si and GaP substrates, while
run 3805 grows only GaP buffer layer on Si and GaP. The thickness and overall growth rates
are calculated using weight measurements. The characterization on these samples will be
introduced in the following chapters.
Table 1 Growth conditions and growth rates of samples grown by HVPE
Run
number
Layers Substrates V/III
ratio
Total
flow(
sccm
)
Tempe
rature
(℃)
Time
(min)
Thickness Overall
Growth
rate
3750 GaAs
(overgr
owth)
Planar
Si(001),
Planar
Si(111),
Patterned
Si(001),
Patterned
Si(111),
Planar
GaAs
20 895 665 15 planar
Si(001):
1.5μm,
planar
GaAs:
1.4μm
planar
Si(001):
5.3μm/h,
planar
GaAs:
4.9μm/h
GaAs
(Buffer
layer)
10 825 439 2
3770 GaP
(overgr
owth)
Planar
Si(001),
Planar
Si(111),
Patterned
Si(001),
Patterned
Si(111),
Planar
GaP
20 895 715 15 planar
GaP:
4.3μm,
planar
Si(001):
3.1μm
planar
GaP:
15.1μm/
h, planar
Si(001):
11.0μm/
h
GaP
(Buffer
layer)
10 825 445 2
14
3800 GaAsP
(overgr
owth)
Planar
Si(001),
Planar
Si(111),
Textured
Si(001),
Planar
GaP
20 895 661 15 textured
Si(001):
4.2μm,
planar
GaP:
5.1μm
textured
Si(001):
14.7μm/
h, planar
GaP:
18.1μm/
h
GaP
(Buffer
layer)
10 825 444 2
3805 GaP
(Buffer
layer)
Planar
Si(001),
Planar
Si(111),
Textured
Si(001),
Planar
GaAs
10 825 448 2
15
4. Characterization techniques
After substrate preparation and epitaxial growth, samples need to be characterized to
analyze their growth result. An optical microscope is always firstly used to get a general view
of their surfaces and find places with interesting features. Then, other characterization methods
might be applied for further study on layer thickness, band structure, defects and composition,
which will be briefly introduced.
4.1 SEM and EDS
Scanning electron microscope (SEM) is a technique used for imaging of micro- or nano-
structures. Unlike optical microscope whose imaging depends on the interaction of visible light
and samples, SEM uses electron beams instead and scans in a raster pattern. Using electron
beam instead can largely improve the resolution of microscope beyond the optical limit. The
resolution R of microscope is confined by Rayleigh criterion
R =0.61λ
NA , (9)
where λ is wavelength and NA is numerical aperture. For optical microscope, the wavelength
of visible light is between 380nm and 780nm, yielding resolution in the order of 100nm.
However, as a de Broglie wave, the electron wavelength is typically several picometers, and
the resolution is down to 1nm or even less, which makes it possible to observe smaller
structures.
The setup of SEM is shown in Figure 4.1. It consists of an electron gun, several electron
lenses and detectors working in the vacuum. The electron beams are emitted and accelerated
at high voltage by the electron gun, and then focused by electromagnetic lenses. A deflection
system is applied to scan over a rectangular area on the sample. When the incident beams strike
to the sample, electrons will either be scattered by atoms, producing backscattering electrons
(BSE), or it will knock secondary electrons (SE) off the sample. The electrons are collected by
a detector and amplified, and both signals of intensity and position are recorded for imaging.
Generally, BSE mode has better resolution, and SE mode gives good topographic information.
Beside electrons, characteristic X-rays are also emitted. Since each element has its own
characteristic peaks on the emission spectrum, this helps to identify the composition of the
sample, as well as element distribution. This is known as energy-dispersive X-ray spectroscopy
or EDS.
(a) (b)
16
Fig.4.1 (a) The setup of SEM and (b) its working principle[13]
The resolution of SEM/EDS can be improved by reducing its working distance. However,
this will reduce its depth of field as well. Thus a trade-off is required in operation.
4.2 AFM
Atomic force microscopy (AFM) is also used in the project for morphology
characterization. Its resolution can be as high as 1nm, and also gives other information like
roughness and topography. Unlike SEM where a vacuum system is required, AFM is able to
work in the ambient atmosphere.
The setup is shown in Figure 4.2. It has a tip attached to a cantilever. The tip is made of Si
or Si3N4. It scans over an area on the sample surface and requires much longer time for imaging
(several minutes). When its apex and surface atoms are close enough, an attractive or repulsive
force will be generated and is strongly dependent on their distance. In contact mode, a
piezoelectric positioning element adds a pushing force to the cantilever and a repulsive force
between tip and sample is generated. It affects deflection of the cantilever, which can be sensed
and sent into a DC feedback amplifier. The signals are converted to topography map by
recording their coordinates in 3 directions.
Fig.4.2 An AFM setup
Compared to contact mode, tapping mode is more commonly used in the measurement. It
utilizes the oscillation of the cantilever at resonance frequency, thus “tapping” the sample
surface. During each time of tapping, the energy loss caused by the contact of tip and surface
will lower the oscillation amplitude. The reduction of amplitude is detected for surface mapping.
AFM also has a feedback loop which is able to maintain the oscillation amplitude constant by
automatically adjusting the distance between tip and sample according to the feedback signal.
Different kinds of artifacts may also appear in the AFM images. For example, if the tip
moves across a steep edge of the sample, hysteresis of the scanning system may cause the image
smoother. Besides, the tip is critical to the quality of image. If it gets damaged, or if its aspect
ratio is not enough, artifacts may also appear.
4.3 XRD and RLM
X-ray diffraction (XRD) is a tool used for analysis of crystalline quality, layer thickness,
17
strain, and composition, etc. The measurement of crystal is based on the Bragg’s law (Figure
4.3) which can be described as an equation
2dsinθ=nλ, (10)
where d is the spacing of two adjacent planes of atoms, θ is incident angle, n is any integer, and
λ is its wavelength. When an incident X-ray strikes a crystal at a certain angle that fulfills the
Bragg’s law, the scattered waves will interfere constructively, and the intensity of beams will be
stronger. This angle is known as Bragg angle. For different crystals, the lattice constant and
crystal orientation are different, yielding different Bragg angles. These angles can be used to
identify the composition of samples.
Fig.4.3 Bragg’s law in crystals[14]
A High Resolution X-Ray Diffraction (HRXRD) is shown in Figure 4.4. It consists of an
X-ray tube, a monochromator, a sample holder, and two detectors. Inside the X-ray tube, a high-
voltage field is applied to accelerate electrons emitted by a filament at one end. The electrons
gain large kinetic energy and when they strike a metal target at another end, X-rays will fly out.
A monochromator is used to select the wavelength of X-ray, allowing only one certain
wavelength to pass through. After that, the X-ray beams strike the crystal samples on the sample
holder. The holder has 3 rotational degrees of freedom (ω, ψ, φ) and is also able to move along
z axis. These 4 parameters need to be optimized before measurement. The beams scattered by
the sample are collected by either double-axis or triple-axis detector.
Fig.4.4 The HRXRD setup. 1. X-ray tube, 2. monochromator, 3. sample holder, 4.
Detectors
There are 2 scanning modes most commonly used: ω scan (also known as rocking curve)
and ω-2θ scan. ω is the angle between incident beam and the sample surface, while 2θ is the
18
angle between incident beam and detectors (Figure 4.5). ω and 2θ can be tuned by rotating the
sample holder and detectors, respectively. ω scan can plot the curve of intensity versus ω, from
which the perfection of crystalline quality can be determined. Any defect such as misfit
dislocation, threading dislocation, or anti-phase domain in the crystal will broaden the peak in
the curve. ω-2θ scan rotates both sample holder and detectors to keep ω and 2θ in coupled
manner (2 ∙ ω = 2θ). In detection of epitaxial films grown on a substrate, it plots both substrate
peak and layer peaks. From these peak positions, information on composition or
strain/relaxation can be calculated.
Both double-axis and triple-axis detectors can be used for ω and ω-2θ scan. The difference
is that triple-axis detector has an analyzer which constrains incoming beams to the detector, so
that only the signal at a certain angle can be observed, while double-axis detector can receive
signal from different angles. In order to avoid the intensity being too high at that angle and
damage the detector, an attenuator is inserted in front of the monochromator.
The collected data is processed by “Philips X’Pert Epitaxy” software, which is able to plot
the curve and calculate lattice mismatch, composition, and evaluate crystalline quality from full
width at half maximum (FWHM) of peaks. The sample with high crystallinity always has sharp
and intense peak, thus its FWHM is smaller. The lattice mismatch and composition of an epitaxy
layer grown on a substrate is obtained from ω-2θ scan. Layer and substrate peak positions 𝜃𝐿
and 𝜃𝑆 are firstly extracted. From Bragg’s law, the lattice constant of layer and substrate in
vertical direction (the direction of epitaxial growth) is calculated by
nλ = 2aL⊥sinθL (11)
nλ = 2aSsinθS . (12)
Thus the vertical lattice mismatch is
(∆a
a)
⊥=
aL⊥−aS
aS=
sinθL
sinθS− 1 . (13)
The relaxed lattice mismatch is
(∆a
a)
Relaxed=
1
2[(
∆a
a)
⊥+ (
∆a
a)
∥] . (14)
Under a critical thickness, the layer is coherently grown on the substrate (Figure 4.5(a)), which
means its lateral lattice mismatch
(∆a
a)
∥= 0. (15)
Another case is that the layer is fully relaxed. If the layer is thick enough, then the lateral and
vertical lattice mismatches take the same value,
(∆a
a)
∥= (
∆a
a)
⊥ . (16)
In either of these two cases, the relaxed lattice mismatch can be calculated. Otherwise, if the
layer is partly relaxed, then reciprocal lattice mapping (RLM) is required to make.
(a) (b)
Layer
Substrate
19
Fig. 4.5 Lattice mismatch of layer and substrate, where (a) is the case of coherent growth and
(b) is the relaxed growth
In ternary semiconductor materials such as GaAs1-xPx and AlxGa1-x As, their composition,
which is represented by molar fraction x, always needs to be determined for analysis of band
structure and lattice constant. This can be derived from Vegard's law, which can be described as
a formula,
aA1−xBx= (1 − x)aA + xaB . (17)
A and B are two elements of a compound 𝐴1−𝑥𝐵𝑥, and 𝑎𝐴1−𝑥𝐵𝑥, 𝑎𝐴 𝑎𝑛𝑑 𝑎𝐵 are their lattice
constants. Then the relaxed lattice mismatch of 𝐴1−𝑥𝐵𝑥 on a substrate is
(∆a
a)
Relaxed=
(1−x)aA+xaB−asubstrate
asubstrate. (18)
Since asubstrate, aA, aB and the lattice mismatch are known, its molar fraction x can be
calculated.
Fig.4.6 Schematic diagram of 2 scanning modes
More detailed information can be extracted from RLM. It is realized by varying ω and
collecting data of multiple sets of ω-2θ scans. A 2D map of reciprocal lattice points is plotted,
from which the lateral and vertical lattice mismatch can be directly read, thus determining the
degree of strain/relaxation in two directions. Other information such as mosaic spread can also
be extracted by the software.
4.4 PL spectroscopy
Photoluminescence (PL) spectroscopy is a powerful characterization method mainly used
for studying material properties such as composition, bandgap, and crystalline quality.[15] It is
based on the phenomenon that photons are emitted from semiconductors after absorption of
higher energy photons (Figure 4.7). When the external photons with energy higher than the
bandgap are absorbed by a semiconductor, electrons will be excited from valence band to
conduction band. Then, after interacting with surrounding molecules, electrons lose a part of
energy and fall to the minimum of conduction band. At last, after a period of time, electrons fall
to valence band again and recombine with holes. The recombination generates new photons
emitted out. From above principle it is obvious that using PL spectroscopy has two requirements:
1. The energy of external photons should be larger than the bandgap of the material so that
2θ ω
Detectors (movable)
X-ray tube (fixed)
Monochromator
Sample
20
electrons can be excited from valence band to conduction band. 2. Semiconductor under
examination should have a direct bandgap. For indirect bandgap materials such as GaP,
momentum provided by phonons are required to assist electrons’ recombination. Hence PL in
indirect bandgap materials could hardly be observed.
Fig.4.7 The process of photoluminescence[16]
The schematic diagram of PL spectroscopy is shown in Figure 4.8. It consists of a light
source, a monochromator, a detector, and several optical prisms and lenses. An argon laser is
usually used as the light source which emits 514 nm green light at a fixed power. The light
directed by prisms and lenses is shot onto the sample. Then, after excitation and emission, the
photons are concentrated by lenses, selected by a monochromator and collected by a detector.
A spectrum of intensity versus emission wavelength is finally plotted. The bandgap of each
semiconductor is unique, so the material and bandgap can be identified by the peak position
and by using the formula
Eg =hc
λ (19)
or
Eg(eV) =1.24
λ(μm) (20)
where λ is the peak position. The peak also gives the information of crystalline quality and
defects. If the quality of the sample is good, the peak will be more intense. Otherwise, the peak
becomes broad and weak.
Fig.4.8 Schematic diagram of PL spectroscopy
4.5 Raman spectroscopy
Valence band
Conduction band
Excited states
Bandgap energy Excitation photon (λ)
Luminescence photon (λ’)
Electron
Photon
Argon laser
Monochromator Detector Electronics
Sample Lens 1 Lens 2
Prism
21
Raman spectroscopy is used for analysis of crystallographic orientation, composition, and
strain of semiconductor materials. It is based on the Raman scattering, where the photons of
incident light are scattered by the molecules, causing their energy shift up or down. Electrons
are excited by incident photons to a virtual level with higher energy. It is not a stable level, so
electrons will fall back to lower levels and emit photons. In this procedure, if electrons start
from the ground level and fall back to it, there will be no change in photon energy. This elastic
scattering is known as Rayleigh scattering. However, if it starts from the ground level and fall
to vibrational levels or on the contrary, from vibrational levels to the ground level, the energy
of emitted photons will no longer be the same as before. If the scattered light has less energy,
then it is Stokes Raman scattering. Otherwise, it is anti-Stokes Raman scattering.
Figure 4.9 gives the schematic diagram of Raman spectroscopy. Like PL spectroscopy, it
also uses argon laser as light source. A microscope is used to observe the place of interest on the
sample and adjust the focus of laser. The scattered light passes through an optical filter, where
the elastically scattered light is filtered out, leaving only Raman scattering part. Then, a
diffraction grating disperses light and projects it onto a CCD detector, thus a Raman spectrum
is plotted. In the measurement, the detector always needs to be cooled down by liquid nitrogen
to reduce the dark current and improve spectrum quality.
Fig.4.9 Schematic diagram of Raman spectroscopy[17]
The Raman spectrum plots the signal intensity versus Raman shift, which is defined as the
difference between the wavenumbers of incident and scattered light,
Raman shift = (1
𝜆0−
1
𝜆1). (21)
𝜆0 and 𝜆1 represent the wavelength of incident and scattered light, respectively. Wavenumber
is directly proportional to energy, and is more convenient to represent energy shift than
wavelength. By analyzing Raman shift of the peaks on the spectrum, which semiconductor
material it is and its crystalline orientation can be identified. The shift of peak position from a
standard also gives information of strain inside the sample. Both tensile and compressive strains
change the crystalline structure and molecular vibration, which shift the peaks to lower and
higher wavenumbers, respectively. By analyzing the degree of shift, the strain can be
quantitatively analyzed. Besides, the crystalline quality is determined by how intense and sharp
the peaks are, which is similar to PL spectroscopy.
Laser
Beam
splitter
Grating
CCD
detector
Raman spectrometer
Sample
Microscope
22
4.6 Ellipsometry
Ellipsometry uses the principle of polarization of light to study material properties such as
refractive index, layer thickness, composition, etc. Polarization is a phenomenon describing the
change in electric field of light. As an electromagnetic wave, light travels in the space with its
electric field transverse to travelling direction. The electric field then can be separated to two
components in two orthogonal directions (Ex and Ey in Figure 4.10), thus separating one wave
into two waves. There are 3 kinds of polarized light commonly used: 1. linearly polarized light,
where the waves travel in phase, 2. circularly polarized light, where the waves are 90°out of
phase but has the same amplitude, and 3. elliptically polarized light, where the phases and
amplitudes are arbitrary but fixed. The linearly and circularly polarized light are the special
cases of the elliptically polarized light.
(a) (b) (c)
Fig. 4.10 Three kinds of polarized light, where arrows are polarization directions. (a) linear
polarization, (b) elliptical polarization, (c) circular polarization[18]
In ellipsometry, such an elliptically polarized light is used as an incident light to shot onto
the sample. As is mentioned before, the wave can be separated into two waves in two orthogonal
directions. Generally in ellipsometry, it is separated into p- and s-waves, the waves parallel and
normal to the plane of incidence. Snell’s law of refraction tells the relationship between incident
and refracted angles when passing through the interface two media (Figure 4.11). It is described
as a formula,
n1sin φ1 = n2sinφ2 (22)
in dielectric materials like glass and water. n1 and n2 represent the refractive indices of two
materials, and φ1, φ2 are the incident and refracted angles, respectively. In other materials such
as metal and semiconductors, Snell’s law takes the general form
Ñ1sin φ1 = Ñ2sinφ2. (23)
where
Ñ1 = n1 − 𝑗k1 (24)
Ñ2 = n2 − 𝑗k2 (25)
are the complex refractive indices comprising a refractive index n and an extinction coefficient
k which represents absorption of the material.
Ex
Ey
Ex
Ey
Ex
Ey
𝜑1
𝜑2
Ñ1
Ñ2
23
Fig.4.11 Snell’s law of light passing through the interface of two media
When the refractive angle is obtained, the reflection coefficients of p- and s- waves can be
calculated by Fresnel equations
r12p
=Ñ2cosφ1−Ñ1cosφ2
Ñ2cosφ1+Ñ1cosφ2 (26)
r12s =
Ñ1cosφ1−Ñ2cosφ2
Ñ1cosφ1+Ñ2cosφ2 (27)
where the refractive coefficients r12p
and r12s are the ratios of amplitude of incident and
reflected light. The reflectance R of two waves is defined as the squire of amplitude ratio,
Rp = |r12p
|2 (28)
Rs = |r12s |2. (29)
When measuring the sample with one or more epilayers grown on the substrate, there are at least
two interfaces, and the reflection coefficient of each layer need to be calculated separately to
obtain the total reflection coefficient. Take one-layer case for example (Figure 4.12), the total
reflection coefficient is given by
rtotp
=r12
p+r23
pexp (−j2β)
1+r12p
r23p
exp (−j2β) , (30)
rtots =
r12s +r23
s exp (−j2β)
1+r12s r23
s exp (−j2β), (31)
where β is the phase change from top to bottom of the epilayer, and is relevant to thickness d
of the epilayer,
β = 2π(d
λ)Ñ2cosφ2. (32)
Therefore, the total reflectance is
Rp = |rtotp
|2 (33)
Rs = |rtots |2. (34)
Fig.4.12 The light reflection of a sample with one epilayer
As an indirect method, the data collected and plotted by ellipsometry are not these material
properties, but phase change Δ and amplitude ratio ψ of the reflected and incident beams.
However, these parameters are related to the reflection coefficient by the formula
tanψ ∙ ejΔ =rtot
p
rtots . (35)
Figure 4.13 is the schematic diagram of ellipsometry. It consists of a light source, a
polarizer, a rotating analyzer and a detector. The polarizer allows only a certain direction of
electric field to pass, so that the unpolarized light emitted from the light source is converted to
linearly polarized light and is reflected by the sample. The reflection causes phase and
𝜑1
𝜑2
Ñ1 air
Ñ2 epilayer
Ñ3 substrate 𝜑3
24
amplitude change of p- and s-waves, turning from linearly to elliptically polarized light. A
continuously rotating analyzer combined with the detector is used to measure polarization
change.
Fig.4.13 The schematic diagram of ellipsometry[20]
In data analysis, a model of complex refractive index and thickness need to build to derive
Δ and ψ curves and fit with the experimental data. The flow chat is shown in Figure 4.14. Take
Cauchy model used for dielectric materials (k=0) for example, the refractive index takes the
form
n(λ) = An +Bn
λ2 +Cn
λ4 , (36)
where 𝐴𝑛, 𝐵𝑛, 𝐶𝑛 are parameters of each layer and need to be determined. The first step is to
assign each of the parameter a number either taken from an optical handbook or merely a guess.
Then, according to Snell’s law and Fresnel equations, the reflection coefficient of each layer is
calculated, yielding the total reflectance of p- and s- waves. Next, phase change Δ and amplitude
ratio ψ are derived by equation (35) and plotted. The Δ and ψ curves are compared with the
experimental data. Parameters of the refractive index and layer thickness are tuned until two
curves fit well. Then the refractive index and thickness are the true values.
Light source Detector
Analyzer
Φ
Polarizer
Sample
25
Fig. 4.14 The flow chat of simulation
Ellipsometry has very high precision in measuring thickness of thin films. The minimum
thickness can be as low as several nanometers, which is very suitable for measuring the buffer
layers in this project grown at low temperature for 2 min.
No
Yes
Selecting a model, setting initial values for Ñ1 Ñ2 Ñ3 𝑑
φ1 φ2 φ3 by Snell’s law
r12p
r12s r23
p r23
s by Fresnel equation
rtotp
rtots
β
Simulation curves of Δ and ψ
Comparing with experimental curves of Δ and ψ
Fit well?
Ñ1 Ñ2 Ñ3 𝑑 are true values
setting new values for
Ñ1 Ñ2 Ñ3 𝑑
26
5. Result analysis
The prepared samples are characterized by the methods including SEM, AFM, XRD, PL
and Raman spectroscopy, etc. introduced in Chapter 4. Table 2 summarizes these methods
applied to each sample, and the result will be elaborated in this chapter.
Table 2 Characterization methods applied to the prepared samples
Run
num-
ber
Layers Substrates SEM/
optical
microscope
EDS AFM XRD RLM PL Raman Ellip-
sometry
3750 GaAs
overlayer
+GaAs
buffer
layer
Planar
Si(001) X
X X
X X
Planar
Si(111) X
X X
X X
Patterned
Si(001) X
X X
X X
Patterned
Si(111) X
X X
X X
Planar
GaAs X
X
X X
3770 GaP
overlayer
+GaP
buffer
layer
Planar
Si(001) X X X X
X
Planar
Si(111) X
X X
X
Patterned
Si(001) X
X X
X
Patterned
Si(111) X
X X
X
Planar
GaP X
X
X
3800 GaAsP
overlayer
+GaP
buffer
layer
Planar
Si(001) X X
X
X
Planar
Si(111) X
X
X
Textured
Si(001) X
X X X
Planar
GaP X
X
X
3805 GaP buffer
layer
Planar
Si(001) X
X X
Planar
Si(111) X
X
Textured
Si(001) X
X
27
Planar
GaAs X
X
5.1 GaAs on Si
In this run, planar and patterned Si substrates, as well as a GaAs(001) substrate are
prepared for low-temperature GaAs buffer layer and high-temperature GaAs overlayer growth.
The GaAs nanopillars are grown on patterned n-type Si(111) and Si(001) substrates, as shown
in Figure 5.1(a)(b). Apparently, the crystalline orientation has great influence on the growth
result of HVPE as expected. On Si(111), the 2-step growth resulted in circular patches formed
at low temperature (439℃) and pillars grown at high temperature (665℃). The diameter of the
pillar is about 4μm, slightly larger than the size of openings (2-4μm) and the diameter of the
patches is about 8μm. The difference in vertical/lateral growth rate could be the coherent result
of change in V/III ratio, total flow, and temperature which has an influence on supersaturation
in HVPE. On Si(001), the growth of GaAs is mainly lateral and forms only circular patches
with diameter of 10-20μm. In some areas, the lateral growth is so fast that the surface is totally
covered by GaAs and thus forming a rough layer. Both patterned samples show a good selective
area growth (SAG) from openings of the substrates, which is a merit of HVPE.
(a) (b)
(c) (d)
28
(e)
Fig.5.1 (a) SEM top view of GaAs nanopillars on patterned n-type Si(111) and (b) Si(001)
substrates, and epilayers on planar (c) Si(111), (d) Si(001), and (e) planar GaAs top view by
Nomarski microscopy.
Beside the patterned substrates, two planar Si(111) and Si(001), as well as a planar GaAs
samples grown at the same condition are also investigated (Figure 5.1(c)(d)(e)). The Si(111)
sample has a rough surface consisting of many pits with diameter less than 1μm, while the
Si(001) sample surface are covered with meandering lines. These lines indicate the presence of
anti-phase domain (APD) introduced in Chapter 2. APD arises from the interface of substrate
and layer, and propagates to the layer surface, forming flat terraces and meandering lines. The
planar GaAs sample has the smoothest surface with the fewest defects. It consists of many
hillocks at the center of which are the sites of nucleation. This nucleation process happens at
the defects on the substrate surface and provides atomic steps to facilitate following growth.
Thus, the growth on defects can be faster than other areas and finally forms hillocks.
AFM gives clear 3D morphology of the same samples (Figure 5.2). The height of pillars
and patches can be extracted from the images, which is about 2.8μm for pillars on Si(111). On
Si(001), the height of patches is about 1.1μm. There are also some artifacts present at the edge
of the patches in Figure 5.2(a), which is due to hysteresis of the AFM scanner when the tip
moves across steep edges. The pits on the planar Si(111) sample are identified as cavities
(Figure 5.2(c)). They are assumed to the result of self-annihilation of APD along <111>
directions inside the layer. There are also some molecules attached on the surface, which are
identified using EDS as carbon. The RMS roughness of sample surface is 29.5nm and because
of APD annihilation, the thickness variation is not large. The 3D image also gives a clear view
of APD on the planar Si(001). Those meandering lines in SEM are the edges of flat terraces.
Because of the presence of terraces, thickness variation of this sample is the greatest among all
of the samples. The RMS roughness reaches 69.3nm.
29
(a) (b)
(c) (d)
Fig.5.2 3D AFM plots of GaAs on patterned n-type (a) Si(111) and (b)Si(001) substrates, and
planar (c) Si(111) and (d) Si(001) substrates
By using double-axis XRD scans, lattice mismatch and crystalline quality of these samples
are studied. (004) and (333) diffraction planes are selected for (001) and (111) samples,
respectively, to obtain a relatively strong signal. Figure 5.3 shows the result of scans, where (a)
and (b) are obtained by ω-2θ scan on both layer and substrate peaks. The lattice mismatch in
growth direction is calculated by measuring their peak positions. Patterned and planar Si(111)
samples have 3.56% and 3.70% mismatch, respectively, and Si(001) samples are slightly larger,
which are 3.97% and 3.78%. The variation of mismatch in different samples can be explained
by different thickness of layers, and different height and diameters of patches or pillars. But
considering that the mismatch between bulk GaAs and Si is 4.09%, the strain in these samples
are close to fully relaxed, which indicates misfit dislocation present as a defect at the GaAs/Si
interface. Figure 5.3 (c) is ω scan on the layer peaks. Full width at half maximum (FWHM) of
the peaks indicates the crystalline quality, and a better quality yields a smaller FWHM.
Substrates always have sharp, intense peaks, and their FWHM are all very small. However, the
layer peaks are quite different in these samples. The GaAs homostructure sample has both layer
and substrate peaks overlapped at the same position as expected, so it forms the most intense
and sharp peak. Among the rest of peaks, planar Si(111) has the lowest FWHM, which is
275arcsec. The better quality of this sample is assumed to be the result of APD annihilation,
and denser atomic steps in <111> directions that improve deposition. Meanwhile, the patterned
Si(111) has the largest FWHM (891arcsec) among all of the samples, which indicates that
growing on the patterned instead of planar Si(111) substrate will greatly lower the crystalline
quality. Compared to Si(111) samples, two Si(001) samples have similar FWHM, which are
707arcsec and 618arcsec. This means using the patterned Si(001) substrate makes no large
difference in crystalline quality, though these samples still have large amount of defects in their
layers that broaden and lower the peaks.
30
(a) (b)
(c)
Fig. 5.3 XRD plot of (a) patterned and planar Si(111) ω-2θ scan, (b) patterned and planar Si(001)
ω-2θ scan, (c) ω scan of all the samples. Scans on (004) and (333) diffraction planes are for
(001) and (111) samples, respectively
Figure 5.4 is the PL spectrum of these samples. A planar GaAs sample grown in a
calibration run is taken as a reference. In the calibration runs, only overlayers are grown on
substrates, while the growth condition and time is the same as this run. The peaks of two planar
GaAs samples are both sharp and intense, which implies a good crystalline quality with the
fewest defects as expected. Among the rest of the samples, planar Si(111) is the best, followed
by planar and patterned Si(001), which have the similar peaks. The patterned Si(111) has the
broadest and weakest peaks because of poor layer quality of the sample. This result is generally
the same as the result of ω scan by XRD in Figure 5.3(c). The peak positions are slightly
different. For two GaAs-substrate samples, they are both 870nm, yielding 1.425eV bandgap
energy. This is the same as theoretical value. However, the rest of samples more or less shift
from this value. The planar and patterned Si(111) samples’ peak positions are at 880.8nm and
871.1nm, respectively. It corresponds to their lattice mismatch measured in XRD, which are
3.70% and 3.56%. A larger lattice mismatch seems to cause peak positions shift towards the
longer wavelength. In Si(001) samples, similar conclusion can be made as well. The peak
position of planar Si(001) is 874.4nm and has 3.78% mismatch, while the patterned Si(001) is
876.2nm and has 3.97% mismatch. This can be explained by strain in mismatched materials
that affects p orbitals of electron, yielding a change in bandgap. In bandgap engineering, strain
is always utilized for tuning the properties of interest.
31
Fig. 5.4 PL spectrum of GaAs growing on Si and GaAs substrates
PL mapping gives the images of luminescence intensity over an area at a certain
wavelength. In Figure 5.5, four Si-substrate samples are investigated by PL mapping at the room
temperature. The planar Si(111), Si(001) and patterned Si(111) samples are scanned over a
30μm × 30μm area with 3μm step size, and patterned Si(001), considering the size of its
patches, is scanned over a 60μm × 60μm area with the same step size. As can be seen from
Figure 5.5(a), the PL mapping of the patterned Si(111) at 869.5nm gives a clear image of
nanopillars and patches with the similar size as measured by AFM. The emission intensity of
pillars are stronger, which indicates a better crystalline quality than in the patches. The blue
areas are covered by SiO2, and the intensity is the weakest. From Figure 5.5(b), the patches on
patterned Si(001) is not easy to be recognized, but there are still some emission peaks which
are assumed to be the center of patches. The overall intensity of planar Si(111) is the strongest
(Figure 5.5(c)), which accords with the result of PL spectrum in Figure 5.4. However, the
intensity variation is still large, and at the center, the intensity is less than a half at the edge.
This implies that defects still exist and unevenly spread on the surface. The similar conclusion
can be made to planar Si(001), where the influence of APD is more severe and yields a lower
intensity.
(a) (b)
32
(c) (d)
Fig. 5.5 PL mapping of GaAs growing on patterned (a) Si(111), (b)Si(001), and planar (c)
Si(111), (d) Si(001) substrates
Raman spectra of GaAs are shown in Figure 5.6. The presence of different modes indicates
different crystalline orientations present at the surface. For the sample having (001) substrates,
a (001) layer is expected, which is shown in Raman spectrum as the presence of LO and absence
of TO. On the contrary, a (111) or (110) layer results in the presence of TO and absence of LO.
The planar Si(001) and two GaAs samples only have LO mode, and the planar Si(111) only has
TO mode in this spectrum, that is what expected. However, two patterned samples have both
LO and TO, and two modes have similar intensity. This indicates that their surfaces have facets
of other orientations.[10] There is almost no LO peak shift from GaAs calibration sample, which
means the strain inside the sample is small.
Fig. 5.6 Raman spectrum of GaAs growing on Si and on GaAs
Hall measurement is performed to the planar Si(001) sample, along with two calibration
runs which grow only overlayers but have the same growth condition. The results are listed in
Table 3. The Hall mobility of planar Si(001) sample (1306cm2/(V⋅s)) is much lower than two
calibration runs(4737cm2/(V ⋅ s) and 5109cm2/(V ⋅ s)). It is the result of higher defect
concentration in the epilayer that largely decrease the mean free time between collisions and
will lead to an inferior device performance. This result is consistent with the result of XRD and
PL.
Table3 Hall measurement to 3607, 3627 and 3750 GaAs runs
Run number Substrate type Carrier concentration
(cm-3)
Hall mobility
(cm2/(V⋅s))
3607 (calibration) SI GaAs 2.98× 1015 4737
33
3627 (calibration) SI GaAs 1.53× 1015 5109
3750 n-type planar
Si(001)
2.99× 1017 1306
5.2 GaP on Si
The planar and patterned n-type (001) and (111) Si, as well as a planar GaP(001) substrate
are prepared for growing GaP buffer layer and overlayer. As can be seen in Figure 5.7(a)(b), the
lateral growth on the patterned Si(111) and Si(001) samples becomes too fast that patches fully
coalesce and form a rough layer on SiO2 mask. On Si(111), nanopillars with diameter around
8μm can be identified, which shows their growth both in lateral and vertical directions. On
Si(001), the lateral growth is so fast that a rough GaP layer is formed. This is similar to the
growth result of GaAs samples. On the planar Si samples (Figure 5.7(c)(d)), terraces and pits
due to APD are present, creating a much rougher surface and larger thickness variation than the
GaP-substrate sample (Figure 5.7(e)).
(a) (b)
(c) (d)
(e)
34
Fig.5.7 (a) The SEM top view of GaP on n-type patterned and (c) planar Si(111), (b) patterned
and (d) planar Si(001) substrates, and (e) the top view of GaP-substrate sample by Nomarski
microscope
EDS analysis on the cross section of planar Si(001) sample shows its element distribution
along growth direction (Figure 5.8). From the image of its cross section, the GaP layer has
thickness variation as expected, but is not that large as GaAs on planar Si(001). The thickness
of the layer is about 4μm. From the line scan result, the contents of Ga, P, and Si abruptly change
at the GaP/Si interface. There is almost no Ga and P signals on the substrate side, and Si on
layer. Hence the penetration of substrate and layer atoms is small.
(a) (b)
(c)
Fig.5.8 (a) Cross section of planar Si(001), (b) line scan site of EDS and (c) element
distribution along growth direction
The 3D images obtained by AFM are shown in Figure 5.9. The height of pillars on
patterned Si(111) are about 3μm. There is a rough GaP layer covering on SiO2, and from AFM
analysis, its RMS roughness is 189nm. On the patterned Si(001), the thickness variation reaches
3.2μm. The terraces on planar Si samples can be more clearly identified. On the planar Si(111),
the thickness variation is 1.5μm, while on planar Si(001) is 1.1μm. The depth of pits is less than
1μm. Considering the layer thickness, these pits are not deep enough to touch the interface.
35
(a) (b)
(c) (d)
Fig.5.9 3D AFM plots of GaP on n-type (a) patterned and (c) planar Si(111), (b) patterned and
(d) planar Si(001) substrates
Triple-axis XRD scans, including ω and ω-2θ scans, measure crystalline quality and lattice
mismatch of the samples in this run and its calibration run 3768 (Figure 5.10). As can be seen
in Figure 5.10 (a)(b), the patterned Si(001) has very poor quality, and yields no layer peak in ω-
2θ scan. The lattice mismatches of the rest of samples are from 0.27% to 0.34%. It is close to
the lattice mismatch between two bulk materials (0.37%), which means the strain in the samples
are close to fully relaxed. From ω scans plotted in Figure 5.10(c), the patterned Si(001) sample’s
signal is still too weak to detect. Two Si(111) samples’ FWHM are basically the same, and have
the best quality among the rest of the samples. The FWHM of planar Si(001) reaches 913arcsec,
more than twice as that of Si(111). This indicates that defects like APD have more severe
influence on GaP growth on (001) plane than (111). The ω scan on the planar GaP(001) shows
overlapping of substrate and layer peaks. Because of homoepitaxy, the peak positions are
basically the same, but because of defects, a broader layer peak with 623arcsec FWHM is
observed. This result is better than planar and patterned Si(001), but is still much too large
comparing with GaP homostructure grown in 3768 calibration run (Figure 5.10(d)), whose
FWHM is 10arcsec. The only difference between these 2 runs is that the calibration run does
not grow buffer layer, having only overlayer directly grown on substrate at the same condition.
This implies that the growth condition of buffer layer is not optimized that it introduces much
more defects in the nucleation process. More investigations will be made in Chapter 5.4, where
only GaP buffer layer is grown on Si substrates.
36
(a) (b)
(c) (d)
Fig.5.10 XRD plot of (a) patterned and planar Si(111) ω-2θ scan, (b) patterned and planar Si(001)
ω-2θ scan, (c) ω scan of all the samples in run 3770, and (d) ω scan of GaP homostructure in
3768 the calibration run. Scans on (004) and (333) diffraction planes are for (001) and (111)
samples, respectively
A Hall-effect measurement is performed at room temperature to GaP-substrate samples
(Table 4). The result shows that the Hall mobility of GaP sample in run 3770 is only 34cm2 V-
1s-1, smaller than GaP in 3768 calibration run (83cm2 V-1s-1). Thus it can be concluded that the
defects originating from buffer layer growth have great influence on electron mobility and affect
the device performance错误!未找到引用源。
.
Table 4 Hall measurement to 3768 and 3770 GaP runs
Run number Substrate type Carrier concentration
(cm-3)
Hall mobility
(cm2/(V⋅s))
3768 (calibration) SI GaP 3.21 × 1016 83
3770 SI GaP 3.7 × 1016 34
Figure 5.11 is the Raman spectrum of the samples. Both GaP samples have a strong LO
peak and a weak TO peak in the spectrum, which indicates almost no (111) or (110) facets on
the surface. The rest of samples all contain both LO and TO peaks, and have facets of other
orientations. Besides, the LO peak positions of Si samples are all close to 400cm−1, which is
almost the same as the sample in GaP calibration run. Small peak shift means strain in these
samples is also negligible.
37
Fig. 5.11 Raman spectrum of GaP on Si and GaP substrates
5.3 GaAsP on Si
The 3% KOH+10% IPA textured Si(001) substrate, along with the untextured, planar
Si(001), Si(111) and GaP(001) substrates, are used to grow GaAsP with GaP as buffer layer.
Figure 5.12 shows the top view and cross section of these samples. The surfaces of the
untextured Si(111) and Si(001) samples are quite rough and contain many grains. The cross
section of Si(001) shows that the thickness of GaAsP layer varies, and in some areas, there is
even no layer growth. The textured sample formed rough bumps. From its cross section (Figure
5.12(b)), it is clear that these bumps are formed by GaAsP grown on the pyramids. Compared
with the untextured sample, its uniformity of layer is improved, and the surface is totally
covered by a layer with thickness ranges from 2.6-4.8μm. The GaP sample has the smoothest
surface and contains only some hillocks observed in other homoepitaxy samples.
(a) (b)
(c) (d)
38
(e) (f)
Fig.5.12 (a) SEM top view of the planar Si(001), (c) textured Si(001), (e) planar Si(111),
the cross section of (b) planar Si(001), (d) textured Si(001), and (f) top view of GaP samples by
Nomarski microscopy
The EDS line scan to the untextured Si(001) shows its element distribution along growth
direction (Figure 5.13). From this figure, the interface of GaAsP and Si is identified, which is
the place where element contents abruptly change. The signal intensity of P is weaker than As,
which implies its lower content in the layer. Ga and Si have the strongest signals above and
below the interface, respectively. This conforms to the fact that they have the largest contents
in the layer or substrate. The Si signal at the layer region and Ga, P, As signals at the substrate
region are very weak, which means the elements penetrating to the other side are quite few.
(a)
39
(b)
Fig. 5.13 EDS analysis on the cross section of planar Si(001). (a) the site been line-scanned
(b) element distribution in vertical direction
In order to further study the crystalline quality and composition, XRD measurement is
performed on these samples (Figure 5.14). The composition of each GaAsP layer is obtained by
ω-2θ scan and summarized in the figure. It shows a slight variation in the content of As and P
elements on different substrates, even though the growth condition for the samples is the same.
For Si-substrate samples, their As content ranges from 0.78 to 0.80 and mismatch from 3.27%
to 3.36%, while for the GaP-substrate sample, As content is 0.75 and mismatch is 2.79%. The
difference in As and P content is explained by the effect known as lattice pulling[26], where the
strain between substrate and layer will change the element content in the layer, allowing its
lattice constant to be closer to substrate’s. Therefore, the GaAsP layer grown on Si will have
similar lattice constant as Si, and on GaP substrate will be similar to GaP. ω scan shows that the
layer of planar Si(001) sample has the largest FWHM, which reaches 4047arcsec. For the
textured sample, it is 1334arcsec. This indicates that texturing has greatly improved the
crystalline quality by reducing its defects, although the quality is still not ideal. The planar
Si(111) and GaP samples have lower FWHM than Si(001) samples, which are 827arcsec and
711arcsec, respectively. Their crystalline quality is the best in this run.
From XRD result of GaAs, GaP, and GaAsP heterostructures, the growth on planar Si(111)
always have better quality. This could be explained by APD self-annihilation and denser atomic
steps along <111> directions than <001> that improves deposition. And for the textured samples,
because of their pyramids contain {111} facets, their crystalline quality, uniformity and growth
rate are also improved.
(a) (b)
Fig.5.14 XRD plot of (a)ω-2θ scan and (b) ω scan on GaAsP/Si samples. Scans on (004)
and (333) diffraction planes are for (001) and (111) samples, respectively
RLM of the textured (001) Si sample is shown in Figure 5.15. It is realized by making a
series of ω-2θ scans with various ω offsets. The results of both symmetric (004) and asymmetric
(224) scans are plotted. Two coordinates Qz and Qx are parallel to the directions of [001] and
[110], respectively. In Figure 5.15(a), two peaks are observed, where the smaller peak is the
substrate’s and the other with broader and less intensive peak is the layer’s. The symmetric scan
can only measure the variations of lattice constant in vertical direction (Qz), which is 3.39%.
40
On the horizontal direction (Qx), the broadening of the layer peak means the presence of
mosaicity rising from the tilt of some crystallites during epitaxial growth. From asymmetric
scan, still the thinner peak is the substrate peak, while the broader is the layer peak. Both lateral
and vertical mismatches can be calculated, which are 3.56% and 3.33%, respectively. The blue
lines are in the directions of [224] and [001], which represent layer peak positions in the fully
relaxed and fully strained cases, respectively. The red line across the center of substrate and
layer peaks is between the blue lines but is much closer to the [224] line. This means that the
strain in the sample is close to fully relaxed. The substrate and layer peaks appear like a tilted
ellipsoid. That is coherent result of mosaic spread and lateral correlation due to a finite lateral
crystallite size. The green lines perpendicular to [224] show the peak broadening due of
mosaicity, and the layer peak is broader in this direction as expected.
(a) (b)
Fig. 5.15 RLM of the textured Si(001) sample by (a) symmetric scan on (004) diffraction plane
and (b) asymmetric scan on (224) plane
PL spectrum of these four samples are given in Figure 5.16. Compared to the untextured
samples, the textured sample has improved light emission intensity, thus its crystalline quality
is improved as well. Three Si-substrate samples have similar peak positions, which are 763.0nm
(textured Si(001)), 764.6nm (planar Si(001)), and 762.0nm (planar Si(111)). This indicates a
slight difference in bandgap energy (1.623eV~1.627eV). The peak position of planar GaP
sample is 728.5nm, which shows apparent difference (1.702eV) to other samples. This is
consistent with the result of XRD ω-2θ scan, where As content is the lowest in the GaP sample.
Fig. 5.16 PL spectrum of GaAsP growing on Si and GaP substrates
[001] [224]
Mosaic spread
41
5.4 GaP buffer layer growth
The GaP, GaAs, and GaAsP samples analyzed above all have buffer layers in a 2-step
epitaxial growth to facilitate the nucleation and reduce defects and residual stress. In order to
better understand the growth result, GaP buffer layers are grown on four different substrates: the
untextured, planar Si(001) and Si(111), textured Si(001) and planar GaAs(001) substrates. Their
SEM images are in Figure 5.17. It can be seen that buffer layers do have some growth at this
condition but is assumed to be very thin. The surfaces are smoother than overlayers’ illustrated
in Chapter 5.2. The planar Si(001) sample has half of the area with no GaP nucleation happening.
Noticed that in GaAsP run mentioned in Chapter 5.3, there are also some large areas in planar
Si(001) with no layer growth, we assume that this is because native oxide formed when substrate
is exposed to ambient atmosphere. This is less severe on Si(111) or textured Si(001) because
oxide layers form much slower along <111>. The Si(111) and planar GaAs samples have a lot
of dots on their surface. By using EDS (Figure 5.18), they are identified as the area with no GaP
growth. The textured sample has a good uniformity. All the pyramids are covered with a thin
layer.
(a) (b)
(c) (d)
Fig.5.17 SEM top view of (a) planar Si(001), (b) planar Si(111), (c) textured Si(001) and (d)
planar GaAs samples
42
(a) (b)
(c) (d)
Fig.5.18 EDS analysis to the planar Si(111) sample, red box in (a) is the area of interest, and
(b)(c)(d) are Si, P and Ga distribution
Ellipsometry is used to measure buffer layer thickness of untextured Si(001) sample and
calculate its growth rate. Since the buffer layer is grown at low temperature for only 2min, its
thickness is presumably below 100nm. Methods like profilometer cannot reach high precision
at this thickness. Therefore, using ellipsometry would be preferable.
As is introduced in Chapter 5.4, ellipsometry is an indirect method for thickness
measurement. It needs a model of complex refractive index to derive amplitude diminution ψ
and phase change Δ, and then fit to the experimental data of ψ and Δ by varying parameters in
the model. The model selected here is based on Forouhi-Bloomer dispersion equations, which
was initially used for amorphous materials[22]. By slightly modifying its formulation, now it can
also be used for crystalline semiconductors and dielectrics. The Forouhi-Bloomer dispersion
equations of complex refractive index is given by
n(E) = n(∞) + ∑B0i
E+C0i
E2−BiE+Ci
qi=1 (37)
k(E) = (E − Eg)2 [∑Ai
E2−BiE+Ci
qi=1 ] (38)
N(E) = n(E) − jk(E) (39)
q is an integer representing how many terms it has. Each term corresponds to a peak of n(E) and
k(E). Eg is the bandgap of the material. n(∞) is the refractive index when energy approaches
43
infinity. Ai, Bi and Ci are three dimensionless parameters relevant to the material properties.
B0i and C0i
are parameters dependent on Ai, Bi Ci and Eg,
B0i=
Ai
Qi(−
Bi2
2+ EgBi − Eg
2 + Ci) (40)
C0i=
Ai
Qi[(Eg
2 + Ci)Bi
2− 2EgCi] (41)
Qi =1
2√4Ci − Bi
2 (42)
An initial guess of these parameters needs to be given to the model, which is directly
correspond to the fitting time. Here these parameters for both Si and GaP are taken from A.R.
Forouhi’s article optical properties of crystalline semiconductors and dielectrics[22] and are
listed in Table 5 and Table 6.
Table 5 Parameters of Forouhi-Bloomer model for Si
i Ai Bi Ci Qi B0i C0i
Eg(eV) n(∞)
1 0.00405 6.885 11.864 0.114864 -0.19968 0.689602 1.06 1.95
2 0.01427 7.401 13.754 0.24556 -0.40167 1.504874
3 0.0683 8.634 18.812 0.41894 -1.70082 7.528832
4 0.17488 10.652 29.841 1.214382 -2.40838 14.639
Table 6 Parameters of Forouhi-Bloomer model for GaP
i Ai Bi Ci Qi B0i C0i
Eg (eV) n(∞)
1 0.00652 7.469 13.958 0.107284 -0.14805 0.555095 2.17 2.07
2 0.14427 7.684 15.041 0.529184 -0.68581 2.890168
3 0.13969 10.237 26.567 0.606595 -1.91729 10.31331
4 0.00548 13.775 47.612 0.417545 -0.28979 2.017527
On the fitting software of ellipsometry, initial values of 𝐴𝑖 , 𝐵𝑖 , 𝐶𝑖, 𝐸𝑔 , n(∞) and
thickness d are input and set variable, the modelling curve being most close to experimental data
will automatically be selected. This happens when d=78nm. The ψ and Δ curves at this thickness
are plotted in Figure 5.19, where the black and red curves are model and experimental data,
respectively. The ψ curves fit well, especially at long wavelength, where the difference is less
than 1º. Two Δ curves have similar shape and peak position, but do not fit well. This could be
because the model is not suitable enough for this sample. Further works can be done by trying
other fitting models such as Tauc-Lorentz or Drude-Lorentz modes and see if their curves can
fit better.
44
(a) (b)
Fig. 5.19 Model and experimental data of (a) ψ and (b) Δ versus wavelength
Figure 5.20 is the Raman spectrum of GaP buffer layer grown on four substrates. Each
sample has two peaks with similar intensity belonging to TO and LO modes. This indicates that
the layers have facets of other orientations on their surfaces. Compared with Raman spectrum
in Figure 5.11, it can be concluded that these facets exist in the whole process of epitaxy, from
buffer layer to overgrowth. The TO and LO peak positions of these samples are the same, which
are 363.0cm-1 and 398.7cm-1, respectively. This value of LO peaks is slightly smaller than the
planar GaP sample grown in calibration runs (the light green curve in Figure 5.11), which is
400.9 cm-1. This 0.54% peak shift is due to strain in the samples. The homostructure of GaP
does not have strain between substrate and layer. But in GaP/Si or GaP/GaAs heterostructure,
under a critical thickness, the strain due to lattice mismatch will be present and shift the Raman
peaks.
Fig. 5.20 Raman spectrum of GaP buffer layer growth
45
6. Conclusion and future work
In this thesis, we investigated the direct heteroepitaxial growth of three compound
semiconductor materials (GaAs, GaP and GaAsP) on different Si substrates by HVPE.
The substrates are prepared before epitaxy. The textured substrates are obtained by wet
etching using 3% KOH+10% IPA for 30min. The patterned substrates for selective-area growth
of nanopillars are processed by PECVD, photolithography and etching. After that, a 2-step
epitaxial growth of III-V compounds is involved by using HVPE, which includes a low-
temperature buffer layer growth for 2min and a high-temperature growth for 15min.
The first run grows GaAs layers and nanopillars on planar and patterned Si(001) and (111)
substrates. The surface of planar samples shows a feature of APD, though because of self-
annihilation, APD in Si(111) is less severe than in Si(001). By applying Hall measurement to
Si(001), its lower Hall mobility indicates that APD as a defect can greatly reduce the device
performance. The patterned samples prove selective-area growth of GaAs on substrates, which
is a merit of HVPE. Different growth rate on Si(001) and (111) also proves the influence of
surface kinetics in HVPE. XRD and PL measurement shows no great change in crystalline
quality of two Si(001) samples, but the planar and patterned Si(111) samples have the best and
worst qualities, respectively. XRD also shows variation of lattice mismatch (3.56%-3.97%), and
generally it is close to the mismatch in two bulk materials, indicating the strain in these samples
is close to fully relaxed. Compared with their lattice mismatch and bandgap, it can also be
concluded that bandgap can be tuned by different strain, thus tuning emission wavelength of
photonic devices. Raman spectrum shows the planar Si(001) only has LO mode, and the planar
Si(111) only has TO mode as expected, while two patterned samples have both LO and TO,
indicating the presence of facets of other orientations on their surfaces.
The second run grows GaP on the same Si substrates. The surface of planar samples shows
APD present on both Si(001) and Si(111). On the patterned samples, the lateral growth is so fast
that GaP covers on SiO2 layer. XRD measurement shows a smaller mismatch (0.27%-0.34%)
than GaAs/Si. Besides, two Si(111) samples have similar crystalline quality and are better than
Si(001). A broader layer peak with 623arcsec FWHM is observed, which is too large comparing
with GaP homostructure grown in its calibration run. This implies that the growth condition of
buffer layer is not optimized that it introduces much more defects in the nucleation process. Hall
measurement also indicate that the defects originating from buffer layer growth greatly lower
electron mobility and affect the device performance. Raman spectrum shows that all the
heterostructure samples have both TO and LO peaks at 363cm-1 and 400cm-1, respectively, thus
containing facets of other orientations.
The third run grows GaAsP on planar Si(001), Si(111) and textured Si(001). The planar
samples are still affected by APD and show thickness variation, while the textured sample has
better uniformity. XRD measures the variation in content of As and P, and their lattice mismatch
varies from 2.79% to 3.36%, though their growth condition is the same. The crystalline quality
of samples in this run are in general worse than previous runs. The FWHM of planar Si(001)
even reaches 4047arcsec, and textured Si(001) is 1334arcsec with improved quality. Both
symmetric (004) and asymmetric (224) scans are applied to textured Si(001), indicating little
strain but defects like mosaicity present in the sample. PL spectrum of three Si-substrate samples
shows similar bandgap energy (1.623eV~1.627eV), which are apparently different from the
46
GaP-substrate sample (1.702eV). Combined with the result of XRD, this difference can be
explained by lattice pulling effect where the growth on different substrates induces different
content in the layer.
The last run only grows GaP buffer layers on substrates. Their SEM images show that
buffer layers do have some growth but are assumed to be very thin and have poor uniformity.
By using Forouhi-Bloomer model in ellipsometry, the thickness of planar Si(001) is calculated
to be 78nm. In Raman spectrum, the TO and LO modes, at 363.0cm-1 and 398.7cm-1, respectively,
indicate the facets of other orientations present even in the nucleation process of crystal.
In general, the growth of three materials on planar Si(111) has better crystalline quality and
uniformity than Si(001), which is due to denser atomic steps, APD self-annihilation along <111>,
and slower formation of native oxide on their surface. Similar features can be also observed in
textured samples, which contain a large number of {111} facets. The patterned Si(111) samples
also show a preference to vertical growth, which shows a potential to grow nanowires with large
aspect ratio. But before that, the problem of low crystalline quality of patterned samples must
be overcome.
More works can be done to further investigate III-V/Si heterostructures in the future. For
example, temperature and V/III ratio can be optimized to improve the growth rate in HVPE.
Defects like APD, misfit dislocation should also be reduced to improve the performance of
photonic devices. In the future, doping will be induced, which will also gain difficulty on
epitaxial growth.
47
7. References
[1] Tomioka, K., Tanaka, Hara, Hiruma, & Fukui. (2011). III-V Nanowires on Si Substrate:
Selective-Area Growth and Device Applications. IEEE Journal of Selected Topics in Quantum
Electronics, 17(4), 1112-1129.
[2] Zhang, Kim, Faleev, & Honsberg. (2017). Improvement of GaP crystal quality and silicon bulk
lifetime in GaP/Si heteroepitaxy. Journal of Crystal Growth, 475, 83-87.
[3] Lin, A., Fejer, M., & Harris, J. (2013). Antiphase domain annihilation during growth of GaP
on Si by molecular beam epitaxy. Journal of Crystal Growth, 363, 258-263.
[4] Barrett, C. S. C., Martin, T. P., Bao, X.-Y., Martin, P., Sanchez, E., & Jones, K. S. (2016).
Determination of Antiphase Domain Boundary Annihilation Rate in GaAs on Si(001) and the
Influence of MOCVD Growth Temperature. ECS Transactions, 72(4), 335-340.
[5] Spirkoska, D., Arbiol, J., Gustafsson, Anders, Conesa-Boj, S., Glas, F., Zardo, I., . . .
Fontcuberta i Morral, A. (2009). Structural and optical properties of high quality zinc-
blende/wurtzite GaAs nanowire heterostructures. Physical Review B (Condensed Matter And
Materials Physics), 80(24), Physical Review B (Condensed Matter and Materials Physics),
2009, Vol.80(24).
[6] Lourdudoss, S., & Kjebon, O. (1997). Hydride vapor phase epitaxy revisited. IEEE Journal of
Selected Topics in Quantum Electronics, 3(3), 749-767.
[7] Stergiakis, S. (2016). Properties of III-V semiconductor materials grown by HVPE
(Dissertation). Retrieved from http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-188788
[8] Gil, E., André, Y., Cadoret, R., & Trassoudaine, A. (2014). Hydride Vapor Phase Epitaxy for
Current III-V and Nitride Semiconductor Compound Issues. In Handbook of Crystal Growth:
Thin Films and Epitaxy: Second Edition (Vol. 3, pp. 51-93). Elsevier.
[9] Crafts, P. (2007). Chapter 2 The role of solubility modeling and crystallization in the design of
active pharmaceutical ingredients. Computer Aided Chemical Engineering, 23, 23-85.
[10] Omanakuttan, G. (2019). Epitaxial III-V/Si heterojunctions for photonic devices (TRITA-SCI-
FOU ; 2019:32).
[11] Soga, T. , Jimbo, T. , & Umeno, M. . (1993). Dislocation generation mechanisms for GaP on
Si grown by metalorganic chemical vapor deposition. Applied Physics Letters, 63(18), 2543.
[12] Naresh-Kumar, Vilalta-Clemente, Jussila, Winkelmann, Nolze, Vespucci, . . . Trager-Cowan.
(2017). Quantitative imaging of anti-phase domains by polarity sensitive orientation mapping
using electron backscatter diffraction. Sci Rep, 7(1), 10916.
[13] Jonathan Atteberry "How Scanning Electron Microscopes Work" 21 April 2009.
HowStuffWorks.com. https://science.howstuffworks.com/scanning-electron-microscope.htm
15 July 2019
[14] Wikipedia contributors. (2019, July 17). X-ray crystallography. In Wikipedia, The Free
Encyclopedia. Retrieved 20:50, July 17, 2019, from
https://en.wikipedia.org/w/index.php?title=X-ray_crystallography&oldid=906663808
[15] Perkowitz, S. (1993). Optical Characterization of Semiconductors: Infrared, Raman, and
Photoluminescence Spectroscopy (Vol. 14, Techniques of Physics). Elsevier Science.
[16] Cordero, Eliana & Supervisors, B & Ploss, Bernd & Sivakov, Vladimir. (2016). Hydrogen
Generation on Silicon Nanostructures. 10.13140/RG.2.2.14474.57284.
[17] B. Fenn, Michael & Xanthopoulos, Petros & Pyrgiotakis, Georgios & Grobmyer, Stephen &
Pardalos, P & L. Hench, Larry. (2011). Raman Spectroscopy for Clinical Oncology. Advances
48
in Optical Technologies. 2011. 10.1155/2011/213783.
[18] Wikipedia contributors. (2019, May 21). Polarization (waves). In Wikipedia, The Free
Encyclopedia. Retrieved 19:29, July 20, 2019, from
https://en.wikipedia.org/w/index.php?title=Polarization_(waves)&oldid=898052028
[19] Aydinci, N. (2014). Processing Technology for Si Based Tandem Solar Cells.
[20] Wikipedia contributors. (2019, June 11). Ellipsometry. In Wikipedia, The Free Encyclopedia.
Retrieved 12:51, July 23, 2019, from
https://en.wikipedia.org/w/index.php?title=Ellipsometry&oldid=901309995
[21] Wikipedia contributors. (2019, October 1). Electron mobility. In Wikipedia, The Free
Encyclopedia. Retrieved 11:56, November 21, 2019, from
https://en.wikipedia.org/w/index.php?title=Electron_mobility&oldid=918999526
[22] Forouhi, & Bloomer. (1988). Optical properties of crystalline semiconductors and
dielectrics. Physical Review. B, Condensed Matter, 38(3), 1865-1874.
[23] Kotulak, Nicole. (2019). Growth and Analysis of Gallium Phosphide on Silicon for Very High
Efficiency Solar Cells.
[24] “Plasma Enhanced Chemical Vapour Deposition (PECVD)”.
https://plasma.oxinst.com/campaigns/technology/pecvd
[25] “Reactive Ion Etching (RIE)”. https://plasma.oxinst.com/campaigns/technology/reactive-ion-
etching
[26] Kong, Albert, Bengoechea‐Encabo, Sanchez‐Garcia, Calleja, & Trampert. (2015). Lattice
pulling effect and strain relaxation in axial (In,Ga)N/GaN nanowire heterostructures grown on
GaN‐buffered Si(111) substrate. Physica Status Solidi (a), 212(4), 736-739.
[27] Wikipedia contributors. (2017, October 18). Hydride vapour phase epitaxy. In Wikipedia, The
Free Encyclopedia. Retrieved 09:47, November 18, 2019, from
https://en.wikipedia.org/w/index.php?title=Hydride_vapour_phase_epitaxy&oldid=80584321
9
[28] Wikipedia contributors. (2019, January 5). Gallium arsenide phosphide. In Wikipedia, The Free
Encyclopedia. Retrieved 11:28, November 18, 2019, from
https://en.wikipedia.org/w/index.php?title=Gallium_arsenide_phosphide&oldid=877008720
[29] Tellurex Corporation. (2015, June 10). Gallium Arsenide Phosphide (GaAsP) Semiconductors.
AZoM. Retrieved on November 18, 2019 from
https://www.azom.com/article.aspx?ArticleID=8489.
[30] John, Susan. (2018). Different Types of in Light Emitting Diodes (LED) Materials and
Challenges- A Brief Review. International Journal for Research in Applied Science and
Engineering Technology. 6. 4418-4420. 10.22214/ijraset.2018.4723.
[31] Li, J., & Zhang, G. (2019). Light-Emitting Diodes Materials, Processes, Devices and
Applications (Solid State Lighting Technology and Application Series, 4).
49
www.kth.se
TRITA-SCI-GRU 2019:387