Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Efficient yellow and green emitting InGaN/GaN nanostructured QW materials and LEDs
(USC Thesis Other)
Efficient yellow and green emitting InGaN/GaN nanostructured QW materials and LEDs
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
1
Efficient Yellow and Green Emitting InGaN/GaN Nanostructured QW
Materials and LEDs
By
Yoshitake Nakajima
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Electrical Engineering)
December 2016
2
Acknowledgements
I would like to express my profound gratitude to my thesis advisor, Prof. P.
Danial Dapkus for his guidance and support during my Ph.D. studies. I appreciate that he
gave me an opportunity to work in his lab during my senior year of undergraduate
studies and inspired me to pursue research in the field of compound semiconductor and
optoelectronics. It is my great fortune to be able to work with such a renowned scientist
and a lab with a tremendous legacy in the field. Not only has he been a great mentor
with broad technical knowledge and experiences, but also a great mentor for always
teaching us life lessons through his experiences and stories. Under his guidance and the
project he provided me, I gained tremendous hands-on experience in material growth,
characterization, device fabrication, and testing among others. I also gained in-depth
knowledge and understanding of the LED technologies.
I would also like to thank Prof. Wu and Prof. Ravichandran for serving on my
defense committee, and Prof. Wu, Prof. Nakano, Prof. Povinelli and Prof. Cronin for
serving on my qualifying exam committee.
I would like to acknowledge my group members for their support during my
studies. I especially thank Yenting Lin for fully taking me under his wing during my first 2
years of Ph.D. studies. He gave me training on most of the equipment and was very
patient in providing hands-on training. He also had great insight into the growth of
nitride materials and I learned a lot from him both in theoretical understanding and
hands-on “tricks” for doing some of the important processes. I would like to
3
acknowledge the following former group members for interesting scientific discussion as
well as casual conversations and support: Lawrence Stewart, Yunchu Li, Suzana Sborlan,
Ting-Wei Yeh, Chunyung Chi, Yenting Lin, Maoqing Yao and Ashakan Seyedi. I would also
like to thank the current student, Mitchell Dreiske, for his contribution in managing the
reactor lab.
I would like to acknowledge the collaborators that contributed to the work in
this dissertation. I would like to thank Matthew Mecklenburg for always helping me with
TEM work and advises me and introduces me to other collaborators like Noah Bodzin at
UCLA who prepared a few FIB samples as well as engineers in FEI who prepared state of
the art FIB samples. I would like to thank prof. Ng for help in setting up the pulsed
integrating sphere measurement.
I would like to thank the following people for their support in various aspect of
my studies: Eliza and Jenny for help with administrative issues, Sean for servicing a
number of pumps, Guillermo for servicing the SEM.
Finally, I would like to thank my parents for all they have done and sacrificed for
raising me. From when I was young, they preached hard work and perseverance as the
key to success. I took it to heart and not matter how hard the life gets, they provided me
with the strength to always prevail. I would like to thank my wife for her emotional
support all these years and always help me put things in perspective and keep me a
great company.
4
Table of Contents
Acknowledgements.......................................................................................................................... 2
Abstract ............................................................................................................................................ 6
Chapter 1 Introduction .................................................................................................................... 7
1.1 Solid state lighting and motivation ........................................................................................ 7
1.2 Basic device structure and operations of a nitride based LED............................................. 14
1.3 Challenges of long wavelength emitting nitride based LEDs ............................................... 16
1.4 Brief overview of MOCVD .................................................................................................... 24
1.5 Flow dynamics of the CCS showerhead reactor .................................................................. 26
1.6 Thin film growth model........................................................................................................ 28
1.7 Thesis organization .............................................................................................................. 35
1.8 Chapter References .............................................................................................................. 36
Chapter 2 Nanostructure Growth .................................................................................................. 39
2.1 GaN nanostructure template growth .................................................................................. 39
2.2 Growth of nanostripes with smooth {11-22} facets ............................................................ 41
2.3 Nanostripes along different directions ................................................................................ 49
2.4 A-plane nanosheet structures .............................................................................................. 51
2.5 Chapter references ............................................................................................................. 55
Chapter 3 Growth and optimization of high In-content InGaN QW on nanostructure templates 58
3.1 Growth of InGaN QW on various nanostructures................................................................ 58
3.2 Optimization of InGaN QW growth on {10-11} the nanostripes ......................................... 65
3.3 Quantum efficiency measurement of the QW grown on the {11-22} nanostripes ............. 69
3.4 Chapter references .............................................................................................................. 74
Chapter 4 Optical and structural characterization of nanostripes MQW structures .................... 76
4.1 Cathodoluminescence characterization of nanostripe InGaN QW structures .................... 77
4.2 Transmission Electron Microscopy (TEM) characterizations ............................................... 83
4.3 Chapter references ............................................................................................................ 107
Chapter 5 Fabrication and characterization of nanoLEDs ........................................................... 108
5.1 Introduction to conventional LED structures and the challenges ..................................... 108
5.2 Development of {10-11} nanostripes based nanoLEDs ..................................................... 115
5.3 Characterization of nanoLEDs ............................................................................................ 130
5.4 External quantum efficiency (EQE) measurement ............................................................. 132
5.5 Extraction of IQE from the integrated electroluminescence ............................................. 138
5
5.6 Chapter References ............................................................................................................ 146
Chapter 6 Localized stacking faults in high In-content InGaN MQW nanostructures ................. 149
6.1 Optical Characterization of amber emitting nanostripes MQW structures ...................... 149
6.2 Structural characterizations of defects in amber emitting nanostripes MQW structures 153
6.3 Stacking fault generation mechanisms .............................................................................. 158
6.4 Chapter References ............................................................................................................ 164
Chapter 7 Conclusions and future studies ................................................................................... 169
7.1 Conclusions ........................................................................................................................ 169
7.2 Future Studies .................................................................................................................... 171
Appendix A ................................................................................................................................... 175
A.1 MOCVD growth system 1: H 2 purifier ................................................................................ 175
A.2 MOCVD reactor system 2: gas handling ............................................................................ 180
A.3 MOCVD system 3: reactor chamber .................................................................................. 186
A.4 References ......................................................................................................................... 190
6
Abstract
Efficient green emitting LEDs and monolithic white light emitting LEDs require
the extension of the range of efficient light emission in the GaN / InGaN materials
system. We demonstrate high efficiency green and yellow light emitting multiple
quantum well (MQW) structures grown on GaN nanostripe templates. The structures
show promise for realizing high-efficiency phosphor - free white LEDs. The nanostripe
dimensions range from 100 nm to 300 nm and have separations that range from 300 nm
to 1 micron. Such small lateral dimensions are chosen to promote the elimination of
threading dislocations from the structures. Nanostripes with various semipolar surfaces
are grown with selective area growth on e-beam lithography patterned c-plane GaN
where the mask openings are oriented between the [10-10] and [11-20] directions.
Photoluminescence (PL) measurement shows that MQWs grown on stripes with (10-11)
surfaces and triangular shape emit the longest peak wavelength and have the best
surface stability. Efficient PL emission peak wavelengths as long as 570 nm are realized
on the triangular stripes with (10-11) surfaces by optimizing the MQW growth
conditions for long wavelength emission. Power dependent PL measurement shows
linear response over more than three orders of magnitude of excitation power,
indicating high radiative recombination efficiency. LED structures that utilize MQWs
grown on nanostripes with (10-11) surfaces were fabricated to further demonstrate the
viability of the approach.
7
Chapter 1 Introduction
1.1 Solid state lighting and motivation
Lighting is an important part of modern society and has a significant economic
and environmental impact on people’s lives. A recent report from the United States (US)
Department of Energy (DoE) shows that lighting accounted for 18 % of total electricity
usage in the US in 2010 [1]. The lighting industry represents a 100 billion dollar global
market and is expected to expand 5 % annually [2]. One rapidly increasing player in this
enormous market is solid state lighting (SSL) which are based on light emitting diodes
(LEDs). Recently, LED lighting has surpassed fluorescent lighting in terms of luminous
efficacy [3], which is a measure of the power conversion efficiency from electricity to
light and how efficient the human eyes perceive the produced light. Incandescent and
fluorescent lighting have a luminous efficacy of around 16 and 70 lm/W respectively [4]
while average commercial LED lighting has a rating of 100 lm/W as of 2014 [3] and it has
the potential to increase to the theoretical limit of around 400 lm/W [1]. The increase in
efficiency and the expected rapid price drop of SSL mean significant savings in electricity
bills for consumers and are expected to drive the market share of SSL up to 60 % by
2025 [1]. As a result, total aggregate energy savings can be as large as 217 terawatt-
hours (TWh), which are equivalent to one-third of the current energy consumption for
lighting [1]. The significant energy savings can improve energy conservation and
generate fewer greenhouse gasses. Therefore, the advancement of SSL can lead to
savings for everyday consumers as well as a better environment and a better planet.
8
The luminous efficacy of SSL must continue to increase to achieve the potential
of SSL, and a deviation from the current approach may be necessary. Figure 1.1 (a)
shows different ways that LEDs can generate white light for SSL and figure 1.1 (b) shows
the spectrum of the white LEDs using the different approaches. Currently, the most
mature and common way to produce white light is blue LEDs with a yellow phosphor
coating, which absorb some of the blue light generated by the LED and convert it to
yellow light. The mixture of the transmitted blue light and converted yellow light gives
the perception of white light. However, due to fundamental Stoke’s shift energy loss in
phosphor down-conversion of blue to yellow light, this approach has a limitation in the
theoretically maximum efficiency, which is around 80%. Another drawback of this
approach is the poor color rendering and cold color temperature due to the lack of light
from the red component of the spectrum. It is possible to use green and red phosphor
coating instead to achieve better color rendering and temperature, but this approach
suffers from even more Stoke’s loss which further limits the attainable luminous
efficacy.
9
Figure 1.1 White light based on LEDs. (a) Three dominant ways to produce white light
based on LEDs. (b) Comparison of the spectrum of ideal sunlight with two LED-based
white light sources [5]
Therefore, an approach that does not utilize the phosphor coating can give rise
to greater efficiency, which can be achieved by mixing LEDs that emit different colors to
generate white light. For example, a mixture of efficient blue, green and red lights can
produce efficient white light, and if each LEDs can operate at a high efficiency, the
overall efficiency, as well as CRI, can surpass those that rely on the phosphor. Philips et
al. show the combination of monochromatic blue (463 nm), green (530 nm), yellow (570
nm) and red (614 nm) light can yield the most luminous efficacy of 408 lm/W while
achieving color rendering index of at least 90 [6]. It was made an emphasis in their study
that the wavelength of red light needs to be “shallow” because the eye response drops
off rapidly as the wavelength increases beyond 610 nm range.
10
Figure 1.2 Quantum efficiency (QE) of InGaN and AlInGaP LEDs as a function of emission
wavelength [7]
One of the major challenges in the multi-LED approach is the so-called green
gap, as illustrated in figure 1.2. State-of-the-art blue and red LEDs can achieve an
internal quantum efficiency of more than 50 %, but green and yellow LEDs are less than
10 %. As the figure 1.2 shows, GaN/InGaN MQW system produces efficient blue light,
but as the emission wavelength increases to the green region, the efficiency decreases.
A similar trend occurs with the AlInGaP system that emits red light efficiently. As the
wavelength becomes shorter into the yellow region, the efficiency decreases
dramatically. Therefore it is necessary to “close” this green gap to produce high-
efficiency RGB white LEDs.
11
Figure 1.3 Band alignment of AlInGaP as Al content increases from left (a) to right (b). (c)
The bandgap energy of the direct band gap and indirect bandgap as a function of alloy
composition. [8]
There is a fundamental limitation of the AlInGaP system to extend the efficient
range of light emission to shorter wavelength because the bandgap of the materials
becomes indirect as the emission wavelength decreases. Figure 1.3 (a) and (b) shows
12
the schematics of the band structures as Al content in the alloy increases from low to
high . Figure 1.3 (b) shows the energy of the direct band gap and indirect bandgap as a
function of the alloy composition of Al. As the Al to Ga ratio surpass 53%, which
corresponding to an emission wavelength of 555 nm, the indirect band gap becomes
smaller than the direct bandgaps. Due to the lower energy of the conduction band that
corresponds to the indirect band gap, the electrons will occupy those states first.
Consequently, the radiative recombination rate decreases significantly for this material
system to emit in the green range.
Figure 1. 4 Bandgap energy versus lattice constant of III-Nitride semiconductor at room
temperature [8]
13
On the other hand, InGaN has a direct bandgap for all In compositions, and it
should, in principle, emit all visible wavelengths efficiently. Figure 1.4 shows the
bandgap energy as a function of the lattice constant of the III-N semiconductor at room
temperature. GaN and InN have a bandgap of around 3.4 eV and 0.7 eV, respectively,
and corresponds to an emission in the UV and infra-red range respectively. However,
there are tremendous challenges to increase the quantum efficiency of the high In-
content InGaN material). There are two major fundamental problems associated with
green gap problem for the nitride system: one is the adverse effect of piezoelectric
fields in InGaN heterostructures and the other is the deterioration of the quality of the
InGaN material as the In composition increases. We will discuss these problems in the
following. However, before we discuss these advanced issues, a basic understanding of
the nitride-based LED structures and their operations are necessary. In the next section,
we first discuss the basic concepts of semiconductor and pn junction. Readers who are
familiar with these concepts should skip to the next section where basic concepts of
nitride based LEDs are discussed.
14
1.2 Basic device structure and operations of a nitride based LED
Figure 1.5 (a) Schematics of the cross-section of nitride based LED device structure. (b)
Schematics of the working of a nitride based pn-junction LED under forward bias (c)
Band diagram of a nitride based pn junction LED under forward bias
In this section, we discuss the basic device structure and operation of a nitride
based LED structure. Figure 1.5 (a) shows the schematics of the cross-section of nitride
based LED device structure. The structure consists of n and p-type GaN as well as a new
layer called InGaN QW layer that is sandwiched between n and p-type GaN. A QW is a
thin layer of material with a thickness in the range of 3 to 10 nm. The bandgap energy of
this layer, InGaN, is smaller than that of the cladding layer material, which is GaN in this
case. The QW has two important functions, one is the confinement of carriers and
another is the determination of the color or the wavelength of the light emission.
Figure 1.5 (b) shows the schematics of the working of a nitride based pn-junction
LED under forward bias. Similar to the silicon pn junction under forward bias discussed
in the previous section, the LEDs are also operated under forward bias that creates
15
excess diffusion for electron and holes to the opposite side. However, one important
difference is that the existence of the QW intercepts the electrons and holes in the
diffusion process and confines the carriers in the quantum well. This is best illustrated in
the band diagram of Figure 1.5 (c). Because the energy of the QW is lower than the
cladding layer for both electrons and holes, once these carriers are captured by the QW,
they relax quickly to the band edge of the InGaN QW. These carriers will then need to
gain significant energy to escape the well, thus making this process unlikely. The high
concentration of electrons and holes confined in the QW leads to large localized
recombination and emission of the light. We will show in chapter 3 that the larger the
concentration of carriers, the more efficient the light emission is, to some extent.
Therefore, the confinement of carriers leads to efficient light emitting device.
In addition, most LEDs utilized what’s called a multiple quantum well (MQW)
structure, where 2 or more QW are placed one after another. MQWs have 2 advantages
over a single QW. Firstly, if the carriers are not captured by the 1
st
QW, then it may be
captured by the 2
nd
or the 3
rd
QW, therefore increasing the probabilities that the
carriers recombine only in the InGaN layer. Another advantage, which will be discussed
more in chapter 5 is that MQW allows the carrier concentration to be lower to achieve
higher light emission efficiency under high injection. We will discuss this issue in more
details in chapter 5.
The wavelength of the light emission depends on the bandgap of the InGaN
material. The effect of the bandgap of the InGaN material in the wavelength emission is
plotted in figure 1.4. Therefore, the emission color of the LED can be tuned by varying
16
the In content of the well. Because InGaN covers emission from UV to infrared, it is the
most promising material for realizing multi-color LEDs.
1.3 Challenges of long wavelength emitting nitride based LEDs
Figure 1.6 (a) Schematics of crystal structure of GaN (b) Schematics of the conceptual
understanding of the origin of the piezoelectric field.
With the basic concepts of the nitride LEDs discussed, we can continue our
discussion for challenges for nitride LEDs to emit long wavelength light. The first
problem associated with the nitride-based LEDs is the adverse effect of piezoelectric
field. Figure 1.6 shows (a) shows the schematics of the GaN crystal structure and (b)
shows the schematics of the origin of the piezoelectric field. GaN has a wurtzite crystal
structure, and the material and device structures are usually grown on the c-plane. The
c-plane has a stacking sequence of … ABABAB… with successive rows of Ga and N with
tetrahedral bonding. Due to the large difference in the electronegativity of Ga and N,
the Ga-N bonding has strong ionicity which leads to partially positively charged N and
17
partially positively charged Ga. Depends on the exact distribution of electrons for the
tetrahedral bonding structure, which requires detailed simulation, there will be a net
polarization that points in the c direction. This naturally occurring polarization is called
spontaneous polarization. When the InGaN is grown on GaN, as it is for heterostructure
devices, the InGaN lattice will deform to accommodate the lattice mismatch strain
between the layers, the deformation of the crystal bonding further alters the electron
distribution altering the total polarization field. This added polarization is called
piezoelectric polarization. The piezoelectric field can be written by:
P
𝑃𝑍
=(
0
0
𝑒 31
(𝘀 𝑥𝑥
+𝘀 𝑦𝑦
)+𝑒 33
𝘀 𝑧𝑧
)
Where 𝘀 𝑥𝑥
, 𝘀 𝑦𝑦
, 𝘀 𝑧𝑧
are the normal strain in the 3 principle directions and 𝑒 31
and 𝑒 33
are piezoelectric coefficients.
The dipoles of the spontaneous and piezoelectric polarization cancel along the c
directions except at the GaN/InGaN interface. The discontinuity of the polarization
induces surface charges on both GaN/InGaN interfaces. The surface charges can be
written as:
𝜎 𝜋 =(𝑃 InGaN
−𝑃 GaN
)∙𝑛
Where 𝑃 InGaN
and 𝑃 GaN
are total polarizations of the two materials and 𝑛 is the normal
direction of the interface plane. Because the spontaneous polarization of InGaN and
GaN are very similar [9], the surface charges are mostly induced by the piezoelectric
18
polarizations. Finally the surface charges induce electric field across the thin InGaN layer
that is thin enough to be a quantum well (QW), called the piezoelectric field.
Figure 1.7 Band alignment of an InGaN/GaN quantum well structure for polar (left) and
nonpolar/semipolar planes (right) [10]
Figure 1.7 shows the adverse effect of this piezoelectric field on the radiative
efficiency of the QWs. The piezoelectric field separates the electron and hole wave
function in the QWs, which decreases the overlap integral of electron and hole wave
functions and leads to reductions in the radiative recombination rate. As the emission
wavelength is required to be longer for green, yellow and red LEDs, the In composition
of the QW must be larger which further increases the piezoelectric field and further
reduces the radiative recombination rate. These phenomena pose a fundamental limit
on the efficiency of long wavelength light emitting QWs grown on c-plane GaN.
19
Due to the problems associated with the piezoelectric field in QWs grown on the
polar surface, research has been focusing on growing MQWs on non-polar and semi-
polar GaN surfaces which result in zero and significantly reduced piezo-electric field,
respectively [9]. Figure 1.7 shows the band diagram of the QWs grown on polar and
non-polar planes. For MQWs grown on non-polar surfaces, the piezoelectric field
becomes zero because there is no resolved field component in the growth direction in
the nonpolar case and thus there is no separation of electron and hole wavefunctions.
Figure 1.7 also shows various low index semipolar planes which can lead to
significantly reduced the piezoelectric field. A. E. Romanov et al. calculated the
piezoelectric field for InGaN QWs grown on these semi-polar planes as a function of the
inclination angles from the c-plane, shown in figure 1.8 [9]. When the inclination angle is
greater than 45 degrees, the piezoelectric field can be reduced to smaller than one-third
of the polar case.
Figure 1.8 Dependence of Piezoelectric field in InGaN grown on semipolar planes on the
inclination angle of the semi-polar plane with respect to the c-plane for various In
composition of (1) 5%, (2) 10 %, (3) 15 %, (4) 20 %. [9]
20
Another challenge of growing high In content QWs is that low growth
temperature is required to incorporate sufficient In in the QW region due to the high
volatility of In in the growth process. However, the NH 3 decomposition rate is
significantly reduced at low growth temperatures which lead to inferior crystalline
quality [11]. Studies have shown that some semipolar crystal planes have higher indium
incorporation rate than others when grown at the same temperature [12]. Therefore,
high In content QWs with higher crystal quality can be grown on the planes with higher
indium incorporation rate because they can be grown at a higher temperature [12]. For
example, figure 1.9 shows the In composition of InGaN grown on various semipolar and
nonpolar planes as a function of the growth temperature. On the m-plane, InGaN with
10 % indium is grown at 840 degrees. By comparison, on the (20-2-1) plane, InGaN with
the same 10 % indium can be grown at 880 degrees. The fact that the InGaN can grow
on the (20-2-1) plane at a higher temperature than other planes for the same In content
means that the InGaN grown on this plane likely has the highest material quality.
Therefore there is an interest to survey different crystal planes, including nonpolar and
semipolar planes, and find the plane with the highest In incorporation rate.
21
Figure 1.9 In composition of InGaN grown on various semipolar and nonpolar planes as a
function of the growth temperature [13]
Therefore, research has been focusing on the growth of nonpolar and semipolar
plane GaN as a template for growth of High In content QWs. Figure 1.10 summarizes the
several ways to obtain nonpolar and semipolar GaN. Currently, most states of the art
studies on semi-polar plane GaN utilize substrates obtained from slicing millimeter thick
bulk GaN in desired orientations from thick epitaxial layers grown on Al 2O 3. Single
crystalline bulk GaN cannot be grown from the melt like other semiconductors because
of its high covalency that forces the material to decompose rather than melt at high
temperature [14]. Bulk GaN can be grown by hydride vapor phase epitaxy (HVPE) on
sapphire with high growth rate, but the high strain results in bowing of the wafer and
may lead to cracking and breaking of the wafer if it is too thick [14]. As a result, the
22
dimension of the non-polar and semi-polar planes are limited to a few millimeters on a
side, which is not practical for mass production. Nonpolar and semipolar plane GaN can
be grown on large and cost effective substrate like non-basal plane sapphire and silicon,
but these suffer from the high density of dislocations [15].
Figure 1.10 Summary of different ways to grow semipolar and nonpolar GaN substrates
Another way to produce semi-polar plane is to use selective area epitaxy (SAE)
on patterned c-plane GaN. A thin amorphous “mask” layer, typically SiN or SiO2 layer is
deposited on the c-plane GaN and periodic openings are generated with
photolithography and etching. When the patterned GaN is loaded back into the MOCVD
reactor for regrowth, the material will only grow in the opening regions, and gas sources
that impinge on the mask region diffuse to the open region and contribute to the
23
growth there. The growth front of emerging structure from the opening is kinetically
limited and usually confined by the slow growing inclined planes that are around 60
degrees from the c-plane. Early work on this kind of structure is limited to micron-size
openings. There are considerable threading dislocations, generated at GaN/Sapphire
interface, that propagate to the GaN surface and penetrate from the opened regions to
the semi-polar surfaces. A study shows that these threading dislocations can be the
origin for misfit dislocations generation in InGaN thin film grown on the semipolar
planes [16] and they may adversely affect the radiative efficiency.
A study shows that when the mask openings are decreased to the nanoscale, the
dislocation screening effect can increase dramatically [17]. The TEM study of over 3000
nanorods grown with less than 100 nm diameter openings shows that they are entirely
free of dislocations. Essentially all threading dislocations are filtered by the nano masks.
Therefore, it is critical to scale down the SAE to nano-scale dimensions to significantly
reduce the threading dislocations that penetrate to the semi-polar planes surfaces.
Based on these considerations, the goal of this study is to explore the possibility
of using non-polar and semi-polar planes GaN obtained from nano-scale SAE as a
template for growing high efficiency green, yellow and red emitting LEDs and provide
the insight for the possibility of monolithic integration of white LEDs.
24
1.4 Brief overview of MOCVD
So far we have discussed the challenges of obtained efficient long wavelength
emitting nitride LEDs and how we are going to solve the problems. However, before we
dwell into details of this dissertation, we would like to give a brief overview of the
MOCVD technology so the readers will be more familiar with the terminology during in-
depth discussions.
Metal organometallic chemical vapor deposition (MOCVD) is a chemical vapor
deposition process that produces epitaxial growth of single crystalline thin films on a
single crystalline substrate to make various optoelectronic devices such as LEDs and
Lasers. For GaN growth specifically, Ga-containing metal organic compound, usually
trimethylgallium, and NH 3 are transported to the reactor chamber with H 2 or N 2 carrier
gas. These precursors diffuse to the substrate surface, which is heated up to around
1000 degrees, while undergoing series of gas phase and surface reactions, finally deposit
on the surface of substrate forming single crystalline materials. Therefore, the MOCVD
system is composed of 2 major components, one is the gas handling system and another
is the reactor chamber system. The gas handling system selects which MO sources are
supplied to the reactor and control their flow rate. The various gasses are delivered to
the reactor chamber for material growth and the unreacted gasses and the byproducts
are removed from the reactor to the exhaust. Readers are referred to appendix A to
study the architecture of our MOCVD reactor.
25
Compared to another popular thin film deposition technique, molecular beam
epitaxy, MOCVD technology has distinct advantages in terms of scalability and
technology transfer to large scale production. MBE deposition requires ultra-high
vacuum reactor chamber, which is very costly to scale up. MOCVD reactors are subject
to the moderate pressure of 10 to 760 Torr, which can be easily obtained even for large
reactors. Therefore, production scale MOCVD reactor can achieve growth on as many as
56, 2-inch wafers in one growth chamber [18]. Even with these large reactor
configurations, a use of vertical flow reactor ensures uniform material depositions
throughout the whole wafer carrier surface. In the following section, we describe in
more details the growth process of MOCVD.
26
1.5 Flow dynamics of the CCS showerhead reactor
Figure 1.11 flow pattern of the CCS showerhead reactor [19]
Figure 1.11 shows the flow pattern of the gaseous species in the closed coupled
showerhead (CCS) reactor chamber, which is a vertical flow reactor configuration [19].
Readers are encouraged to study appendix A for more information. This vertical flow
reactor design is advantageous because the deposition rate can be uniform across the
wafer. Both the horizontal and vertical component of the flow velocities decrease to
zero at the substrate surface to satisfy the no-slip condition [19]. This “plane-stagnation
point” flow dynamics can be solved exactly, and an important result is that the
boundary layer thickness is independent of the position. The boundary layer is defined
27
to be the layer from the substrate surface where the gas flow velocity is zero to a
distance from the substrate where the gas velocity is close to the free stream velocity.
Because the concentration boundary layer thickness is directly related to the velocity
boundary layer, it is also independent of the position [19]. Therefore, in a mass
transport limited regime, the growth rate can be uniform across the wafer.
Figure 1.12 MOCVD growth processes [18]
Figure 1.12 shows the schematics of the MOCVD growth processes. The gaseous
species are diffused to the surface due to the concentration gradient discussed earlier.
As the gas species diffuse closed to the surface, the temperature also increases and
leads to homogeneous pyrolysis. For group III containing compounds like TMG, they
undergo the following reactions at a temperature above 550 C [20]:
𝐺𝑎 (𝐶 𝐻 3
)
3
→ 𝐺𝑎𝐶 𝐻 3
+2(𝐶 𝐻 3
)
28
where two methyl groups are liberated from the TMG. The bond strength of NH 3 is
large, and it does not undergo homogeneous pyrolysis at a typical growth temperature
of GaN. Then the Ga source and N source undergoes heterogeneous pyrolysis as they
are adsorbed on the surface:
𝐺𝑎 (𝐶 𝐻 3
)
3
→ 𝐺𝑎 + 𝐶 𝐻 3
2𝑁 𝐻 3
→ 2𝑁 +3𝐻 2
The adsorbed N and Ga atoms can diffuse to a growth step and be incorporated, or they
can desorb and evacuate from the growth surfaces.
1.6 Thin film growth model
So far we gave an overview of the MOCVD reactor systems and described the
growth processes qualitatively. The actual growth processes are complex with
numerous reactions both in the gas phase and on the surface that happens
simultaneously. We will get into a more detailed discussion in the specific chapters to
discuss these details if applicable. In the following, we apply the conceptual
understanding of MOCVD growth to a more quantitative description of the thin film
growth of GaN. This thin film model predicts the growth rate relatively well and gives
readers a taste of how the precursors diffuse and react and grow on the substrate.
Maxwell-Boltzmann distribution, which gives the velocity distribution of gas
particles, in 1D is
29
𝑓 𝑣 (𝑣 )𝑑𝑣 =√
𝑚 2𝜋 𝑘 𝐵 𝑇 𝑒 −
𝑚 𝑣 2
2𝑘 𝐵 𝑇 𝑑𝑣
The average velocity of gas particles in 1 D is
〈𝑣 〉=∫ 𝑣 𝑓 𝑣 (𝑣 )𝑑𝑣 ∞
−∞
=√
𝑚 2𝜋 𝑘 𝐵 𝑇 ∫ 𝑣 𝑒 −
𝑚 𝑣 2
2𝑘 𝐵 𝑇 𝑑𝑣 ∞
−∞
=
√
𝑘 𝐵 𝑇 2𝜋𝑚
=
√
𝑅𝑇
2𝜋𝑊
=3638
√
𝑇 𝑊 (𝑐𝑚 /𝑠 )
Where W is the molecular weight of the precursor. Therefore this expression gives the
average rate at which precursors arrive at the substrate surface through diffusion. The
assumption in the derivation was that the diffusion only occurs in the vertical direction.
Then the first order heterogeneous reaction rate k r can be given by
𝑘 𝑟 = 𝛾 〈𝑣 〉=3638𝛾 √
𝑇 𝑊 (𝑐𝑚 /𝑠 )
where 𝛾 is the sticking coefficient which is the reaction probability of group III
precursors at the surface, which is assumed to be 1. For the growth temperature of
1155 C, the reaction rate for TMGa is 12829 cm/s.
The continuity of masses states that the flux of precursors that arrive on the
surface must equal the rate of destruction by the heterogeneous reaction,
𝑘 𝑟 𝑋 0
=𝐷 [
𝑑𝑋 𝑑𝑧 ]
𝑧 =0
30
Where X 0 is the mole fraction of the precursor at the surface z=0. Assume a linear
variation of the precursor concentration, we get
𝑘 𝑟 𝑋 0
=
𝐷 (𝑋 ∞
−𝑋 0
)
𝛿 𝑐
𝑋 ∞
is the mole fraction of the precursor far away from the surface. The mole fraction of
the precursors in the ambient can be calculated as follows. Figure 1.13 shows the
schematics of the flow rates of various precursors during the GaN thin film growth. Note
that MO sources are usually in the liquid form, and they are stored in a stainless steel
vessel called bubblers. There is an inlet tube that is submerged underneath the liquid
level and the outlet tube is above the liquid level. A carrier gas, either H 2 or N 2 is
supplied from the inlet, and it creates “bubbles” in the vessel. The vapor of the MO
sources are carried out of the outlet tube, and eventually flow to the reactor. The
bubblers is submerged in a cooler bath to control the vapor pressure of the MO sources.
The mole fraction of the MO source in the reactor can be calculated as follows We take
the TMGa flow as a example
Flow rate of TMGa through the bubblers are
𝐹 𝑇𝑀𝐺𝑎 =
𝑝 𝑇𝑀𝐺𝑎 𝑝 𝐻 2
𝐹 𝐻 2
=
𝑝 𝑇𝑀𝐺𝑎 𝑝 𝑏𝑢𝑏𝑏𝑙𝑒𝑟 −𝑝 𝑇𝑀𝐺𝑎 𝐹 𝐻 2
=
56.5
1234−56.5
32=1.54 𝑠𝑐𝑐𝑚
31
Where 𝑝 𝑇𝑀𝐺𝑎 is the vapor pressure of TMGa, which can be found in a standard MO
source vapor pressure chart with a cooler bath temperature of -2.5 C.
Mole fraction of TMGa in the reactor is therefore
𝑋 𝑇𝑀𝐺𝑎 =
𝐹 𝑇𝑀𝐺𝑎 𝐹 𝑡𝑜𝑡𝑎𝑙 =2.65×10
−4
Do the same for NH 3 and Si 2H 6
𝑋 𝑇𝑀𝐺𝑎 𝑋 𝑁𝐻 3
𝑋 𝑆𝑖 2𝐻 6
𝑋 𝐻 2
2.5×10
−4
0.41 3.4×10
−5
0.59
Figure 1.13 Schematics of the flow rates of various precursors during GaN growth
32
𝛿 𝑐 , the concentration boundary layer can be calculated from hydrodynamic
treatment [21],
𝛿 𝑐 ~
√
𝐷 ∞
𝜔
𝐷 ∞
is the diffusion coefficient of the Group III precursor at room temperature and 𝜔 is
the disk spin rate. 𝛿 𝑐 is solved numerically to be 1.0 cm for TMGa at 𝜔 =1000 𝑟𝑝𝑚 [21]
and 0.7 cm for Ga at 1200rpm [22]. From the scaling relation, we plug in our spin rate
50rpm and get 𝛿 𝑐 =4.5 𝑐𝑚 for TMGa and 𝛿 𝑐 =3.4 𝑐𝑚 for Ga. Then we can take an
average to be our boundary layer thickness
𝛿 𝑐 =4 𝑐𝑚
This value only depends on the disk spin rate and the reactant; therefore this value can
be treated as a constant.
Now we need to know the diffusion coefficient in the continuity equation. The
diffusion coefficient is a strong function of pressure and temperature.
𝐷 =𝐷 𝑟𝑒𝑓
(
𝑝 𝑟𝑒𝑓 𝑝 )(
𝑇 𝑇 𝑅𝑒𝑓 )
1.7
D ref is the diffusion coefficient of the Group III precursor (0.39 and 0.40 cm
2
/s for TMGa
and TMAl respectively [21] ) at p ref = 760 Torr and T ref = 300k . We know the pressure is
200 Torr everywhere in the reaction chamber, but the temperature varies with the
distance from the susceptor and so will be the diffusion coefficient. One approximation
is to take the diffusion coefficient at T(𝛿 𝑐 /2) as the effective diffusion coefficient for the
33
continuity equation [22]. Therefore we need to know the temperature variation. The
temperature boundary thickness scale with
𝛿 𝑇 ~√
𝑘 ∞
𝜌 ∞
𝑐 𝑝 ∞
𝜔
where 𝜌 ∞
, 𝑐 𝑝 ∞
, and 𝑘 ∞
, are the mixture density, specific heat, thermal conductivity and
reactant diffusivity (taken to be that of the group III precursor), respectively, evaluated
at the inlet.
The temperature boundary thickness is numerically determined to be 1.32 cm
[21], for the case where the total pressure is 140 torr, inlet mole fractions for hydrogen
and ammonia are 0.4868 and 0.5111 and a disk temperature of 1323 K. For our case,
the mole fractions are 0.41 and 0.59 for ammonia and hydrogen respectively.
The transport data for these two gasses are found on the website
1
and they are
summarized in the table below (200Torr, 300K). Mixture properties are calculated for
[22] and our case from the mole fractions.
𝜌 (kg/cm
3
) 𝑘 ∞
(W/(m K)) 𝑐 𝑝 ∞
(kJ/(kg K))
NH 3 0.189 0.022 2.19
H 2 0.021 0.168 14.32
Mixture (41:59) 0.090 0.108 9.35
Mixture (49:51) 0.103 0.096 8.38
1
http://www.engineeringtoolbox.com
34
Therefore, the temperature boundary thickness for our case is
𝛿 𝑇 =7 𝑐𝑚
Therefore
𝑇 (
𝛿 𝑐 2
)=1098 𝐾
Therefore,
𝐷 =𝐷 𝑟𝑒𝑓
(
𝑝 𝑟𝑒𝑓 𝑝 )(
𝑇 𝑇 𝑅𝑒𝑓 )
1.7
=13.5 𝑐 𝑚 2
/𝑠
Now, from the continuity equation, we can calculate the X 0
𝑋 0
=
1
1+
𝛿 𝑐 𝑘 𝑟 𝐷 𝑋 ∞
=6.9𝑒 −8
The growth rate can by calculated by [21]
𝐺 =(
𝑊 𝑠𝑜𝑙𝑖 𝑑 𝜌 )𝑘 𝑟 𝑋 0
𝑐
Where W solid is the molecular weight (83.79 g/mole for GaN), 𝜌 is the solid density (6.15
g/cm), c is the total (z-independent) concentration of species (in moles/cm
3
), which
obeys the ideal gas-law scaling relation
𝑐 =𝑐 𝑟𝑒𝑓
(
𝑝 𝑝 𝑟𝑒𝑓
)(
𝑇 𝑟𝑒𝑓 𝑇 )
Where c ref is (1 mole)/(22400 cm
3
).
35
Therefore the growth rate is calculated to be
𝐺 =3 𝐴 /𝑠
The actual growth rate is 2 microns/hr which is around 5.5A/s. The calculation is
surprisingly close to the actual growth rate considering the simplicity of the model and
the approximations made.
1.7 Thesis organization
This thesis is organized as follows. Chapter 2 provides details of the growth of
nanostructures with various semipolar and nonpolar planes. Specifically, we looked at
the nanostripes with the {10-11} and {11-22} facets and nanosheet with {11-20} facets.
The growth mechanism of these structures was explained in terms of the surface
reconstruction and surface kinetics. In chapter 3, we compare these nanostructures in
terms of their indium incorporation rate and selected the nanostripes with the {10-11}
nanostripes as the most promising candidate for further growth optimization of the QW
and device structures. In chapter 4, we provide a detailed analysis of structural and
optical properties of the QW structures grown on the nanostripes with the {10-11}
facets using nanoscale and atomic scale characterizations. In Chapter 5, we developed
the growth and fabrication processes of the LED structures and demonstrated amber,
yellow and green emitting nanoLEDs. Their optoelectronic properties are studied. In
Chapter 6, we investigated and identified the cause for reduced luminescence
36
efficiencies of amber emitting nanoLEDs. In chapter 7, we conclude the thesis and
provide opportunities for future studies.
1.8 Chapter References
[1] N. Bardsley, S. Bland, L. Pattison, M. Pattison, K. Stober, F. Welsh and M. Yamada,
"Solid-State Lighting Research and Development Multi-Year Program Plan," U.S.
Department of Energy, 2014.
[2] T. Baumgartner, F. Wunderlich, D. Wee and A. Jaunich, "Lighting the way:
Perspectives on the global lighting market," McKinsey&Company, 2012.
[3] O. Comstock, "LED light bulbs keep improving in efficiency and quality," U.S.
Energy Information Administration, 4 11 2014. [Online]. Available:
http://www.eia.gov/todayinenergy/detail.cfm?id=18671.
[4] J. R. Brodrick, "U.S. Lighting Market Characterization Volume I: National Lighting
Inventory and Energy Consumption Estimate," Navigant Consulting, Inc., 2002.
[5] S. Pimputkar, J. S. Speck, S. P. DenBaars and S. Nakamura, "Prospects for LED
lighting," Nature Photonics, vol. 3, pp. 180 - 182, 2009.
[6] J. M. Phillips, M. E. Coltrin, M. H. Crawford, A. J. Fischer, M. R. Krames, R. Mueller-
Mach, G. O. Mueller, Y. Ohno, L. E. S. Rohwer, J. A. Simmons and J. Y. Tsao,
"Research challenges to ultra-efficient inorganic solid-state lighting," Laser &
Photonics reviews, vol. 1, no. 4, pp. 307-333, 2007.
[7] A. Khan, "Semiconductor photonics: Laser diodes go green," Nature Photonics,
vol. 3, pp. 432-434, 2009.
[8] E. F. Schubert, T. Gessmann and J. k. Kim, Light emitting diodes, John Wiley &sons,
Inc., 2005.
[9] A. E. Romanov, T. J. Baker, S. Nakamura and J. S. Speck, "Strain-induced
polarization in wurtzite III-nitride semipolar layers," JOURNAL OF APPLIED
PHYSICS, vol. 100, p. 023522.
37
[10] K. Yamane, "Scaling Semi -Polar Substrates," 22 4 2014. [Online]. Available:
http://www.compoundsemiconductor.net/article/91897-scaling-semi-polar-
substrates.html.
[11] W. Van der Stricht, I. Moerman, P. Demeester, L. Considine, E. Thrush and J.
Crawley, "MOVPE growth optimization of high quality InGaN lms," MRS Internet
Journal of Nitride Semiconductor Research, vol. 2, no. 16, 1997.
[12] Y. Zhao, Q. Yan, C.-Y. Huang, S.-C. Huang, P. S. Hsu, S. Tanaka, C.-C. Pan, Y.
Kawaguichi, K. Fujito, C. G. Van de Walle, J. S. Speck, S. P. DenBaars, S. Nakamura
and D. Feezell, "Indium incorporation and emission properties of nonpolar and
semipolar InGaN quantum wells," Applied Physics Letters , vol. 100, p. 201108,
2012.
[13] Y. Zhao, Q. Yan, C. Huang, S. Huang, P. Hsu, S. Tanaka, C. Pan, Y. Kawaguchi, K.
Fujito, C. V. d. Walle, J. Speck, S. DenBaars, S. Nakamura and D. Feezell, "Indium
incorporation and emission properties of nonpolar and semipolar InGaN quantum
wells," Applied Physics Letters, vol. 100, p. 201108, 2012.
[14] V. Avrutin, D. J. Silversmith, Y. Mori, F. Kawamura, Y. Kitaoka and H. Morkoc,
"Growth of Bulk GaN and AlN: Progress and Challenges," Proceedings of the IEEE,
vol. 98, no. 7, 2010.
[15] Q. Sun, B. H. Kong, C. D. Yerino, T.-S. Ko, B. Leung, H. K. Cho and J. Han,
"Morphological and microstructural evolution in the two-step growth of nonpolar
a-plane GaN on r-plane sapphire.," Journal of Applied Physics, vol. 106, no. 12, p.
123519, 2009.
[16] E. C. Young, F. Wu, A. E. Romanov, A. Tyagi, C. S. Gallinat, S. P. DenBaars, S.
Nakamura and J. S. Speck, "Lattice Tilt and Misfit Dislocations in (11-22) semipolar
GaN heteroepitaxy," Applied Physics Express, vol. 3, p. 011004, 2010.
[17] R. Colby, Z. Liang, I. H. Wildeson and D. A. Ewoldt, "Dislocation Filtering in GaN
Nanostructures," Nano letters, vol. 10, pp. 1568-1573, 2010.
[18] "Close Coupled Showerhead® 3x2 inch · 6x2 inch," Aixtron, [Online]. Available:
http://www.aixtron.com/fileadmin/documents/matrix/r_d_systems/CCS_3x2_6x
2_inch.pdf..
[19] Z. C. Feng, in III-nitride: semiconductor materials, Imperial College Press, 2006, p.
76.
38
[20] G. B. Stringfellow, "Organometallic vapor-phase epitaxy: theory and practice,"
Academic Press, 1999, p. 232.
[21] W. Breiland, M. E. Coltrin, J. R. Greighton, H. Q. Hou, H. K. Moffat and J. Y. Tsao,
"Organometallic vapor phase epitaxy," Materials Science and Engineering, vol. R:
Reports 24, pp. 241-274, 1999.
[22] M. E. Coltrin and C. C. Mitchell, "Mass transport and kinetic limitations in MOCVD
selective-area growth".
[23] T. Wernicke, L. Schade, C. Netzel, J. Rass, V. Hoffmann, S. Ploch, A. Knauer, M.
Weyers, U. Schwarz and M. Kneissl, "Indium incorporation and emission
wavelength of polar, nonpolar and semipolar InGaN quantum wells,"
Semiconductor science and techonolgy, vol. 27, p. 024014, 2012.
[24] D. Neamen, An introduction to semiconductor devices, New York, NY: McGraw-
Hill, Inc, 2006.
39
Chapter 2 Nanostructure Growth
2.1 GaN nanostructure template growth
Figure 2.1 shows the fabrication and growth processes of the nanostructure templates
Figure 2.1 shows the schematics of the fabrication and the growth processes of
the GaN nanostructure templates. The c-plane GaN substrate is grown on a sapphire
substrate in a close-coupled showerhead (CCS) reactor to a thickness of 5 microns.
These substrates exhibit nearly featureless surfaces and the dislocation densities have
been measured previously in our group to be in the10
9
/cm
2
range. Approximately 20
nm of SiN is subsequently deposited on the GaN layers with plasma enhanced chemical
vapor deposition (PECVD). The built-in recipe is used for the PECVD deposition. Electron
40
beam lithography is used to define a periodic array of stripe openings with
approximately 100 nm width and pitch ranging from 300 to 1000 nm. The PMMA
patterns are transferred to the underlying SiN by using CF 4 Reactive ion etching The RIE
etching condition is 100 W and 100 mTorr of pressure. The etch rate is around 2 nm/s.
The residual PMMA was stripped by first sonicating the sample in Acetone for 10 min.
The sample is further cleaned by oxygen asher for 10 min under 150 mTorr and 100 W
for the plasma setting. The sample is then cleaned in 1 M KOH solution at 60 C for 10
min. After the cleaning process, the sample is first transferred to a beaker with DI water
to rinse off the KOH and then transferred to a beaker with IPA. Finally, the patterns are
dried with N 2 gun. The patterned substrate is loaded into the same MOCVD reactor for
nanostructure growth. The shape of the structures can be controlled to have inclined
and intersecting walls or nearly vertical walls by varying the MOCVD growth conditions
[1]. The crystallographic planes of the surfaces of the nanostructures depend on the
shapes of the structures and the in-plane orientations of the stripe patterns. For stripes
patterns that are oriented in the [11-20] direction, triangular prism structures with {10-
11} surface emerge under a wide range of growth conditions. For the stripe patterns
that are oriented in the [1-100] direction, triangular prism structures with {11-22}
surfaces emerge at a low growth temperature in the 900 ˚C range and rectangular prism
structure with {11-20} and (0001) surfaces emerge at a high growth temperature of
more than 1000 ˚C. These structures are illustrated in the step (5) of figure 2.1.
In general, the growth behaviors of the nanostructures are similar to those of the
microstructures, which have been studied extensively for epitaxial lateral overgrowth
41
(ELO) [2] [3]. In comparison, we intend to use the various surfaces expressed by the
structures as templates for QW and LED structure growth, thus placing a premium on
the surface quality of these facets. While structures with {10-11} and {11-20} facets are
shown to have smooth surface morphology, the growth of structures with {11-22}
surfaces are difficult to control and usually leads to rough surface morphology [3] [4].
Here we demonstrate and explain the growth conditions necessary for smooth {11-22}
surfaces. In this work, the emphasis is placed on achieving high vertical to horizontal
aspect ratio in the structures rather than on achieving small aspect ratios and rapid
coalescence as is the case in ELO studies that employ non-polar A-planes. Here, we
demonstrate and explain the growth conditions necessary to achieve high aspect ratio
vertical a-plane nanosheet structures.
2.2 Growth of nanostripes with smooth {11-22} facets
Figure 2.2 top down SEM images of (a) nano-pyramid (b) {10-11} nanostripes (c) {11-22}
nanostripes
First, we develop the growth conditions necessary for the growth of nanostripes
with smooth {11-22} surfaces. For systematic comparison, nanostripes with {10-11}
(a)
(b) (c)
42
surfaces grown on patterned substrates with stripe oriented in the [11-20] direction as
well as those with circular patterns are studied concurrently. The growth temperature is
maintained below 1000 ˚C to ensure the formation of a triangular prism structure while
the growth is varied over a wide range of conditions. For a growth mask with periodic
circular patterns, nano-pyramid structures with the {10-11} surfaces emerge under
normal growth conditions as shown in Fig. 2.2 (a). Nanostripe structures with the same
crystallographic planes can be achieved by using line patterns that are oriented in the
<11-20> direction, as shown in Fig.2.2 (b). An advantage of using line patterns is that
nanostructures with different planes can be achieved by orienting the pattern to
different directions. For example, if we orient the pattern to the <10-10> direction,
nanostructures with the {11-22} surfaces can be achieved, as shown in Fig.2.2 (c). Figure
2.3 shows a schematic drawing of the different low index semi-polar and non-polar
planes. The {11-22} planes are inclined planes between c-plane and a-plane. Likewise,
the {10-11} planes are inclined planes between the c-plane and the m-plane.
Figure 2.3 Schematics of the various low index semipolar and nonpolar planes. Source:
http://www.compoundsemiconductor.net/
43
As can be seen in Fig2.2, there is a significant difference in the surface
morphology of the {11-22} and {10-11} nanostripes. While the {10-11} nanostripes show
smooth surface morphology, the surfaces of the {11-22} nanostripes are significantly
rougher with surface undulation along the <10-10> direction. This surface roughness is
non-ideal for subsequent growth of InGaN QWs and p-GaN because defects can
nucleate on the surface kinks, especially at the valleys of rough surfaces [5].
Furthermore, the surface roughness may affect the uniformity of the QWs in terms of
their In incorporation and thickness, which degrade the LED characteristics. Therefore,
we developed the growth conditions that allow the growth of smooth {11-22} surfaces.
We studied the effect of V/III ratio, temperature, pressure, and carrier gas on the
surface morphology of the {11-22} nanostripes. We found that the type of the carrier
gas, i.e. H 2 or N 2 had the most significant effect on the surface morphology while all
other parameters studied only had a minor effect on the surface morphology. Figure 2.4
shows the top-down SEM images of the {10-11} and {11-22} nanostripes grown under N 2
and H 2 ambient. The surface roughness of the {11-22} facets is dramatically improved as
the carrier gas is changed from the N 2 to H 2. The {10-11} nanostripes had a smooth
surface for all growth conditions, showing superior surface stabilities.
44
Figure 2.4 top-down SEM images of the nanostripes grown in N 2 ambient for (a) {10-11}
nanostripes and (b) {11-22} nanostripes and the same that are grown in H 2 ambient for
(c) {10-11} nanostripes and (d) {11-22} nanostripes
In the following, we examine the growth mechanisms of the {11-22} nanostripes
to explain the dramatic effect of carrier gas on the surface morphology. A series of
annealing experiments were performed after the growth of smooth {11-22} nanostripes
in H 2 ambient to study the surface stability under different ambient. In all the annealing
conditions, a flow rate of 3 slm of NH 3 is flowing to prevent decomposition of the
nanostripes. Figure 2.5 (a) shows the top-down SEM image of the nanostripes that were
annealed in N 2 ambient for 10 min at 925 C. We can observe significant etching of
materials and the etching seems to stop until the {10-11} facets are exposed. Figure 2.5
45
(b) shows the SEM images of the stripes that were subject to the same conditions as in
Fig. 2.5 (a) but the temperature are decreased to 790 C. We observe similar etching
Figure 2.5 top-down SEM images of the {11-22} nanostripes subjected to annealing in (a)
NH 3 and N 2 for 10 min at 925 C, (b) NH 3 and N 2 for 10 min at 790 C, (c) NH 3 and H 2 for 10
min at 925 C and (d) NH 3, N 2 and H 2 (2 slm) 10 min at 925 C
effect as in Fig 2.5 (a), but the degree of the etching is significantly reduced.
Figure 2.5 (c) shows results for annealing the stripes in H 2 ambient for 10 min at 925 C.
The smooth surface is maintained in this case, further demonstrating that this surface is
stable under H 2 ambient. Figure 2.5 (d) shows the results for an annealing condition
46
where the carrier gas is N 2, but 2 slm of H 2 are introduced through the makeup lines. The
smooth surface morphology is maintained in this case, showing that as long as some H 2
is flowing, the surface can be stabilized.
So far, the experimental data suggests that the carrier gas significantly affects
the surface morphology of the {11-22} surface and the surface is destabilized in N 2
ambient due to an etching effect. One of the major differences between N 2 and H 2
ambient in GaN growth is the reactivity of H 2 at the surface. N 2 is extremely inert, and its
participation in the growth can be considered negligible. H 2, on the other hand, has
been shown to alter the GaN surface significantly [6]. An in-situ surface reflectivity
difference measurement showed that H 2 carrier gas above 800 C changes the GaN
surface from N-rich to Ga-rich [7]. It is postulated that H 2 at elevated temperature react
with surface N, forming gaseous NH 3 that desorbs from the surface. This principle is
used to explain the experimental data observed in this study, as explained in more
details as follows.
47
Figure 2.6 Schematics of the surface configuration of the {11-22} surfaces and their
interactions with the ambient gasses under (a) N2 ambient and (b) H2 ambient.
Fig. 2.6 shows the schematics of the surface configuration of the {11-22} surfaces
and their interaction with the gas ambient. The yellow, blue, red and green atoms
represent the surface Ga, surface N, adsorbed N and N adatom respectively. The surface
Ga and N are those that exist on the top most layer of the ideal cleavage of the
crystallographic plane. Theoretical calculations have shown that the ideal {11-22}
surface has high surface energy and can be stabilized by forming covalent bonds with N
adatoms [8]. It should be noted, that H adatoms are formed on some of the surface Ga,
N and N adatoms where appropriate to reduce the surface energy. These H adatoms are
omitted from the schematics to simplify the picture to show the most significant
features of the surface.
48
Let us go back to Fig 2.5 (a) where it showed significant destabilization of the
surface under NH 3 and N 2 annealing condition. Because N 2 is inert as discussed
previously, it is likely that NH 3 is responsible for the surface destabilization. NH 3
molecules are pyrolyzed heterogeneously on the surface to supply adsorbed N on the
surface. We think that this adsorbed N can diffuse to the vicinity of the N adatoms,
binding with the N adatoms and desorb as N 2 gas, as shown in Fig. 2.6 (a). Once the N
adatoms are removed from the surface, the surface becomes significantly destabilized,
and surface Ga can easily desorb, leading to significant etching [8]. The removal of the
materials continues until the {10-11} surfaces appear at which point the surface is
stabilized. Because the rate of these reactions, from heterogeneous pyrolysis, reaction
rates, and desorption rates all depends exponentially on the temperature, the etching
effects were reduced at a lower temperature, as shown in Fig. 2.5 (b).
Fig 2.6 (b) shows the schematics of the surface processes of the reconstructed
{11-22} surface in H 2 ambient. H 2 can remove the adsorbed N atoms, significantly
reducing the densities of the adsorbed N atoms. Alternatively, the process can be
thought of as follows. The following simple reaction equation describes the
heterogeneous pyrolysis of NH 3:
2𝑁𝐻
3
↔ 2𝑁 𝑎𝑑
+3𝐻 2
Based on the principle of mass action, the equilibrium of shifts to the left side of the
equation when the H 2 partial pressure increases, enhancing the production of NH 3 and
reducing the N adsorbed on the surface. The reduction of N adsorbed reduces the etching
effect observed in N 2 ambient, and the surface is stabilized. The picture painted here is
49
that NH 3 keeps supplying the N adsorbed in both ambients. The difference is that in H 2
ambient, H 2 removes the N adsorbed constantly from the surface, preventing the N adsorbed
from binding with N adatom and destabilizing the surface. Figure 2.5 (d) shows that in N 2
ambient, the addition of small amount of H 2 can also stabilize the surface. This
observation further proves that N 2 do no participate in these processes and H 2 is the
dominant factor in the growth of the {11-22} nanostripes.
2.3 Nanostripes along different directions
We also studied the growth of nanostripes that are oriented in various
directions between the two low index directions to study higher index semi-polar
planes. It is possible that some of these semipolar planes have high indium
incorporation rate and may be suitable as templates for the growth of high In-content
QWs. Stripe patterns with orientations between the [12-10] and [1-100] in 2-degree
increments were fabricated for these growth studies. Due to the 6 fold symmetry of the
hexagonal lattice of GaN, a total variation of 30 degrees from the [12-10] and [1-100] is
equivalent to scanning all directions. Fig. 2.7 shows the nanostripes with various
orientations grown under H 2 ambient. When the stripes are oriented less than 4 degrees
from the [11-20] direction, the high index planes quickly grow to extinction and stripes
are terminated with the {10-11} surfaces with larger lateral growth. As the angle is
increased further, the stripes start to align with the patterned directions, but the surface
breaks into steps consisting of {10-11} planes. This shows that these high index
semipolar planes have a high surface energy density, and they are not stable against the
50
formation of low index planes.
Figure 2.7 Top-down SEM images of nanostripes with orientations between the [11-20]
and [1-100] directions with 2-degree increment. The offset angle of the stripe orientation
to the [11-20] direction is shown on the top left of the images. The stripe with 0˚
orientation has {10-11} surfaces and the stripe with 30˚ orientation has {11-22} surfaces.
When the stripes are oriented less than 4 degrees from the [11-20] direction, the high
index planes quickly grow to extinction and stripes are terminated with the {10-11}
surfaces with larger lateral growth. As the angle is increased further, the stripes start to
align with the patterned directions, but the surface breaks into steps of {10-11} planes.
As the stripes are oriented close to the [1-100] direction, the stripes exhibit smoother
surface morphology, similar to the surface of {11-22} planes.
51
Another observation is the difference between the structure near the ends of
the stripes and that near the center of the stripes. Near the ends of the stripes, the
structure is defined by the slowest growing plane, which is the {10-11} plane. Therefore,
if sufficient time is given and the spacing between the stripes are large enough, the high
index planes will grow to extinction, just like those stripes with the smaller angle from
the [11-20] direction and the final structure would resemble an asymmetric pyramid.
Interestingly, as the stripes are oriented close to the [1-100] direction, the stripes
exhibit smoother surface morphology, similar to the surface of {11-22} planes. These
planes may also be suitable for the growth of QWs.
2.4 A-plane nanosheet structures
We also explored the growth of tall vertical structures with large nonpolar a-
plane sidewalls that are grown on patterns with [1-100] direction. Previous studies that
examined ELO [10] have shown that the horizontal growth rate is usually faster than the
vertical growth rate for stripes oriented in this direction. In our previous studies on GaN
nanorod growth, we found that vertical growth rate can be enhanced dramatically by
decreasing the NH 3 flow rate to 6 sccm at a high growth temperature of 1150 C [9].
Similar experiments are done in the growth on the stripe patterns. Fig. 2.8 (a) and (b)
shows inclined view SEM images of the nanostructures with NH3 flow rate of 100 and 6
sccm respectively. We measure the width of the stripes from the top down SEM images
(not shown) and calculate the height of the stripes from the measurement from the
inclined image and some geometrical manipulation. The height of the structure
52
increases from 408 nm to 760 nm and the width decreased from 640 nm to 520 nm. The
vertical to horizontal aspect ratio increased from 0.6 to 1.5 as the NH3 flow rate
decreased.
Figure 2.8 Nanosheet structures grown under NH 3 flow rate of (a) 100 sccm and (b) 6
sccm. Surface reconstruction of the GaN (0001) plane under (c) N-rich and (d) Ga-rich
growth conditions, corresponding to N adatoms and Ga adatoms reconstructions,
respectively.
We attribute the increase in the vertical growth to the difference in the surface
reconstruction of the c-plane surface under the different growth conditions. Theoretical
calculations of the surface reconstruction under MOCVD growth conditions show that
under normal high V/III ratio conditions and high growth temperature, the surface is
stabilized by the (2 × 2) N-adatom configuration, shown in (c), where the N adatoms are
53
bonded to three surfaces Ga, and the last dangling bond is terminated by the H [10]. The
N-H termination is a stable configuration due to the strong bonding, and when a growth
step propagates to the N adatoms, this strong bonding needs to be broken for the
growth to proceed. Due to this kinetic barrier, the growth rate of the c-plane can be
reduced significantly. However, as the NH3 flow rate decreases, the N coverage on the
surface decreases and the surface can transition to a (2 × 2) Ga adatom configuration
where Ga adatoms bond with three of the four Ga on the surface, shown schematically
in (d) [10]. Because the H passivation effect is removed in this case, the c-plane growth
rate can increase significantly. We should note that further decrease in the NH3 flow
rate results in significant decrease in the grown material and some regions lead to
termination of the growth. When NH3 flow rate is not sufficient, GaN can decompose
easily at such a high growth temperature.
We were able to further increase the vertical growth rate by increasing the
disilane flow. Fig. 2.9 shows the aspect ratio of nanosheet increased from 1.5 to 3.0 as
disilane flow rate increased from 1.5 to 4.5 nmol/min. However, a further increase in
the disilane flow leads to decrease in the aspect ratio. It is possible that under the
extreme Ga-rich condition where these structures are grown, the supply of N may
become important in the growth. It has been reported that increase in the disilane flow
rate leads to accumulation of Si near the surface [11]. It is possible that Si layer acts as a
surfactant layer and enhances the sticking coefficient and incorporation of N into the
materials. The increased N incorporation also means that the Ga desorption rate will
decrease because the strong covalent bonds that are created between Ga and N,
54
compared to metallic bonding formed among Ga atoms. However, if Si concentration is
above a threshold concentration, it is not stable against the formation of SiN, and the
growth rate can decrease [11]. There is evidence of facet termination near the ends of
the stripes, which are not identified to be any of the low index planes. We can minimize
these edge effects by growing the nanosheet on long stripe patterns. Figure 2.9 (b)
shows the middle part of the nanosheet grown on 100-micron long stripe patterns and
the same aspect ratio is achieved.
Figure 2.9 Nanosheet structures grown on (a) 10-micron long pattern and (b) 100-micron
long pattern. The aspect ratio was increased by increasing the disilane flow rate.
To summarize this chapter, we have developed the growth processes and
explained the growth mechanisms for nanostripes with various crystallographic planes
and nanosheet with the non-polar {11-20} sidewall facets. We were able to growth the
{11-22} nanostripes with smooth surface morphology. Annealing experiments were
performed to show that H 2 can remove adsorbed N so that the surface can be stabilized
and therefore the existence of H 2 is paramount in the growth of the {11-22} nanostripes.
55
Nanostripes that are aligned to all directions are achieved as potential candidates for
the next studies. Finally, we have developed growth conditions and achieved nanosheet
structures with a-plane sidewalls where vertical to the horizontal aspect ratio of is 3:1.
To our knowledge, such a high aspect ratio nanosheet structures with a-plane sidewall
are achieved for the first time. In the next chapter, we compared these different
structures in terms of their In incorporation rates and investigate their suitability for the
growth of high In-content QWs and LEDS.
2.5 Chapter references
[1] T.-w. Yeh, Y.-T. Lin, B. Ahn, L. S. Stewart, P. D. Dapkus and S. R. Nutt, "Vertical
nonpolar growth templates for light emitting diodes formed with GaN
nanosheets," Applied Physics Letters, vol. 100, p. 033119, 2012.
[2] K. Hiramatsu, K. Nishiyama, A. Motogaito, H. Miyake, Y. Iyechika and T. Maeda.,
"Recent Progress in Selective Area Growth and Epitaxial Lateral Overgrowth of III‐
Nitrides: Effects of Reactor Pressure in MOVPE Growth," physica status solidi (a),
vol. 176, no. 1, pp. 535-543, 1999.
[3] H. Marchand, J. P. Ibbetson, P. T. Fini, S. Keller, S. P. DenBaars, J. S. Speck and U. K.
Mishra, "Mechanisms of lateral epitaxial overgrowth of gallium nitride by
metalorganic chemical vapor deposition," Journal of Crystal Growth, vol. 195, no.
1, pp. 328-332, 1998.
[4] W. Feng, V. V. Kuryatkov, S. A. Nikishin and M. Holtz, "Selective area epitaxy of
InGaN quantum well triangular micro rings with a single type of side wall facets,"
Journal of cyrstal growth, vol. 312, pp. 1717-1720, 2010.
[5] F. Wu, Y. Zhao, A. Romanov, S. P. DenBaars, S. Nakamura and J. S. Speck, "Stacking
faults and interface roughening in semipolar ( 20-2-1) single InGaN quantum wells
for long wavelength emission".
56
[6] D. Koleske, A. Wickenden, R. Henry, J. Culbertson and M. Twigg, "GaN
decomposition in H2 and N2 at MOVPE temperatures and pressures," Journal of
Crystal Growth, vol. 223, pp. 466-483, 2001.
[7] Y. Kobayashi and N. Kobayashi, "Influence of N2 carrier gas on surface
stoichiometry in GaN MOVPE studied by surface photoabsorption," Journal of
Crystal Growth, Vols. 189-190, pp. 301-304, 1998.
[8] T. Akiyama, T. Yamashita, K. Nakamura and T. Ito, "Ab initio-Based Study for
Adatom Kinetics on Semipolar GaN (11-22) Surfaces," Japanese Journal of Applied
Physics, vol. 48, p. 120218, 2009.
[9] Y.-T. Lin, T.-W. Yeh, Y. Nakajima and P. D. Dapkus, " Catalyst-Free GaN Nanorods
Synthesized by Selective Area Growth," Advanced Functional Materials, vol. 24, no.
21, pp. 3162-3171, 2014.
[10] C. G. Van de Walle and J. Neugebauer, "Role of hydrogen in surface
reconstructions and growth of GaN," Journal of Vacuum Science & Technology B,
vol. 20, no. 1640, p. 1491545, 2002.
[11] A. L. Rosa, J. Neugebauer, J. E. Northrup, C.-D. Lee and R. M. Feenstra, "Adsorption
and incorporation of silicon at GaN (0001) surfaces," Applied physics letters, vol.
80, p. 1452785, 2002.
[12] A. Tyagi, Y.-D. Lin, D. A. Cohen, M. Saito, K. Fujito, J. S. Speck, S. P. DenBaars and S.
Nakamura, "Stimulated Emission at Blue-Green (480 nm) and Green (514 nm)
Wavelengths from Nonpolar (m-plane) and Semipolar (11-22) InGaN Multiple
Quantum Well Laser Diode Structures," Applied Physics Express 1, p. 091103, 2008.
[13] K. Hiramatsu, K. Nishiyama, M. Onishi, H. Mizutani, M. Narukawa, A. Motogaito, H.
Miyake, Y. Iyechika and T. Maeda, "Fabrication and characterization of low defect
density GaN using facet-controlled epitaxial lateral overgrowth (FACELO)".
[14] M. Funato, T. Kotani, T. Kondou, Y. Kawakami, Y. Narukawa and T. Mukai, "Tailored
emission color synthesis using microfacet quantum wells consisting of nitride
semiconductors without phosphors," Applied Physics Letters, vol. 88, p. 261920,
2006.
[15] Y. Kawakami, K. Nishizuka, D. Yamada, A. Kaneta, M. Funato, Y. Narukawa and T.
Mukai, "Efficient green emission from ( 11 2-2) InGaN/GaN quantum wells on GaN
microfacets probed by scanning near field optical microscopy," Applied Physics
Letters, vol. 90, p. 261912, 2007.
57
[16] B. Leung, Q. Sun, C. D. Yerino, J. Han and M. E. Coltrin, "Using the kinetic Wulff plot
to design and control nonpolar and semipolar GaN heteroepitaxy," Semiconductor
Science and Technology, vol. 27, p. 024005, 2012.
[17] T. Zywietz, N. Jorg and S. Matthias, "Adatom diffusion at GaN 0001 and 000-1
surfaces," Applied Physics Letters, vol. 73, no. 487, p. 121909, 1998.
[18] T. Ito, T. Akiyama and K. Nakamura, "Ab initio-based approach to reconstruction,
adsorption and incorporation on GaN surfaces," Semicond. Sci. Technol., vol. 27, p.
024010, 2012.
[19] C. G. Van de Walle and J. Neugebauer, "First-Principles Surface Phase Diagram for
Hydrogen on GaN Surfaces," PHYS ICAL RE V IEW LETTERS, vol. 88, no. 6, p.
066103, 2002.
[20] A. Munkholm, G. Stephenson, J. Eastman, C. Thompson, P. Fini, J. Speck, O. Aucillo,
P. Fuoss and S. DenBaars, "Surface Structure of GaN(0001) in the Chemical Vapor
Deposition Environment," PHYSICAL REVIEW LETTERS, vol. 83, no. 4, p. 741, 1999.
58
Chapter 3 Growth and optimization of high In-content InGaN QW
on nanostructure templates
3.1 Growth of InGaN QW on various nanostructures
A structure containing a single InGaN QW is grown on the nanostripes and
nanosheet structures. Because the nanostripes with the {11-22} facets are sensitive to
the growth ambient, care needs to be taken when the carrier gas is switched to N 2 for
the QW growth. We found that if we stopped the flow of H 2 from the makeup line
before the ramp-down of the temperature, the surface etching effect similar to what is
observed in the annealing experiment in the previous chapter occurs. The process to
change the carrier gas is described in the previous chapter. Therefore, H 2 must continue
to flow during the temperature ramp down to the QW growth temperature; then the
carrier gas is switched to the N 2 and the QW growth proceeds. After the growth of QW,
the carrier is switched to H 2 immediately to grow the quantum barrier. This way, the
smooth surface of the {11-22} nanostripes can be preserved. The specific growth
conditions are described as follows. The QW is grown at 800 ˚C in N 2 ambient and the
quantum well barriers are grown at 925 C in H 2 ambient. The TEG, TMIn, and NH 3 flow
rates are 20 μmol/min, 100 μmol/min, and 9 slm respectively.
We use photoluminescence (PL) to study the optical properties of the InGaN
SQW grown on different structures. The PL spectra are acquired by exciting the samples
with a 325 nm HeCd laser with a spot size of around 50 by 50 μm
2
at a power density of
around 1kW/cm
2
.
59
Figure 3.1 shows the PL spectra of the SQW grown on nanostripes and
nanosheet structures under the same condition. For the nanostripes structures, only
stripes with selected orientation are plotted to make the graph less crowded. The line
shapes of the PL spectra for those spectra that are not plotted are similar to the ones
shown in the figure.
Figure 3.2 shows the peak wavelength of PL emission from the SQW grown on
nanostripes and nanosheet structures, extracted from Figure 3.1. The data is plotted
against the orientation of the stripe with respect to the [11-20] direction. The SEM
images of some of the structures are shown as insets. The SQW grown on the
nanostripes with {10-11} nanostripes emit at around 500 nm, significantly longer than
those grown on the {11-22} nanostripes that emit at around 465 nm and the a-plane
nanosheet that emit at around 435 nm.
Figure 3.5 PL of SQW grown on (a) nanostripes along different orientations and (b) on a-
plane nanosheet
60
Figure 3.2 Peak wavelength of the PL spectra of SQW grown on nanostripes and
nanosheet structures. The data is plotted against the orientation of the stripe with
respect to the [11-20] direction. The SEM images of some of the structures are shown as
insets.
The PL emission wavelength of InGaN QWs depends on the interplay between
the indium composition of the InGaN QW, the thickness of the QW and the piezoelectric
field. The effect of piezoelectric field is expected to be small in QWs grown on semipolar
61
and nonpolar planes because of the significantly reduced piezoelectric field [1]. The
indium composition affects the bandgap of the QW materials and the band edge
emission wavelength. The QW thickness affects the quantization of the states in the QW
and modifies the emission energy slightly. The indium composition of the QW then has
a 1
st
order effect and the QW thickness has a 2
nd
order effect on the emission
wavelength [2]. For example, a theoretical calculation of QW transitions based on single
band effective mass approximation shows that increasing the QW thickness from 4 to 40
nm only results in at most a 9 nm shift in the QW emission wavelength [2]. Because the
wavelength differences among different structures are around 40 nm, it is reasonable to
conclude that the difference in the emission wavelength is mostly a result of the
difference in the In composition in the QWs grown on the different structures.
Therefore, the QW grown on the {10-11} nanostripes should have the highest indium
content.
However, these data simply cannot lead to the conclusion that the nanostripes
with the {10-11} facets have the highest In incorporation rate. The Indium composition
of the QW depends on the complicated interplay between the growth conditions and
the surface kinetics. For example, because the dimensions of these structures are
different, the mass transport of the gas species or the growth rate enhancement is
different for different structures under the same growth conditions. Also, the surface
kinetics are different for different planes. We present in the following the simplest
model that can describe the growth mechanisms of the InGaN QW.
62
The rate equation of a simplified model of the growth processes of InGaN can be
described as follows:
𝜕 [𝜃 𝐼𝑛
𝑐 𝐶 𝑠 ]
𝜕𝑡
=𝐹 𝐼𝑛
(1−𝜃 𝐼𝑛
𝑐 −𝜃 𝐺𝑎
𝑐 )−𝑘 𝐼𝑛
𝑑𝑒
𝜃 𝐼𝑛
𝑐 −𝑘 𝐼𝑛
𝑖𝑛
𝜃 𝐼𝑛
𝑐
𝜕 [𝜃 𝐺𝑎
𝑐 𝐶 𝑠 ]
𝜕𝑡
=𝐹 𝐺𝑎
(1−𝜃 𝐼𝑛
𝑐 −𝜃 𝐺𝑎
𝑐 )−𝑘 𝐺𝑎
𝑑𝑒
𝜃 𝐺𝑎
𝑐 −𝑘 𝐺𝑎
𝑖𝑛
𝜃 𝐺𝑎
𝑐
where 𝜃 𝐼𝑛
𝑐 and 𝜃 𝐺𝑎
𝑐 are the coverage of the surface sites by In and Ga respectively, 𝐶 𝑠 is
the density of surface sites, 𝐹 𝐼𝑛
and 𝐹 𝐺𝑎
are the flux of the In and Ga species
respectively, 𝑘 𝑑𝑒
is the desorption rate, and 𝑘 𝑖𝑛
is the incorporation rate. We assume
group III species govern the growth. If the thickness and the composition of the QW can
be precisely determined, we can have the following additional equations:
𝐺𝑟𝑜𝑤𝑡 ℎ 𝑟𝑎𝑡𝑒 𝑜𝑓 𝐼𝑛𝑁 =𝑘 𝐼𝑛
𝑖𝑛
𝜃 𝐼𝑛
𝑐
𝐺𝑟𝑜𝑤𝑡 ℎ 𝑟𝑎𝑡𝑒 𝑜𝑓 𝐺𝑎𝑁 =𝑘 𝐺𝑎
𝑖𝑛
𝜃 𝐺𝑎
𝑐
Still, there are six unknown with four equations, which are not sufficient to solve the
system. Besides, the precise determination of the In composition in these
nanostructures are extremely challenging. Therefore, even this simple model cannot be
solved because of the lack of knowledge of some of the coefficients in the rate equation.
Instead, we optimized the InGaN growth conditions for long wavelength emission
individually for these structures, varying TMIn, TEG, flow rate ratio, total flow rate,
growth time, growth temperature, growth pressure. As will be shown shortly, it was not
possible to push the wavelength emission beyond 500 nm for the {11‐22} nanostripes.
63
The main reason is that H 2 is required to grow the QB layer so as to maintain the
integrity of the structure, but the H 2 also removes In from the QW, thereby blue‐shifting
the emission structure from what it could achieve. On the other hand, the {10‐11}
nanostripes can be grown in the N 2 ambient and longer wavelength emission was
achieved. Less optimization was done on the nanosheet structure for the following
reasons. First, the growth condition for this structures can be considered extreme and
difficult to control. Because the QW thickness and composition are highly dependent on
the dimensions of the templates, the reproducibility of the template is critical for the
reproducibility of the QWs. The growth condition of extremely low V/III ratio introduces
high density of N vacancies or carbon incorporation which results in strong yellow band
emission, also observed in nanorods grown with continuous mode [3]. In the case of
blue emitting QWs, the QW emission spectrum can be separated from the yellow band
as shown in Fig.3.1. However, our purpose is to grow QW that emits in the green, yellow
a red regime, and their emissions can be difficult to distinguish from the yellow band.
Secondly, different from nanorod and m‐plane nanosheet structures [4], the a‐plane
nanosheet structures has non‐negligible c‐plane top surfaces, which is not ideal for the
growth of high In QW structures because of the large piezoelectric field that exists in the
material that inevitably grows on these planes.
Moreover, a study of MQW grown on various semipolar and nonpolar planar
substrates also showed that {10-11} surface has the longest emission wavelength [1].
This longer emission wavelength for planar substrates was explained as being caused by
higher Indium composition and higher indium incorporation on the {10-11} surfaces
64
since the QW thickness was similar for the various orientations. The same reasoning can
be used to explain the longer emission wavelength of QWs grown on {10-11} facets of
nanostripes although more extensive studies are necessary to draw final conclusions
about the indium incorporation rate of various facets of nanostructure growth.
The higher In incorporation rate on the {10-11} facets than others may be
explained by the differences in the atomic configurations of the surfaces and how
strongly In is adsorbed on these surfaces. What is unique about the {10-11} surface
among the various low index planes is the ability for adsorbed cations, like In and Ga, to
form 3 coordinated bonds with surface N, as shown in figure 3.3 [5]. The three strong
covalent bonds can reduce the desorption rate of the In adatoms. The only other plane
that can form such a bonding configuration is on the N-polar (000-1) plane [6]. The {11-
22} and {11-20} planes can form 2 coordinated and 1 coordinated bonding with the
adsorbed cations, respectively [7] [8]. Therefore, In is expected to be more strongly
adsorbed on the {10-11} than the {11-22} and the {11-20} surfaces. As discussed in
Chapter 1, the higher Indium incorporation rate of the {10-11} plane allows the growth
of InGaN at a higher temperature for the same Indium composition compared to those
grown on planes with smaller indium incorporation rate. As a result, the QW grown on
the {10-11} facets can have higher crystal quality especially for long wavelength emitting
materials and devices.
65
Figure 3.3 The top and side view of the schematics of the bonding configuration of Ga
adatoms on the {10-11} planes. Source: Northrup et al., Applied Physics Letters, 74, 1999
3.2 Optimization of InGaN QW growth on {10-11} the nanostripes
Because of the long emission wavelength that occurs on these structures,
nanostripes with the {10-11} facets are chosen as the most promising candidate for
further investigation to demonstrate efficient green emission using the nanostructure
geometry. Figure 3.4 shows the initial optimization of InGaN QW grown on the {10-11}
nanostripes.
66
Figure 3.4 PL spectra of InGaN grown on the {10-11} nanostripes under different growth
conditions. The Legend shows the difference in the growth conditions among the three
curves. All other parameters are identical, like the pressure is 300 Torr, TEG flow rate of
30 sccm, Growth time of 120 s,
The green curve shows the spectrum of the InGaN grown with the conditions
that are based on the conditions optimized for QW growth on the planar structures. The
peak QW emission wavelength is around 486 nm. We can observe that the yellow band
emission in the 550 nm to 600 nm range has a peak intensity that is comparable to the
QW emission intensity. The PL intensity ratio of GaN band-edge emission, to the QW
emission, and to the yellow band emission is 2.25: 1.1: 1. The blue curve shows the PL of
InGaN sample grown at 800 C, a 40 C increase from the first sample. The TMIn flow rate
0
100
200
300
400
500
600
350 400 450 500 550 600
PL Intensity (a.u.)
Wavelength (nm)
PL spectra of InGaN QWs grown on {10-11}
nanostripes
760 C, TMIn = 50 sccm,
NH3 = 5 slm
800 C, TMIn = 100
sccm, NH3 = 5 slm
800 C, TMIn = 100
sccm, NH3 = 9 slm
67
is doubled from 50 sccm to 100 sccm to compensate for the reduced In incorporation as
a result of an increase in the growth temperature. The QW emission wavelength of this
sample is 486 nm. The peak PL intensity ratio of GaN: QW: YL is 2.18:1.5:1. Compared to
the previous sample, the QW emission increased relative to the GaN band-edge
emission and the yellow band. This is because that the QW growth temperature is
increased, leading to higher QW material quality. However, because the QB growth
conditions are not modified, the yellow band emission is still significant. The red curve
shows that the PL of the InGaN QW and QB sample grown with an NH 3 flow rate of 9
slm. The emission wavelength is redshifted to 506 nm with respect to the previous
sample. The peak GaN: QW: YL of this sample is 15: 3.5: 1. The band edge and the QW
emission are significantly increased with respect to the yellow band emission. As
discussed previously, increase in the NH 3 flow rate can decrease the N-vacancies
densities or the carbon incorporation, which are responsible for the yellow band
emission.
Figure 3.5 shows the growth conditions that significantly increased the emission
wavelength. The red curve in the figure is the same as the red curve shown in Figure 3.4.
The blue curve shows the PL of InGaN sample grown under the same conditions as the
previous sample, except that the QB is grown in the N 2 ambient rather than the H 2
ambient. The emission wavelength increased to 570 nm and the PL intensity also
increased. The green curve shows the PL of InGaN sample grown in the same run as the
blue curve sample, but the difference is that the stripe spacing is reduced from 1000 nm
to 300 nm. The decrease in the spacing between the nanostripes means that laser light
68
excites the nanostripes QWs more and less laser light penetrate to the bulk of the GaN.
Because of the increase in the active area, the PL intensity was further increased.
Figure 3.5 PL spectra of InGaN grown on the {10-11} nanostripes under different growth
conditions. The Legend shows the difference in the growth conditions among the three
curves.
The H 2 that is present during the QB growth likely removed some In from the
QW, thereby making the incorporation of In difficult. The reason why we grew the QB in
the H 2 ambient is that we have shown that optimal growth of the {11-22} nanostripes
require is achieved in an H 2 ambient, and this was the initial conditions from which more
growth optimization was performed. This experiment shows how significant and
important it is to grow the QB in N 2 ambient and the stability of the {10-11} nanostripes
in N 2 ambient is one of the reasons why we chose the {10-11} over the {11-22}
0
200
400
600
800
1000
1200
1400
1600
1800
2000
350 450 550 650
PL intensity (a.u.)
Wavelength (nm)
PL spectra InGaN QWs grown on {10-11}
QB in N2 ambient,
Stripe spacing = 300
nm
QB in N2 ambient,
stripe spacing = 1000
nm
QB in H2 ambient,
stripe spacing = 1000
nm
69
nanostripes. Another note is that we were not able to further increase the wavelength
emission of QW grown on the {10-11} nanostripes without a significant drop-off in the
PL intensity. The problem with the amber emitting QWs will be discussed in one of the
later chapters.
3.3 Quantum efficiency measurement of the QW grown on the {11-22}
nanostripes
In order to quantify the quantum efficiency of the yellow emitting QWs, we set
up a power dependent PL system that has been used to extract the quantum efficiency
of the materials, which is developed as follows [9] [10]. The model is based on the rate
equation:
𝐺 =𝐴𝑛 +𝐵 𝑛 2
+𝐶 𝑁 3
Where G is the carrier generation rate, A, B and C are the Shockley-Read-Hall (SRH),
radiative, and Auger recombination, respectively, and n is the carrier concentration. The
internal quantum efficiency is defined by
𝜂 =
𝐵 𝑛 2
𝐴𝑛 +𝐵 𝑛 2
+𝐶 𝑛 3
The radiative recombination rate is related to the measured PL intensity by:
𝐼 𝑃𝐿
=𝑎𝐵 𝑛 2
70
Where a is a constant that takes into account the volume of the excited active region
and the overall collection efficiency of the luminescence. We substitute the expression
for the radiative recombination into the rate equation:
𝐺 =𝐴 √
𝐼 𝑃𝐿
𝑎𝐵
+
𝐼 𝑃𝐿
𝑎 +
𝐶 𝐼 𝑃𝐿
3
2
[𝑎𝐵 ]
3
2
Now, G can be expressed as a function of the laser excitation:
𝐺 =𝑥 𝑃 𝑙𝑎𝑠𝑒𝑟
𝑥 =
(1−𝑅 )𝛼 𝐴 𝑠𝑝𝑜𝑡 ℎ𝜈
Where R is the reflection coefficient, 𝛼 is the absorption coefficient of the active region,
𝐴 𝑠𝑝𝑜𝑡 is the area of the laser spot size, and ℎ𝜈 is the photon energy of the laser.
Therefore, we can obtain the following relations:
𝑃 𝑙𝑎𝑠𝑒𝑟 =
𝐴 √𝑎𝐵
√𝐼 𝑃𝐿
+
1
𝑥𝑎
𝐼 𝑃𝐿
+
𝐶 𝑥 [𝑎𝐵 ]
3
2
𝐼 𝑃𝐿
3
2
The coefficients of the three terms on the right-hand side can be substituted by fitting
parameters :
𝑃 𝑙𝑎𝑠𝑒𝑟 =𝑃 1
√𝐼 𝑃𝐿
+𝑃 2
𝐼 𝑃𝐿
+𝑃 3
𝐼 𝑃𝐿
3
2
Where 𝑃 𝑙𝑎𝑠𝑒𝑟 is the laser power, 𝑃 1
, 𝑃 2
, and 𝑃 3
are fitting parameters. A series of PL
spectra are obtained for different laser excitation power. Then the integrated PL
intensities are plotted against the laser excitation power. The data is fitted according to
71
the equation above and the fitting parameters are obtained. The quantum efficiency is
calculated from the following:
𝜂 =
𝐵𝑛
2
𝐴𝑛 +𝐵 𝑛 2
+𝐶 𝑛 3
=
𝐼 𝑃𝐿
𝑃 2
𝑃 𝑙𝑎𝑠𝑒𝑟
Figure 3.6 shows the schematics of the layout of the power dependent PL
measurement. The laser output first goes through a (neutral density) ND filter mounted
on a rotational wheel and then another ND filter on a flip mount. The optical density of
the filters that are mounted on the wheels are 0.5, 1.0, 1.5, 2.0, 2.5, and 3.0, and that of
the one mounted on the flip mount is 0.3. The optical density is related to the
attenuation of the signal with the following relationships:
I
out
𝐼 𝑖𝑛
=10
−𝑂𝐷
For example, an optical density of 3.0 means that the laser power is attenuated to 0.001
of the original power. When ND filters are put in series, their optical density adds.
Therefore, this setup can generate a combination of 12 laser powers from 0.5 I in to 5 x
10
-4
I in by different combinations of the 1
st
and the 2
nd
filter. We should note that
because the ND filters attenuate the signal by reflection, there will be an effect of
multiple reflections that complicates the system. Therefore, we limit the number of ND
filters to two. After the filters, we installed a flip mirror that can direct the light to an
integrating sphere where the laser power can be measured. This calibration is done for
every measurement for accurate results. The rest of the configuration is the same as a
72
regular PL setup and it is self-explanatory. The power density of the unattenuated laser
is around 1 kW/cm
2
and the spot size is around 50 by 50 μm
2
.
Figure 3.6 Schematics of the power dependent PL set up
Figure 3.7 (a) shows the PL spectra obtained on the yellow emitting InGaN QW
structures grown on the {10-11} nanostripes under different laser excitation. The curves
are plotted on the log scale to show all the PL spectra clearly. The QW emission is
around 2 orders of magnitude higher than the GaN band-edge emission, indicating that
the laser is mostly exciting the QW region. Figure 3.7 (b) shows the integrated PL
intensities extracted from the spectra shown in Figure 3.7 (a), and they are plotted
against the excitation power.
73
Figure 3.7 (a) Power dependent PL spectra of yellow emitting InGaN structures grown on
the {10-11} nanostripes. (b) Integrated PL intensities with respect to the normalized
excitation power. The dotted line shows conceptually what the curve should look like if
non-radiative recombination processes are significant.
Figure 3.7 (b) shows a linear increase of the emitted intensity for over three
order of magnitude increase of excitation power, indicating a dominant radiative
recombination process over Shockley-Read-Hall and Auger non-radiative recombination
processes. Conceptually, if the two non-radiative recombination processes are
significant, the PL intensity will be reduced under high and low excitation as plotted in
the dotted lines in Figure 3.7 (b). In other works that utilized this method, only a
variation of 2 orders of magnitude of laser power saw a non-linear PL response. In this
work, we varied the laser power by 5000 fold, and we still observed linear relationships,
which indicates high radiative recombination efficiency [9] [10]. The PL linewidth is
measured to be around 60 nm, which is comparable to the linewidth of typical c-plane
LED luminescence spectra in the green-yellow range [11] . Because the spot size of the
74
laser excitation is around 50 by 50 μm
2,
the PL spectra represent the average
luminescence of many stripes. Therefore, the narrow linewidth obtained show the high
uniformity of the QW materials grown on the nanostructures. In the next chapter, we
will show detailed nanoscale and atomic scale optical and structural characterization of
the InGaN QW materials.
3.4 Chapter references
[1] T. Wernicke, L. Schade, C. Netzel, J. Rass, V. Hoffmann, S. Ploch, A. Knauer, M.
Weyers, U. Schwarz and M. Kneissl, "Indium incorporation and emission
wavelength of polar, nonpolar and semipolar InGaN quantum wells,"
Semiconductor science and techonolgy, vol. 27, p. 024014, 2012.
[2] X. Zhang, "Pattern dependent lateral epitaxial overgrowth of gallium nitride by
metalorganic chemical vapor deposition," Doctoral dissertation , p. 142, 2001.
[3] Y.-T. Lin, T.-W. Yeh, Y. Nakajima and P. D. Dapkus, " Catalyst-Free GaN Nanorods
Synthesized by Selective Area Growth," Advanced Functional Materials, vol. 24, no.
21, pp. 3162-3171, 2014.
[4] T.-w. Yeh, Y.-T. Lin, B. Ahn, L. S. Stewart, P. D. Dapkus and S. R. Nutt, "Vertical
nonpolar growth templates for light emitting diodes formed with GaN
nanosheets," Applied Physics Letters, vol. 100, p. 033119, 2012.
[5] K. Hiramatsu, K. Nishiyama, A. Motogaito, H. Miyake, Y. Iyechika and T. Maeda.,
"Recent Progress in Selective Area Growth and Epitaxial Lateral Overgrowth of III‐
Nitrides: Effects of Reactor Pressure in MOVPE Growth," physica status solidi (a),
vol. 176, no. 1, pp. 535-543, 1999.
[6] T. Zywietz, N. Jorg and S. Matthias, "Adatom diffusion at GaN 0001 and 000-1
surfaces," Applied Physics Letters, vol. 73, no. 487, p. 121909, 1998.
[7] J. E. Northrup, L. T. Romano and J. Neugebauer, "Surface energetics, pit formation,
and chemical ordering in InGaN alloys," Applied Physics Letters, vol. 74, p. 123837,
1999.
75
[8] J. E. Northrup, "GaN and InGaN (11-22) surfaces: Group-III adlayers and indium
incorporation," Applied physics letters, vol. 95, p. 133107, 2009.
[9] Y.-S. Yoo, T.-M. Roh, N. Jong-Ho, S. Son and Y.-H. Cho, "Simple analysis method for
determining internal quantum efficiency and relative recombination ratios in light
emitting diodes," Applied Physics Letters, vol. 102, p. 211107, 2013.
[10] Q. Gao, D. Saxena, F. Wang, L. Fu, S. Mokkapati, Y. Guo , L. Li, J. Wong-Leung, P.
Caroff, H. H. Tan and C. Jagadish, "Selective-Area Epitaxy of Pure Wurtzite InP
Nanowires: High Quantum Efficiency and Room-Temperature Lasing," Nano
Letters, vol. 14, no. 9, pp. 5206-5211, 2014.
[11] S. Nakamura, M. Senoh, N. Iwasa and S.-i. Nagahama, "High-Brightness InGaN
Blue, Green and Yellow Light-Emitting Diodes with Quantum Well structures,"
Japanese Journal of Applied physics, vol. 34, pp. 797-799, 1995.
[12] J. E. Northrup and J. Neugebauer, "Theory of GaN (10-10) and (11-20) surfaces,"
Physical Review B, vol. 53, no. 6, p. 477, 1996.
76
Chapter 4 Optical and structural characterization of nanostripes
MQW structures
In the previous chapter, we discussed the growth optimization of long
wavelength emitting MQW structures and demonstrated high radiative recombination
efficiency using power dependent PL measurements. However, one significant problem
associated with QWs grown on non-planar templates, commonly observed in micron-
scale structures, is the varying emission wavelength from the top to the bottom of the
structures, which can be attributed to the differences in the mass transport and the
growth kinetics of In and Ga precursors that causes the variation in the indium
composition in the QWs [14] This effect is more severe for growth of high In content
QWs as the gradient of In composition is larger and as much as 50 nm variation of
wavelength has been observed for green emitting QWs grown on micron-sized
pyramidal stripe structures [15]. The non-uniformity of In composition causes a
blueshift in the emission wavelength of LEDs grown on non-planar templates [1]. When
there is a gradient in the In composition, the region with higher In composition turns on
first and as the current increases, the region with lower indium turns on and dominates
the emission.
In this chapter, we examine the optical and structure properties of high In-
content InGaN QW materials grown on the nanostripe structures. Cathodoluminescence
(CL) was used to study the nanoscale optical properties, and transmission electron
microscopy (TEM) was used to study the structural properties of the nanostripe MQW
structures.
77
4.1 Cathodoluminescence characterization of nanostripe InGaN QW structures
In the previous chapter, we used PL to characterize the optical characteristics of
the nanostripe MQW structures. However, because the dimension of the nanostripe is
in the nm range, a higher resolution characterization method is required to quantify the
spatial uniformity of the active region. Cathodoluminescence measurement is an ideal
characterization method to measure the nano-scale optical properties. The
cathodoluminescence is caused by recombination of carriers excited by the scattering of
incident high energy electron beams in the sample. Because the electron beam can be
focused to nm scale, we can excite a localized region of the sample and acquire the
optical characteristics of the nm-scale region of interest.
Figure 4.1 Schematic of the CL system within a SEM
78
Figure 4.1 shows the CL system inside a SEM. A parabolic mirror with an aperture is
inserted inside the SEM. The electron beam passes through an aperture in the parabolic
mirror and excites the sample, which is placed at the focal point of the parabolic mirror.
The cathodoluminescence was then collimated by the parabolic mirror and the light is
guided out of the SEM to the spectrometer.
Cathodoluminescence (CL) measurements were performed on the sample
described in the last chapter where we showed high radiative recombination efficiency.
It consists of three pairs of QWs on {10-11} facets of a nanostripe structure for which
the PL emission wavelength is around 570nm. Again, the growth conditions of the
nanostripe templates are 3 slm of NH3, 8 sccm of TMG, growth pressure of 200 torr,
temperature of 975 C and H 2 carrier gas. A prism structure with triangular cross-
sectional shape is grown out of the mask opening regions. Both sides of facets are {10-
11} planes and the top (0001) plane is pinched off. Three pairs of QWs are then grown
on the facets of the nanostripes. The growth conditions for the InGaN QW are 9 slm of
NH 3, 20 sccm of TEG, 240 sccm of TMI, growth pressure of 300 torr, temperature of 800
C, and N 2 carrier gas. The quantum barrier (QB) is grown under the same conditions as
the QW except that there is no TMI flow and the growth temperature is 875 C.
79
Figure 4.2 CL line scan performed on MQW structures grown on stripes with 300 nm and
500 nm spacing and the results are shown in figure 4.1 (a) and 1 (b) respectively. In each
of the figures, a SEM image is shown at the top and the CL spectra are shown at the
bottom. In the top-down view SEM images, the region with dark contrast corresponds to
the mask region and the region with brighter contrast corresponds to the nanostripes.
The top and bottom of the nanostripes are indicated by the arrows in figure 4.1 (a). The
electron beam is scanned horizontally, in the <1-100> direction, across arrays of stripes
shown in the SEM images with scanning step size of around 15 nm. Five lines scans are
performed for each sample, and the distance between each line scans are around 650
nm in the vertical <11-20> direction. At each scan location, a spectrum is obtained which
is plotted vertically in the color map and the intensity is represented by color. The white
vertical lines represent the edges of the stripes.
80
Our SEM-CL measurements are performed in a Hitachi S4800 SEM that is
equipped with a Horiba CL system. The acceleration voltage and the probe current
setting in the SEM are 5 kV and 20 μA respectively. CL line scans have been performed
perpendicular to the stripe orientations to investigate the spatial dependence of the
optical emission properties within one stripe and among different stripes.
Figure 4.2 shows the CL line scans performed on MQW structures grown on
stripes with 300 nm and 500 nm spacing and the results are shown in figure 4.2 (a) and
4.2 (b), respectively. In each of the figures, a SEM image is shown at the top, and the CL
spectra are shown at the bottom. In the top-down view SEM images, the region with
dark contrast corresponds to the mask region and the region with brighter contrast
corresponds to the nanostripes. The top and bottom of the nanostripes are indicated by
the arrows in figure 4.2 (a). The electron beam is scanned horizontally, in the <1-100>
direction, across arrays of stripes shown in the SEM images with scanning step size of
around 15 nm (the horizontal distance between 2 excitation locations is 15 nm). Five
lines scans are performed for each sample, and the distance between each line scans
are around 650 nm in the vertical <11-20> direction. In other words, an area of
approximately 3 by 3 microns is scanned at high resolution in the horizontal direction
and with low resolution in the vertical direction to take advantage of the symmetry in
the stripe orientation. At each scan location, a spectrum is obtained which is plotted
vertically in the color map where the intensity is represented by color. The white vertical
lines represent the edges of the stripes.
81
The emission intensity generally decreases from the peak to the edges of the
stripes. This can be understood by the fact that near the edges of the stripes there is
less emitting material. However, in some regions, there are slight reductions in the CL
intensity near the tip of the stripes, causing an appearance of two lobes in the color
map. This issue is unresolved but could be attributed to the fact that there is more
surface recombination near the tip of the stripes. Also, some stripe locations have
significantly reduced CL signal and these locations are indicated the red dotted circles.
These could be a result of threading dislocations that reach the nanostripe surfaces or
some other unknown localized defects. However, these non-radiative recombination
centers are in the minority and most stripes have strong CL intensity.
Because there are fairly large discrepancies in the absolute emission intensity of
the CL spectra at each location, the differences of the emission wavelength at each
location is difficult to visualize. Figure 4.3 shows the same CL line scan measurement
shown in figure 4.2 but the spectra are normalized to unity at each location. In other
words, the integral of each spectrum is set to unity. The line that traverses each CL
spectrum shows the wavelength at the maximum CL intensity for each spectrum.
82
Figure 4.3 CL line scan of the same nanostripes MQW structures shown in figure 4.2, but
the spectrum at each location is normalized to the same peak intensity. (a) Nanostripes
with 300 nm spacing and (b) nanostripes with 500 nm spacing. A line that goes across
the CL mapping represents the wavelength of the peak intensity in the spectra.
The emission wavelengths for both stripes are around 560 nm. The stripes with
300 nm spacing have generally uniform wavelength emission with fluctuation in peak
wavelength in the order of less than 10 nm within each stripe and among different
83
stripes. The emission from stripes with 500 nm are less uniform and the fluctuations in
the wavelength emission are within 20 nm. These data show that the MQW grown on
nanostripe structures is significantly more uniform than those grown on micrometer-
sized stripes structures where more than 50 nm of wavelength difference is observed
from tip to edge of the stripe [2]. As the size of the templates is reduced to sub-micron
scale, the dimensions may become smaller than the diffusion length of the precursors
and therefore the variation of mass transport profile may become negligible. Therefore,
the small sizes of these templates eliminate the concern of non-uniform active region
growth on non-planar templates and the luminescence spectra can be as narrow as
those for a planar structure.
4.2 Transmission Electron Microscopy (TEM) characterizations
In this section, we use TEM to analyze the structural properties of the MQW
structure in the atomic scale. TEM is a microscopy technique where e-beam is
transmitted through an ultra-thin sample with a thickness typically less than 100 nm. As
the e-beam pass through the thin sample, it interacts with the crystal lattice and gets
scattered as it exits the sample. The transmitted and scattered e-beam is then
manipulated by the electron optics behind the sample and a magnified image is
projected onto an imaging device. TEM images give atomic scale resolution of the
lattices and can be used to further study the structural properties of the nanostripes
MQW structures.
84
Figure 4.4 Schematics of TEM operation in (a) TEM mode and (b) STEM mode [3]
Figure 4.4 shows the schematics of the two of the most important operation
modes in a TEM: one is the HRTEM phase contrast imaging and the other is the STEM Z-
contrast imaging. In phase contrast imaging, a parallel beam illuminates the sample.
Some electrons are diffracted by the crystal lattice by an angle alpha while some are not
scattered. An electromagnetic lens located behind the specimen focuses the exit beam.
On the image plane, either constructive or destructive interference between the
scattered beam and un-scattered beams occur, which gives the contrast of the lattice.
85
In Z-contrast imaging, a finely focused e-beam called a probe beam with a spot
size in the order of a couple of angstroms is scanned across the sample. When the probe
is scanned to a location close to an atomic column, it is strongly scattered as it exits the
sample and it is detected by an annular detector. When the probe is scanned to a
location without an atomic column, it is not scattered and it cannot be detected
because of the shape of the annular detector. The larger the Z number of the atomic
column, the larger the scattering cross-section as well as larger scattering strength. As a
result, the atomic column with larger Z number gives the larger signal and brighter
contrast.
Both of the operation modes requires an ultra-thin specimen. The thin sample
was prepared by focused ion beam (FIB) method in this study. Because our
nanopatterns are generated by e-beam lithography, only a small regions in the order of
300 by 300 microns are patterned. Therefore, standard TEM sample preparation
techniques such as ion milling and electropolishing cannot be used as they require the
area of interest to be located throughout the wafer. FIB enables us to select a region of
interest precisely.
There are two main processes associated with the FIB and TEM sample
preparation. One is ion milling and another is carbon or platinum deposition. Milling
occurs when high energy and finely focused ion beam is incident on the substrate and
causes sputtering of the material. The deposition occurs via FIB assisted chemical vapor
deposition. For example, Pt-containing precursors can be introduced into the chamber.
When a specific area of the substrate are subject to FIB scanning, the precursors that
86
are adsorbed on the surface will decompose to non-volatile Pt, which will stay on the
surface and volatile components which will desorb and leave the ambient. These layers
are important for some of the processes, as described in the following.
Figure 4.5 Schematics of process steps of TEM sample preparation by using FIB liftoff method [4]
Figure 4.5 shows the schematics of the process steps of TEM sample preparation by
using FIB liftoff method and each step is described as follows:
87
1. Deposit Carbon on the substrate surface at the area of interest. The dimension
of this carbon layer is usually in the order of 10 microns by 2 microns.
2. Use ion milling to mill two sides of the carbon layer. The depth of the milling is in
the order of 10 microns.
3. Mill the bottom and right part of the thin material. On the left side, only mill half
of the material so the thin piece is still attached to the bulk
4. Insert a probe and attach the tip of the probe to the right side of the thin piece.
The tip of the probe is attached by depositing Pt around the tip of the probe
5. Mill the rest of the left part of the suspended film so that the thin piece is
detached from the bulk.
6. Attach the thin piece on a TEM grid by depositing Pt near the junction
7. Detach probe from the thin piece by milling
8. Final thinning on the TEM grid and surface cleaning.
One of the significant disadvantages of the FIB method is the beam damage it
induces during the ion milling process. The rule of thumb is 1 kV of FIB acceleration
voltage induces 1 nm of damage. Therefore, it is crucial that in the final thinning
process, the low kV ion beam is used to minimize the beam damage. Another important
process step in FIB of nanostripes structures is the ex-situ Pt deposition on the
structures. As discussed earlier, the in-situ carbon deposition requires the ion beam
bombardment in the initial stage. We found that, for non-planar structures, the carbon
is deposited on the bottom and it slowly fills up the gap. As the top part of the stripes
88
are not protected in the beginning, they could be damaged. Therefore, 20 nm of carbon
and 20 nm of Pt are deposited on the nanostripes sample by sputtering prior to the FIB
treatment.
Figure 4.6 STEM HAADF images of cross-section on nanostripes with the {10-11} facet.
(a) the whole nanostripes region, (b) magnified image of the region enclosed by the
dotted line in (a), (c) magnified image of the region enclosed by the dotted line in (b)
Figure 4.6 shows the STEM HAADF images of the cross-section of the nanostripe
MQW structures. Figure 4.6 (a) shows the cross-section of the entire nanostripe MQW
structures. 3 pairs of QWs are clearly visible and are indicated by the arrows. The
contrast of the QWs are due to the difference in the average Z number of the materials,
discussed previously, and InGaN having a larger average Z number shows brighter
contrast. The materials on top of the nanostripes structures are ex-situ carbon, ex-situ
Pt, and in-situ carbon, successively. Figure 4.6 (b) shows the magnified image of the
region enclosed by the dotted line in figure 4.6 (a). We observed a thin region of dark
contrast located near the apex of the nanostripes. A dark contrast in the STEM image
89
can be either a region with lower Z number or it could be related to defects such as
threading dislocations. Figure 4.6 (c) shows a magnified image of this region with dark
contrast, also indicated by the region enclosed by the dotted line. No defects are
observed in this region and the contrast should be the difference in Z number. We
conclude that this region has minimal In incorporation.
There could be two reasons for the reduced In incorporation near the apex of
the nanostripes. One reason is related to the growth kinetics where the In desorption is
enhanced near the tip of the structure. However, this explanation seems to contradict
with our previous studies on MQW structures grown on nanorods and nanosheets
structures where the c-plane top surface is not pinched off. In the Z-contrast images
shown in those studies, there was no obvious decrease in the In composition in the QW
grown on the top c-plane surface [5] [6]. The other reason could be that the In
incorporation near the apex causes significant enhancement of strain. In order to reduce
the strain, the In was removed from the region and desorbed from the surface.
Figure 4.7 Important steps in the strain simulation of using TiberCAD software
90
We used a commercial software, TiberCAD, to simulate the strain distribution
and clarify the growth mechanisms of InGaN QW near the apex of the nanostripe
structures. The simulation is based on continuum elasticity theory implemented by the
finite element method. Figure 4.7 shows the important steps to simulate the structure.
First, we define the geometry of the nanostripes QW structures. We note that in our
case, a 2D simulation is sufficient because the 3
rd
dimension in the [11-20] direction can
be considered to be infinite in the nanostripes structure. Figure 4.7 (a) shows the
geometry of an SQW grown on the nanostripes structures. In these simulations, we do
not consider the interaction between the overgrown region and the growth mask as
their interaction mechanisms is not well understood. The software also does not have
the capability to simulate this interaction and this is a topic that we can explore in future
studies. Then we generate a mesh using GMSH software, an open-source software that
automatically generates a triangular mesh, shown in figure 4.7 (b). We set the bottom of
the structures to have a fixed grid point because this is the area that is grown epitaxially
on the substrate. All other grid points are allowed to relax to achieve the minimum
energy. The next step is to simulate the strain using the TiberCAD software. In the
program, we set the simulation parameters such as the physical constants of each
region, for example, the In composition, the lattice constants and the stiffness
constants, etc. Then the output was visualized by Paraview, another open-source
software package for viewing scientific data.
91
Figure 4.8 FEM simulation of strain in nanostripes MQW structures near the apex. (a)
strain energy density, (b) strain XX, (c) strain YY, and (d) strain XY.
Figure 4.8 shows the simulated strain distribution near the apex of MQW
structures grown on nanostripes. The QW thickness is 5 nm and the QB thickness is 15
nm. The indium composition is set to 30% in the QW region. Figure 4.8 (a) shows the
strain energy density map. It shows that the strain energy density is lower near the apex
than on the sides of the pyramid, except for at the very tip, where the strain is larger
than on the sides. Figure 4.8 (b) to (c) shows the distribution of the three strain
components, which are the normal strain in the x and y-direction, e xx and e yy, and the
92
shear strain, e xy, respectively. We observe that the strain distribution is significantly
altered near the apex of the structure.
In order to interpret the strain distribution, we calculate the deformation of the
lattices based on the displacement field, which is also calculated by the software. One
difficulty of plotting the deformation is that the displacement vectors are calculated in a
triangular mesh but the visualization of the lattice deformation requires a rectangular
grid. We were able to extract the calculated displacements by saving the data file in
a .csv format and a .vtk format in Paraview. In the .csv format, the coordinate of the
nodes and the displacement field data are included. However, in order to convert it to
the rectangular grid, we need to know the connectivity of the nodes. We then open
the .vtk file. There is one segment of the data file with 3 columns where each row
correspond to the coordinate ID of the nodes. Using this information, we can convert
the data from a triangular grid to rectangular grid using the PDE toolbox in Matlab with
the command tri2grid. We set the new lattice to consist of square lattice where the
lattice constant in the x and y-direction are equal. Using a square lattice helps us to
visualize the deformation.
93
Figure 4.9 Schematics of the lattice deformation of the nanostripes MQW structure near
the apex.
Figure 4.9 shows the lattice deformation of the nanostripes MQW structure near
the apex. We note that we deliberately increase the deformation by 10 fold so that the
deformation is more obvious. MQW grown on the side of the nanostripes have a similar
strain state as those grown on the planar structures. It is mostly dominated by the shear
strain and the InGaN lattice is squeezed diagonally to fit the semi-polar plane of the
94
GaN. However, near the top the structure, the shear strain is significantly reduced and
replaced by the 2 normal strains. Near the first GaN/InGaN interface, the InGaN is
constrained more in the x-direction and allow to relax in the y direction. This is because
the significant constraint placed by the underlying GaN. Near the other interface, which
is the InGaN/GaN interface, the strain is dominated by the y direction while the x
direction is allowed to relax. This is because this region is constrained by the GaN
barriers on the two sides of the QW while it is less constrained in the x direction.
The strain simulation shows that the strain energy density is lower near the
GaN/QW interface while is larger near the QW/GaN interface. However, the lack of In
incorporation runs along a line from the GaN/QW interface to QW/GaN interface, and
therefore, this phenomenon cannot be explained by a strain argument. We then suggest
that the kinetics of growth may explain the lack of In incorporation near the tip. Firstly,
we understand that the In cooperation rate on the c-plane is lower than other planes
due to the bonding configuration discussed in chapter 2. Therefore, the c-plane should
inherently have less In incorporation. Secondly, the c-plane, in this case, is very small, in
the order of a couple of atomic columns, therefore, the adatoms have fewer atoms to
bond to near the apex, which may lead to a larger In desorption rate. The difference
between this study and those done on the nanorods and nanosheets mentioned earlier
is that the c-plane area on those structures is significantly larger than that of the c-plane
on these nanostripes. Therefore the growth rate of the c-plane on the nanorod and
nanosheet structures is significantly larger and the In can be trapped in the InGaN layer
as it grows. However, the growth rate on the pinched off pyramid is limited by the
95
growth rate of the inclined plane, which is significantly lower than that of the c – plane,
as we showed in our previous studies [6] [5]. Therefore, if the In desorption rate
exceeds the growth rate of the InGaN film, the indium incorporation near the apex can
be significantly reduced.
In the following, we do some further analysis of the simulation results to
highlight some inhomogeneous strain distribution in the nanostripes structures. A more
intuitive way to represent the strain distribution is setting the normal strain components
to lie along and perpendicular to the nanostripes facets. In the nanostripe case, we can
set the principle direction x’ and y’ to be in the {10-11} and the {11-23} direction. The
transformation of the strain components by the change of the coordinate system are
derived by Romanov et al., with the following expressions [7]:
(
𝑒 𝑥𝑥
′
𝑒 𝑦𝑦
′
𝑒 𝑥𝑦
′
)=(
cos
2
𝜃 sin
2
𝜃 sin2𝜃 sin
2
𝜃 cos
2
𝜃 −sin2𝜃 −sin2𝜃 2
sin2𝜃 2
cos2𝜃 )(
𝑒 𝑥𝑥
𝑒 𝑦𝑦
𝑒 𝑥𝑦
)
Where 𝑒 𝑥𝑥
′
, 𝑒 𝑦𝑦
′
, and 𝑒 𝑥𝑦
′
are the strain components in the new coordinate system.
96
Figure 4.6 Normal strain (a) e xx’ and (b) e yy’ in the coordinate system with x’ and y’
parallel and perpendicular to the facet surface.
Figure 4.10 shows the normal strain 𝑒 𝑥𝑥
′
, and 𝑒 𝑦𝑦
′
in the new coordinate system.
In the figure we also indicate the directions of the new coordinate system with respect
to the regular coordinate. The QWs near the apex are relaxed in both e xx and e yy, as the
strain map shows reduced strain magnitude compared to the side of the stripes.
Consequently, the GaN QB on top of the QW near this region experiences tensile strain
in the x’ direction to accommodate the mismatch. The expansion of the GaN QB in this
region means that the subsequent QWs grown on this region experience reduced strain.
97
Figure 4.7 exx’ for 1
st
, 2
nd
and 3
rd
QW as a function of vertical distance y from the 3
rd
QW. exx for GaN barrier layer underneath the 3 QWs are also plotted
Figure 4.11 shows the exx’ for the 1
st
, 2
nd
and 3
rd
QW as a function of the vertical
distance from the apex of the 3
rd
QW. The exx’ values for the GaN QB layer underneath
the QW is also plotted. For the 2
nd
and 3
rd
QW, the compressive strain is reduced near
the apex because the GaN QB layer underneath experience tensile strain. The strain
distribution can be explained conceptually as follows. The InGaN QW near the apex is
partially relaxed, which means that the GaN barrier grown on top needs to expand.
Then because the GaN in the region now has a larger lattice constant, the subsequent
QW grown on it experience less strain.
98
In order to verify the simulation as well as the implication of the difference in the
strain for the 3 QWs, we analyze the lattice images of this structure. The strain
distribution can be obtained from geometrical phase analysis (GPA) of the lattice images
[8]. This method is used to analyze high-resolution lattice images to obtain spatial
dependence of strain values. We consider the lattice image to be composed of 2
periodicities, in our case is (10-10) and (0001) vectors. We estimate the position of the
lattice fringes by taking the inverse Fourier transform of the two diffraction spots in the
reciprocal space. The displacement of the lattices can be obtained from the phase of the
filtered lattice fringe images and these displacement vectors are used to calculate
various strain components.
We begin by writing the image intensity of a lattice image at a position r,
assuming that the image is made up of 2 periodicities:
I(𝑟⃗)= ∑𝐻 𝑔 (𝑟⃗)exp(2𝜋𝑖 𝑔⃗⃗∙𝑟⃗)
2
𝑔
Where H g is the Fourier component of the reciprocal space vectors that is close
to 𝑔⃗. H g is a complex number with the following expression:
𝐻 𝑔 (𝑟⃗)=𝐴 𝑔 (𝑟⃗)exp(𝑖𝑃 𝑔 (𝑟⃗))
If we place a mask around the diffraction spots ±g, meaning to place a filter on the
reciprocal image at regions around the diffraction spots ±g and everywhere else was
forced to zero. Then the inverse Fourier transform of this image is:
99
B
g
(𝑟⃗)=𝐻 𝑔 (𝑟⃗)exp(2𝜋𝑖 𝑔⃗⃗∙𝑟⃗)+𝐻 −𝑔 (𝑟⃗)exp(−2𝜋𝑖 𝑔⃗⃗∙𝑟⃗)
Because B
g
(𝑟⃗) is real, the expression can be written as
B
g
(𝑟⃗)=2𝑅𝑒 [𝐻 𝑔 (𝑟⃗)exp{2𝜋𝑖 𝑔⃗⃗∙𝑟⃗}]
Re-writing this expression in terms of A and P, we have
B
g
(𝑟⃗)=2𝐴 𝑔 (𝑟⃗)cos{2𝜋𝑖 𝑔⃗⃗∙𝑟⃗+𝑃 𝑔 (𝑟⃗)}
Now, for an ideal set of lattice fringes with constant lattice spacing, the phase and the
amplitude are spatially independent. We write that for an ideal case:
B
g
𝑖𝑑𝑒𝑎𝑙 (𝑟⃗)=2𝐴 𝑔 cos{2𝜋𝑖 𝑔⃗⃗∙𝑟⃗+𝑃 𝑔 }
If a region has a slightly different reciprocal lattice vector, which can be caused by
deformation of the lattice, we write the expression for the lattice with a displacement
vector 𝑢⃗⃗ :
B
g
𝑑𝑒𝑓𝑜𝑟𝑚𝑒 𝑑 (𝑟⃗)=2𝐴 𝑔 cos{2𝜋𝑖 𝑔⃗⃗∙𝑟⃗−2𝜋𝑖 𝑔⃗⃗∙𝑢⃗⃗}
Therefore, we obtain that
𝑃 𝑔 (𝑟⃗⃗)=−2𝜋𝑖 𝑔⃗∙𝑢⃗⃗
When the g 1=(10-10) is aligned to the x-direction and g 2=(0001) is aligned to the y-
direction, then
𝑃 𝑔 1
(r)=−2𝜋𝑖 g
1
𝑢 𝑥
𝑃 𝑔 2
(r)=−2𝜋𝑖 g
2
𝑢 𝑦
100
Therefore, the phase of the filtered image directly gives the displacement field, 𝑢⃗⃗. We
note that g 1 and g 2 can be obtained from a reference region where no deformation is
considered. Then the strain can be obtained from the following expressions [8]:
e=(
𝑒 𝑥𝑥
𝑒 𝑥𝑦
𝑒 𝑦𝑥
𝑒 𝑦𝑦
)=
(
𝜕 𝑢 𝑥 𝜕𝑥
𝜕 𝑢 𝑥 𝜕𝑦
𝜕 𝑢 𝑦 𝜕𝑥
𝜕 𝑢 𝑦 𝜕𝑦
)
101
Figure 4.8 (a) HRTEM images of the nanostripe MQW structure near the apex. (b) e xx (c)
e yy (d) e xy strain map.
102
Figure 4.12 shows the strain map calculated using the geometrical phase
analysis. Figure 4.12 (a) shows the HRTEM image of the nanostripes MQW structure
near the tip of the structure. Figure 4.12 (b) shows the normal strain in the x-direction.
We should note that in this figure, the strain is calculated with respect to the lattice
spacing of GaN. A positive value in the figure means that the lattice constant in the (10-
10) direction is larger than that of the GaN in those regions. We observe an expansion in
the lattice constant in the QW region. However, near the tip, the lattice spacing in the x-
direction is similar to that of GaN. Figure 4.12(c) shows the strain in the normal y
direction that is to see the lattice deformation in the (0001) direction. There is some
enhancement in the strain near the apex of the structure. Figure 4.12 (d) shows the
shear strain, which we will explain shortly.
We have noted that the figure 4.12 is not physical strain because it does not take
into account the InGaN QW. In other words, the strain in figure 4.12 is calculated
assuming that the materials are GaN everywhere. The physical strain can be calculated
from the following expressions:
e
xx
𝑝 ℎ𝑦𝑠𝑖𝑐𝑎𝑙 =𝑒 𝑥𝑥
+
𝑎 𝐺𝑎𝑁
−𝑎 𝐼𝑛𝐺𝑎𝑁 𝑎 𝐺𝑎𝑁
e
xx
𝑝 ℎ𝑦𝑠𝑖𝑐𝑎𝑙 =𝑒 𝑥𝑥
+
𝑐 𝐺𝑎𝑁
−𝑐 𝐼𝑛𝐺𝑎𝑁 𝑐 𝐺𝑎𝑁
e
xy
𝑝 ℎ𝑦𝑠𝑖𝑐𝑎𝑙 =𝑒 𝑥𝑦
e
zz
𝑝 ℎ𝑦𝑠𝑖𝑐𝑎𝑙 =
𝑎 𝐺𝑎𝑁
−𝑎 𝐼𝑛𝐺𝑎𝑁 𝑎 𝐺𝑎𝑁
103
Where a and c are the lattice constant of g = (10-10) and g = (0001) respectively.
However, we do not know the In composition precisely at each location, which prevents
us from calculating the physical strain precisely. We estimate the In composition in the
following way: Just like the strain calculations we did for figure 4.12 where we assume
that the material is GaN everywhere, we do a series of calculation assuming that the
material is InGaN everywhere, where the In composition is swept from 0.01 to 1. Then
we calculate the strain energy density at each location for each In composition using the
following expression:
dU=
1
2
(𝑒 𝑥𝑥
𝜎 𝑥𝑥
+𝑒 𝑦𝑦
𝜎 𝑦𝑦
+𝑒 𝑧𝑧
𝜎 𝑧𝑧
+𝑒 𝑥𝑦
𝜎 𝑥𝑦
)
where the 𝜎 is the stress tensor, which can be calculated from the following equations:
𝜎 𝑥 𝑥 =𝐶 11
𝑒 𝑧𝑧
+𝐶 12
𝑒 𝑥𝑥
+C
13
𝑒 𝑦𝑦
𝜎 𝑥𝑥
=𝐶 12
𝑒 𝑧𝑧
+𝐶 11
𝑒 𝑥𝑥
+C
13
𝑒 𝑦𝑦
𝜎 𝑦𝑦
=𝐶 13
𝑒 𝑧𝑧
+𝐶 13
𝑒 𝑥𝑥
+C
33
𝑒 𝑦𝑦
𝜎 𝑥𝑦
=𝐶 44
𝑒 𝑥𝑦
Where C is the elastic stiffness tensor.
Now strain energy density map is obtained for each In composition. At each
location, we choose the In composition that results in the smallest strain energy density.
104
Figure 4.13 (a) In composition map (b) the e xx strain map (c) e yy strain map (d) e xy strain
map
Figure 4.13 (a) shows the estimated In composition map and Figure 4.13 (b), (c)
and (d) shows the physical strain map of e xx, e yy, and e xy respectively. The lack of In
composition near the apex of the QWs agrees with the Z-contrast images shown in
Figure 4.6. The strain components generally agree with the simulation, except for near
105
the tip of the stripe where In composition, in this case, is significantly lower, which
results in a large eyy just near the tip of the stripe.
Figure 4.14 (a) e’ xx and (b) e’ yy of MQW nanostripes structure.
We again use the transformation of the coordinate to interpret the results,
which is shown in figure 4.14. We find that there is some significant difference between
the e’ xx components on the left and on the right of the stripe. On the right, InGaN QW
has large compressive strain while the GaN has a small tensile strain, which agrees with
the simulation. However, on the left, the InGaN QW is less compressively strained, while
GaN QB is more tensely strained than on the right. Closer examination shows that the
QW near the tip on the left is relaxed, which could be the reason why the GaN barrier is
more tensely strained. This agrees with the simulation results that the more InGaN QW
is relaxed, the more GaN is strained. However, the difference is the symmetry of the
106
strain distribution is skewed to the left and this issue is an interest for further
investigation.
To summarize this chapter, we used CL measurement to show the high spatial
uniformity of light emission within one nanostripe and among different stripes. These
results show significant improvement in uniformity compared to MQW grown on micron
scale structures. We also used TEM to study the structural properties of the MQW
structure grown on the nanostripes. We showed no defects in the structures which
agree with the high radiative recombination efficiency shown in chapter 3. We observed
an interesting feature near the apex of the structures where the In incorporation is
significantly reduced. We conducted strain simulation of a similar structure and showed
that the strain is not significantly increased near the apex of the structure. Therefore,
we attribute the lack of In composition to the kinetics of the growth. The combination of
high In desorption rate on c-plane, high adatoms desorption rate at kinks, and the slow
growth rate could be the reason for the lack of In incorporation. We also used
geometrical phase analysis to obtain the strain map from TEM lattice images and we
showed that the GaN barrier layers can be strained due to partially relaxed InGaN near
the tip of the stripes, which agree with the simulation results.
107
4.3 Chapter references
[1] S.-P. Chang, J.-R. Chang, K.-P. Sou, M.-C. Liu, Y.-J. Cheng, H.-C. Kuo and C.-Y. Chang,
"Electrically driven green, olivine, and amber color nanopyramid light emitting
diodes," Optics Express, vol. 21, no. 20, pp. 23030-23035, 2013.
[2] W. Feng, V. V. Kuryatkov, A. Chandolu, D. Y. Song, M. Pandikunta, S. A. Nikishin
and M. Holtz, "Green light emission from InGaN multiple quantum wells grown on
GaN pyramidal stripes using selective area epitaxy," Journal of Applied Physics, vol.
104, p. 103530, 2008.
[3] E. Abe, "Electron microscopy of quasicrystals - where are the atoms?," Chem Soc
Rev, pp. 6787 - 6798, 2012.
[4] B. Myers, "TEM sample preparation with the FIB/SEM," [Online]. Available:
http://www.nuance.northwestern.edu/docs/epic-
pdf/TEM%20Sample%20Preparation_BDM_2009_abridged.pdf.
[5] T.-w. Yeh, Y.-T. Lin, B. Ahn, L. S. Stewart, P. D. Dapkus and S. R. Nutt, "Vertical
nonpolar growth templates for light emitting diodes formed with GaN
nanosheets," Applied Physics Letters, vol. 100, p. 033119, 2012.
[6] T.-W. Yeh, Y.-T. Lin, L. S. Stewart, P. D. Dapkus, R. Sarkisian, J. D. O'Brien, B. Anh
and S. R. Nutt, "InGaN/GaN Multiple Quantum Wells Grown on Nonpolar Facets of
Vertical GaN Nanorod Arrays," Nano Letters, vol. 12, no. 6, pp. 3257-3262, 2012.
[7] A. E. Romanov, T. J. Baker, S. Nakamura and J. S. Speck, "Strain-induced
polarization in wurtzite III-nitride semipolar layers," JOURNAL OF APPLIED PHYSICS,
vol. 100, p. 023522.
[8] M. Hytch, "Geometrical phase analysis of high resolution electron microscope
images," Scanning Microscopy, vol. 11, p. 53, 1997.
[9] S. Khatsevich, D. H. Rich, X. Zhang, W. Zhou and P. D. Dapkus, "Temperature
dependence of excitonic recombination in lateral epitaxially overgrown
InGaN/GaN quantum wells studied with cathodoluminescence," Journal of Applied
Physics, vol. 95, p. 1832, 2004.
[10] P. Potts, A handbook of Silicate Rock Anaylsis, New York, 1987.
108
Chapter 5 Fabrication and characterization of nanoLEDs
5.1 Introduction to conventional LED structures and the challenges
In the previous chapters, we have developed the growth conditions for high-
efficiency long wavelength emitting InGaN QW structures on the {10-11} nanostripes. In
this section, we present the development and demonstration of nanoLED structures. In
the following, we briefly introduce some basic concepts in the conventional c-plane
InGaN/GaN LEDs structures before we present the work on nano-LEDs
Figure 5.1 shows the fabrication process of the conventional nitride-based LED
structures grown on (0001) sapphire. After the growth of InGaN/GaN MQWs, what
follows immediately is the growth of AlGaN electron blocking layer and p-type GaN. The
function of the electron blocking layer will be described shortly. After the growth, the
wafer is subject to N 2 thermal annealing to activate the p-type GaN. This activation step
is actually one of the major breakthroughs of the nitride-based LEDs. Nakamura et.al.
Found that as grown p-GaN acceptors are passivated by H and a thermal annealing in N 2
ambient at temperatures above 700 degrees can remove the atomic hydrogen from the
acceptor-H neutral complexes [1]. The resistivity was reduced from 10
6
Ω∙cm to around
2 to 8 Ω∙cm after the N 2 thermal annealing step [1].
109
Figure 5.1 Fabrication process of conventional c-plane nitride-based LEDs.
The next step of the process is mesa etching of the LED wafer to expose the N-
GaN, illustrated in Figure 5.1 (b). ICP RIE is the most popular way to etch GaN because of
its scalability, high etch rate and low surface damage [2]. The plasma are generated
remotely and they are introduced to the reactor chamber by diffusion, thus achieving
high etch rate while producing low surface damage [2]. A mixture of Cl 2, H 2 and Ar is
commonly used as the precursors [2]. Photolithography is used to define the mesa
region, and the wafer is etched selectively by ICP-RIE.
110
After the mesa etching, the wafer is ready for contact deposition. ITO is usually
used as a transparent contact and a spreading layer for p-GaN because of low resistivity
(~10
-4
Ω∙cm) and high visible light transmittance (> 80%) [3]. However, depositing ITO
directly onto p-GaN results in Schottky contacts and the resulting operating voltage was
too large for practical device applications [4]. The contact resistance can be reduced
significantly by the growth of n+-InGaN/GaN short period superlattice capping layer and
the carriers can be injected into the p-GaN through tunneling [4]. Another way to
reduce the contact resistance is by depositing a thin Ni layer before the ITO deposition.
After thermal annealing, the p-GaN/Ni/ITO structures showed ohmic characteristics
while maintaining a high transparency [5]. Finally, Cr/Au is deposited on both N-GaN and
the ITO layer as a contact pad, shown in figure 5.1 (d). A forward bias is then applied to
the LEDs. As the electrons and holes diffuse across the junction, they are captured by
the QW and recombination occurs. When the carrier concentration in the QW is large
enough, the radiative recombination dominates and light is generated from the QW
region. The details of the various recombination mechanisms are discussed shortly.
111
Figure 5.2 Comparison of Auger recombination and electron leakage within a LED energy
diagram [7]
One of the major problems associated with nitride based LEDs is the efficiency
droop which is the decrease in the radiative recombination efficiency with increasing
current injection. The origins of the efficiency droop can be a combination of Auger
recombination and electron overflow, which are summarized in figure 5.2. All three
contributions will be discussed in the following one at a time.
First, we discuss the effect of current overflow on the efficiency droop. Under
high current injection, some electrons go through the active regions without
recombination and leak to the p-side. A common way to reduce the electron leakage is
to grow a layer of p-type AlGaN around 20 nm thick right after the MQW growth.
Because the band gap of AlGaN is larger than that of the GaN, there is a discontinuity in
the conduction band at the interface. When electrons transport to that interface, the
built-in barrier prevents the electrons from flowing over to the p-GaN side. However,
there are a few problems with the AlGaN. It will present barriers for holes to flow from
112
the p side to the active region and therefore reduces the hole injection efficiency. For
nanostructure growth specifically, because the diffusion length for the Al adatoms is
very short, the polycrystalline deposition on the mask region could be a potential
problem. Also, the AlGaN deposited on the susceptor will also be difficult to remove.
Therefore, care should be taken when AlGaN is grown for nanostructure growth.
Secondly, we discuss the contribution of the Auger recombination to the
efficiency droop. Auger recombination has a cubic power dependence on the carrier
density. When the carrier density is high, Auger recombination starts to dominate and
radiative recombination is suppressed. In order to decrease the current density inside
one QW, MQWs are usually grown to divide carriers into multiple wells and reduces the
carrier density in each well and thus reduces the Auger recombination. However,
because of the low mobility of holes in the nitride system, most of the holes are trapped
in the first QW closest to the p region. Therefore, most carriers recombine in that first
well and carriers cannot be divided evenly among all wells and the effect of having
MQW is reduced [6].
A recent study by Zhou et al. show that a staircase quantum barrier structure can
improve the carrier distribution among different quantum wells [6]. Figure 5.3 (a) show
the novel structure, where from the n side, the quantum barrier In content gradually
increases. With this structure, hole experience lower barriers as they transport from the
p-side to the n-side. The better transport of holes significantly increases the hole
concentration in wells further from the p-side. Figure 5.3 (b) shows that the droop
113
phenomenon is greatly reduced. Similarly, the structure can be grown on nanostripes to
improve the carrier distribution in the MQWs. A comparison between this structure and
original structure can verify the feasibility of this approach.
Figure 5.3 Schematic diagram of LEDs with thin GaN QBs and InGaN stair case QBs (left),
EL properties comparison of GaN-QB and InGaN SC-QB LEDs (right) [6]
Another possible way to decrease the current density in the quantum wells is to
increase the well thickness or grow a double heterostructure (DH). For quantum wells
that are grown on the c-plane, the presence of large piezo electric field separates the
electron and hole wave functions. Therefore, the quantum well needs to be sufficiently
thin to increase the overlap integral. For nanostripes with semi-polar surfaces, the
piezoelectric field is expected to be significantly lower so thicker QWs or even double
heterostructure may be realized. Gardner et al. show that a 13 nm double
heterostructure LED reaches maximum quantum efficiency at 200 A/cm
2
compared to
10 A/cm
2
for multiple quantum well with 2.5 nm thickness [18]
114
Figure 5.4 Relative quantum efficiency of two 2.5 nm QW (a) six 2.5 nm QW (b) and 13
nm DH (c) LEDs vs current density [18]
Double heterostructures can be grown on the nanostripe template. However,
there is a concern that if the thickness of the active region exceeds the critical thickness,
defects will be generated which may reduce the radiative efficiency significantly.
Therefore, this strategy can work for low Indium content quantum wells and may not be
suitable for long wavelength emitting LEDs.
115
5.2 Development of {10-11} nanostripes based nanoLEDs
Figure 5.5 (a) Schematics of the cross-section of the nanoLED structure. (b) SEM image of
the nanoLED cross-section.
In this section, we present the development and the demonstration of the
nanoLEDs grown on the {10-11} nanostripes. We show the challenges we faced during
the development of the nanoLEDs and how we overcome these. The demonstration of
the nanoLEDs is necessary to provide a proof of principle of the nanoLED approach. In
116
the end of this section, we quantify the external quantum efficiency of the nanoLEDs
and highlight some limitations of this approach.
Figure 5.5 (a) shows the schematics of the cross-section of the LED structure. The
growth conditions for the nanostripes and MQW structures are described in the
previous chapters. Subsequent to the MQW growth, p-type GaN is grown at 875 C in N 2
ambient with NH 3 flow rate of 9 slm, TEG flow rate of 40 to 80 sccm, and CP 2Mg flow
rate of 90 sccm (bath temperature of 60 C). Post-growth annealing was performed in a
furnace in N 2 ambient at 780 C for 15 min to activate p-type doping to remove the H
passivation of the acceptors as discussed in the previous section. The flow rate of the N 2
is around 100 sccm. We put the sample inside the Furnace at room temperature and
then turn on the furnace. There is around 400 C offset of furnace setting temperature to
the actual temperature. After calibration, we found that the setting on the furnace is
usually 330 C to attain an actual temperature of around 780 C. The time that is takes to
ramp up the temperature from the room temperature to 780 C is usually 20 min, and
the actual annealing time at the setting temperature is 15 min. Therefore, we turn off
the furnace 35 min after we turn on the Furnace. Finally, we wait till the furnace to cool
to the room temperature and take out the sample. Figure 5.5(b) shows the cross-section
of one of the device structures, showing p-type contrast.
The fabrication of p-type contact requires 2 step metallization. The first step is
Ni/Au ohmic contact, and the 2
nd
step is Ti/Au contact pad and spreading layer. We first
describe the overall process and then give more important details later. Ni/Au with a
nominal thickness of 10 and 20 nm are deposited by e-beam evaporation on the
117
photolithography defined area followed by lift-off using Acetone sonication. Then the
sample is annealed in O 2 ambient at 500 C in the furnace. The flow rate of the O 2 is 100
sccm and the temperature setting is 125 C. Ti/Au (10/100 nm) p-type contact pad with
“fingers” that are extended to the nanostripes device regions is deposited to spread the
current to a 300 by 300 μm
2
patterned area, as shown in figure 5.5 (d). We deposit the
contact pad outside of the nanostructure regions to avoid probing directly on the
nanostripes and damaging the structures. For n-type contact, part of SiN growth mask
next to the device region is removed to expose the n-type GaN, and Indium is pressed
onto the exposed area to be used as N-type contact pad without annealing.
As discussed in chapter 1, we have 8 nanoLEDs samples with an area of 300 by
300 μm
2
, as shown in figure 5.5 (c). Each sample has a “skirt” region with 30 microns in
width and 40 microns away from the LED area. This skirt region is a wide window region
which is designed to absorb excess precursors from the mask regions (outside the skirt
region) and prevent them from reaching the nano pattern region. Without the skirt
region, there will be a significant gradient in growth rate from outer to inner regions of
the nanostructures pattern due to the effect of diffusion of precursors from the outside.
Originally, the skirt region was continuous, meaning the skirt region had a shape of a
hollow square. However, we found that if the “fingers” go over the skirt regions, it
introduces significant leakage current as the IV characteristics are ohmic rather than
rectifying. Therefore, we revised the skirt region design to have apertures in them so
that the figure will not be electrically connected to the skirt region. Readers may also
notice that there are apertures not only on the top side where the spreading fingers
118
come in but also on all sides. The reason for this design is to have symmetric growth and
increase the growth uniformity.
We also found that ohmic contact metal must fully cover the 300 by 300 μm
2
patterned region, but not overlap the skirt regions. If the ohmic contact does not fully
cover the sample area, we suspect that the region that is not covered by metal can be
oxidized during O 2 annealing process and turn to n-type GaN. When the spreading layer
is in contact with these regions, it leads to high nonradiative leakage current.
Finally, the thickness of the metals is important. We found that ohmic contact
with a nominal thickness of 10 and 20 nm for Ni and Au, respectively are necessary to
have good spreading along the stripes. The Ti/Au of 10 and 100 nm are necessary to
spread the current to all the stripes.
119
Figure 5.6(a) and (b) show the IV characteristics of nanoLED devices with 300 and 500
nm spacing, respectively. (c) and (d) show the cross-sectional SEM of the nanoLED
devices with 300 nm and 500 nm spacings, respectively
However, even with these preliminary design improvements, the device failed to
work. Figure 5.6 (a) and (b) show the IV characteristics of early nanoLEDs on a regular
scale and semilog scale respectively. Both samples had large reverse bias leakage
current and premature turn-on voltage. The nanoLEDs with 500 nm pitch even had
larger current under reverse bias than forward bias. Figure 5.6 (c) and (d) shows the
cross-sectional SEM images of the nanoLEDs with 300 nm spacings and 500 nm spacings
respectively. For both of these images, p-type contrast can be observed, and the PN
junction is established. However, there are a couple of concerns. Firstly, we suspect that
120
there may be leakage current that goes through the SiN mask. The design thickness of
deposited SiN is 20 nm. The measured thickness of the SiN is around 12.5 nm. The
growth processes at high temperature may have annealed the SiN layer and led to the
reduction of the SiN thickness. Regardless, the 12.5 nm of SiN should be sufficient to
prevent tunneling of carriers. Another observation is that the p-GaN thicknesses are 40
nm and 60 nm for the nanoLEDs with 300 and 500 nm spacing respectively. These
thicknesses are thinner than the regular p-GaN thickness of approximately 200 nm used
for devices grown on c-plane GaN [8].
Figure 5.7 (a) IV characteristics of 3 nanoLEDs on semilog. (b) Cross-sectional SEM of
nanoLEDs with a p-GaN thickness of 120 nm and SiN thickness of 60 nm. (c) micrograph
of green emitting nanoLEDs.
121
Figure 5.7 (a) shows the IV curves of nanoLEDs with an improved design. The
blue curve shows the results from the figure 5.6 (a) and the spacing of this device is 500
nm. The green and red curves are new IV data based on improved structures which will
be described shortly. First, we increased the SiN thickness from 20 nm to 60 nm to
decrease the leakage current. The increased SiN thickness requires longer CF 4 RIE etch
time. We changed the e-beam resist from A2 to A4 resist to increase the PMMA
thickness so that it can withstand the increased RIE etching time. The number after the
letter “A” of the PMMA specification is related to the concentration of the PMMA in the
solvent: the larger the number, the larger the concentration and the thicker the
resulting film. The etching time increased from 12 s to 45 s to fully open the stripe
openings. The rest of the growth and fabrications follow the same procedures.
We found that the reverse bias leakage current was reduced by two orders of
magnitude. The turn-on voltage also increased, and the IV curve becomes more
asymmetric compared to the blue curve. Again, the blue curve is the same from figure
5.6 (b) for the nanostripes with the 500 nm spacing. The SiN thickness for this sample
was 12.5 nm and the p-GaN thickness was 60 nm. It is very likely that there was indeed
leakage current through the thin SiN mask in the previous samples that created low
shunt resistance. By increasing the shunt resistance, we also reduced the current under
low forward bias, which is evident in the green IV curve. There is a “knee” at around 1 V
where junction current starts to dominate over the shunt current. Even though the
leakage current is reduced, the device failed to light up.
122
In the next iteration, we increased the p-GaN thickness from 60 nm to 120 nm
and the IV characteristics of this sample is plotted as the red curve in figure 5.7 (a).
Figure 5.7(b) shows the cross-sectional SEM image of this sample and Figure 5.7(c)
shows the micrograph of this sample lighting up green. This sample is the first
demonstration of the green nanoLEDs for this project.
The IV curve shows a 2 order of magnitude reduction in the reverse bias leakage
current as well as a larger turn on voltage. We can observe in figure 5.7 (b) that the
stripe nanostructures are wider and have coalesced. The ohmic contact is completely
isolated from the growth mask, which may explain the reduction in the leakage current.
There are two reasons for the appearance of this wider structure. As discussed
previously, the PMMA thickness was increased to withstand longer etching time to etch
the thicker SiN. However, we found that it was necessary to increase the e-beam
exposure dose to fully expose the thicker PMMA and have a clean pattern transfer. The
increase in the e-beam dosage inevitably increases the width of the stripe openings,
which in this case is around 250 nm. As a result, the nanostripe templates are in closer
proximity to one another than those grown on a thinner mask. Also, because we
increased the p-GaN thickness, the overall structures is wider, and it caused the grown
structures to coalesce.
The important observation is that the LED emitted light when we increased the
p-GaN thickness. Whether the LED emits light or not depends on the electron and hole
carrier densities in the QW and the radiative recombination efficiencies. When a PN
junction LED is under forward bias, the carrier concentration increases in both QWs and
123
leads to radiative recombination of the carriers and emission of light. However, if the p-
type GaN thickness is thinner than the depletion width, the p-type GaN layer will be
completely depleted and under forward bias, the electron is directly injected into the p-
type contact – the device will behave like a Schottky barrier. In the limit that p-GaN
thickness is zero, the device essentially becomes a Schottky diode, which is a majority
carrier device and the light emission mechanisms becomes completely different from a
PN junction LED.
The first LED demonstrated was based on SiC Schottky diode [9].
Historically, the first LED demonstrated was based on SiC Schottky diode [9].
Figure 5.8 shows the schematics of the band diagram of a Schottky diode under
different bias conditions. Under forward bias, the Fermi level is close to the conduction
band and therefore, the hole concentration is significantly lower, and therefore the
radiative recombination rate will be extremely low. Under larger forward bias, holes can
be injected into the semiconductor regions by tunneling through the surface barrier and
recombine radiatively with the n-type majority carriers in the QW. Therefore, the
Figure 5.8 Schematics of the band diagram of a Schottky diode under (a) zero bias, (b)
forward bias and (c) larger forward bias. Source: Light-Emitting Diodes by E. Fred
Schubert, Cambridge University Press
124
operating voltage of SiC Schottky diode ranged between 10 V to 110 V [9]. Interestingly,
we observed weak light emission when the nanoLEDs is operated at 20 V of forward bias
for the sample with a p-GaN thickness of 40 to 60 nm, while no light emission was
observed at a normal operating voltage of around 5 V. Therefore, there is a strong
evidence that when p-type GaN is too thin, the device functions as a Schottky diode and
the light emission efficiency is extremely low.
In the following, we calculate the depletion width of the nanoLEDs to validate
our understanding further. The depletion width on the p-type side of a PN junction can
be calculated as follows:
𝑥 𝑝 =√
2𝘀 𝑞 (𝑉 0
−𝑉 𝑎 )
𝑁 𝐷 −
𝑁 𝐴 +
1
𝑁 𝐴 +
+𝑁 𝐷 −
Where 𝑥 𝑝 , is the depletion width on the p side, 𝘀 is the dielectric constant, 𝑞 is the
elementary charge, 𝑉 0
is the built-in voltage, 𝑉 𝑎 is the applied bias, 𝑁 𝐷 −
is the ionized
donor concentration on the n-side of the space charge region, and 𝑁 𝐴 +
is the ionized
acceptor concentration on the p-side of space charge region. The built-in voltage is
calculated by:
V
0
=
𝑘𝑇
𝑞 ln(
𝑁 𝐴 +
𝑁 𝐷 −
𝑛 𝑖 2
)
Where 𝑘 is the Boltzmann constant, T is the temperature, 𝑛 𝑖 is the intrinsic carrier
concentration. The intrinsic carrier concentration can be found by:
125
𝑛 𝑖 =√𝑁 𝐶 𝑁 𝑉 exp(−
𝐸 𝑔 2𝑘𝑇
)
Where 𝑁 𝐶 is the effective density of states in the conduction band, 𝑁 𝑉 is the effective
density of states in the valence band, and 𝐸 𝑔 is the band gap. The effective density of
states for wurtzite GaN is:
𝑁 𝐶 =4.3 ×10
14
×𝑇 3/2
𝑁 𝑉 =8.9 ×10
15
×𝑇 3/2
All parameters are known except for the 𝑁 𝐴 and 𝑁 𝐷 , but their concentration can vary
from 10
16
to 10
19
range.
We need to point out that the ionized acceptor concentration of p-type GaN is
much lower than the doping level of Mg due to the high ionization energy of the Mg
acceptor level, which is in the range of 200 meV [12]. The bounded hole concentration
can be found from semiconductor statistics from the following equation [13]:
𝑝 𝑎 =
𝑁 𝑎 1+
1
4
𝑒 −
𝐸 𝑎 −𝐸 𝑓 𝑘𝑇
Where 𝑝 𝑎 is the bound hole concentration at the acceptor sites, 𝑁 𝑎 is the acceptor
concentration, and 𝐸 𝑑 and 𝐸 𝑓 are the acceptor energy level and the fermi level,
respectively. The charge balance equation in a semiconductor can be written as [13]:
𝑛 +𝑛 𝑑 +𝑁 𝑎 =𝑝 +𝑝 𝑎 +𝑁 𝑑
126
Where n and p are the free electron and hole concentration respectively, n d and p a are
the bound electron and hole concentration at the donor and acceptor sites, respectively
and 𝑁 𝑎 and 𝑁 𝑑 are the concentration of acceptors and donors, respectively. In p-type
GaN, 𝑁 𝑑
, 𝑛 𝑑 and n can be considered negligible. The charge balance equation can be
written as:
𝑁 𝑎 =𝑝 +𝑝 𝑎
Consider in the non-degenerate regime:
𝑝 =𝑁 𝑣 𝑒 −
𝐸 𝑓 −𝐸 𝑣 𝑘𝑇
Where E v is the valance band energy. Therefore, the charge balance equation can be
written as follows:
𝑁 𝑎 =𝑝 +
𝑁 𝑎 1+
1
4
𝑒 −
(𝐸 𝑎 −𝐸 𝑣 )−(𝐸 𝑓 −𝐸 𝑣 )
𝑘 𝑇
We make a substitution:
𝑝 1
=𝑁 𝑣 𝑒 −
𝐸 𝑎 −𝐸 𝑣 𝑘𝑇
Then the charge balance equation can be written as
𝑁 𝑎 =𝑝 +
𝑁 𝑎 1+
1
4
𝑝 1
𝑝
After some mathematical manipulation, we arrive at the following equation:
127
4
𝑝 1
𝑝 2
+𝑝 −𝑁 𝑎 =0
The solution of this equation is as follows:
𝑝 =
−1±√1+
16
𝑝 1
𝑁 𝑎 8
𝑝 1
=−
𝑝 1
8
+
𝑝 1
8
√1+
16
𝑝 1
𝑁 𝑎
Figure 5.9 ionized acceptor concentration vs acceptor concentration in p-type GaN for
Ea=200 meV
Figure 5.9 shows the free carrier concentration (ionized acceptor concentration)
as a function of the acceptor concentration in p-type GaN when the acceptor level is 200
128
meV from the top of the valence band. We observe 2 orders of magnitude smaller
ionized acceptor concentration than the acceptor concentration.
However, in the space charge region, the free carrier concentration can be
considered negligible and there is a spatial dependence of the acceptor level due to the
band bending in the space charge region. We again write the ionized acceptor
concentration as follows:
𝑁 𝑎 +
=𝑁 𝑎 −
𝑁 𝑎 1+
1
4
𝑒 −
𝐸 𝑎 −𝐸 𝑓 𝑘𝑇
The valance band and the acceptor level bend downwards from the p-type to the n-
type, meaning that −(𝐸 𝑎 −𝐸 𝑓 ) becomes larger in the space charge region. In the limit
of −(𝐸 𝑎 −𝐸 𝑓 )≫𝑘𝑇 , the denominator in the above expression infinite, and
consequently,
𝑝 ≈𝑁 𝑎 𝑑𝑜𝑝𝑖𝑛𝑔
Considering the built-in potential of around 3 eV of GaN pn junctions, most of the space
charge region satisfy −(𝐸 𝑎 −𝐸 𝑓 )≫𝑘𝑇 condition, and therefore the acceptors in the
space charge region is completely ionized, even though the acceptors are only partially
ionized in the bulk p-GaN layer. Therefore, in the calculation of the space charge width,
we assume the acceptors are completely ionized in the space charge region
129
Figure 5.10 Contour plot of the depletion width on the p – side for different combinations
of ionized donor and acceptor levels.
Figure 5.10 shows the contour plot of the depletion width on the p-side of the
space charge region as a function of the ionized donor and acceptor concentrations in
the space charge region. For an n-type doping in the order of 10
18
cm
-3
, the p-type
doping must be in the order of 10
18
cm
-3
to have a depletion width that is thinner than
40 nm. This level of ionized acceptor level can be achieved in p-GaN grown on c-plane
GaN under high temperature of around 1000 C in H 2 carrier gas [10]. However, our p-
GaN is grown under a lower temperature of 875 C and in N 2 ambient to protect the high
In content QW. In these conditions, the p-GaN can be unintentionally doped n-type with
130
C impurities or N vacancies. Therefore, the net positive charge concentration in the
space charge region can be decreased to the mid-10
17
cm
-3
range. In this case, the p-
GaN thickness must be larger than 100 nm to avoid complete depletion of p-GaN.
Further studies of the doping concentration in the p-GaN layer of the nanostripes LED
structures are required to draw final conclusions.
5.3 Characterization of nanoLEDs
With the demonstration of the nanoLEDs that emit in the yellow range, we
present the optical characteristics of the LEDs. PL setup is modified so that
electroluminescence can be measured without moving any optical components. The
device is accessed by microprobes. A LabView program is written to automate the entire
measurement. However, the following manual procedures are required to set up the
measurement. First, we drive the LEDs at medium current like 50 mA to align the system
to get the maximum signal. Because alignment may take some time, we do not wish to
drive the current too high for too long and risk burning the device. The stage should be
moved in all three directions iteratively until the signal cannot be increased further.
At this stage, we want to test the range of EL intensity we can measure. First, we
drive the current at the highest drive current we wish to measure, for example, 100 mA.
Some LEDs emit strongly and saturate the detector. The conventional way to reduce the
signal is by decreasing the slit width on the spectrometer. However, we found that the
image of the LED chip area that forms on the slit is large, in the order of a couple of
131
mm
2
. Therefore, if we close the slit width to a very small value, only light from a small
area of the LED chip is collected by the spectrometer. Because we would like to know
the characteristics of the light emission from all chip area, we need to keep the slit
open. Another way to reduce the signal is to decrease the applied bias on the
photomultiplier tube to reduce the gain. The range of the voltage that we tune is limited
to between 1000 V and 1300 V to ensure the photomultiplier functions properly.. Then
we can start the Labview programs which automatically acquires the EL spectra at
different drive current.
Figure 5.11 (a) EL spectra of yellow emitting nanostripes LED under the drive current
from 10 mA to 100 mA (b) Peak wavelength and (c) linewidth extracted from the spectra
shown in (a) and plotted against the drive current
132
Figure 5.11 (a) shows the EL spectra of yellow emitting nanostripes LED under
the drive current from 10 mA to 100 mA and (b) and (c) show the peak wavelength and
linewidth extracted from the spectra shown in (a) and plotted against the drive current.
There is a 7 nm blue shift of peak wavelength from 10 to 100 mA of drive current. The
linewidth of 48 nm at around 80 mA of drive current is comparable to that of planar
LED structures [31].
5.4 External quantum efficiency (EQE) measurement
In the previous section, we demonstrated promising EL characteristics of the
nanoLEDs, including small linewidth and small blue-shift. Another important
characteristic of a LED is its EQE. The EQE measurement of the nanoLEDS is necessary
because it is the primary objective of this dissertation, which is to develop high-
efficiency long wavelength emitting nitride nanoLEDs. In order to measure the EQE of a
LED, we need to insert the LED into the integrating sphere to capture as much light as
we can. However, the devices fabricated so far are limited to the probe level, and we
need to package it like commercial LEDs. We have attempted to wire bond the devices
to a TO can, however, there were major sticking issues that the wire would not bond to
the contact pads. One reason is that the contact pad thickness is in order of 100 nm,
which is thin for wire bonding. Also, the contact pad is very rough because it is
deposited on a film of rough polycrystalline GaN, and this roughness can be problematic
for wire bonding. Instead of developing a new packaging technology, we quickly
133
developed a manual procedure to mount the LEDs to a TO can and obtained some
preliminary data.
Figure 5.12 (a) shows the mounting of our nanoLED devices to a TO can. First, we
increase the size of the contact pad from 100 μm
2
to 1 mm
2
by designing a new
photolithography mask. The chip is secured on the TO can with a double sided tape.
Then, we place a piece of thin gold wire on the gold pad, and we press a small piece of
Indium onto the gold wire. Consequently, the gold wire is sandwiched between the In
and the contact pad, and the In acts like glue. The other end of the gold wire is attached
to one of the terminals of the TO can using the same procedures. The n-contact was
connected to the other terminal of the TO can following the same procedure. Figure
5.12 (b) shows the working LEDs attached to aTO can, demonstrating that this packaging
process worked. The IV characteristics were also verified, and no additional series
resistance was observed.
Figure 5.12 (a) Picture of a LED device mounted on a TO can (b) LED on a TO can light up
134
Next, we present the quantitative characterization of the external quantum
efficiency of these nanoLEDs. We set up an automated pulsed integrating sphere
measurement setup, shown schematically in Figure 5.13. We used the Keithley 220
Programmable current source to trigger the LDP-3811 current source. When examining
the response time of the LEDs, the typical rise time was in the 500 μs range, but the
maximum pulse repetition interval of the LDP-3811 is 1 ms, which equates to a
minimum duty cycle of around 50%. Typically we need duty cycles of less than 10 % to
avoid heating of the device. Keithley 220 programmable current source can generate a
trigger with minimum repetition interval of 3 ms, which is more appropriate for use as
the trigger source. However, the maximum current output of the Keithley 220 is 100 mA
while that of the LDP-3811 is 500 mA. Therefore, we still use the LDP_3811 as the
current source because of the range of the drive current attainable.
135
Figure 5.13 Pulsed integrating sphere measurement set up
The output of the current source drives the LEDs which are inserted into the
integrating sphere. The oscilloscope and the box car average were triggered by the
“trigger out” output of the current source. The output of the photodetectors is
amplified by the Pico ammeter, which is a transconductance amplifier that amplifies a
current signal and transforms it to voltage. The rise time of the picoammeter is 1 ms,
which is sufficiently fast to record the actual pulse, as can be seen in the oscilloscope
reading in figure 5.13. Without the amplification, the signal to noise ratio is extremely
136
small, especially at a low drive current. One of the reasons is that the integrating sphere
itself attenuates the signal by around 3 orders of magnitude. The existence of the
amplifier increases the response time of the device, but it can be compensated for by a
longer repetition interval. The analog output of the pico-ammeter is connected to the
boxcar average that integrates the signal. The oscilloscope is used to monitor the
trigger, device response and the gate delay, which is positioned at the flat part of the
device response curve, shown in the figure. The output of the boxcar is connected to a
multimeter to read the voltage.
The photocurrent out of the photodetector is calculated using the following
formula:
𝐼 𝑝 ℎ𝑜𝑡𝑜 =𝑉 𝑚 𝑆 𝑏𝑜𝑥𝑐𝑎𝑟 𝑆 𝑝𝑖𝑐𝑜𝑎𝑚𝑚𝑒 𝑡 𝑒𝑟
Where 𝐼 photo
is the photocurrent measured by the photodetector on the integrating
sphere, 𝑉 𝑚 is the voltage measured by the multimeter, 𝑆 𝑏𝑜𝑥𝑐𝑎𝑟 , is the sensitivity setting
on the boxcar, and 𝑆 𝑝𝑖𝑐𝑜𝑎𝑚𝑚𝑒𝑡𝑒𝑟 is the sensitivity setting on the picoammeter. 𝑆 𝑏𝑜𝑥𝑐𝑎𝑟 is
a ratio of input voltage to output voltage, therefore unit-less. 𝑆 𝑝𝑖𝑐𝑜𝑎𝑚𝑚𝑒𝑡𝑒𝑟 is the ratio of
input current to output voltage and has a unit of A/V. The integrated light output can be
calculated as follows:
𝑃 𝐿𝐸𝐷 =𝐼 𝑝 ℎ𝑜𝑡𝑜 ∫
𝑃 𝑛𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 (𝜆 )
𝑅 (𝜆 )
𝑑𝜆 𝜆
Where 𝑃 𝐿𝐸𝐷 is the integrated light output power out of the LED and has a unit of Watt,
𝐼 𝑝 ℎ𝑜𝑡𝑜 is the photocurrent calculated from the previous equation, 𝑃 𝑛𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 (𝜆 ) is the
137
normalized spectral intensity of the LED light output, and 𝑅 (𝜆 ) is the responsivity of the
calibrated integrating sphere system and has a unit of A/W. Finally the external
quantum efficiency is calculated with following:
𝐸𝑄𝐸 =
𝑃 ℎ𝑜𝑡𝑜𝑛𝑠 𝑒𝑚𝑖𝑡𝑡𝑒𝑑 𝑐 ℎ𝑎𝑟𝑔𝑒 𝑐𝑎𝑟𝑟𝑖𝑒𝑟𝑠 𝑖𝑛𝑗𝑒𝑐𝑡𝑒𝑑 =
𝑃 𝐿𝐸𝐷 /ℎ𝜈 𝐼 /𝑒
Where ℎ𝜈 is the photon energy, I is the drive current.
With the procedures outlined above, we were able to measure the EQE of blue,
green, yellow and amber emitting nanoLEDs. Figure 5.14 shows the integrated EL power
and the EQE of these LEDS. Figure 5.14 (c) shows the photograph of LED emission of
blue, green, yellow and amber emitting nanoLEDs. The blue, green, and yellow LEDs
have similar output power and EQEs, which increase as drive current increase and start
to saturate at high current. The emission is in the 250 μW range and the EQE in the 0.1%
range at 100 mA of drive current. The performance of the amber LEDs is one order of
magnitude lower despite the fact that their emission wavelength is only around 30 nm
longer than the yellow light. The limiting factors for the amber emitting nanoLEDs will
be discussed in the next chapter. Another note is that we also measured a commercial
blue LED in the same system, and the EQE is around 20%. Therefore, the measurement
setup should be relatively well calibrated.
138
Figure 5.14 (a) Integrated EL power and (b) the EQE of blue, green, yellow and amber
emitting nanoLEDS. The legends show the color and peak wavelength of each LEDs. (c)
Photograph of light emission from blue, green, yellow and amber emitting LEDs from left
to right
5.5 Extraction of IQE from the integrated electroluminescence
Because the packaging of these nanoLEDs is in the very preliminary stage, with
no encapsulation and other measures to increase the light extraction, these EQE values
are not indicative of the potential of the nanoLEDs. One way to assess the efficiency is
to measure the IQE. We can we utilize the rate equation model similar to what we used
in Chapter 3 to estimate the IQE of the material by fitting the EL vs current curve. Once
139
the IQE is determined, we can define the limitations such as the light extraction
efficiencies.
The rate equation for carriers in semiconductor QWs by electrical carrier
injection is given by [12]:
𝜂 𝑗 𝐼 𝑞𝑉
𝑄𝑊
=𝐴 𝑁 𝑄𝑊
+𝐵 𝑁 𝑄𝑊
2
+𝐶 𝑁 𝑄𝑊
3
+
𝐽 𝑡 ℎ𝑒𝑟𝑚𝑖𝑜𝑛𝑖𝑐 𝑒 𝑞 𝐿 𝑧
Where 𝜂 𝑗 is the injection efficiency, 𝐼 is the drive current, 𝑉 𝑄𝑊
is the volume of the QW,
𝑞 is the elementary charge, 𝑁 is the carrier density, and 𝐴 (𝑠 −1
) , 𝐵 (𝑐 𝑚 3
𝑠 −1
) , and
𝐶 (𝑐 𝑚 6
𝑠 −1
) are the SRH, radiative, and Auger recombination respectively, 𝐽 𝑡 ℎ𝑒𝑟𝑚𝑖𝑜𝑛𝑖𝑐 𝑒 is
the carrier escape from the QW to the QB layer, and 𝐿 𝑧 is the QW thickness, and The
thermionic emission has the following expression
J
thermionic
𝑒 =
4πq(k
B
𝑇 )
2
ℎ
3
𝑚 𝑒 ∗
exp(−
𝐸 𝑏𝑎𝑟𝑟𝑖𝑒𝑟 𝑒 −𝐹 𝑒 𝑘 𝐵 𝑇 )
Where 𝐸 𝑏𝑎𝑟𝑟𝑖𝑒𝑟 𝑒 is the barrier energy level, 𝐹 𝑒 is the fermi energy, h is the planck’s
constant. The fermi energy can be calculated from the carrier density in the QW, which
is expressed by the following.
N
QW
=
m
n
∗
𝑘𝑇
𝜋 2
ℏ
3
𝐿 𝑧 ∑ln[1+𝑒 𝐹 𝑒 −𝐸 𝑖 𝑘𝑇
]
𝑖
Where m
n
∗
is the effective electron mass, and 𝐸 𝑖 is the quantized energy levels in the
QWs. Because the sum term is difficult to manipulate, we assume that there is only one
effective level in the QW.
140
N
QW
=
m
n
∗
𝑘𝑇
𝜋 2
ℏ
3
𝐿 𝑧 ln[1+𝑒 𝐹 𝑒 −𝐸 𝑖 𝑘𝑇
]
Now, the fermi level can be expressed in terms of the carrier concentration:
𝐹 𝑒 =𝐸 𝑖 +𝑘𝑇 ln(𝑒 (𝑁 𝑄𝑊
𝜋 2
ℏ
3
𝐿 𝑧 𝑚 𝑛 ∗
𝑘𝑇
)
−1)
We plug in the fermi energy back to the expression for the thermionic emission:
J
thermionic
𝑒 =
4πq(k
B
𝑇 )
2
ℎ
3
𝑚 𝑒 ∗
exp(−
𝐸 𝑏𝑎𝑟𝑟𝑖𝑒𝑟 𝑒 −𝐸 𝑖 𝑘 𝐵 𝑇 )(𝑒 (𝑁 𝑄𝑊
𝜋 2
ℏ
3
𝐿 𝑧 𝑚 𝑛 ∗
𝑘𝑇
)
−1)
We then plug this expression back into the rate equation:
𝜂 𝑗 𝐼 𝑞𝑉
𝑄𝑊
=𝐴 𝑁 𝑄𝑊
+𝐵 𝑁 𝑄𝑊
2
+𝐶 𝑁 𝑄𝑊
3
+
4πq(k
B
𝑇 )
2
ℎ
3
𝑚 𝑒 ∗
exp(−
𝐸 𝑏𝑎𝑟𝑟𝑖𝑒𝑟 𝑒 −𝐸 𝑖 𝑘 𝐵 𝑇 )(𝑒 (𝑁 𝑄𝑊
𝜋 2
ℏ
3
𝐿 𝑧 𝑚 𝑛 ∗
𝑘𝑇
)
−1)
The coefficient of the current density was replaced by x.
𝑥 =
𝜂 𝑗 𝑞 𝑉 𝑄𝑊
From here, we follow the same procedures discussed in chapter 3 where IQE is
extracted by using parameter fittings [13]. The EL intensity is scaled by:
𝑁 𝑝 ℎ
=
𝑃 𝐸𝐿
ℎ𝜈 𝐴 𝑎𝑟𝑒𝑎 𝑑 =𝑎𝐵 𝑁 𝑄𝑊
2
Where 𝑁 𝑝 ℎ
is the number of photons generated per seconds per volume, 𝑃 𝐸𝐿
is the
measured integrated EL power, ℎ𝜈 is the photon energy, 𝑎 is the extraction efficiency.
141
Then current can be written as:
𝐼 =
𝐴 𝑥 √𝑎𝐵 ℎ𝜈 𝐴 𝑎𝑟𝑒𝑎 𝑑 √𝑃 𝐸𝐿
+
1
𝑥𝑎 ℎ𝜈 𝐴 𝑎𝑟𝑒𝑎 𝑑 𝑃 𝐸𝐿
+
𝐶 𝑥 (𝑎𝐵 ℎ𝜈 𝐴 𝑎𝑟𝑒𝑎 𝑑 )
3/2
𝑃 𝐸𝐿
3
2
+
4πq(k
B
𝑇 )
2
ℎ
3
𝑚 𝑒 ∗
exp(−
𝐸 𝑏𝑎𝑟𝑟𝑖𝑒𝑟 𝑒 −𝐸 𝑖 𝑘 𝐵 𝑇 )
(
𝑒 (
𝜋 2
ℏ
3
𝐿 𝑧 𝑚 𝑛 ∗
𝑘𝑇
𝑥 √𝑎𝐵 ℎ𝜈 𝐴 𝑎𝑟𝑒𝑎 𝑑 √𝑃 𝐸𝐿
)
−1
)
Changing the coefficients of the fitting parameters and we have:
𝐼 =𝑃 1
√𝑃 𝐸𝐿
+𝑃 2
𝑃 𝐸𝐿
+𝑃 3
(𝑃 𝐸𝐿
)
3
2
+𝑃 4
(𝑒 𝑃 5
√𝑃 𝐸𝐿
−1)
Where P 1 and P 5 have units of eV
-1/2
, P 2 and P 3 have units of, eV
-1
, eV
-3/2
, respectively,
and P 4 has a unit of A.
In this expression, the 3
rd
and the 4
th
term represent the Auger recombination
and carrier escape components of the current, and they are responsible for current
behavior in the high injection regime. Because of the ambiguity of the fitting
coefficients, the ratio of the contribution from these two terms cannot be determined.
We showed in the power dependent PL measurement that Auger recombination in
these QW was negligible. The excitation power density ranges from 0.06 W/cm
2
to 300
W/cm
2
. The absorption coefficients of the 325 nm laser on GaN is in the order of 10
5
cm
-
1
which corresponds to absorption layer thickness of around 100 nm. Because the
thickness of MQW is in the order of 50 nm, and the fact that QW region should have
higher absorption coefficients show that majority of the laser light is absorbed by the
142
QW region. The drive current of 10 mA to 100 mA corresponds to current densities of
10 A/cm
2
to 100 A/cm
2
based on the chip area of around 0.1mm
2
. Therefore, the carrier
concentration generated by the maximum PL excitation is higher than the largest drive
current densities and the auger recombination in the high current injection regime
should be minimal. Therefore, we drop the 3
rd
term in the rate equation:
𝐼 =𝑃 1
√𝑃 𝐸𝐿
+𝑃 2
𝑃 𝐸𝐿
+𝑃 4
(𝑒 𝑃 5
√𝑃 𝐸𝐿
) The data for the yellow LED was fitted with the
equation, and the fitting parameters P 1, and P 2 and P 4 and P 5 were 2.395 ± 0.16 (eV
-1/2
),
139.3 ± 6.1 (eV
-1
), 0.0008931 ± 0.0004306 A and 188.2 ± 19.6 (eV
-1/2
) respectively.
Finally, the IQE can be calculated as:
𝜂 =
𝑃 2
𝑃 𝐸𝐿
𝐼 =
𝑞 𝜂 𝑗 𝑎 ℎ𝜈 𝑃 𝐸𝐿
𝐼
Figure 5.15 (a) shows the measured and the fitted drive current as a function of
the integrated EL power, showing a close fit of the data to the model. Figure 5.15 (b)
shows the IQE extracted from the fitting parameter P 2. The maximum efficiency is
around 35 % at a current of around 120 mA.
143
Figure 5.15 (a) Current vs. Integrated EL power for the yellow emitting nanoLED. The
measured values are indicated by blue dots, and the fitting curve is shown in orange. (b)
the IQE of the nanoLED vs. current
Figure 5.16 (a) shows the contribution of the SRH, Radiative, and Auger
recombination processes to the total recombination rate. At low current, SRH
recombination dominates, but as the current increases, the SRH decreases rapidly while
the overflow increases significantly. Figure 5.16 (b) shows the same as the Figure 5.16
(a), but all the values are extrapolated for higher current, using the fitting parameters.
We observe that the SRH non-radiative recombination becomes negligible at high
injection and the current overflow ultimately limit the IQE.
144
Figure 5.16 (a) the contribution of the SRH, Radiative, and Auger mechanisms to the
carrier recombination based on the fit to the measured data. (b) the same but
extrapolated to higher current based the fitting parameters.
The fitting parameter of P 2 is 139.3 (eV
-1
), and we can substitute it back to the
expression, and we have:
𝜂 𝑗 𝑎 =
𝑞 ℎ𝜈 𝑃 2
=0.3 %
145
This model estimates a light extraction efficiency of 0.3%, assuming that the
injection efficiency is unity. The poor extraction efficiency of these nanoLEDs can be a
result of preliminary device design. First, the current nanoLEDs structures are far inferior
in terms of light extraction from the chip compared to the state of the art LEDs which
have sophisticated packaging and encapsulation treatment to maximize the light out
[16]. Patterned sapphire is often used to enhance the extraction of light at the
GaN/sapphire interface so the light can escape from the bottom, and the devices can
also be diced so that the light can escape from the side of the chips. A back reflector or a
transparent mount is used to collect these light that is escaped from the side and the
bottom. Also, encapsulation by epoxy can increase the light escape by reducing
refractive index contrast. Without these treatments, light can only escape from an
escape cone due to the total internal reflection. An accurate estimation of extraction
efficiency requires advanced modeling due to the non-planar geometry and diffraction.
If we consider a planar case, the extraction efficiency can be calculated from the
following:
η
extraction
=
𝐴 𝑒𝑠𝑐𝑎𝑝𝑒 𝑐𝑜𝑛𝑒 𝐴 𝑠𝑝 ℎ𝑒𝑟𝑒
=
2𝜋 𝑟 2
(1−𝑐𝑜𝑠𝜙 )
4𝜋 𝑟 2
=
1−cos(𝑎𝑟𝑐𝑠𝑖𝑛 (
1
𝑛 𝐺𝑎𝑁
))
2
=4.5%
It is evident from figure 5.12 (b) that the whole LED chip lights up, which is probably due
to the light that is guided along the film and substrate by total internal reflections. We
note that the surface of the chip outside of the nanostripes areas are covered by
polycrystalline deposition and as a results some light can be scattered out as they are
guided along the wafer. We also note that the majority of the chip areas are covered by
146
the large metal pads which absorb the light. As the light travels along the wafer, some
portion of the light is absorbed by the materials and reduces the total light out.
Also, the top contact on the nanoLEDs is Ni/Au with a nominal thickness of 10/20
nm. The same metal film co-deposited on a glass substrate and going under the same O 2
annealing treatment had a transmission of around 30 % in the green, and yellow range.
We note that some portion of the light is reflected back to the chip area and may
eventually escape from the structure. Therefore, the upper limit of the light extraction
should be around 1% which is in the same order of the light extraction extracted from
the IQE model. The discrepancies may come from the current injection loss at some
other locations such as threading dislocations and nanostripes surfaces. The effect of
these leakage paths is an interest for future studies. In the next chapter, we focus on the
amber emitting QMW structures and identify the defect structures that limit its
efficiency.
5.6 Chapter References
[1] S. Nakamura, N. Iwasa, M. Senoh and T. Mukai, "Hole compensation mechanism of
p-type GaN films".
[2] R. Shul, G. McClellan, S. Casalnuovo, D. Rieger, S. Pearton, C. Constantine, C.
Barratt, R. Karlicek, C. Tran and M. Schurman, "Inductively coupled plasma etching
of GaN," Applied Physics Letters, vol. 69, p. 1119, 1996.
147
[3] C.-Y. Ho, T.-Y. Tu, C.-C. Wang and Y. Kang, "Investigation of post-annealing indium
tin oxide for future electro-optical device application," Recent Researches in
Telecommunications, Informatics, Electronics and Signal Processing, p. 154.
[4] C. Chang, S. Chang, Y. Su, C. Kuo, W. Lai, Y. Lin, Y. Hsu, S. Shei, J. Tsai, H. Lo, J. Ke
and J. Sheu, "High Brightness InGaN Green LEDs With an ITO n++ SPS uper
contact," IEEE TRANSACTIONS ON ELECTRON DEVICES, vol. 50, no. 11, p. 2208,
2003.
[5] Y. Lin, S. Chang, Y. Su, T. Tsai, C. Chang, S. Shei, C. Kuo and S. Chen.
[6] K. Zhou, M. Ikeda, J. Liu, S. Zhang, D. Li, L. Zhang, J. Cai and H. W. H. B. W. H. Yang,
"Remarkably reduced efficiency droop by using staircase thin InGaN quantum
barriers in InGaN based blue light emitting diodes," Applied Physics Letters, vol.
105, p. 173510, 2014.
[7] S. Li and J. Piprek, "New Insight Into The Efficiency Droop Of GaN-Based LEDs," 4 2
2011. [Online]. Available:
http://www.compoundsemiconductor.net/article/87576-new-insight-into-the-
efficiency-droop-of-gan-based-leds.html.
[8] C.-H. Liao, C.-Y. Chen, H.-S. Chen, K.-Y. Chen, W.-L. Chung, W.-M. Chang, J.-J.
Huang, Y.-F. Yao, Y.-W. Kiang and C.-C. Yang, "Emission Efficiency Dependence on
the p-GaN Thickness in a High-Indium InGaN/GaN Quantum-Well light_emitting
Diode," IEEE Photonics Technology letters, vol. 23, no. 23, p. 1757, 2011.
[9] E. F. Schubert, T. Gessmann and J. k. Kim, Light emitting diodes, John Wiley &sons,
Inc., 2005.
[10] U. Kaufmann, M. Kunzer, M. Maier, H. Obloh, A. Ramakrishnan, B. Santic and P.
Schlotter, "Nature of the 2.8 eV photoluminescence band in Mg doped GaN,"
Applied Physics Letters, vol. 1326, p. 72, 1998.
[11] H. Obloh, K. Bachem, U. Kaufmann, M. Kunzer, M. Maier, A. Ramakrishnan and P.
Schlotter, "Self-compensation in Mg doped p-type GaN grown by MOCVD," Journal
of Crystal Growth, vol. 195, p. 270Ð273, 1998.
[12] H.-Y. Ryu, H.-S. Kim and J.-I. Shim, "Rate equation analysis of efficiency droop in
InGaN light-emitting diodes," APPLIED PHYSICS LETTERS, vol. 95, p. 081114, 2009.
[13] Y.-S. Yoo, T.-M. Roh, N. Jong-Ho, S. Son and Y.-H. Cho, "Simple analysis method for
determining internal quantum efficiency and relative recombination ratios in light
emitting diodes," Applied Physics Letters, vol. 102, p. 211107, 2013.
148
[14] H. Zhao, G. Liu, J. Zhang, R. A. Arif and N. Tansu, "Analysis of Internal Quantum
Efficiency and Current Injection Efficiency in III-Nitride Light-Emitting Diodes,"
JOURNAL OF DISPLAY TECHNOLOGY, vol. 9, no. 4, p. 212, 2013.
[15] Q. Dai, Q. Shan, J. Wang, S. Chhajed, J. Cho, E. Schubert, M. H. Crawford, D. D.
Koleske, M.-H. Kim and Y. Park, "Carrier recombination mechanisms and efficiency
droop in GaInN/GaN light-emitting diodes," APPLIED PHYSICS LETTERS, vol. 97, p.
133507, 2010.
[16] S. Yamamoto, Y. Zhao, C.-C. Pan, R. B. Chung, K. Fujito, J. Sonoda, S. P. DenBaars
and S. Nakamura, "High-Efficiency Single-Quantum-Well Green and Yellow-Green
Light-Emitting Diodes on Semipolar (20-21) GaN Substrates," Applied Physics
Express, vol. 3, p. 122102, 2010.
[17] S. Nakamura, M. Senoh, N. Iwasa and S.-i. Nagahama, "High-Brightness InGaN
Blue, Green and Yellow Light-Emitting Diodes with Quantum Well structures,"
Japanese Journal of Applied physics, vol. 34, pp. 797-799, 1995.
149
Chapter 6 Localized stacking faults in high In-content InGaN MQW
nanostructures
6.1 Optical Characterization of amber emitting nanostripes MQW structures
In the previous chapter, we have shown that the amber emitting LEDs had
significantly lower EL output power compared to yellow, green and blue emitting
nanoLEDs. In this chapter, we identify the nature of the defects that limit the efficiency
of the high indium content MQW structures and clarify its generation mechanisms using
nanoscale and atomic scale structural and optical characterizations.
Figure 6.1(a) and 6.1 (b) show the schematics of the MQW structures grown on
nanostripes with the {10-11} facets. The nanostripe templates are grown at 975 C and
pressure of 200 Torr in H 2 ambient for 360 s. TMG and NH 3 are used as precursors for
the Ga and N sources, and their flow rates were set to 8 sccm and 3 slm, respectively. A
prism structure with triangular cross-sectional shape is grown in the mask opening
regions. Both facets of the pyramid are {10-11} planes and the top (0001) plane is
pinched off. Three pairs of QWs are then grown on the facets of the nanostripe
pyramids at a pressure of 300 torrs and temperature of 800 C in N 2 ambient during the
same growth. TEG and TMI are used to supply Ga and In sources, respectively and TMI
flow rate is adjusted to tune the In composition in the QWs while NH 3 and TEG flow
rates are maintained at 9 slm and 20 sccm, respectively. The quantum well barrier (QB)
is grown under the same conditions as the QW except that there is no TMI flow and the
growth temperature is increased to 875 C. The resulting MQW nanostripe structures are
150
examined by a PL, cathodoluminescence (CL) and transmission electron microscopy
(TEM).
Figure 6.1 (a) Schematics of an array of GaN MQW nanostripe cross-section (b) zoomed-
in schematics of one GaN MQW nanostripe cross-section (c-d) Top down SEM images of
nanostripes MQW structures with 500 nm and 1000 nm spacing, respectively and TMI
flow of 120 sccm for the QW growth. (e-f) Top down SEM images of nanostripes MQW
151
structures with 500 nm and 1000 nm spacings, respectively and TMI flow of 480 sccm
for the QW growth (g) PL spectrum taken for the samples (c)-(f)
Figure 6.1 (c)-(f) show the top-down SEM images of nanostripe MQW structures
that are grown with TMI flow rate of 240 and 480 sccm and spacing of 500 nm and 1000
nm. The PL spectra measured on the corresponding samples are plotted in figure 6.1(g).
The PL spectra are acquired by exciting the samples with a 325 nm HeCd laser at a
power density of around 1kW/cm
2
and a spot size of around 50 by 50 μm
2
. The
contribution of the yellow band to the QW emission spectra in these samples is
insignificant as we have optimized the growth conditions for the barrier layers to
suppress the yellow band emission. As the stripe spacing or the TMI flow rate increases,
the surface of the nanostripes becomes significantly rougher especially near the bottom
of the stripes and their PL intensities are significantly reduced with a small red shift. The
stripes with 1000 nm pitch have larger QW thickness than the stripes with 500 nm pitch
that are grown in the same run due to a larger growth rate enhancement effect as a
result of larger diffusion of growth species from the mask region to the nanostripes.
Therefore, these data suggest that enhanced strain in the QW region, either through
higher In composition or larger QW thickness, deteriorates the surface morphology and
reduces luminous efficiency.
Figure 6.2 shows a high-resolution CL intensity mapping correlating the region of
the rough surface to the reduced luminescence efficiency. Figure 6.2 (a) and (c) shows
the top down view SEM images of stripes with 500 nm and 750 nm spacing, respectively.
152
The stripe with 750 nm spacing has rough surface near the bottom on one side (north
side) and has a smooth surface on the other side (south side). The stripe with 500 nm
spacing has smooth surfaces on both facets. Figure 6.2 (b) and (d) plots the CL mapping
where intensity at each location represents the spectrally integrated CL intensity from
500 nm to 630 nm.
Figure 6.2 (a) and (b) shows the top down view SEM images of stripes with 500 nm and
750 nm spacing. (c) and (d) shows the CL intensity mapping with spectral integration
from 500 nm 630 nm of the stripes with 500 nm and 750 nm spacing respectively
For both stripes, the intensity profiles have some non-uniformity in the <11-20>
directions. These could be a result of In composition or quantum well thickness
153
fluctuation. By comparison, there is a significant decrease in the intensity in the regions
where rough surfaces are observed for stripes with 750 nm pitch shown in figure 6.2 (d).
The black dotted line indicates the transition from smooth to rough surface morphology
and the extension of the line to the CL mapping coincides with the drop-off in CL
intensity. However, wherever a smooth surface is observed and for the stripes with 500
nm pitch, a similar drop off in the intensity is not observed. Therefore, the rough surface
observed in the SEM corresponds to regions of non-radiative recombination which
significantly reduce the luminescence efficiency. TEM is used to characterize these
regions further.
6.2 Structural characterizations of defects in amber emitting nanostripes MQW
structures
Figure 6.3 show the cross-sectional TEM images of nanostripes with 750 nm
spacing. The TEM sample is prepared by first depositing 20 nm of carbon and 20 nm Pt
as a protection layer and the cross-section is prepared by focused ion beam (FIB) milling
and lift off. Two beam conditions are used to characterize the defects. Figure 6.3 (a) and
(b) show the TEM images of the entire stripe structure cross-section taken with g = [1-
100] and the [11-20] zone axis, respectively. Figure 6.3 (c) and (d) show enlarged images
of regions enclosed by the white dashed lines indicated in Figure 6.3 (a) and (b)
respectively. A high density of extended defects is observed near the bottom and on
both sides of the stripe for the images taken with g = [1-100] while the defects are not
154
visible when the TEM images are taken with g = [0002]. The defects originate from the
1
st
QW and the density and the number of defects increases for the 2
nd
and the 3
rd
QW.
Most extended defects terminated at the surface which, we believe, caused the surface
instability observed with SEM.
Figure 6.3 TEM images of the nanostripes MQW structures taken with (a) g =[1-100] and
(b) g= [0002]. (c) and (d) are the magnified images of (a) and (b), respectively, in regions
enclosed by the white dotted lines.
155
The extended defects observed can be either basal plane stacking faults (BPSF)
or the threading segments of misfit dislocations. The misfit segment of the dislocations
that should lie on the InGaN/GaN interface are not observed and as a result, the
observed defects do not originate from these dislocations. In terms of BPSFs, there are
three types of BPSF for nitride materials, I 1, I 2 and I 3 types with R = 1/6[2-203], R = 1/6[2-
200], and R=1/2[0001] respectively where R is the displacement vector. The invisibility
criterion for the defects is g ∙ R equals to zero or an integer. The defects were observed
with g = [1-100] but disappeared in g = [0002], indicating that the stacking faults are
either type I 1 or I 2.
Figure 6.4 (a) TEM lattice image of the bottom part of the nanostripes MQW structures
(b) Filtered HRTEM image of the stacking fault in the region enclosed by the dashed lines
in figure 6.4 (a). Blue, green and red dots represent the A, B and C lattices respectively.
The blue, green and red dotted line represent the locations where A, B and C lattices
align to. The yellow dotted line shows the location of the stacking fault. It is clear the
lattices below the stacking fault align with A and B lattices while the lattices above the
(a) (b)
156
stacking fault align with B and C lattices, representing a change of stacking sequence
from ..ABAB.. to ..BCBC..
Figure 6.4 (b) shows a filtered HRTEM image near the vicinity of one of the
stacking faults shown in the dashed rectangle figure 6.4 (a). The blue, green and red
dots represent the A, B and C lattices respectively. The blue, green and red dotted line
represent the locations where A, B and C lattices align to. The yellow dotted line shows
the location of the stacking fault. It is clear the lattices below the stacking fault align
with A and B lattices while the lattices above the stacking fault align with B and C
lattices, representing a change of stacking sequence from ..ABAB.. to ..BCBC.. The
stacking sequence is … ABABABCBCBC … which are the characteristics of type I 1 stacking
faults. The generation of type I 1 stacking faults is also observed in planar green emitting
MQW and LED structures grown on the {10-11}, {20-2-1}, and {10-10} planes [1] [2] [3].
They are caused by the removal of one half basal plane which releases compressive
strain in the QW region [3]. The stacking sequence of the type I 2 stacking fault is …
ABABABCACACA … where two consecutive faults occur. The generation of this type of
stacking fault requires more energy than the type I 1 and therefore are less commonly
observed [3]. Nevertheless, Tischer et al. have observed type I 2 stacking faults in the
micron-scale GaN stripe structures with the {10-11} facets [4] [5]. The generation of
these stacking faults is attributed to the thermal stress during the cool down after the
growth [5] [6]. By comparison, the type I 1 stacking faults observed in this study are
generated to relieve the excessive compressive strain in the InGaN QW. Significantly
fewer stacking faults are observed for nanostripe MQW structures with thinner QW
157
thicknesses or reduced In content in the QW (not shown), further confirming that
stacking fault generation is a result of strain relaxation.
Figure 6.5 HRTEM images of QW region near the (a) top, (b) middle, and (d) bottom of
the MQW nanostripe structures. (c) shows the HRTEM image of the region that
transition from open to the masked region.(e) shows the QW thickness variation from
the top to the bottom of the stripes (f) shows the peak emission wavelength variation
from top to the bottom of the stripes.
158
6.3 Stacking fault generation mechanisms
The origin and the mechanisms for the localized stacking fault generation are
investigated in the following studies. Figure 6.5 (a) (b), and (d) show the lattice images
of the 1
st
QW near the tip, middle and the bottom of the stripes, where the QW
thickness is measured to be 6.0 nm, 4.7 nm, and 7.0 nm respectively. Figure 6.5 (e)
shows the continuous variation of the QW thickness extracted from lower magnification
images, and it is plotted against the distance from the apex to the bottom of the stripe.
The QW thickness is the largest near the apex and the bottom of the stripes, and the
thickness decreases as it moves towards the middle. The differences in the growth rate
of the QW layer across the facets can be attributed to the gas phase diffusion and/or
the surface diffusion of the precursors from the mask regions to the nanostripe
template. Shioda et al. conducted detailed studies of the gas-phase diffusion in the
selective area growth of InGaN on c-plane GaN [7]. At 800 C, the growth rate
enhancement profile is similar to that of the GaN at the same temperature and the
diffusion length, D/k s is extracted to be around 10 μm (D is the precursor vapor-phase
mass diffusivity and k s is the surface incorporation rate constant). Because the diffusion
length is significantly larger than the facet size of the nanostripes, the gas phase
diffusion is expected to play a small role in the growth rate variation.
We conclude that surface diffusion of the precursors caused the differences in
the QW thickness. The surface diffusion length is usually in the nm length scale. As a
result, there is an additional growth rate enhancement that is confined near the bottom
159
of the stripes. The increase in the QW thickness near the tip of the stripes can be
explained by the existence of the small (0001) plane. Our previous studies on MQW
structures grown on nanorods and nanosheets structures show that the growth rate and
the resulting QW thickness on the c-plane are significantly larger than that of an
adjacent {10-11} plane due to the migration of precursors from the {10-11} to the (0001)
plane via surface diffusion [8] [9].The difference, in this case, is that the area of c-plane
is significantly less, and therefore the difference in the QW thicknesses on the two
facets is less. However, the growth mechanisms should be similar.
Another attribute that determines the strain variation is the In composition non-
uniformity. Figure 6.5 (f) shows the emission wavelength as a function of distance from
the apex to the bottom of the stripes based on CL measurements on the representative
stripes. The emission wavelength variation is within 5 nm, and it resembles the QW
thickness variation. Therefore, it is reasonable to conclude that the In composition
across the facets is relatively uniform, and the small wavelength shift is caused by the
QW thickness variation.
From the analysis above, it is clear that QW near the middle of the stripe should
experience smaller strain due to the thinner QW thickness. However, there is not a
significant difference in the QW thickness near the apex compared to near the bottom
of the stripes and the slight 16 % larger QW thickness near the bottom may not
sufficiently explain the high density of stacking fault generation.
160
Close examination of the lattice images reveals the existence of around 3-degree
downward lattice tilt near the bottom of the stripes, indicated in figure 6.5 (d), which
may explain the significant localized defect generation. By comparison, lattice tilt is not
present in the top or the middle regions of the stripes. Figure 6.5 (c) shows the lattice
image near the interface between the mask and window regions and it shows that
lattice tilt originates from the GaN template and occurs as soon as the GaN material is
overgrown on top of the mask region. Lattice tilt in the overgrown regions is commonly
observed in epitaxial lateral overgrowth studies of various materials using selective area
growth [10]. The tilt has been attributed to the thermal strain during the cool down and
or adhesion bonding between the overgrown material and the growth mask [10]. An in-
situ x-ray measurement show that the tilt occurs in GaN selective area epitaxy at the
growth temperature, and the additional tilt caused by the thermal strain during the cool
down was small [11]. Therefore, the downward tilt has been attributed to the adhesion
bonding of the overgrown materials with the mask materials.
161
Figure 6.6 Lattice image of the overgrown region of the nanostripes structures. The
green arrows show the termination of the (0001) planes into the SiN mask.
Figure 6.6 shows the lattice image of the overgrown region of the GaN
nanostripes on SiN. The c-plane lattices clearly terminate into the SiN, further
demonstrating that the interfacial bonding between GaN and SiN caused the tilt of the
lattices in the overgrown region. Lattice bending as a result of thermal strain does not
lead to termination of the lattice planes into the growth mask [12]. The interfacial
bonding between the overgrown region and the SiN has significant implications on the
strain variations near the bottom of the nanostripes.
Figure 6.7 (a) shows the lattice image of the bottom part of the QW grown on
the nanostripes. The white dotted line indicates the interface between QB and QW. The
red lines trace the lattice plane from the left side to the right side of the QW. The line 1
162
and line 3 trace the (0001) and (1-100) planes, respectively, which pass through the QW
on the sidewall of the nanostripes. We can observe an up shift for line 1 and a right shift
for line 2. Figure 6.7 (b) shows the schematics of the deformation of the InGaN QW in
this region. Because the lattice spacing of InGaN is larger than GaN, when they are
grown epitaxially on GaN, they need to contract according to the lattice spacing of the
GaN. For semipolar planes, InGaN needs to contract in the diagonal directions, which
leads to dominant shear strain over normal strains [13]. As a result, there is a relative
shift in the QB layers on two sides of the QW.
In comparison, the strain distribution of the InGaN QW near the bottom is
different from those on the side. The line 2 and 4 trace the lattices near the bottom of
the QW. There was no relative shift of lattice of line 2, or the (0001) plane while there is
a right shift in the lattice for line 4. Figure 6.7 (c) shows the schematics of the InGaN
deformation near the bottom. Because of the interfacial bonding between InGaN and
SiN, it prevents the InGaN from deforming to its ideal shape to have an upward shift.
Consequently, the InGaN are forced to contract in the yy directions. This agrees with the
stacking fault generation near the bottom because the I 1 type fault can relieve the strain
in the yy direction.
163
Figure 6.7 (a) the lattice image of nanostripes QW structures, near the bottom of the
stripes. The white dotted line indicates the location of the QW. The red lines trace some
specific lattice planes to show a shift in the planes. (b) and (c) show the schematics of the
deformation of the InGaN QW structures near the side and near the bottom of the
nanostripes respectively.
In conclusion, based on the CL and TEM observations made in this chapter, we
conclude that the optical and structural properties of amber emitting MQW structures
grown on {10-11} facets of nanostripes are strongly affected by inhomogeneous strain
164
created by surface material transport and lattice tilt. As the QW thickness or the In
composition increases, we observe that stripes with rough surface morphology near the
bottom of the stripes begin to occur and dominate the morphology of the structures. CL
mapping and TEM analysis show that these regions correspond to non-radiative
recombination centers caused by the formation of stacking faults generated from the
QW regions near the bottom that propagate to the surface. HRTEM images show that
the stacking fault is I 1 type which is formed by removal of one-half basal plane to relieve
the compressive strain in the InGaN QW. In these structures, the QWs at the bottom of
the nanostructure facet are thicker than near the top, probably due to the enhanced
precursor diffusion from the mask region. In addition, the QWs experience additional
compressive strain in the yy direction due to bonding between the overgrown region
and the mask material. We attribute the thicker QW and the additional strain as the
cause for the stacking fault generation only near the bottom while the rest of the
structure is generally defect-free. The additional strain caused by the wing tilt may
ultimately limit the critical thickness of the layers near the bottom.
6.4 Chapter References
[1] Z. Wu, T. Tanikawa, T. Murase, Y.-Y. Fang, C. Q. Chen, Y. Honda, M. Yamaguchi, H.
Amano and N. Sawaki, "Partial strain relaxation by stacking fault generation in
165
InGaN multiple quantum wells grown on (11-01) semipolar GaN," Applied Physics
Letters, vol. 98, p. 051902, 2011.
[2] F. Wu, Y. Zhao, A. Romanov, S. P. DenBaars, S. Nakamura and J. S. Speck, "Stacking
faults and interface roughening in semipolar ( 20-2-1) single InGaN quantum wells
for long wavelength emission".
[3] F. Wu, Y.-D. Lin, A. Chakraborty, H. Ohta, S. P. DenBaars, S. Nakamura and J. S.
Speck, "Stacking fault formation in the long wavelength InGaN/GaN multiple
quantum wells grown on m -plane GaN," Applied Physics Letters, vol. 96, p.
231912, 2010.
[4] I. Tischer, M. Feneberg, M. Schirra, H. Tacoub, R. Sauer, K. Thonke, T. Wunderer, F.
Scholz, L. Dieterle, E. Muller and D. Gerthsen, "Stacking fault-related luminescence
features in semi-polar GaN," Phys. Status Solidi B, vol. 248, no. 3, pp. 611-615,
2011.
[5] I. Tischer, M. Feneberg, M. Schirra, H. Yacoub, R. Sauer, K. Thonke, T. Wunderer, F.
Scholz, L. Dieterle, E. Muller and D. Gerthsen, "I2 basal plane stacking fault in GaN:
Origin of the 3.32 eV luminescence band," PHYSICAL REVIEW B, vol. 83, p. 035314,
2011.
[6] D. N. Zakharov, Z. Liliental-Weber, B. Wagner, Z. J. Reitmeier, E. A. Preble and R. F.
Davis, "Structural TEM study of nonpolar a-plane gallium nitride grown on (11-20)
4H-SiC by organometallic vapor phase epitaxy," Physical Review B, vol. 71, p.
235334, 2005.
[7] T. Shioda, M. Sugiyama, Y. Shimogaki and Y. Nakano, "Selectiveareametal-
organicvapor-phaseepitaxyofInN, GaN and InGaN covering whole composition
range," Journal of Crystal Growth, pp. 2809-2812, 2009.
[8] T.-w. Yeh, Y.-T. Lin, B. Ahn, L. S. Stewart, P. D. Dapkus and S. R. Nutt, "Vertical
nonpolar growth templates for light emitting diodes formed with GaN
nanosheets," Applied Physics Letters, vol. 100, p. 033119, 2012.
[9] T.-W. Yeh, Y.-T. Lin, L. S. Stewart, P. D. Dapkus, R. Sarkisian, J. D. O'Brien, B. Anh
and S. R. Nutt, "InGaN/GaN Multiple Quantum Wells Grown on Nonpolar Facets of
Vertical GaN Nanorod Arrays," Nano Letters, vol. 12, no. 6, pp. 3257-3262, 2012.
[10] Z. Zytkiewicz, "Laterally overgrown structures as substrates for lattice mismatched
epitaxy," Thin Solid Films, vol. 412, p. 64–75, 2002.
166
[11] P. Fini, A. Munkholm, C. Thompson, G. Stephenson, J. Eastman, M. V. Ramana
Murty, O. Auidello, L. Zhao, S. DenBaars and J. Speck, "In situ, real-time
measurement of wing tilt during lateral epitaxial overgrowth of GaN," Applied
Physcis Letters, vol. 76, no. 26, 2000.
[12] T. Zheleva, W. Ashmawi and K. Jones, "Pendeo-Epitaxy versus Lateral Epitaxial
Overgrowthof GaN: A Comparative Study via Finite Element Analysis," phys. stat.
sol. (a), vol. 176, p. 545, 1999.
[13] A. E. Romanov, T. J. Baker, S. Nakamura and J. S. Speck, "Strain-induced
polarization in wurtzite III-nitride semipolar layers," JOURNAL OF APPLIED PHYSICS,
vol. 100, p. 023522.
[14] T. Wernicke, L. Schade, C. Netzel, J. Rass, V. Hoffmann, S. Ploch, A. Knauer, M.
Weyers, U. Schwarz and M. Kneissl, "Indium incorporation and emission
wavelength of polar, nonpolar and semipolar InGaN quantum wells,"
Semiconductor science and techonolgy, vol. 27, p. 024014, 2012.
[15] S. Yamamoto, Y. Zhao, C.-C. Pan, R. B. Chung, K. Fujito, J. Sonoda, S. P. DenBaars
and S. Nakamura, "High-Efficiency Single-Quantum-Well Green and Yellow-Green
Light-Emitting Diodes on Semipolar (20-21) GaN Substrates," Applied Physics
Express, vol. 3, p. 122102, 2010.
[16] Y. Zhao, S. H. Oh, F. Wu, Y. Kawaguchi, S. Tanaka, K. Fujito, J. S. Speck, S. P.
DenBaars and S. Nakamura, "Green Semipolar (20-2-1) InGaN Light-Emitting
Diodes with Small Wavelength Shift and Narrow Spectral Linewidth," Applied
Physics Express, vol. 6, p. 062102, 2013.
[17] H. Sato, A. Tyagi, H. Zhong, N. Fellows, R. B. Chung, M. Saito, K. Fujito, J. S. Speck,
S. P. DenBaars and S. Nakamura, "High power and high efficiency green light
emitting diode on free-standing semipolar (1122) bulk GaN substrate," Phys. stat.
sol. (RRL), vol. 1, no. 4, pp. 162-164, 2007.
[18] H. Zhong, A. Tyagi, N. Fellows, R. Chung, M. Saito, K. Fujito, J. Speck, S. DenBaars
and S. Nakamura, "Demonstration of high power blue-green light emitting diode
on semipolar (11-22) bulk GaN substrate," Electronics Letters, vol. 43, no. 15, pp.
825 - 826, 2007.
[19] K. Fujito, S. Kubo and I. Fujimura, "Development of bulk GaN crystals and
nonpolar/semipolar substrates by HVPE," MRS bulletin, vol. 34, no. 05, pp. 313-
317, 2009.
167
[20] K. Iso, H. Yamada, H. Hirasawa, N. Fellows, M. Saito, K. Fujito, S. P. DenBaars, J. S.
Speck and S. Nakamura, "High brightness blue InGaN/GaN light emitting diode on
nonpolar m-plane bulk GaN substrate.," Japanese Journal of Applied Physics, vol.
46, no. 10, p. 960, 2007.
[21] Q. Sun, B. H. Kong, C. D. Yerino, T.-S. Ko, B. Leung, H. K. Cho and J. Han,
"Morphological and microstructural evolution in the two-step growth of nonpolar
a-plane GaN on r-plane sapphire.," Journal of Applied Physics, vol. 106, no. 12, p.
123519, 2009.
[22] R. Colby, Z. Liang, I. H. Wildeson and D. A. Ewoldt, "Dislocation Filtering in GaN
Nanostructures," Nano letters, vol. 10, pp. 1568-1573, 2010.
[23] Y. Nakajima, Y. Lin and P. Dapkus, " Efficient yellow and green emitting InGaN/GaN
nanostructured QW materials and LEDs," Phys. Status Solidi A., 2016.
[24] Y.-T. Lin, T.-W. Yeh, Y. Nakajima and P. D. Dapkus, " Catalyst-Free GaN Nanorods
Synthesized by Selective Area Growth," Advanced Functional Materials, vol. 24, no.
21, pp. 3162-3171, 2014.
[25] R. A. Leute, D. Heinz, J. Wang, T. Meisch, M. Muller, G. Schmidt, S. Metzner, P.
Veit, F. Bertram, J. Christen, M. Martens, T. Wernicke, M. Kneissi, S. Jenisch, S.
Strehle, O. Rettig, K. Thonke and F. Scholz, "Embedded GaN nanostripes on c-
sapphire for DFB lasers with semipolar quantum wells," Phys. Status Solidi B , vol.
253, no. No.1 , pp. 180-185, 2016.
[26] C. Miao, Y. Honda, M. Yamaguchi and H. Amano, "Growth of InGaN/GaN multiple
quantum wells on size-controllable nanopyramid arrays," Japanese Journal of
Applied Physics, vol. 53, p. 030306, 2014.
[27] K. Wu, T. Wei, D. Lan, X. Wei, H. Zheng, Y. Chen, H. Lu, K. Huang, J. Wang, Y. Luo
and J. Li, "Phosphor-free nanopyramid white light-emitting diodes grown on {10-
11} planes using nanospherical-lens photolithography".
[28] Y.-H. Ko, J.-H. Kim, S.-H. Gong, J. Kim, T. Kim and T.-H. Cho, "Red Emission of
InGaN/GaN Double Heterostructures on GaN Nanopyramid Structures," ACS
Photonics, vol. 2, pp. 515-520, 2015.
[29] R. A. Leute, J. Wang, T. Meisch, J. Biskupek, U. Kaiser and F. Scholz, "Blue to true
green LEDs with semipolar quantum wells based on GaN nanostripes," Phys. Status
Solidi C, Vols. No. 4-5, pp. 376 - 380, 2015.
168
[30] K. Wu, T. Wei, H. Zheng, D. Lan, X. Wei, Q. Hu, H. Lu, J. Wang, Y. Luo and J. Li,
"Fabrication and optical characteristics of phosphor-free InGaN nanopyramid
white light," Journal of Applied Physics , Vols. 115,, p. 123101, 2014.
[31] Y.-J. Li, J.-R. Chang, S.-P. Chang, K.-P. Sou, Y.-J. Cheng, H.-C. Kuo and C.-Y. Chang,
"InGaN/GaN multiple-quantum-well nanopyramid-on-pillar light-emitting diodes,"
Applied Physics Express, vol. 8, p. 042121, 2015.
[32] Y.-H. Ra, R. Navamathavan, J.-H. Park and C.-R. Lee, "Coaxial InxGa1−xN/GaN
Multiple Quantum Well Nanowire Arrays on Si(111) Substrate for High-
Performance Light-Emitting Diodes," Nano letters, vol. 13, p. 3506−3516, 2013.
[33] A. K. Rishinaramangalam, M. Nami, M. N. Fairchild, D. M. Shima, G. Balakrishnan,
S. Brueck and D. F. Feezell, "Semipolar InGaN/GaN nanostructure light-emitting
diodes on c-plane sapphire," Applied Physics Express, vol. 9, p. 032101, 2016.
[34] S. Sundaram, R. Puybaret, Y. El Gmili, P. Bonanno, K. Pantzas, G. Orsal, D. Troadec ,
Z.-H. Cai, G. Patriarche, P. Voss, J. Salvestrini and A. Ougazzaden, "Nanoscale
selective area growth of thick, dense, uniform, In-rich, InGaN nanostructure arrays
on GaN/sapphire template," JOURNAL OF APPLIED PHYSICS, Vols. 116,, p. 163105,
116.
[35] S. Sundaram, Y. El Gmili, R. Puybaret, X. Li, P. Bonanno, K. Pantzas, G. Patriarche, P.
Voss, J. Salvestrini and A. Ougazzaden, "Nanoselective area growth and
characterization of dislocation-free InGaN nanopyramids on AlN buffered Si(111)
templates," Applied Physics Letters, vol. 107, p. 113105, 2015.
[36] I. H. Wildeson, R. Colby, D. A. Ewoldt, Z. Liang, D. N. Zakharov, N. J. Zaluzec, R. E.
García, E. A. Stach and T. D. Sands, "III-nitride nanopyramid light emitting diodes
grown by organometallic vapor phase epitaxy," Journal of applied physics, vol. 108,
p. 044303, 2010.
169
Chapter 7 Conclusions and future studies
7.1 Conclusions
In this dissertation, we have developed and demonstrated nanostructure III-nitride-
based LEDs that emit efficiently in the yellow and green spectral regions and provided
proof-of-concept of its potential to solve the “green gap” problem in the LED
technology. Along the way, we have overcome a number of technological challenges
associated with nanostructures growth, fabrication, and characterization and made
novel structures as well as clarifying the physical mechanisms of various processes.
We started our journey with the growth of nanostructure with different
semipolar and nonpolar facets, namely nanostripes with the {10-11} and {11-22} facets
and the nanosheet with the {11-20} facets. We were able to develop growth conditions
that yield {11-22} nanostripes with a smooth surface and {11-20} nanosheet structure
with high vertical to horizontal aspect ratio, both of which are novel structures. We
explained their growth mechanisms in terms of the surface reconstruction and kinetics
and highlighted the significant effect of carrier gas and V/III ratio on the growth kinetics
of these structures.
We then investigated different nanostructures in terms of their In incorporation
rate. Our results show that the nanostripes with the {10-11} facets had the highest In
incorporation rate, which agrees with studies conducted on planar substrates. We
attributed the unique surface bonding configuration on the {10-11} nanostripes to the
170
enhancement of the In incorporation on this surface. We, therefore, focused on
optimizing the growth conditions of the QWs on the nanostripes with the {10-11}
surface and demonstrate efficient emission up to 575 nm with strong radiative
recombination efficiency.
We also conducted nanoscale and atomic scale characterization of the MQW
structure grown on the {10-11} nanostripes. We used the cathodoluminescence
measurement to show that the wavelength emission is uniform within one stripe and
across several stripes. This uniformity is a significant improvement on the MQW grown
on micron-scale structures and as large as 50 nm of wavelength shift was observed
within one stripe. TEM studies clearly showed no structural defects in these stripes. We
observed a unique feature where the In incorporation is significantly reduced near the
very tip of the structure. We used strain simulation to show that the In incorporation
should actually be favored in those locations. Therefore, we attributed the growth
kinetics to the lesser indium incorporation at the tip. We also gave an interpretation to
the strain simulations and highlighted inhomogeneous strain distributions in the MQWs
grown on the nanostripes.
We then developed and demonstrated nanoLEDs based on the MQW structures
grown on the {10-11} nanostripes. We showed that the thickness of the growth mask
needs to be large enough to significantly reduce the leakage current through the mask.
We also showed that the thickness of p-GaN layer must be large enough to avoid
complete depletion of the p-GaN layer. We demonstrated nanoLEDs with narrow
emission linewidth and minimal wavelength shift as the drive current increases. These
171
characteristics are comparable to those grown on c-plane GaN, indicating high spatial
uniformity of the active layer grown on the nanostripes despite the non-planar
geometry. We also measured the EQE of the nanoLEDs and showed a maximum IQE of
around 35 %, and the electron overflow ultimately limits the efficiency. The light
extraction efficiency is also identified as an area for significant improvement as the
device structures are still in a preliminary stage and no light extraction technology has
been applied.
Finally, we investigated amber emitting MQW structures grown on the {10-11}
stripes and showed that stacking fault generations near the bottom of the stripes
significantly limit its efficiency. We showed I 1 type stacking faults originate from the first
QW and they extend to the surface causing surface roughness. From the TEM studies,
we concluded that the interfacial bonding between the overgrown region and the
growth mask increases the strain in the QW near the bottom of the stripes and
therefore becomes a driving force for the stacking fault generation.
7.2 Future Studies
We have demonstrated the great potential of the nanostripes LEDs to solve the
green gap problem and realize the monolithic integration of all nitride based multicolor
white LEDS. Therefore, there are exciting opportunities for future work.
172
In terms of blue, green and yellow emitting nanoLEDs, we identified electron
overflow as a major limiting factor of the nanoLED performance. We can incorporate an
AlGaN electron blocking layer to reduce the electron overflow. However, polycrystalline
deposition on the mask and subsectors could be potential problems associated with Al
contacting selective area growth.
Figure 7.1 Relative EQE of two m-plane LEDs grown on free-standing m-plane (11-00)
GaN substrates with a three layer SEI: one with and one without the EBL. The inset
shows the schematics for the LED with a 10 nm EBL(p-Al 0.15Ga 0.85N) (left), Calculated
overflow electron current/total electron current as a function of applied forward voltage
across the 6 nm thick In0.20Ga0.80N active region for the LED without a SEI and without
an EBL, and the other LED with a one-layer In0.10Ga0.90N SEI and without an EBL. Solid
lines are guides to the eye. The inset shows the band diagrams for the LEDs with and
without (dashed line) the SEI (right) [19]
A recent study by Ni et al. shows that when an InGaN electron injector layers are
grown before the growth of the active region, the “hot” electrons can be cooled
sufficiently so that the p-GaN itself act as an electron blocking layer because electrons
have already relaxed to the energy of the InGaN electron injector layer [19]. Figure 7.1
shows the Relative EQE of two m-plane LEDs grown on free-standing m-plane (11-00)
173
GaN substrates with an electron injection layer: one with and one without the EBL. The
inset shows the schematics for the LED with a 10 nm EBL(p-Al 0.15Ga 0.85N). They show
that when the electron injector layer is present, the AlGaN becomes unnecessary, i.e.
they find similar droop behavior with and without the AlGaN layer. Figure 7.1 (b) shows
the ratio of the calculated overflow electron current to the total electron current as a
function of applied forward voltage for the sample with and without the injection layer.
The results show that the electron overflow can be improved significantly. Therefore, an
electron injector layer can be grown before the MQW growth on the nanostripes to
reduce the electron overflow under high current injection.
Another limiting factor of the LEDs is the light extraction efficiency. First, ITO
should be used as the p-type ohmic contact spreading layer. In order to form an ohmic
contact, either a n+ InGaN/GaN superlattices are required after the p-GaN growth or a
thin layer of Ni should be deposited before the ITO, followed by O 2 annealing. These
have been discussed in chapter 5. The light extraction from the top of the surface can be
enhanced greatly by using ITO. Other standard light extraction technology could also be
implemented, for example, the use of wire bonding to limit the contact pad area, the
use of encapsulation, and mounting the LED on a reflector.
For amber emitting nanoLEDs, we need to significantly reduce the stacking fault
generations. As we identified the interactions between the overgrown region and the
growth mask to be the cause for the stacking fault generations, we can use a mask like
174
graphene with low surface energy. If efficient amber emitting nanoLEDs can be
demonstrated, we can realize the monolithic integration of white LEDs.
175
Appendix A
A.1 MOCVD growth system 1: H 2 purifier
Most of the studies undertaken in this dissertation revolve around the growth of
the InGaN-based material and device structures. Therefore a basic understanding of the
MOCVD system and the growth mechanisms may be an interesting topic for readers
who would like to learn more about the growth processes. The next three sections focus
on the hardware of the MOCVD growth reactors.
Figure A.1 the layout of the H 2 purifier system in operation mode
176
This section introduces the H 2 purifier system which is one of the most important
and the most fragile part of the whole reactor system. Figure A.1 shows the layout of
the H 2 purifier system, which is used to provide ultrapure hydrogen to the Group III
source bubblers (e.g., TMGa, TMIn, etc.) as well as delivering group IV gaseous species
into the reactor chamber. It consists of two major components: the palladium
membrane purifier on the bottom left of the layout and the V-purge system on the
bottom right of the diagram. In the hydrogen purifier, the palladium membrane only
allows hydrogen to pass through at an elevated temperature of 400 C. Therefore on one
side of the membrane is the source hydrogen, supplied from H 2 cylinders, and the other
side is the pure hydrogen. The bleed line is used to continuously purge the impurities
that are rejected by the palladium membrane. The V-purge system is used to protect
hydrogen purifier from failing when the reactor is accidently shut down (e.g., Power
outage, etc.). When the reactor is shut down, the temperature in the purifier will drop
rapidly. In the cooling process, if there is hydrogen present in the palladium membrane,
the membrane will tend to crack and cause leakage through the membrane. Therefore,
the V-purge system is there to pump out hydrogen in the palladium membrane from the
pure side to the exhaust line when the reactor shuts down. Consequently, there are two
major operating modes for the purifier system, the operation mode, and the flushing
mode.
The purifier system is usually in the operation mode unless the reactor is shut
down, in which case the system turn into the flushing mode. In the operation mode,
only hydrogen is used while only nitrogen is used to flush the system in the flushing
177
mode. Therefore, we need both hydrogen and nitrogen line to go into the purifier, and
we can use a pair of valves to control which gas goes into the purifier with
pneumatically controlled bellows valves. The bellows valves can be normally closed or
normally open. They are controlled by the signal from the solenoid, which is connected
to the reactor DC power supply. The convention for the state of the valves is that
“normal” means no signal from the solenoid. Therefore when the reactor DC is on, the
solenoid is on, and there is a signal to the valves, which is not “normal.” Therefore for
the hydrogen line, we need normally closed valve (104) and for the nitrogen line
normally open valve (103). In figure A.1, green valves mean they are open, and red
valves mean they are closed. Blue lines indicate the flow of H 2 and the red line indicate
the flow of bleed line gas. For the wires that connect the electronics, solid lines mean
the power is on, and dotted lines mean there is no power.
For the bleed line, there are two parallel paths to the exhaust. On one path there
are a normally open bellow valve 104-1 and a manual valve. This path is open when the
system is in flushing mode to allow a large flow rate of nitrogen to purge the purifier.
The other path is controlled by a needle valve. This path is open during the operation
mode which allows small flow rate to bleed out the H 2.
On the pure side, it first diverges to the two paths: one goes to the V-purge
system to the right, the other goes up, and the line diverges again. The one that goes to
the right goes through the hygrometer, which is used to monitor the moisture level,
pressure control valve and finally to the rest of the reactor systems. The one that goes
178
to the left is used for the shared Hydrogen line. The work is still in progress to share the
pure hydrogen with the Arsenide Reactor.
Figure A.2 Layout of the H 2 purifier in the flushing mode.
Figure A.2 shows the layout of the H 2 purifier in the flushing mode. The V-purge
system consists of two parts: one is the educator pump, and the other is the electrical
circuit part that controls the solenoid. First, the reactor DC signal is connected to the
relay, and the output of the relay is normally closed. So when the reactor is down, which
is “normal,” the output is “closed” or “shorted” and the Time Delay Relay is activated.
Now the time delay relay uses the “Interval Mode,” which means that when its input is
179
turned on, the output will turn on immediately and hold it for a user specified time or if
the input turns off before the specified time the output will also turn off. See the
manual for the time delay relay for more information. Therefore, the Time Delay Relay
will be on for some specific time. Its output is connected to the solenoid via an AC to DC
converter because the power from the UPS is in AC and solenoid need a DC signal. Then,
the solenoid will be on for some time which means the three normally closed valves will
all be open. Large-flow-rate Nitrogen flows from P+ to P+/P- in the educator which
create a vacuum at the P- terminal, thus pumping out pure hydrogen from the
Palladium cell in the purifier. Therefore, the purifier can be protected. After some
specified time when most hydrogen is pumped out, all three valves are closed again.
For completeness, in the following, we show the procedures for doing a leak test
on the H 2 purifier. Even though the V-purge can protect the H 2 purifier system from
failing, leakage can develop over time. Figure A.3 shows the layout of the H 2 purifier
system under leak test mode. We detach the reactor DC signal from the reactor solenoid
to the leak test solenoid in the middle. The system becomes flushing mode. We also
detach the signal line between valves 102 and 103 and connect 102 with middle
solenoid so that it is open. We create a vacuum from the reactor and put Helium as the
source (and turn off nitrogen). If there is a leak, we will not be able to get a good
vacuum.
180
Figure A.3 Layout of the H 2 purifier under leak test mode
A.2 MOCVD reactor system 2: gas handling
In this section, we describe the gas handling part of the MOCVD system. This
part takes the majority of the volume of the reactor chamber, and it selects, controls
and delivers the various growth species to the reactor chamber. This part of the reactor
controls various parameters of the growth procedures, for example, which sources go
into the reactor at what flow rate, etc. The input of the system is purified hydrogen
supplied from the hydrogen purifier from the right-hand side of the gas handling
181
system, and purified nitrogen is supplied from the top of the system. The working of the
hydrogen purifier is explained in the previous section and that of the nitrogen is very
straightforward as it just passes through a catalytic purifier, thus merits no explanation.
The hydrogen lines and nitrogen lines are colored blue and green respectively.
Figure A.4 Schematics of gas handling systems
There are three main parts in the system: carrier lines, source lines, and make up
lines which all select either hydrogen or nitrogen as input gas that is controlled by a pair
of opposite valves. The system is usually flushed by nitrogen so the valve on the
nitrogen line is normally open and therefore the valve on the hydrogen line is normally
Control System
Carrier lines
Make up lines
Source lines
Vent lines
Waste
sweep line
Alternative
source
lines
182
closed. The locations of these valves are noted in the diagram above. As can be seen in
the diagrams, there are a few nitrogen lines that do not fall into the above three
categories. They are vent lines, waste sweep lines, and alternative source lines. These
lines flow nitrogen all the time, and we will address them when appropriate. There are
also leak test lines, colored orange, for the purpose of leak test.
There are group III sources and group V sources for the MOCVD reactor. The
group V sources, like NH 3, PH 3, and AsH 3, are in the gas phase, and they can be supplied
to the reactor from a gas cylinder directly. However the group III sources such as TMGa
are all in the liquid phase. Therefore we need to “carry” the sources to the reactor by
passing hydrogen or nitrogen through a container called bubblers which contain the
liquid source. The inlet tube is submerged in the liquid, and the outlet tube is above the
liquid level. Figure A.5 shows the schematics of the gas handling system when the
growth species are supplied to the reactor. We use the NH 3 line source and
triethylgallium (TEGa) lines as an example for the growth of GaN.
The operation of the NH3 source will be studied first. Initially, there were no gas
lines for the NH3 because the reactor was first built as a GaAs reactor. When the reactor
is modified to a GaN reactor, the NH3 line was added to the PH3 line and two manual
valves are attached to control which gas proceed to flow to the reactor. PH3 line is
disconnected from the source bottle because this reactor is for nitride only.
183
Figure A.5 layout of the gas handling system. Red lines indicate the flow of growth
species.
First, the manual valves for NH3 have to be open, and the PH3 manual valve has
to be closed. The computer program usually controls the other valves. The “M” row of
the growth command controls the flow of the source line, more specifically valves 71,72,
and 73 for PH3. When the value is “M,” then those valves take on the normal condition,
and the carrier gas will purge the line. When the value is “C,” then those valves will be
activated and phosphine, if there is any, will flow in the lines. However, this value for
PH3 does not matter because the manual valve for PH3 is closed. We are basically
184
treating the PH3 line as if it was the NH3 line. Therefore, to open the NH3 line, we
simply open the manual valve.
If we trace the line further, it will encounter two Mass Flow Controllers (MFCs).
One MFC has a maximum flow rate of 10,000 sccm and the other has a maximum flow
rate of 100 sccm. Two MFCs allows us to use a large range of NH 3 flow rate. Finally, we
will get to the manifolds. Usually, the line is connected to the Vent line which will just go
to the exhaust. However, if there is a “.” in front of the flow rate number in the
program, the manifold will switch to the carrier line which goes to the reactor.
The group III source lines function similarly. For TEGa, the carrier gas is always
nitrogen, so the manual valve for the source line is always closed. There are two MFCs
right above the bubblers. The one on the left is the inlet controller and the one on the
right is the dilution control. When the value of the “M” row for TEG is on, valve 38, 40,
and 41 will change their states. The gas that passes through the inlet control will go
through the bubbler and go to valve 38 and go right. The gas that passes through the
dilution control go to valve 38 directly, diluting the source gasses. Now we trace up the
line, and it diverges to two: one goes to pressure control (PC), and the other go to the
manifold. The PC controls the pressure in the bubbler. The other end of the PC goes to
the Waste seep line mentioned earlier, which eventually goes to the exhaust. If we want
the pressure of the bubbler to decrease, we release more gas to the waste sweep and
vice versa. On the way to the manifold, there is another MFC that controls the amount
of diluted source that goes into the reactor. The manifold connects to the carrier line
185
when there is a dot before the value in the program. Otherwise, the gas will be
connected to the vent.
We have already mentioned the functions of carrier lines. They basically “carry”
the source gas to the reactor. The makeup lines are there to provide pressure balance
between the carrier line and the vent line. Notice that there is a pressure gauge
between the vent line and the carrier line to measure the pressure difference between
these two lines. If there is a pressure difference, the makeup line automatically supplies
more gas to either vent or carrier line to counter the difference. Also, R1 lines, which
flow Si2H6, is connected with the makeup lines. The C valve in the “M” row will activate
valves 91 and 92
Also, notice that there are some lines from PC that are connected to EPISON.
EPISON is used to measure the gas concentration. Normally only the concentration of
TMG is measured.
For completeness, we show the procedures to do the leak test for the bubblers.
The leak test is required after replacing or installing a new bubbler. The purpose of the
leak test is to check the seal of the VCR connection of the bubbler lines. Figure A.6
explains the procedures to replace the group III source bubbler. We Take TEGa as an
example. First, we need to purge the line with nitrogen so that there is no source left in
the line. This will prevent the source from being released to the atmosphere. Manually
switch to “Vac” so that all valves 38, 41, and 42 are open. The MO source is purged out
through diffusion overnight. Replace the bubbler and purge overnight again to remove
186
moistures. After replacing the bubbler, we need to do a leak test for the bubbler
connections. The “VAC” state remains while we switch off valve 45. Then we do the leak
test by opening the valve 44.
Figure A.6 the layout of the gas handling system under leak test mode.
A.3 MOCVD system 3: reactor chamber
So far we have discussed the hydrogen purifier and the gas handling system. The
gaseous sources eventually flow to the reactor chamber for material growth. There are
two reactor chambers: the one on the left is for GaN and the one on the right is for
187
GaAs. Because there are two reactors, there are two sets of valves that control the gas
flow to the reactors and two sets of filters and pumps for each reactor. For GaN reactor,
two carrier lines and two flush lines go into the reactor. All these lines are controlled by
the first manifold on the control panel, and they can be either hydrogen or nitrogen. See
the previous section on gas handling system for more details. Neither carrier lines nor
flush lines usually go to the GaAs reactor, which is slowly purged by a small amount of
nitrogen. The yellow color of the valve V7 means that the valve is barely open.
Figure A.7 Gas line layout around the reactor chamber
Both the Vent line and the reactor chamber are usually connected to the exhaust
via particle trapping filters (Arrow means that the gas line goes to the exhaust).
However, both lines are switched to pump before the growth starts. If the vent line was
Carrier line controls
Flush line controls Pump Control for
vent line
Pump Control for
Chamber
188
not switched to the pump, the pressure of the vent line would be a lot larger than the
pressure of the carrier line, which has the same pressure as the reactor chamber.
Consequently, when the manifolds suddenly switch from “run” to “vent,” the gas will
flow backward from the vent line to the source bubblers, which has the same pressure
as the carrier line when it was in the “run” mode. Therefore, the two three-way valves
have to be open either to vent or to pump. There is also a throttle valve that ultimately
controls the pressure in the reactor.
Now we focus on the reactor chamber where the growth of the materials occurs.
Figure A.8 shows the schematic layout of the close-coupled showerhead reactor
chamber. The precursors are introduced into the reactor from the showerhead located
on the top. The group III and group V precursors and injected into the chamber
separately from alternating orifices distributed throughout the showerhead. This design
is advantageous because it minimizes the pre-reactions between the group III and group
IV precursors. The substrates are placed on graphite susceptor which is heated by a PBN
coated graphite heater. The temperatures are monitored by a thermal couple which is
placed directly underneath the susceptor and an optical probe located at the top. The
showerhead is cooled by an external chiller and the temperature of the showerhead is
set to 30 C. The chamber wall is cooled by the cooling water of the building. The
exhausted is located near the bottom of the chamber.
189
Figure A.8 Schematics of the Closed coupled showerhead layout [1]
190
A.4 References
[1] "Close Coupled Showerhead® 3x2 inch · 6x2 inch," Aixtron, [Online]. Available:
http://www.aixtron.com/fileadmin/documents/matrix/r_d_systems/CCS_3x2_
6x2_inch.pdf..
Abstract (if available)
Abstract
Efficient green emitting LEDs and monolithic white light emitting LEDs require the extension of the range of efficient light emission in the GaN/InGaN materials system. We demonstrate high efficiency green and yellow light emitting multiple quantum well (MQW) structures grown on GaN nanostripe templates. The structures show promise for realizing high‐efficiency phosphor‐free white LEDs. The nanostripe dimensions range from 100 nm to 300 nm and have separations that range from 300 nm to 1 micron. Such small lateral dimensions are chosen to promote the elimination of threading dislocations from the structures. Nanostripes with various semipolar surfaces are grown with selective area growth on e‐beam lithography patterned c‐plane GaN where the mask openings are oriented between the [10‐10] and [11‐20] directions. Photoluminescence (PL) measurement shows that MQWs grown on stripes with (10‐11) surfaces and triangular shape emit the longest peak wavelength and have the best surface stability. Efficient PL emission peak wavelengths as long as 570 nm are realized on the triangular stripes with (10‐11) surfaces by optimizing the MQW growth conditions for long wavelength emission. Power dependent PL measurement shows linear response over more than three orders of magnitude of excitation power, indicating high radiative recombination efficiency. LED structures that utilize MQWs grown on nanostripes with (10‐11) surfaces were fabricated to further demonstrate the viability of the approach.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Nanorod-based InGaN/GaN core-shell nanoLEDs
PDF
GaN nanostructures grown by selective area growth for solid-state lighting
PDF
Efficiency droop in indium gallium nitride light emitters: an introduction to photon quenching processes
PDF
GaN power devices with innovative structures and great performance
PDF
Integrated photonics assisted electron emission devices
PDF
Integration of cost effective luminescent materials into organic light emitting diodes (OLEDs) and molecular approaches to improve the device efficiency
PDF
Optimizing nanoemitters using quasi-aperiodicity and photoelectrochemical etching
PDF
Light-matter interactions in engineered microstructures: hybrid opto-thermal devices and infrared thermal emission control
PDF
Multilayer grown ultrathin nanostructured GaAs solar cells towards high-efficiency, cost-competitive III-V photovoltaics
PDF
Nanostructure electronic band modeling for solar cell and lighting applications
PDF
Semiconductor devices for vacuum electronics, electrochemical reactions, and ultra-low power in-sensor computing
PDF
Integrating material growth and device physics: building blocks for cost effective emerging electronics and photonics devices
PDF
2D layered materials: fundamental properties and device applications
PDF
Printed electronics based on carbon nanotubes and two-dimensional transition metal dichalcogenides
PDF
Optical studies in photonics: terahertz detection and propagation in slot waveguide
PDF
Nanophotonic light management in thin film silicon photovoltaics
PDF
GaAs nanowire optoelectronic and carbon nanotube electronic device applications
PDF
Growth, characterization of gallium arsenide based nanowires and application in photovoltaic cells
PDF
Light management in nanostructures: nanowire solar cells and optical epitaxial growth
PDF
Low-dimensional material based devices for neuromorphic computing and other applications
Asset Metadata
Creator
Nakajima, Yoshitake
(author)
Core Title
Efficient yellow and green emitting InGaN/GaN nanostructured QW materials and LEDs
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Electrical Engineering
Publication Date
12/06/2016
Defense Date
12/15/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
green gap,III-N,III-nitride,LEDs,MQW,nanostructure,OAI-PMH Harvest,sempipolar plane
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Dapkus, P. Daniel (
committee chair
), Ravichandran, Jayakanth (
committee member
), Wu, Wei (
committee member
)
Creator Email
ynakajim@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c16-675533
Unique identifier
UC11336364
Identifier
etd-NakajimaYo-4959.pdf (filename),usctheses-c16-675533 (legacy record id)
Legacy Identifier
etd-NakajimaYo-4959.pdf
Dmrecord
675533
Document Type
Dissertation
Rights
Nakajima, Yoshitake
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
green gap
III-N
III-nitride
LEDs
MQW
nanostructure
sempipolar plane