Tài liệu hạn chế xem trước, để xem đầy đủ mời bạn chọn Tải xuống
1
/ 142 trang
THÔNG TIN TÀI LIỆU
Thông tin cơ bản
Định dạng
Số trang
142
Dung lượng
2,38 MB
Nội dung
CHARACTERIZATION AND NUMERICAL SIMULATION OF GALLIUM
NITRIDE-BASED METAL-OXIDE-SEMICONDUCTOR HIGH ELECTRON
MOBILITY TRANSISTOR WITH HIGH-K GATE STACK
LOW KIM FONG EDWIN
(B. ENG (HONS.), NATIONAL UNIVERSITY OF SINGAPORE)
A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF ENGINEERING
DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE
ACKNOWLEDGEMENTS
The author would like to acknowledge his heartfelt gratitude to his supervisors,
Associate Professor Tan Leng Seow and Dr. Yeo Yee Chia as well as his mentor from
the Data Storage Institute, Dr. Lee Hock Koon. Their guidance and attitude towards
excellence has always been a source of inspiration. Their care and concern for the
author makes this journey to be one of not only academic in nature but also spiritual.
Their knowledge and integrity has taught the author that only when one possesses both
can success be possible.
In addition, the author would also like to take this opportunity to thank the mentors at
GlobalFoundries Singapore. Special thanks go to Dr. Lap Chan, Dr. Ng Chee Mang
and Mr. Leong Kam Chew. Without this scholarship opportunity, the author would not
have been able to go on this enlightening journey.
The author would also like to thank the staff and friends in Silicon Nano Devices
Laboratory (SNDL) and Data Storage Institute (DSI) for their help and companionship.
It has been a challenging journey but enlightening nonetheless.
Last but not least, the author would like to thank his friends and family for putting up
with him throughout this journey. Without their encouragement and support, this
journey would not have been possible. Special mention is given to the author’s
girlfriend, Charmaine Yeong, for enduring these difficult times with the author
together. Thank you.
I
CONTENTS
ACKNOWLEDGEMENTS
I
CONTENTS
II
SUMMARY
V
LIST OF TABLES
LIST OF FIGURES
LIST OF SYMBOLS AND ABBREVIATIONS
Chapter 1 Introduction
VIII
IX
XV
1
1.1
Background
1
1.2
Scope and Purpose
5
1.3
Organisation of Thesis
8
Chapter 2 Theory of GaN-based Heterojunction Structures and Devices
2.1
GaN Crystalline Structure
2.2
Polarisation effect in AlGaN/GaN Heterostructure
9
9
10
2.2.1
Spontaneous Polarisation
11
2.2.2
Piezoelectric Polarisation
15
2.3
High Electron Mobility Transistor
17
2.4
Metal-Oxide-Semiconductor High Electron Mobility Transistor
18
Chapter 3 Device Fabrication and Characterisation
3.1
Mask Design
21
21
3.1.1
Mesa Isolation
22
3.1.2
Source-Gate/Drain-Gate Separation
22
II
3.1.3
3.2
Other Test Structures
23
Device Fabrication Procedure
29
3.2.1
Surface Passivation
33
3.2.2
Diamond-like Carbon (DLC) Layer
34
3.3
Electrical Characterisations of the MOSHEMTs
37
3.3.1
Effect of Surface Passivation on the ID-VGS Characteristics
38
3.3.2
Effect of Surface Passivation on ID-VDS Plot
39
3.3.3
Effect of DLC Liner on C-V characteristics
40
3.3.4
Effect of DLC liner on the ID-VGS Characteristics
41
3.3.5
Effect of DLC compressive stress liner on the ID-VDS Characteristic 43
Chapter 4 Numerical Simulation Fundamentals
4.1
Simulation Basics
45
45
4.1.1
Device Structure
49
4.1.2
Carrier Transport
56
4.1.3
Mobility Model
58
4.2
Effect of different AlGaN Layer thickness
63
4.3
Effect of Interface Traps on Device Operations
73
4.3.1
Effect of Surface Passivation
Chapter 5 Simulation of the Effect of Stress on AlGaN/GaN MOSHEMT
77
80
5.1
Device Structure and Stress Model
81
5.2
Effect of Stress on Electrical Characteristics
87
5.3
Effect of Stress on Carrier Concentration
89
5.4
Effect of Stress on Carrier Mobility
91
III
Chapter 6 Conclusion
6.1
Possible Future Work
94
96
References
97
Appendix A
111
Appendix B
117
List of Publications
124
IV
SUMMARY
In this work, the fabrication of the AlGaN/GaN MOSHEMT was performed and the
process was preceded by the design of the necessary mask set. In addition to the
devices that were required for characterisation, several other support and test structures
were designed and included into the mask layout. These structures included the Van
Der Pauw structure, Vernier Alignment Scale structure, Transmission Line Model
(TLM) structure, Step Coverage structure and devices with long gate widths.
In the fabrication of the devices, the splits of the experiments with different
configurations were discussed with focus on the objective of these processes and their
characterisation. Electrical measurement was the main mode of characterisation in this
work. Methods such as the C-V, ID-VGS and ID-VDS measurement were performed on
experimental devices. The enhancements in device performance due to certain
fabrication processes employed in the experiments were evident from the results of the
electrical characterisations.
In order to better understand these enhancements, a numerical simulation project was
undertaken and the results of these simulations are presented in this thesis. Sentaurus
TCAD simulation suite was used for the device simulations. The simulation physics
and models were discussed before some preliminary simulation results were presented.
The simulation results presented was generally in line with the device physics
understood from various literatures and in some cases shed light on the possible
V
mechanism of the performance enhancement of the devices by some of the fabrication
techniques introduced in the experiments.
Firstly, simulation of the effect of different thickness of AlGaN barrier layer and their
effect on device performance was performed and this gave results that were in line
with the literature as described by Ibbetson in [1].
Next, the effect of interface states on device performance was simulated. The reduction
of donor-like states in the interface between the AlGaN and high-k gate dielectric
caused the 2DEG density to reduce accordingly, resulting in a lower drive current.
However, this is not consistent with the surface passivation technique discussed in the
fabrication process. This process was shown in the electrical characterisation to
improve the drive current by more than 50%. Thus, a different mechanism was
proposed in the simulation.
Acceptor-like traps and donor-like traps were postulated to be present at the abovementioned interface. The hypothesis is that the surface passivation technique would
tend to passivate more acceptor-like traps. This effect was simulated and the results of
the simulation show a close resemblance of the effect seen in the characterisation of
the experimental devices.
Finally, the effect of a Diamond-Like Carbon (DLC) stress liner on the MOSHEMT
was also simulated and it was shown that this compressive stress liner not only
VI
provided a compressive stress in the gate region but also a coupled tensile stress in the
region beyond the gate region. This change in carrier distribution resulted in a positive
shift in threshold voltage as well as a slight enhancement in drive current. It is noted
that in this simulation, the stress was simulated using a separate simulation program
and the resulting polarisation charge was input into the Sentaurus TCAD simulation
suite separately, thus the effect of change in mobility due to band structure
deformation [2] and change in trap concentration [3] were not simulated accordingly.
For this reason, this simulation could only be a qualitative description of the actual
effect of stress on AlGaN/GaN MOSHEMTs.
VII
LIST OF TABLES
Table I.
A comparison of semiconductor material parameters at
300K is presented [4].
Table II.
2
Competitive Advantage of GaN devices in terms of the
requirements of different types of devices is presented
[5].
Table III.
A comparison of the lattice constants, spontaneous and
piezoelectric constants/coefficients of GaN and AlN [6].
Table IV.
53
The default coefficients for GaN in the Masetti model
[7].
Table V.
3
60
The default coefficients for Silicon in the Lombardi
Model [7]. These coefficients are used for the simulation
due to a lack of these values for GaN in the literature.
Table VI.
63
The configurations of device used for simulation with
varying AlGaN thickness.
65
VIII
LIST OF FIGURES
Figure 1.1. Illustration of a typical structure of a Heterojunction Field Effect Transistor.
4
Figure 2.1. A wurtzite structure of GaN crystal (Ga-faced) and its various planes [20].9
Figure 2.2. A Wurtzite crystal of GaN (N-face) with lattice constants c and a [9].
11
Figure 2.3. The GaN crystal is often represented by the tetrahedral stick and ball figure
- P0 refers to the polarisation vector in the structure [10].
12
Figure 2.4. A representation of the polarisation profile of a nitride-based
semiconductor heterostructure [10].
13
Figure 2.5. A graphical representation of the spontaneous polarisation in nitride-based
alloys according to Vegard's Law interpolation [26].
15
Figure 2.6. The tetrahedral stick and ball representation of GaN undergoing tensile
strain [10].
16
Figure 2.7. The energy band diagram of AlGaN/GaN Heterojunction.
19
Figure 2.8. A typical illustration of a Metal-Oxide-Semiconductor High Electron
Mobility Transistor.
20
Figure 3.1. a) A typical transistor design for the purpose of DC characterisation. W, L
and LD-G referred to the gate width, gate length and drain-gate separation respectively.
b) The width of a mesa island would dictate the gate width of a transistor.
22
Figure 3.2. The design of a Van Der Pauw Structure in this mask set design.
23
Figure 3.3. The design of the Vernier Alignment Scale Structure used in this mask set
design.
24
IX
Figure 3.4. The design of the Transmission Line Model (TLM) Structure used in this
mask set design.
25
Figure 3.5. An example of a typical Resistance vs. Contact spacing plot extracted from
TLM measurement.
26
Figure 3.6. An 3D illustration of the gate metal being deposited over a step of the mesa
island.
27
Figure 3.7. The design of the step coverage test structure used in this mask set design.
27
Figure 3.8. The design of the long gate width transistor design.
28
Figure 3.9. The starting layer structure of HEMT wafer before the fabrication process.
29
Figure 3.10. The fabrication process of AlGaN/GaN MOSHEMT [31].
30
Figure 3.11. An illustration of the device schematic of a completed AlGaN/GaN
MOSHEMT.
32
Figure 3.12. An illustration of the device schematic of an AlGaN/GaN MOSHEMT
with a DLC stress liner.
33
Figure 3.13. The average compressive stress induced in the channel due to 40 nm DLC
simulated for various LG: 300 nm, 500 nm and 700 nm [31].
36
Figure 3.14. The effect of the surface passivation technique on the ID-VGS plot of
MOSHEMTs with LG = 2 µm. Incorporation of In situ SiH4 passivation improved
subthreshold swing to 100 mV/decade and gm of 95 mS/mm [31].
38
Figure 3.15. The effect of surface passivation on ID-VDS plot of MOSHEMT with LG =
2 µm. ID,sat of 624 mA/mm measured at VD = 10 V and VG = 4 V was observed.
39
X
Figure 3.16. The C-V characteristics of a control device and the device with 40 nm
DLC liner deposited on a MOSHEMT with LG = 300nm.
40
Figure 3.17. The effect of the DLC stress liner (Compressive Stress) on the ID-VGS
charactieristics of a MOSHEMT with LG = 300 nm. Low SS of 110 mV/decade and
high gm of 88 mS/mm were achieved for the device with DLC liner [31].
42
Figure 3.18. The effect of the DLC Stress liner (Compressive Stress) on the ID-VDS
characteristic with LG = 300 nm [31].
43
Figure 4.1. A flowchart showing the process of device simulation (Names in bracket
denote name of program used for that purpose).
46
Figure 4.2. An example of a Meshing Strategy for a 2D device in Sentaurus simulation
program [7].
48
Figure 4.3. The schematic diagram of the simulated structure in Sentaurus TCAD.
49
Figure 4.4. An illustration of the interfaces with its polarisation and the respective
induced polarisation charges.
52
Figure 4.5. The graph of the 2DEG density measured as a function of Al0.34Ga0.66N
barrier thickness @ room temperature. Solid dots are experimental data, curve
represent least square fit of equation (4.22) up to 15 nm [1].
64
Figure 4.6. Simulated ID-VGS characteristics of devices (VDS = 0.1 V) with different
AlGaN thickness. The threshold voltage are, VTH(20 nm) = -4.20 V, VTH(15 nm) = 3.25 V, VTH(10 nm) = -2.25 V, VTH(5 nm) = N.A.
66
Figure 4.7. The electron density mapping in device with 5nm thick AlGaN layer.
67
Figure 4.8. The electron density mapping in device with 10 nm thick AlGaN layer. 68
Figure 4.9. The electron density mapping in device with 15 nm thick AlGaN layer. 68
XI
Figure 4.10. The electron density mapping in device with 20 nm thick AlGaN layer. 69
Figure 4.11. A plot showing the simulated 2DEG density as a function of Al0.25Ga0.75N
barrier thickness. The 2DEG density ranges from 8.69×104 cm-2 @ 5nm to 5.90×1012
cm-2 @ 20nm AlGaN layer thickness.
70
Figure 4.12. The energy band diagram of AlGaN/GaN MOSHEMT with 5 nm thick
AlGaN layer extracted from the simulation.
71
Figure 4.13. The energy band diagram of AlGaN/GaN MOSHEMT with 20 nm thick
AlGaN layer extracted from the simulation.
72
Figure 4.14. A schematic diagram showing where energy level of donor-like surface
states reside in a) thin AlGaN layer that is < critical thickness b) thicker AlGaN layer >
critical thickness [1].
73
Figure 4.15. The energy level of the trap in a device with 20 nm thick AlGaN layer
being programmed in this simulation run.
74
Figure 4.16. A ID-VGS plot of simulated devices with different donor-like trap
concentration.
75
Figure 4.17.The simulated space charge data on AlGaN/GaN MOSHEMT at gate
region extracted from the simulation.
76
Figure 4.18. The ID-VGS plot of devices with different concentration of acceptor-like
trap.
78
Figure 4.19. The ID-VDS plot of devices with different concentration of acceptor-like
traps.
79
XII
Figure 5.1. The schematic view of the AlGaN/GaN MOSHEMT with a compressive
Diamond-Like Carbon (DLC) liner. The 2-DEG density decreases with application of
compressive stress.
81
Figure 5.2. The schematic diagram of the device being simulated in Sentarus TCAD
and Abaqus for the effect of compressive stress liner. The X and Y-axes denote the
device geometry that was used in this simulation.
82
Figure 5.3. A map depicting the StressXX distribution that is experienced in a
simulated AlGaN/GaN device with LG = 300 nm due to 40nm of DLC compressive
stress liner.
83
Figure 5.4. A plot of the average StressXX distribution that is experienced in the
AlGaN layer of the device structure shown in Figure 5.2.
85
Figure 5.5. a) Polarisation induced charges at the interface of AlGaN/GaN due to
compressive stress by the DLC stress liner, dashed horizontal line denotes the uniform
polarisation charge in a non-stressed device. b) A corresponding device schematic
illustrating the device geometry.
86
Figure 5.6. The ID-VGS plot of the MOSHEMT device with and without the DLC stress
liner. The gate length of the devices, Lg = 300nm.
87
Figure 5.7. The effect of compressive stress from DLC stress liner on the ID-VDS plot of
simulated device with Lg = 300nm.
88
Figure 5.8. A plot of the simulated 2DEG density in the x-direction of the device at a
depth of 1 nm below the AlGaN/GaN interface. A reduction of electron density is
observed in the channel region under the gate while a slight increase in electron
density is observed outside the gate region.
90
XIII
Figure 5.9. A plot of the simulated electron mobility value along the x-axis of the
MOSHEMT with and without stress, at a depth of 1 nm below the AlGaN/GaN
interface.
91
Figure 5.10. A plot of the simulated vertical electric field along the x-direction of the
MOSHEMT with and without stress, at a depth of 1nm below the AlGaN/GaN
interface.
93
XIV
LIST OF SYMBOLS AND ABBREVIATIONS
ABBREVIATIONS
DESCRIPTION/ EXPANSION
GaN
Gallium Nitride
AlGaN
Aluminium Gallium Nitride
TaN
Tantalum Nitride
HFET
Heterojunction Field Effect Transistor
MODFET
Modulation Doped Field Effect Transistor
HEMT
High Electron Mobility Transistor
MOSFET
Metal-Oxide-Semiconductor Field Effect
Transistor
MESFET
Metal-Semiconductor Field Effect Transistor
2-DEG
2-Dimensional Electron Gas
CFOM
Combine Figure of Merit
SiC
Silicon Carbide
AlN
Aluminium Nitride
GaAs
Gallium Arsenide
i-GaN
Intrinsic Gallium Nitride
DC
Direct Current
RIE
Reactive Ion Etching
MOCVD
Metal Organic Chemical Vapour Deposition
PDA
Post Deposition Anneal
FCA
Filtered Cathodic Arc
DLC
Diamond-like Carbon
XV
TLM
Transmission Line Model
VA
Vacuum Annealing
C-V
Capacitance-Voltage
DUT
Device Under Test
SYMBOL
DESCRIPTION/ EXPANSION
Eg
Band Gap
vsat
Saturation Velocity
χ
Thermal Conductivity
ε
Dielectric constant
µe
Electron Mobility
LS-G
Source-Gate Length
LD-G
Drain-Gate Length
Rc
Contact Resistance
LT
Transfer Length
Vth
Threshold Voltage
Cox
Oxide Capacitance
VGS
Gate Voltage
VDS
Drain Voltage
ID
Drain Current
SS
Sub-threshold Swing
PPE
Piezoelectric Polarisation
XVI
PSP
Spontaneous Polarisation
ϵ
Strain
Polarisation Charge
ψ
Electrostatic Potential
Lg
Gate Length
NA
Density of ionised acceptor
ND
Density of ionised donor
XVII
Chapter 1 Introduction
1.1
Background
Group III-Nitride semiconductors have long been hailed as promising materials for
optoelectronic devices, high power or high temperature electronic devices as well as
for high-frequency applications. This is due largely to the intrinsic properties of these
nitride semiconductors. Thus, the semiconductor industry has shown particular interest
in Gallium Nitride (GaN)-based transistors. The global market for GaN-based devices
experienced a 260% growth in 5 years from 1998 to 2003 [8]. However this growth is
largely limited to the optoelectronics market as the microelectronics market for GaN is
still at its infancy.
Several other III-V compound semiconductors, such as Arsenide-based and
Phosphide-based semiconductors, are also being aggressively researched on. However,
GaN with its excellent material properties and electron transport properties prove to be
a suitable material for general electronics with high temperature, high power or high
frequency requirements. Table I compares the various parameters of GaN, other III-V
compound semiconductors, and Silicon. The combined figure of merit (CFOM) also
shows that it is one of the most suitable materials for high temperature, high power or
high frequency applications [4].
1
Table I. A comparison of semiconductor material parameters at 300 K [4].
Property
Bandgap Eg (eV)
Si
GaAs
4H-SiC
GaN
1.12
1.42
3.25
3.40
0.25
0.4
3.0
4.0
1350
6000
800
1300
1.0
2.0
2.0
3.0
1.5
0.5
4.9
1.3
11.8
12.8
9.7
9.0
1
8
458
489
Breakdown field EB
(MV/cm)
Electron mobility µ
2
(cm /V s)
Maximum velocity vsat
7
(10 cm/s)
Thermal conductivity
χ (W/cm K)
Dielectric constant ε
CFOM = (χ ε µ vsat EB)
/ ( χ ε µ vsat EB)Si
CFOM: Combined figure of merit for high temperature/high power/high frequency
applications.
GaN with a large bandgap (Eg = 3.40 eV), large critical breakdown field of 4.0 MV
cm-1, coupled with good electron transport properties (theoretical electron mobility, µe,
of up to 2000 cm2 V-1s-1 [9] and a peak saturation velocity, vsat, of 3.0×107 cm s-1 at
room temperature) and good thermal conductivity makes it the basic material in the
nitride class of material that is typically used for all device layers that requires fast
2
Table II. Competitive Advantage of GaN devices in terms of the requirements of different types of
devices [5].
Needs
Enabling Feature
High Power
Wide Bandgap, High Field
Performance Advantage
Compact, Ease of
Impedance Matching
Eliminate/Reduce need for
High Voltage
High Breakdown Field
voltage conversion
Operation
Bandwidth, µ-wave/mmHigh Frequency
High Electron Velocity
wave
High dynamic range
Low Noise
High Gain, High Velocity
receivers
High Temperature
Wide Bandgap
Rugged, Reliable
Operations
Direct Bandgap
Technology Leverage
Enabler for Lighting
carrier transport or a high breakdown voltage. Table II presents the competitive
advantages that GaN devices offer.
Several types of devices had been fabricated using GaN with the objective of
exploiting its outstanding intrinsic property for transistor usage. Research groups
around the world are still trying to develop the best baseline process to build devices
3
Figure 1.1. Illustration of a typical structure of a Heterojunction Field Effect Transistor.
such as GaN based Metal-Oxide-Semiconductor Field Effect Transistor (MOSFET)
and Metal-Semiconductor Field Effect Transistor (MESFET). However, due to the
immaturity of the technology of material growth [8], it is still a challenge to make high
quality epitaxial layer on various substrates. Thus GaN layers often come with high
density of defects, primarily due to the difference in lattice constant of GaN and the
substrate it is grown on. This makes surface channel devices generally more
susceptible to anomalies due to these defects.
In order to overcome this challenge, the Heterojunction Field Effect Transistor (HFET)
is a generally preferred candidate for the fabrication of GaN-based devices (Figure 1.1)
as it presents a buried channel which is less susceptible to these defects. These
transistors are also commonly known as Modulation Doped Field Effect Transistors
(MODFETs) or High Electron Mobility Transistors (HEMTs). The MODFET
generally refers to a HFET where one of the layers is doped to supply the channel
4
layer, which is generally undoped, with carriers to form the conductive channel. This
conductive channel is usually confined in a quantum well due to bandgap
discontinuity and consequently, band-bending that occurs in such structures. It is the
nature of the confinement of this conductive channel that this channel is often known
as the 2-Dimensional Electron Gas (2DEG). This buried 2DEG channel eliminates the
effect of impurity scattering due to the absence of dopants in the channel. This allows
the achievement of high electron mobility. However, the disadvantage of such a
device would be a reduction of gate control as the gate would generally be further
away from the channel, as compared to a surface-channel device.
In a GaN-based HFET, doping of a supply layer is not necessary as there is presence
of a strong polarisation effect [10] in an AlGaN/GaN heterojunction. This polarisation
effect allows the formation of the 2DEG without any doping involved. The
polarisation effect and the operation of the HEMT device will be further elaborated in
the next chapter.
1.2
Scope and Purpose
Having discussed the advantages of GaN semiconductor material as being a suitable
candidate for high power and high temperature purposes, a GaN-based HEMT would
thus be the choice candidate for electronic devices capable of withstanding high power
and high temperature. However, the GaN HEMT does face some challenges despite
the various advantages.
5
Firstly, the GaN growth technology is still at its infancy and GaN epitaxial layers
required in HFETs often have high concentration of bulk and surface defects. These
defects would include dislocations and possible dangling bonds on the barrier surface.
Although the 2DEG is buried, charges trapped by these defects could cause Coulomb
scattering to take place in the channel. In addition, it could also degrade highfrequency performance by reducing the cut-off frequency [11, 12] and even affect the
breakdown voltage as the defects might enhance impact ionisation due to the electric
field concentration at the gate electrode edge [13]. This could pose serious degradation
problems in high power and high frequency devices.
Secondly, there is a substantial gate leakage current present in such a structure. Unlike
the MOSFET structure, the absence of a gate dielectric would be detrimental to the
gate current leakage levels. Comparing this gate leakage current to an ideal Schottky
gate device shows that in the AlGaN/GaN HEMT device, the Schottky gate leakage
current is very much higher than that of an ideal Schottky gate reverse current
requirements [14, 15]. Thus a MOSHEMT would be able to overcome these
difficulties faced by a HEMT device.
There is also a problem of self-heating in such structure. It was mentioned previously
that GaN has good thermal conducting properties. However, as the GaN epitaxial layer
is most often grown on sapphire, being the most cost effective substrate available
currently, heat might be not be efficiently channelled away through the sapphire
substrate which is a poor thermal conductor. Solutions have been offered such as
6
changing the substrate to silicon (Si) or silicon carbide (SiC) [16] as these substrates
have lattice constant that are closer to that of GaN and are able to conduct heat better
than sapphire. However, the growth process of GaN on such substrates is usually more
difficult or expensive.
One of the more pressing challenges that the GaN-based HEMT faces is the realisation
of enhancement-mode devices. GaN HEMTs are usually depletion-mode devices
which are less desirable in electronic circuit applications. In order to realise
enhancement-mode devices, several different techniques has been proposed. These
include recessed gate technique [17], surface treatment [18] and an addition of a cap
layer [19] to alter the band structure. In this thesis, experimental devices with some of
these techniques used in its fabrication will be characterised. In addition, simulations
will also be carried out to better understand these processes.
In this thesis, characterization and simulation studies were performed to study the
possible performance enhancement of GaN-based HEMTs for applications in highpower electronics. Simulation work was performed to understand the effect of
different parameters on the electrical performance of the device. Effect of various
parameters such as barrier layer thickness, concentration and energy level of interface
traps as well as thermal effects were investigated. The simulation of the effect of stress
due to the DLC liner on the performance of the GaN MOSHEMTs is also done.
7
1.3
Organisation of Thesis
This thesis is organised into 6 main chapters. Chapter 2 provides insights on the
theoretical background of heterojunction structures and introduces the different
devices that are characterised and simulated in this work. Chapter 3 presents the
characterisation techniques that are used in this project as well as the relevant
characterisation results that had been performed. Chapter 4 explains the fundamental
models that are used in this simulation work and the initial simulation work done. It
also discusses the capability of the simulation software to implement the theoretical
models that had been presented in the literature. Chapter 5 describes the simulation
work that was done to explore the effect of stress on GaN-based HEMTs, which
among other things, could possibly lead to eventually fabricating of an enhancementmode device in future. Chapter 6 concludes this thesis and suggests future research
directions.
8
Chapter 2 Theory of GaN-based Heterojunction Structures and
Devices
In order to gain useful insights in the work performed here, it is essential to first have
a good understanding of the GaN material as well as phenomena that are observed in
GaN-based heterostructures.
2.1
GaN Crystalline Structure
It is not the scope of this thesis to present an extensive description of the GaN crystal
structure. However, to achieve a meaningful discussion in this work, a concise
understanding of the GaN crystalline structure is necessary.
GaN can exist in three different crystalline structures, namely, wurtzite, zincblende
and rocksalt structures. The wurtzite structure is thermodynamically most stable for
bulk GaN and is shown in a stick and ball representation in Figure 2.1.
Figure 2.1. A wurtzite structure of GaN crystal (Ga-faced) and its various planes [20].
9
The c-plane is the most common plane that GaN-based devices are built on, due to its
polar nature and existence of the strongest polarisation effect in that plane. The
direction that is perpendicular to the c-plane is known as the c-direction and is also
known as the (0001) direction.
In this thesis, all the GaN-based crystal structure used in devices is of the wurtzite
structure configuration.
2.2
Polarisation effect in AlGaN/GaN Heterostructure
Polarisation effects in crystalline structure had long been observed in GaN or in
wurtzite structure due to the non-centro-symmtrical structure. However it was R. D.
King-Smith’s theoretical calculations [21] that provided detail study of these
polarisation effects quantitatively, especially those in GaN-based heterostructure. It is
this polarisation effect that allows the GaN-based HFET to be fabricated with
nominally intrinsic layers in its heterostructure, i.e. without any deliberate doping.
However, this brings about the debate of where the carriers that are present at the
interface of these devices come from, given the intrinsic nature of its heterostructure
layers.
Observations of the 2DEG density having a close dependence on the thickness and
composition of the AlGaN barrier layer has led many to believe that these carriers may
originate from the surface donor states of the AlGaN surface. It is useful to note that
10
these polarisation effects alone do not cause the formation of the 2DEG [1]. Instead, it
influences the concentration and distribution of these carriers [22].
There are fundamentally two kinds of polarisation effects that exist in a GaN-based
heterostructure; namely, spontaneous polarisation and piezoelectric polarisation.
These will be discussed in detail in the following sub-sections.
2.2.1
Spontaneous Polarisation
Polarisation is dependent on the polarity of the crystal structure. This is especially so
for spontaneous polarisation (polarisation at zero strain), which is primarily due to
bonds between the cation (Ga) site and the anion (N) site along the c-direction (0001)
in a GaN crystal structure (Figure 2.2) being non-centro-symmetric in structure [23].
(0001)
Figure 2.2. A Wurtzite crystal of GaN (N-face) with lattice constants c and a [9].
11
Figure 2.3. The GaN crystal is often represented by the tetrahedral stick and ball figure - P0
refers to the polarisation vector in the structure [10].
It is beyond the scope of this thesis to fully discuss polarisation charges in the
extensive manner, which is presented in [24] [25], [21]. However, it is possible to
adopt a simplified understanding that allows for a comprehensive calculation of
relevant charges that are induced by these polarisation effects.
The polarisation vector in a GaN crystal can be clearly illustrated by the tetrahedral
stick and ball figure as shown in Figure 2.2. The directions of the polarisation in the
structure are defined as such due to the electron clouds being closer to the N atoms
[10]. This difference in electronegativity together with the non-centro-symmetrical
structure causes a polarisation dipole to exist between the cation (Ga) site and anion
(N) site in the GaN stick and ball model shown in Figure 2.3. The horizontal
component of the polarisation is usually assumed to have cancelled each other out. It
12
is useful to note that in the discussion of spontaneous polarisation, the vertical
component of the polarisation (under no strain) in the three diagonal bonds have been
taken into account.
The polarisation dipole itself would not have constituted a polarisation charge existing
at the interface of the heterojunction. Instead it is the difference in polarisation across
the interface that would result in a polarisation charge at this heterointerface.
Depending on the composition of the adjacent layers that are deposited, the difference
in electronegativity across the heterointerfaces can be very different, thus resulting in
different polarisation. This difference in polarisation between two nitride
semiconductors will manifest itself as a polarisation charge across the heterointerface.
Figure 2.4. A representation of the polarisation profile of a nitride-based semiconductor
heterostructure [10].
13
The polarisation charges due to spontaneous polarisation are not always well-defined
as it requires the polarisation between two phases to be connected by a pseudomorphic
boundary. Such a transition between two phases can be found in GaN-based
heterojunctions. For example (Figure 2.4), another nitride-based semiconductor, an
AlN layer, is grown pseudomorphically on GaN to form such a heterojunction. The
resulting polarisation induced charge, governed by Gauss’s Law [22], can then be
given by
,
where
( 2.1 )
refers to the gradient of polarisation in space, i.e. difference in polarisation
across the heterointerface.
Spontaneous polarisation in various nitride-based semiconductors can be easily
calculated by making use of a Vegard-like rule. The use of Vegard’s Law in
polarisation is a valid estimation as shown in [26]; this includes ternary or quaternary
nitride alloys. In quaternary nitrides, Vegard’s interpolation would be estimated by,
(
(
)
(
)
)
(
)
(
( 2.2 )
)
(
).
This equation can also be loosely represented by Figure 2.5.
14
Figure 2.5. A graphical representation of the spontaneous polarisation in nitride-based alloys
according to Vegard's Law interpolation [26].
2.2.2
Piezoelectric Polarisation
Spontaneous polarisation is also commonly known as the polarisation that exists
without strain. However, in nitride-based heterostructures, there is rarely a case of
zero strain. The difference in lattice constant (Figure 2.5) or a difference in thermal
coefficient of expansion between the GaN layer and AlGaN layer would give rise to
either a compressive or tensile strain to be present in the epitaxial layers that are
grown.
Due to the non-centro-symmetrical nature of nitride-based crystal structure, this stress
would alter the polarisation due to the non-vertical triple bonds that exist in these
nitride-based crystal structures. To illustrate this point more clearly, it is useful to
consider a perfect tetrahedral stick and ball figure of GaN (Figure 2.6).
15
Figure 2.6. The tetrahedral stick and ball representation of GaN undergoing tensile strain [10].
When the GaN layer undergoes a tensile strain, there is a reduction of the vertical
polarisation component from the three diagonal bonds and thus it would represent a
change of polarisation. This difference in polarisation would then result in a
polarisation charge which is confined to the interface of the heterojunction.
Unlike the spontaneous polarisation where the polarisation is dependent solely on the
alloy type and composition of the alloy, piezoelectric polarisation would require more
information on different parameters to determine its resulting polarisation. These
parameters would include strain, piezoelectric coefficients and elastic constants, all of
which might be directly or indirectly affected by the composition of nitride alloys.
Theoretical calculations of both the spontaneous and piezoelectric polarisation and its
induced charges will be elaborated in Chapter 4 where such values will be used for
simulation purposes.
16
2.3
High Electron Mobility Transistor
After a discussion of the various polarisation effects, it is appropriate that the
discussion extends to the kind of devices that would utilise the benefits of such
polarisation charges that are present in a heterojunction. As mentioned previously, the
MODFET was first conceived to make use of the heterojunction characteristics. The
MODFET was more commonly a GaAs-based device. However, GaAs has very weak
polarisation effect compared to GaN and thus modulation doping was the defining
factor for the GaAs MODFET.
When the polarisation factor in GaN was discovered, people started to pay more
attention to the heterojunction field effect transistor (HFET) as a general class of
devices. Also known as the High Electron Mobility Transistor (HEMT), the ability to
induce a conductive 2 DEG in intrinsic GaN heterostructure layers of the device has
ignited interest of many people. In the rest of this thesis, we will be working on the
AlGaN/GaN HEMT structure that is currently the de facto norm for GaN-based
HEMTs.
The High Electron Mobility Transistor (HEMT) presents a few advantages over the
usual Metal-Oxide-Semiconductor Field Effect Transistor (MOSFET). In the HEMT
structure, the conduction band offset due to the heterojunction causes a triangular
quantum well to form just after the GaN/AlGaN interface (Figure 2.7). Carriers that
originate from surface donor states of the AlGaN surface [1] are thus confined in this
17
quantum well. The buried 2-DEG channel eliminates the effect of impurity scattering
due to absence of dopants in the channel, thus allowing for an enhancement in the
mobility of such a device.
2.4
Metal-Oxide-Semiconductor High Electron Mobility Transistor
The HEMT structure may be the choice device for GaN-based transistors. However,
like most devices, it is not without its own unique problems. It is observed that the
HEMT structure is similar to a MESFET structure where the Schottky gate is in
contact directly with the semiconductor. This inherently causes a problem of a
relatively high gate leakage current. A device without a gate dielectric would also
prove to be more susceptible to breakdown as compared to the MOSFET. In addition,
the HEMT, also tends to experience a phenomenon known as current collapse. Current
collapse is the reduction of drain current after certain conditions that a transistor is
subjected to. It is commonly seen when a GaN-based transistor is subjected to a high
drain-source bias. This could also lead to an increase in knee voltage [61].
18
AlxGa1-xN
GaN
Triangular
Quantum
Well
Figure 2.7. The energy band diagram of AlGaN/GaN Heterojunction.
Current collapse could be caused by a few possible reasons: 1) deep levels in the
barrier layer under the metal gate, 2) deep levels in the interface and/or in the buffer
layer, and 3) surface states [27]. The effect is notably caused by electrons being
trapped by these traps that reside in the area between the drain and the gate during
high voltage operations. This induces a virtual gate formation between the drain and
the gate, thus causing the phenomenon of drain current reduction.
In order to overcome these challenges, a gate dielectric is inserted between the gate
metal and the AlGaN layer. This addition of the gate dielectric in the structure gives
rise to a structure known as the Metal-Oxide-Semiconductor High Electron Mobility
Transistor (MOSHEMT) as shown in Figure 2.8.
19
Figure 2.8. A typical illustration of a Metal-Oxide-Semiconductor High Electron Mobility
Transistor.
The MOSHEMT structure is able to effectively reduce gate leakage current and in
thus improving the drive current capacity. In addition, the way the gate dielectric is
deposited allows the surface traps to be partially passivated to reduce the effect of
current collapse.
In this thesis, the MOSHEMT structure is the dominant structure that is being used for
various studies. The details on the various materials used as well as composition of the
various alloys will be discussed in detail in the chapters that follow.
20
Chapter 3 Device Fabrication and Characterisation
This chapter documents results of the design fabrication and characterisation of
AlGaN/GaN MOSHEMTs. Before the fabrication process is presented, a discussion of
the various considerations in the design of the device and other test structures is done.
The fabrication of the device is then described followed by the characterisation
techniques. Characterisation of AlGaN/GaN MOSHEMTs was performed in order to
study the effect of different techniques in fabrication, such as device passivation and
deposition of Diamond-Like Carbon (DLC) stress liner. Device fabrication and
characterisation was a collaborative work with graduate students Liu Xinke and Liu
Bin. While the author’s main contribution in this project is the mask design and
certain lithography steps, the author would like to acknowledge Liu Xinke’s
contribution in the fabrication process particularly in development of mesa etching
process, gate stack formation process, surface passivation process and the contact
formation process. The author would also like to acknowledge the contribution of Liu
Bin in the development of the DLC deposition process.
3.1
Mask Design
Several mask set designs were fabricated in the course of this work. There were a few
basic considerations that require close attention in order to design a useful and
functional mask set. Figure 3.1a shows a typical device design used for DC device
characterisation.
21
b)
a)
Gate
Gate Width
Drain
Source
Figure 3.1. a) A typical transistor design for the purpose of DC characterisation. W, L and LD-G
referred to the gate width, gate length and drain-gate separation respectively. b) The width of a
mesa island would dictate the gate width of a transistor.
3.1.1
Mesa Isolation
The formation of the mesa is an important step that requires several considerations.
The mesa isolation defines the gate width (Figure 3.1b) as well as the external
resistance of the device.
3.1.2
Source-Gate/Drain-Gate Separation
Apart from having a variety of devices with different gate dimensions such as gate
length and gate width, one of the factors that has to be taken into account for this mask
set was the Source-Gate (LS-G) and Drain-Gate (LD-G) separation. The device design or
layout depends on the application. For high power or high voltage application, the
separation between the drain and gate is an important parameter. In high voltage
22
applications, the drain voltage can be very high, e.g. 500 V. On the contrary, the gate
voltage would be relatively lower, e.g. 5 V. The potential difference between the gate
and drain would give rise to a very high electric field which could cause the device to
breakdown. In order to reduce this electric field between the gate and drain, it would
be useful to increase the Drain-Gate separation. However, an increase in this length
would also result in an increase in extrinsic series resistance. Thus, it would be
necessary to optimise this separation.
3.1.3
Other Test Structures
Apart from the devices that are being designed on the mask set, it is useful to have an
assortment of test structures in the same die. These structures are important in process
debugging as well as characterization of various device or material parameters.
100um
Figure 3.2. The design of a Van Der Pauw Structure in this mask set design.
23
3.1.3.1 Van der Pauw Structure
The Van Der Pauw Structure (Figure3.2) allows for the characterisation of the electron
mobility using the principles of the Hall Effect [28]. In the case of the GaN-based
HEMT structure, the drive current would be linearly dependent on the electron
mobility, in addition to the electron density in the 2DEG.
3.1.3.2 Vernier Alignment Scale Structure
Misalignment of the various mask layers during fabrication could be an issue if the
magnitude of misalignment is beyond tolerance. This could cause a device to fail due
to a short circuit or an open circuit depending on the direction and magnitude of the
misalignment. The Vernier Alignment Scale (Figure 3.3) structure allows for a quick
and effective evaluation of the degree of misalignment between two layers.
Figure 3.3. The design of the Vernier Alignment Scale Structure used in this mask set design.
24
3.1.3.3 Transmission Line Model (TLM) Structure
The Transmission Line Method is a common method used to determine contact
resistance which can be also used to calculate the sheet resistance, specific contact
resistance as well as the transfer length [29],[30].
The TLM structure consists of an active region (mesa) with metal contacts placed at
positions spaced apart at different distances as shown in Figure 3.4.
In order to obtain the contact resistance, the resistance between two pads at each
separation distance are measured by a simple I-V measurement. After all the
resistances are measured, the data is used to plot a resistance versus contact spacing
graph, as shown in Figure 3.5.
The intercept on the Y axis (Total Resistance) would give the value of two times of
the contact resistance (2 Rc) while the intercept of the X axis would provide the value
of two times the transfer length (2 LT).
100um
Figure 3.4. The design of the Transmission Line Model (TLM) Structure used in this mask set
design.
25
Total resitance RT)
1600
1400
1200
1000
800
600
400
-100
0
100 200 300 400
Contact Spacing d (m)
500
Figure 3.5. An example of a typical Resistance vs. Contact spacing plot extracted from TLM
measurement.
The sheet resistance and specific contact resistance can then be calculated by equation
(3.1) and (3.2) respectively;
( 3.1 )
( 3.2 )
where W is the width of the contact pads on the TLM structure.
3.1.3.4 Step Coverage Structure
It is observed, in the device design, that there are several process steps in the
fabrication that would require deposition of materials over an etched step. This is
clearly illustrated at the step where Tantalum Nitride (TaN, material for the metal gate)
is deposited over an island after the mesa isolation etching, as shown in Figure 3.6.
26
Gate Metal
Mesa Isolation
Gate Dielectric
Figure 3.6. An 3D illustration of the gate metal being deposited over a step of the mesa island.
A break in the gate metal could occur if the etched depth of the mesa isolation is too
deep. This breakage can be difficult to detect through the optical microscope. The step
coverage structure (Figure 3.7) was design to effectively determine if there is any
discontinuity in the gate metal deposition.
100um
Figure 3.7. The design of the step coverage test structure used in this mask set design.
27
The thin green lines are representation of the gates with various gate lengths while the
red boxes are the mesa isolations. If there is any discontinuity in the gate metal
deposition, a simple I-V measurement across the two corresponding pads would show
a high resistance reading.
3.1.3.5 Long Gate Width Device Structure
GaN-base devices are most often used for high power applications. In such high
power applications, the devices are usually designed with very long gate width, in an
interdigitated structure. In order to have an indication of the device performance with
long gate width, this device design (Figure 3.8) was included as a test structure. This
device structure would give a gate width of approximately 3.05 mm.
100um
Gate
Drain
Source
Figure 3.8. The design of the long gate width transistor design.
28
After ascertaining that all the test structures are compatible with the device fabrication
process, the design was sent out for the fabrication of the mask set.
3.2
Device Fabrication Procedure
The fabrication process of the test devices will be discussed in this section. The
fabrication process starts with blank wafers shown in Figure 3.9.
The thickness of the intrinsic GaN layer is usually between 1-5 µm, while the
thickness of the AlGaN layer is in the range of 5-30 nm. This thickness depends on the
application of the devices and thus different layer engineering would cater to different
applications. The aluminium mole concentration of the AlGaN is 0.25 in our
experiments and simulations. In this work, the AlGaN layer is 20 nm thick while the
GaN layer is 2 µm thick. A summary of the fabrication process is shown in Figure
3.10.
Figure 3.9. The starting layer structure of HEMT wafer before the fabrication process.
29
Figure 3.10. The fabrication process of AlGaN/GaN MOSHEMT [31].
The AlGaN/GaN epitaxial wafer was subjected to a round of lithography and Cl2based reactive ion etching (RIE) to form the mesa isolation. After the active region
was formed, the wafer was subjected to pre-gate cleaning. The wafer was washed
using HCl to remove any native oxide, followed by a dip in (NH4)2S as an ex-situ
surface passivation process.
After the pre-gate cleaning, the wafer was loaded into a Metal Organic Chemical
Vapour Deposition (MOCVD) multi-chamber gate cluster system. In one of the
chambers, in-situ passivation was performed [32]. The wafer was baked in the
MOCVD chamber at a temperature of 300 °C under high vacuum (~1×10-6 Torr)
(Vacuum Annealing). This would decompose any remaining native oxide that might
be left over during the pre-gate cleaning process. A silane (SiH4) treatment at a
temperature of 400 °C for 1 minute at a pressure of 5 Torr was performed, for the
30
purpose of surface passivation. The flow rates of SiH4 and N2 were 60 and 250 sccm
respectively. High-k dielectric (either Aluminium Oxide (Al2O3) or Hafnium
Aluminium Oxide (HfAlO)) was then deposited on the wafer without breaking
vacuum. A Post Deposition Anneal (PDA) was then performed to improve the quality
of the high-k dielectric film. This PDA was performed at 500 °C in N2 ambient for 60
s.
TaN was then sputtered on the wafer, followed by a photolithography step to define
the gate pattern by plasma etching while wet etching (in dilute HF) was done to
remove the HfAlO in source/drain regions. Ohmic contacts were then formed by
depositing a layer of Titanium (30 nm) followed by a layer of Aluminium (70 nm)
which were patterned by a lift-off process. Finally an alloying process was carried out
to complete the process of the ohmic contact formation. This was done at a
temperature of 650 °C in N2 ambient for a period of 30 s. The resulting structure is
shown in Figure 3.11.
31
20nm
2µm
Figure 3.11. An illustration of the device schematic of a completed AlGaN/GaN MOSHEMT.
After the alloying process, the baseline structure is completed and ready for
characterization. In order to study the effect of stress on these GaN-based MOSHEMT
structures, a layer of Diamond-like Carbon (DLC), 40 nm thick, was deposited using a
Filtered Cathodic Arc (FCA) System on some of the devices. A lift-off process was
done to define the pattern of the DLC layer.
Characterization was performed mainly on three sets of devices. These three sets of
devices are: 1) Devices with no surface passivation (which will act as control samples),
2) Devices with surface passivation and 3) Devices with surface passivation and DLC
layer. The final structure with DLC liner layer is shown in Figure 3.12. The effect of
these differences in fabrication steps on the device performance will be discussed in
the following sections.
32
20nm
2µm
Figure 3.12. An illustration of the device schematic of an AlGaN/GaN MOSHEMT with a DLC
stress liner.
3.2.1
Surface Passivation
Surface passivation is one of the most important steps in modern device fabrication.
Even in silicon-based devices, surface treatment is important in passivating the
dangling bonds at interfaces. Without surface passivation, these traps will trap carriers
during operations and cause undesirable effects such as the reduction in drive current
as well as threshold voltage (Vth) shift [33].
Surface passivation studies of Gallium Arsenide (GaAs) [34] provided an important
guide to the surface passivation studies of GaN as both are Gallium-based material. In
[34], a SiH4 and Ammonia surface passivation technique was used to improve the
33
interface of GaAs-based device. However, it was found in [32] that Ammonia was not
beneficial to the purpose of surface passivation in GaN, thus only Vacuum Annealing
and SiH4 treatment was used for surface passivation of GaN-based devices.
GaN MOSHEMTs that were subjected to Vacuum Annealing (VA) at 300 °C and
SiH4 treatment at 400 °C. These treatments effectively decompose any native oxide
and passivate any dangling bonds or any other kind of defects that may manifest as
interface states at the interface between the high-k dielectric (HfAlO) and the AlGaN
barrier layer [32]. This work sets out to characterise two groups of samples: 1) devices
with no surface passivation (control samples) and 2) device with the above mentioned
passivation process. Results of electrical measurements will be compared and a short
discussion will follow.
The passivation studies were performed on long channel devices thus the gate length
of the devices characterised for this study is 2 µm.
3.2.2
Diamond-like Carbon (DLC) Layer
Previous studies had shown that stress would affect the device performance of
AlGaN/GaN HEMTs [35, 36]. Due to the difference in lattice constants between the
AlGaN layer and the GaN buffer layer, intrinsic tensile stress exists in the AlGaN
layer. This intrinsic tensile stress causes an increase in 2DEG concentration due to a
change in piezoelectric polarisation across the heterojunction as discussed in chapter 2.
34
In [36], additional tensile stress was induced by inducing a mechanical stress on the
device. This additional tensile stress increases the piezoelectric polarisation and thus
the 2DEG concentration increases accordingly. This increase in 2DEG concentration
improves the conductivity of the channel and enhances the DC performance of the
transistor. However, an increase in the 2DEG concentration degraded the transient
performance in [36] and this was attributed to the increase in electron density due to
increase in piezoelectric polarisation. This would result in a more negative threshold
voltage. Thus, in order to fabricate an AlGaN/GaN enhancement mode transistor, one
should reduce the 2DEG concentration with a trade-off of a lower conductivity [35].
DLC was first deposited on p-channel silicon transistors to induce a high compressive
stress on the channel so as to improve the performance of these transistors [37]. In the
fabrication process of the AlGaN/GaN MOSHEMT described in the previous section
[38], it was mentioned that a layer of DLC was deposited to induce stress on the
device (Figure 3.12). DLC, with an intrinsic stress of 6 GPa [37], would be effective at
providing a compressive stress in the AlGaN/GaN transistor. This compressive stress
would effectively reduce the intrinsic tensile stress present due to the lattice constant
difference between AlGaN and GaN. A reduction in the intrinsic tensile stress would
reduce the piezoelectric polarisation in the structure and thus reduce the 2DEG
concentration. This could be beneficial in the objective of achieving an enhancement
mode device [31].
35
Devices with and without DLC layers will be subjected to DC characterization to
study the effect of such a DLC liner. In order to maintain consistency in the
experiment, these two groups of devices will be put through the same surface
passivation process mentioned previously.
It was observed that the effect of a 40nm DLC stress liner had greater effect on short
channel devices as the average stress induced on the 2DEG is generally greater for
devices with shorter channels [31]. The average stress was simulated using Abaqus
simulation suite and the result is shown in Figure 3.13. This Abaqus simulation was
Average Channel Stress (MPa)
done by Liu Bin.
-200
-300
-400
-500
-600
-700
-800
-900
300
400
500
600
Gate Length LG (nm)
700
Figure 3.13. The average compressive stress induced in the channel due to 40 nm DLC simulated
for various LG: 300 nm, 500 nm and 700 nm [31].
36
In view of the effect of the DLC liner on short channel devices, the study on DLC
deposition on AlGaN/GaN MOSHEMT was done primarily on short channel devices.
The characterised devices in this work have a gate length of 300nm.
3.3
Electrical Characterisations of the MOSHEMTs
A few typical electrical characterization techniques were used to characterise the
performance of the devices. The results of the characterisation are presented in the
following section. The layer structure and fabrication details are the same in all
devices as described in section 3.2.
37
Effect of Surface Passivation on the ID-VGS Characteristics
Drain Current ID (A/mm)
2
10
1
10
0
10
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10
250
Device without passivation
Device with passivation
VD = 5 V
200
150
VD = 1 V
100
50
-8
-6
-4
-2
0
2
Gate Voltage VG (V)
4
0
Transconductance gm (mS/mm)
3.3.1
Figure 3.14. The effect of the surface passivation technique on the ID-VGS plot of MOSHEMTs
with LG = 2 µm. Incorporation of In situ SiH4 passivation improved subthreshold swing to 100
mV/decade and gm of 95 mS/mm [31].
The VA and SiH4 surface passivation techniques are expected to remove any native
oxide present on the AlGaN surface as well as passivate the interface states that exist
between the AlGaN layer and the HfAlO high-k dielectric layer. Figure 3.14 shows
the effect of the surface passivation on the ID-VGS plot of the device with a gate length
of 2 µm.
The surface treatment reduces the off-state leakage current by about 2 orders of
magnitude. Due to the reduction in off-state leakage current and the increase in onstate current, the Ion/Ioff ratio sees an improvement of about 2 orders. In addition to
the reduced off-state leakage, the sub-threshold swing is also improved from 150
38
mV/decade to less than 100 mV/decade. This improvement could be due to the
suppression of trap-assisted tunneling current through the high-k dielectric which has
effectively passivated the interface states. The threshold voltage moves towards a
slightly more negative value as the improved interface might have resulted in a higher
2DEG density.
3.3.2
Effect of Surface Passivation on ID-VDS Plot
It is observed in the previous section that surface treatment reduces the off-state
leakage current, increases the on-state drain current and improves the sub-threshold
(SS) in general. That effect is also seen in the ID-VDS plot of the device with gate
Drain Current ID (mA/mm)
length of 2 µm, as shown in Figure 3.15.
900
Device without passivation
Device with passivation
750
VG,max = 4 V, Step = - 1 V
600
450
300
150
0
0
2
4
6
8
10 12
Drain Voltage VD (V)
14
Figure 3.15. The effect of surface passivation on ID-VDS plot of MOSHEMT with LG = 2 µm. ID,sat
of 624 mA/mm measured at VD = 10 V and VG = 4 V was observed.
39
From Figure 3.15, it is observed that the saturation drain current experience a 53%
enhancement for devices with in situ VA and SiH4 treatment as compared to the
control device. The drain current enhancement was taken at VGS - VT = 7 V and VD =
15 V. This result is consistent with what that obtained in the ID-VGS plot and the
surface passivation proves to be an effective technique in enhancing the performance
of AlGaN/GaN MOSHEMTs.
3.3.3
Effect of DLC Liner on C-V characteristics
Capacitance C (pF)
120
Control
Device with DLC
100
80
60
40
20
0
-8
-7
-6 -5 -4 -3 -2
Gate Voltage VG (V)
-1
0
Figure 3.16. The C-V characteristics of a control device and the device with 40 nm DLC liner
deposited on a MOSHEMT with LG = 300nm.
40
The capacitance of a device is defined as
(3.3)
where dQ is the magnitude of the change in charge stored in a capacitor as a function
of dV, the change in bias across the capacitor [39]. By integrating the C-V plot, we are
able to determine the charge induced under the gate over a fixed range of sweep of the
bias voltages.
Figure 3.16 presents the C-V characteristics of the control AlGaN/GaN MOSHEMT
and one with a 40 nm of DLC liner, both with gate lengths of 300 nm. It is observed
that there is a distinct change in 2DEG concentration, where the density of the 2DEG
in the device with the DLC liner is reduced from 1.04×1013 cm-2 to 8.97×1012 cm-2.
This reduction in 2DEG density might have effectively caused the threshold voltage to
become more positive, and thus the application of a DLC liner to the MOSHEMT is
an effective way in working towards the achieving of an enhancement mode device.
More details on the threshold voltage will be discussed in the next section.
3.3.4
Effect of DLC liner on the ID-VGS Characteristics
The DLC liner was meant to be a stressor to induce compressive stress in the channel
region. This compressive stress should relieve some of the intrinsic tensile stress in the
AlGaN layer and thus reduce the piezoelectric polarisation in the heterostructure. The
effect of this stress liner on the ID-VGS plot of the device with gate length of 300nm is
shown in Figure 3.17.
41
Drain Current ID (A/mm)
250
Control
Device with DLC
0
10
200
-1
10
150
-2
10
VD = 5 V
-3
10
VD = 1 V
100
-4
10
50
-5
10
-6
10
-8
-6
-4
-2
0
Gate Voltage VG (V)
2
4
0
Transconductance gm (mS/mm)
1
10
Figure 3.17. The effect of the DLC stress liner (Compressive Stress) on the ID-VGS charactieristics
of a MOSHEMT with LG = 300 nm. Low SS of 110 mV/decade and high gm of 88 mS/mm were
achieved for the device with DLC liner [31].
The DLC compressive stress liner has caused the threshold voltage to shift from -5.2V
to -4.5V. In addition, the on-state current is observed to have increased. The shift in
threshold voltage is consistent with the C-V measurement which suggests that there is
a reduction in 2DEG density. It is observed that a low SS is attainable for this device
(110mV/decade), thus it causes no degradation to the sub-threshold performance. The
increase in on-state current could be due to the coupled tensile stress that is
experienced in the contact regions. This issue would be further investigated in the
following chapters.
42
Effect of DLC compressive stress liner on the ID-VDS Characteristic
Drain Current ID (mA/mm)
3.3.5
900
Control
Device with DLC
750 L = 0.3 m, V
= 4 V, Step = -2 V
G
G,max
600
450
300
150
0
0
2
4
6
8
10 12
Drain Voltage VD (V)
14
Figure 3.18. The effect of the DLC Stress liner (Compressive Stress) on the ID-VDS characteristic
with LG = 300 nm [31].
The main objective of depositing DLC liner was to attempt to achieve an enhancement
mode device and an enhancement in drive current was not expected. However, in the
ID-VGS plot, it was observed that there was a slight enhancement in the on-state current.
The ID-VDS plot of the same device with a gate length of 300 nm (Figure 3.18)
provides further verification that there is enhancement in transistor performance and
this could be due to the coupled tensile stress experienced in the contact region as
mention above. A 36% enhancement of saturation drain current is observed for the
device with a DLC liner as compared to the control device at (VGS - VT) = 6 V and a
drain voltage VD of 12 V.
43
From the observations in the electrical characterisation of these experimental devices,
it is observed that the surface passivation process would lead to an enhancement in
drive current as well as an improved sub-threshold swing. However, the impact on the
threshold voltage observed in the ID-VGS plot is not as significant. On the other hand,
the presence of a DLC stress liner caused a positive shift in threshold voltage as well
as an enhancement in the drive current. The solution to an enhancement device is not
single-fold. It would require a multi-fold solution whereby DLC is one of the many
possible solutions that could lead to an enhancement operation device.
In order to gain a better understanding of the mechanisms that lead to these changes
and improvement in device characteristics, device simulations would be useful and
would validate the experimental observations.
44
Chapter 4 Numerical Simulation Fundamentals
In this chapter, the underlying physics of the phenomena observed in the experimental
AlGaN/GaN MOSHEMTs will be studied device simulation. Through device
simulation, certain internal device variables which cannot be observed directly from
device terminal I-V measurements such as the 2DEG density, energy band diagrams,
etc., can be observed clearly. It is thus useful to make use of device simulation to
study the different splits of experiments on the AlGaN/GaN MOSHEMT. In this work,
a commercial numerical simulation software, Synopsys® Sentaurus TCAD, was used
to study the effect of various conditions that might affect the performance of an
AlGaN/GaN MOSHEMT.
4.1
Simulation Basics
Numerical simulations are typically complex mathematical calculations based on
several different models that that can be described by mathematical equations. In order
to make use of such models one needs to understand how simulation program
typically make use of these models on the device.
45
Device Geometry Construction
(Sentaurus Structure Editor)
Meshing
(Sentaurus Mesh)
Running of Simulation
(Sentaurus Device)
View Results
(Tecplot SV & Inspect)
Figure 4.1. A flowchart showing the process of device simulation (Names in bracket denote name
of program used for that purpose).
A flowchart of the device simulation process is shown in Figure 4.1. Numerical
simulations usually begin with the construction of the device geometry. In Sentaurus
TCAD, users can choose to do a device geometry construction using the Graphical
User Interface (GUI) or a batch code scripting scheme (i.e. programming) to do the
construction. This process is done in a program in the Sentaurus TCAD suite known
as the Sentaurus Structure Editor. This editor is capable of 1D, 2D and 3D structure
construction. However, only 2D construction will be implemented in this work as it is
sufficient for this study due to the fact that the AlGaN/GaN MOSHEMT is a planar
device.
46
Apart from the geometry construction, other parameters such as doping profiles, alloy
composition, boundary conditions and definition of meshing grids must be specified at
this stage of the device construction. Boundary conditions would dictate how a
particular model will behave at the boundaries of different regions of the device.
For these models to be applied to the mathematical equations that describe the
physical behaviour of the device, meshing of the device needs to be performed.
Meshing refers to the process whereby the differential equations governing the
operation of a device are approximated by a set of difference equations by dividing the
device into a large number of discrete elements. The carrier distribution and electric
potential of each element is described by a set of difference equations based on the
models chosen to describe the operation of the device. These equations (together with
the boundary conditions) are then solved iteratively to find the potential, carrier
concentration and other parameters in each element. The meshing definition is done in
Sentaurus Structure Editor, but the meshing process is performed by a sub-program
known as Sentaurus Mesh. A numerical solver is then used to solve these equations
and in the Sentaurus suite, this solver is known as Sentaurus Device.
47
Figure 4.2. An example of a Meshing Strategy for a 2D device in Sentaurus simulation program
[7].
The meshing of the device need not be uniform throughout the entire device as there
might be regions that are more important than others, such as the channel or interface
region. These are regions where rapid changes of the carrier concentration or potential
are experienced. Different meshing strategies can then be used to describe the device
to optimise the simulation process while ensuring accuracy. An example of a meshed
device is shown in Figure 4.2. It is observed in Figure 4.2 that the channel region and
interfaces between various regions are more tightly meshed as mentioned previously.
48
4.1.1
Device Structure
300 nm
500 nm
500 nm
20 nm
x
y
Figure 4.3. The schematic diagram of the simulated structure in Sentaurus TCAD.
The device structure simulated in this work is designed to match the actual device
fabricated in the first part of the project as closely as possible. As such, the same layer
structure is specified in the device simulated.
The X-Y axis in Figure 4.3 is the reference frame that is used to refer to the device
geometry in this thesis. The Y = 0 line refers to the interface between the AlGaN and
GaN layer. The X = 0 line refers to the centre of the device (i.e. centre of the gate).
The actual device shown in Figure 3.11 featured a sapphire substrate. In the simulated
structure (Figure 4.3), the substrate has been omitted. This will not only simplify the
construction process of the simulation structure, but will also reduce the duration of
the simulation process. This omission would not have a major impact on the results as
the boundary condition at the bottom of the simulated structure is an insulating
49
boundary; which is consistent with the sapphire substrate being an insulator. A few
other factors were taken into consideration during the construction of the device
structure. The considerations are discussed in the sections below.
Source/Drain region
The actual size of the source and drain contact regions is in the range of tens of
microns. However it would be impractical to simulate a device with a 50µm source
and drain region as that would require a very long time for the simulation process to
complete. In order to reduce the simulation duration but still maintain the accuracy of
the simulation, the length of the source/drain region was set at a length of 5µm. It
would be reasonable to make such a simplification as practically all the drain and
source currents will flow into or out of the 5 um regions only. Therefore the remaining
part of the drain and source contacts can be omitted. In the experimental device, the
contacts need to be larger so that they can be probed during characterisation process.
In this simulation work, the effect of changing the LS-G and LD-G is not studied, thus
the two lengths were set to be the same as the basic design of the fabricated devices. In
this case, LS-G and LD-G were both set at 500 nm.
The device structure was specified with ohmic contacts in the drain and source regions
with a contact resistance (RC) of 100Ω. The contact resistance in the fabricated device
had values in that range and thus the simulated device was given this value.
50
Gate Region
In the previous section, characterisation was done on experimental devices with gate
lengths of 2 µm and 300 nm. In order to maintain consistency of the simulation and to
allow comparisons to be made, the simulated gate lengths were specified to be in that
range as well.
In the fabrication process, Tantalum Nitride (TaN) was the metal used for the metal
gate. In the simulation, it is not necessary to define the material used for the gate.
However, it is necessary to define the workfunction of the metal used for the gate.
Thus the workfunction for the gate in this simulation work has been defined to be 4.4
eV which is the value of the workfunction of TaN in literature.
Definition of the Induced Polarisation Charge
In Chapter 2, the polarisation effect of the AlGaN/GaN heterostructure was discussed.
It was evident that this polarisation, though not the sole factor, was one of the more
important factors in the formation of the 2DEG. It is thus important that such an effect
is accurately accounted for in the device simulations.
In order to accurately represent this induced polarisation charge in the various
interfaces that might experience this polarisation effect, the polarisation charge is
calculated according to [22] and input into the simulation accordingly. Two important
interfaces are identified as being the ones that have dominant effect on the 2DEG
density. They are the interface between the AlGaN layer and the GaN layer and the
51
interface between the high-k dielectric and the AlGaN layer. One can refer to Figure
4.4 to better understand where these interfaces are. The GaN layer is much thicker
than the AlGaN layer and thus it is assumed to be totally relaxed, thus only
spontaneous polarisation exists in the GaN layer.
Figure 4.4. An illustration of the interfaces with its polarisation and the respective induced
polarisation charges.
This section will describe the various equations that are used to quantify the
polarisation effect so that they can be used to calculate the effective induced sheet
charges [10, 22, 23]. Piezoelectric polarisation induced along the c-axis can be
described by equation (4.1)
(
where
),
(4.1)
is the strain along the c-axis, and is represented by equation (4.2)
(
In equation (4.1),
and
)
(4.2)
are the in-plane strain and are assume to be equal,
represented by equation (4.3)
52
(
)
(4.3)
where e33 and e31 are the piezoelectric coefficients while a and c are the lattice
constants of the strained layer.
Table
III.
A
comparison
of
the
lattice
constants,
spontaneous
and
piezoelectric
constants/coefficients of GaN and AlN [6].
Coefficient/Constant
AlN
GaN
a0 (Å)
3.112
3.189
c0 (Å)
4.982
5.185
PSP (C/m2)
-0.081
-0.029
e33 (C/m2)
1.46
0.73
e31 (C/m2)
-0.60
-0.49
The relationship between lattice constants in the Wurtzite AlGaN structure is given by
(
(
)
)
(4.4)
where C13 and C33 are elastic constants. A list of the coefficient values can be found in
Table III.
Combining equations (4.1) to (4.4), an equation for piezoelectric polarisation as a
function of lattice constant is obtained in equation (4.5),
(
)
(
)
(4.5)
The list of coefficients and constants given in Table III are for the binary nitrides of
GaN and AlN. In order to obtain the coefficients and constants for AlxGa1-xN,
53
Vegard’s Law is extended here to interpolate the necessary values. For example, the
e33 and e31 piezoelectric coefficients of AlGaN can be obtained by equation (4.6),
[
(
)
(
(
)]
)
(4.6)
This extrapolation is only valid for a small amount of strain as determined in [40]. As
the strain increases, the piezoelectric coefficient might be significantly affected by the
internal strain component. The elastic constants C13 and C33, on the other hand, can be
calculated [22] by,
( )
(
( )
(
)
(4.7)
)
(4.8)
The non-strained lattice constant of AlGaN can also be calculated using Vegard’s Law
interpolation.
The spontaneous polarisation has been determined experimentally by [6] and can be
described by equation (4.9),
(
)
(4.9)
With equation (4.5) and (4.9), we will be able to obtain the piezoelectric and
spontaneous polarisations of different regions of the structure. The polarisation
induced charges are associated with the gradient of the polarisation in space as
governed by Gauss’s Law, given by equation (4.10),
(4.10)
Assuming these interfaces are abrupt and thus the change of polarisation across the
heterostructure is abrupt, the resulting gradient of polarisation would simply be the
54
difference of the polarisations across the interface. Thus the polarisation induced
charge density can be described by
(
)
(
)
(4.11)
where P is the total polarisation (i.e. spontaneous and piezoelectric). The value of the
induced bound sheet charge can thus be obtained by dividing equation (4.11) by the
value of electronic charge, q. The calculated values of these induced sheet charges are
-3.13841×1013
C/cm2
(High-k/AlGaN
interface)
and
1.328115×1013
C/cm2
(AlGaN/GaN interface).
Specification of Interface and Bulk Traps
The immaturity of the III-nitride technology coupled with the lattice mismatch
between AlGaN and GaN crystal structure results in a structure where significant
amount of structural defects are present [8]. These structural defects would effectively
translate into traps. In addition, in [1], donor-like surface states have been identified as
a possible source of carriers that form the 2DEG. Thus an accurate trap concentration
definition is necessary to ensure an accurate simulation.
It is still relatively difficult to characterisation of traps in III-nitrides and there is a
paucity of literature describing those traps that are present. From some works, such as
[41- 43], the density of the bulk traps in most III-nitride material was found to be in
the range of 1015 – 1018 cm-3. In this simulation, the bulk trap density in both AlGaN
and GaN was set to 5×1017 cm-3 with a cross section of 1×10-15 cm-2. By taking an
55
average of the values from various literature [41, 43, 44], these traps are positioned at
1eV above mid-gap and are defined to be acceptor-like.
The interface states present a more challenging task at hand as these would affect the
2DEG directly according to [1]. It is assumed that the AlGaN/GaN interface does not
have any traps and thus the main interface traps would be those at the high-k
dielectric/AlGaN interface. References [41- 43] show that interface states are usually
in the range of 1012-1013 cm-2 eV-1, thus the concentration of the interface state density
specified in this simulation was set at 3.0×1013 cm-2 eV-1 and is set to be donor-like
(0.2eV above mid-gap of AlGaN) traps as described in [1]. In [32], it was observed
that surface passivation can reduce interface state to a level of 1010 – 1011 cm-2eV-1.
This effect would also be simulated in the following chapters.
More details of the simulation parameters can be found in Appendix A where the
simulation script is also listed in detail.
4.1.2
Carrier Transport
The Sentaurus TCAD simulation suite supports several different carrier transport
models for semiconductors. They include:
1. Drift – Diffusion Model,
2. Thermodynamic Model,
3. Hydrodynamic Model,
4. Monte Carlo Method.
56
In this work, these models were studied carefully and the Drift –Diffusion model was
chosen as it provides relatively fast simulation runs as well as an acceptable level of
accuracy. In addition, the temperature effect of the AlGaN/GaN MOSHEMT was not
studied in this work, thus the Thermodynamic and Hydrodynamic model were not
chosen.
Drift Diffusion Model
The Drift – Diffusion model is one of the most basic carrier transport model in
semiconductor physics. The current densities for electrons and holes are given in
equation (4.12) and (4.13),
⃗⃗⃗
,
(4.12)
⃗⃗⃗
,
(4.13)
where µn and µp are the electron and hole mobilities respectively and Φn and Φp are
the electron and hole quasi-Fermi potentials respectively [7].
The Poisson equation [equation (4.14)], electron continuity equation [equation (4.15)]
and hole continuity equation [equation (4.16)], based on the drift-diffusion model, are
numerically solved with the appropriate carrier mobility models in place. These
equations are
s.E s2 q( p n N ) ,
n 1
.J n (G R)
t q
and
(4.14)
(4.15)
57
p 1
.J p (G R)
t
q
.
(4.16)
In the equations above, E is the electric field vector, εs is the semiconductor
permittivity and ψ is the electrostatic potential. The term q(p-n+N) would give the
total space charge where n and p are the electron and hole carrier concentrations
respectively and N is the electrically active net impurity concentration. The terms G
and R are the rates of generation and recombination respectively.
It is useful to note that in the AlGaN/GaN MOSHEMT structure, the dominant carriers
are electrons which make up the 2DEG layer at the interface of the AlGaN and GaN.
However for a comprehensive simulation run, the continuity equation for holes is
included in the simulations.
4.1.3
Mobility Model
The Sentaurus TCAD simulation suite uses a modular approach in its definition of
carrier mobilities, thus it is possible to include several models which are in effect in a
device. In this work, four kinds of mobility models were identified as relevant and
were incorporated into the simulation process. These models include the constant
mobility model, Masetti model (doping-dependent mobility), Lombardi model
(interface degradation mobility) and Canali Model (high-field saturation mobility).
The Sentaurus TCAD simulator makes use of the Matthiessen’s rule [30] to combine
the contributions of each model. The values of the various coefficients and constants
58
were provided by Synopsys which were based on values available in the literatures.
When such values are not available from literature, the values of GaAs are generally
used as a guideline [7]. A brief discussion of the models follows.
Constant Mobility Model
The constant mobility model [45] is a default model for all simulations performed in
Sentaurus TCAD. This model is dependent only on lattice temperature and is given by
(
)
(4.17)
where µL refers to the mobility due to bulk phonon scattering and is an exponent
which can be found from literature. In this work, µL(electrons) = 1200 cm2/Vs and
µL(holes) = 20 cm2/Vs. Since variation of temperature is not taken into account for this
simulation run, T=300 K and the mobility is generally fixed at its default µL value.
Masetti Model (Doping-dependent mobility)
Although the structure that is used to fabricate the AlGaN/GaN MOSHEMT consists
of intrinsic semiconductor in all the layers, there is usually the presence of
unintentional doping in GaN-based semiconductors. This unintentionally doped GaN
was believed to be n-type due to nitrogen vacancies [46]. Thus due to the presence of
this unintentional doping, there is a degradation of mobility due to this doping effect.
This could be simulated by placing a light N-type doping of 3.0×1016 cm–2 in the GaN
layer.
59
This model accounts for the degradation in mobility due to impurity scattering in the
semiconductor [47]. This model is described by
(
)
(
)
(4.18)
(
)
The values of the various coefficients and constants in equation (4.18) are given in
Table IV, whereas µconst refers to the constant mobility value given in the Constant
Mobility model.
Table IV. The default coefficients for GaN in the Masetti model [7].
Electrons
Holes
Units
85
33
cm2/V s
75
0
cm2/V s
50
20
cm2/V s
6.50×1015
5.00×1015
cm3
9.50×1016
8.00×1016
cm3
7.20×1019
8.00×1020
cm3
α
0.55
0.55
-
β
0.75
0.7
-
Symbol
60
Canali Model (High-field saturation mobility)
Presently, the AlGaN/GaN devices are primarily used for high power and high voltage
applications due to their large bandgap as well as high electron saturation drift
velocity. This drift velocity would be the limiting factor of the mobility when a device
experiences a high electric field condition. Thus it is reasonable to include the high
field dependence model as one of the mobility models. The Canali Model [48] is
based on the Caughey-Thomas model [49] but with temperature-dependent
parameters up to 430 K. The Canali mobility model is given by
(
where
)
⁄
(4.19)
refers to the parallel high field experienced and µlow refers to the low field
mobility, which in this case refers to the constant mobility defined previously. β is a
temperature dependent exponent which in this case is given as 1.7 (provided by
Synopsys) as the temperature effect is not taken into account here. vsat refers to the
electron saturation drift velocity of GaN (3.0×107 cm s-1) which is the critical value
that determines this mobility value.
Lombardi Model (Interface degradation mobility)
The Canali model describes the effect of the lateral electric field that is experienced by
the carriers. These carriers however, experience not only the lateral electric field, but
also the vertical (or gate) electric field. The effect of the vertical electric field on
carrier mobility is attributed to acoustic phonon scattering and surface roughness
scattering [45]. This is modelled by the Lombardi model which combines the bulk
61
mobility together with the effect of acoustic phonon scattering and surface roughness
scattering by Matthiessen’s rule [45].
The acoustic phonon scattering is given by
(
)
(4.20)
(
)
while surface roughness scattering is given by
(
)
(4.21)
[
]
where Eref refers to the reference field of 1V/cm to ensure a unitless numerator. A*
refers to an exponent which can be calculated by an equation found in [50]. At the
point of time this thesis is being written, there is no known literature that gives the
value of the coefficients in the Lombardi model for GaN or GaAs. Thus the
coefficients used are those of silicon and are given in Table V.
62
Table V. The default coefficients for Silicon in the Lombardi Model [7]. These coefficients are
used for the simulation due to a lack of these values for GaN in the literature.
Symbol
C
A*
4.2
Electrons
Holes
Units
4.75×107
9.925×106
cm/s
5.80×102
2.947×103
cm5/3V-2/3s-1
1
1
cm-3
0.125
0.0317
-
1
1
-
5.82×1014
2.0546×1014
cm3
2
2
-
5.82×1030
2.0546×1030
V2cm-1s-1
Effect of different AlGaN Layer thickness
Ibbetson demonstrated in [1], the effect the AlGaN layer thickness has on the 2DEG
density. He showed that the AlGaN thickness has a direct correlation to the 2DEG
density up to a certain thickness (Figure 4.5).
63
Figure 4.5. The graph of the 2DEG density measured as a function of Al0.34Ga0.66N barrier
thickness @ room temperature. Solid dots are experimental data, curve represent least square fit
of equation (4.22) up to 15 nm [1].
He postulated that the 2DEG density would increase with AlGaN layer thickness up to
a thickness of 15 nm according to equation (4.22), after which there would be
significant relaxation in the AlGaN layer and thus a slight reduction in polarisation
and consequently, the 2DEG density:
(
)
(4.22)
where σpz refers to the polarisation-induced charges at the AlGaN/GaN interface and
the surface and tcr refers to the critical thickness at which 2DEG is induced. The
critical thickness according to [1] is 35 Å as seen from Figure 4.5.
64
In this part of the project, this correlation between the 2DEG and AlGaN barrier
thickness is simulated to see if the simulation would behave like Ibbetson’s
experimental data. The simulation was done on a device with the following
configurations (Table VI).
Simulation Results
Various simulation runs were performed in order to study the effect of the thickness of
AlGaN on device performance in general. AlGaN layer thickness was set at 5, 10, 15
and 20 nm while the other parameters were kept constant to ensure consistency in the
results.
The study was done by observing a few of the electrostatic and physical results that
were obtainable from the simulation program. ID-VGS sweeps were simulated on the
devices and the results can be observed in Figure 4.6.
Table VI. The configurations of device used for simulation with varying AlGaN thickness.
Device Parameters
Value
AlGaN thickness
Vary from 5 – 20 nm
GaN thickness
2 µm
High-k Dielectric thickness
7 nm
(Al2O3)
Gate Length (Lg)
300 nm
Source-Gate/Drain-Gate Length
500 nm
65
Drain Current ID (uA/um)
120
5nm
10nm
15nm
20nm
100
80
60
40
20
0
-6
-5
-4
-3
-2
Gate Voltage VG (V)
-1
0
Figure 4.6. Simulated ID-VGS characteristics of devices (VDS = 0.1 V) with different AlGaN
thickness. The threshold voltage are, VTH(20 nm) = -4.20 V, VTH(15 nm) = -3.25 V, VTH(10 nm) = 2.25 V, VTH(5 nm) = N.A.
An evident change in the threshold voltage is observed from Figure 4.6. The threshold
voltage increases from -4.20 V (device with 20 nm AlGaN layer thickness) to -2.25 V
(device with 10 nm AlGaN layer thickness) with decreasing AlGaN layer thickness.
The device with a 5 nm AlGaN layer thickness did not yield any drain current as it
might not have any 2DEG induced as it might be below the critical thickness required
for carrier induction.
The 2DEG density can be observed from results that can be extracted from the
simulation run. Figure 4.7 to Figure 4.10 show the 2DEG density of each device in the
channel region (under the gate).
66
High-K
Dielectric
(Al2O3)
Al0.25Ga0.75N
(5nm)
GaN Buffer
Layer
Source
Drain
Figure 4.7. The electron density mapping in device with 5nm thick AlGaN layer.
The interface between the AlGaN layer and the GaN layer is situated at Y=0 and thus
the 2DEG should exist at the region just below Y=0. The electron density in devices
with 5 nm AlGaN layer appears to be very low. This is in stark contrast with devices
with AlGaN layer thickness of ≥10 nm (Figures 4.8-4.10).
In Figures 4.7- 4.10, the diagram on the left refers to the electron density mapping that
is observable at zero bias; i.e. VD, VS and VG are at 0 V. On the right side, this diagram
represents the electron density mapping observable at VS = 0 V, VD = 0.5 V and VG = 5 V (mapping obtained at the end of ID-VGS sweep). Thus the devices should all be in
the off-state.
67
Figure 4.8. The electron density mapping in device with 10 nm thick AlGaN layer.
Figure 4.9. The electron density mapping in device with 15 nm thick AlGaN layer.
68
Figure 4.10. The electron density mapping in device with 20 nm thick AlGaN layer.
It is however difficult to make a quantitative comparison between the 4 figures, thus
the electron density between Y=0 and Y=0.01 um (10 nm) was integrated to obtain a
number for quantitative comparison. This quantitative comparison is presented in
Figure 4.11.
69
-2
12
Electron Density (×10 cm )
6
5
4
3
2
1
0
4
6 8 10 12 14 16 18 20 22
AlGaN Layer Thickness (nm)
Figure 4.11. A plot showing the simulated 2DEG density as a function of Al0.25Ga0.75N barrier
thickness. The 2DEG density ranges from 8.69×104 cm-2 @ 5nm to 5.90×1012 cm-2 @ 20nm AlGaN
layer thickness.
It is observed that the experimental results (Figure 4.5) of Ibbetson in [1] may not
have the same trend as the simulated result after the 12 nm mark, as shown in Figure
4.11. This is due to the fact that in this simulation performed, the stress relaxation
component was not included. It was observed that in thicker AlGaN layers, there
might be changes (reduction) in stress on the GaN layer, influencing the piezoelectric
polarisation [51]. This resulted in a reduction of 2DEG density. This is not accounted
for in the simulation. Thus the saturation/reduction in 2DEG density after a certain
thickness of AlGaN layer is not observed in this simulation.
In addition, the lower Al mole fraction in the simulations (x = 0.25) would also result
in a lower drive current [52] as compared to that of the experimental results in
70
Ibbetson’s work (x = 0.34). However, a general trend that 2DEG density increases
with increasing AlGaN thickness (before stress relaxation sets in) is observed from
when AlGaN thickness is 4.5 nm to 12 nm.
In addition to electron density mapping, it is also possible to extract the band diagram
in each simulation run. From the band diagram extracted (Figure 4.12), it is observed
that when the AlGaN layer thickness is 5 nm, the triangular quantum well is not
present, as compared to the band diagram (Figure 4.13) for the device with 20 nm
AlGaN layer which shows the triangular quantum well that would confine the carriers
that form the 2DEG.
High-K Dielectric
AlGaN Layer
Energy Level (eV)
GaN Buffer Layer
4
3
2
1
0
-1
-2
-3
-4
EC
EF
EV
-10
0
10
20
30
Y-Axis (nm)
40
50
Figure 4.12. The energy band diagram of AlGaN/GaN MOSHEMT with 5 nm thick AlGaN layer
extracted from the simulation.
71
High-K Dielectric
AlGaN Layer
Energy Level (eV)
GaN Buffer Layer
4
3
2
1
0
-1
-2
-3
-4
EC
EF
EV
-20 -10
0 10 20
Y Axis (nm)
30
40
50
Figure 4.13. The energy band diagram of AlGaN/GaN MOSHEMT with 20 nm thick AlGaN layer
extracted from the simulation.
The thickness of the AlGaN layer would indirectly affect the 2DEG density due to the
energy level of the traps being positioned differently at different AlGaN layer
thickness. The position of the energy level of these traps would affect the ionization
state of these traps and thus affect the 2DEG density directly as described by Ibettson.
72
4.3
Effect of Interface Traps on Device Operations
Figure 4.14. A schematic diagram showing where energy level of donor-like surface states reside
in a) thin AlGaN layer that is < critical thickness b) thicker AlGaN layer > critical thickness [1].
In section 4.2, the effect of AlGaN thickness on the 2DEG density was demonstrated.
That effect on the 2DEG density clearly translated to a change in the drive current.
It is thus useful to observe the effect of the concentration of these donor-like states at
the AlGaN / high-k dielectric interface and how it affects the 2DEG and other
electrical characteristics.
In all previous simulation runs, the donor-like traps are positioned slightly above the
Fermi level as shown in Figure 4.15. The trap energy was defined by
[
(
)]
(4.23)
where E0 is defined to be 0.2 eV (i.e. 0.2 eV above mid-gap level).
73
High-K Dielectric
AlGaN Layer
Energy Level (eV)
GaN Buffer Layer
4
3
2
1
0
-1
-2
-3
-4
Traps Level
EC
EF
EV
-20 -10
0 10 20
Y Axis (nm)
30
40
50
Figure 4.15. The energy level of the trap in a device with 20 nm thick AlGaN layer being
programmed in this simulation run.
In this set of simulation runs, two different simulations were performed. These two
different runs are differentiated solely by the concentrations of the traps that are
placed at the interface between the AlGaN layer and high-k gate dielectric. The
configurations of these simulations are a) Trap concentration = 3.0×1013 cm-2 eV-1 and
b) Trap concentration = 5.0×1013 cm-2 eV-1. All of these traps are assumed to exist at a
single energy level (0.2 eV above mid-gap level) as shown in Figure 4.15.
74
Drain Current ID (uA/um)
140
120
13
-2
-1
Trap Conc. = 3.0×10 cm eV
13
-2
-1
Trap Conc. = 5.0×10 cm eV
100
80
60
40
20
0
-10
-8
-6
-4
-2
Gate Voltage VG (V)
0
Figure 4.16. A ID-VGS plot of simulated devices with different donor-like trap concentration.
The ID-VGS plot (Figure 4.16) would give a first indication of the qualitative
comparison of the device performance in terms of the 2DEG density. It is observed
that the device with a higher donor-like trap concentration had a threshold voltage that
is much more negative which would most like translate to a much higher induced
2DEG density.
By performing a similar integration of the electron density in the first 10 nm of the
GaN layer just below the AlGaN/GaN interface, we obtain the 2DEG density for the 2
devices. The integration yielded a 2DEG density of 5.90×1012 cm-2 for the device with
a trap concentration of 3.0×1013 cm-2 eV-1 while the device with a trap concentration
of 5.0×1013 cm-2 eV-1 yielded 2DEG density of 6.00×1012 cm-2.
75
The difference in 2DEG density seems to suggest that there might be a saturation point
at which an increase in donor-like traps would have little impact on the 2DEG density.
However, the change in threshold voltage does not justify the small change in 2DEG
density. Upon more analysis of the simulation, it seems that in a device with increased
donor-like density, this additional ionised charge would manifest itself as a space
charge in the upper region of the AlGaN layer. The evidence of this is shown in the
data for space charge (in the device at the gate region) extracted from the simulation
(Figure 4.17).
Trap Density = 5.0×1013 cm-2 eV-1
2DEG Region
Trap Density = 3.0×1013 cm-2 eV-1
AlGaN Layer
GaN Layer
Figure 4.17.The simulated space charge data on AlGaN/GaN MOSHEMT at gate region
extracted from the simulation.
76
This manifestation of charge in the AlGaN layer could lead to the more negative
threshold voltage as seen in Figure 4.16. However, the actual mechanism of this
phenomenon is largely unknown as it requires more study.
4.3.1
Effect of Surface Passivation
The previous simulation result does not show consistency with the experimental
results of the surface passivation. In the results of the experimental devices, the
surface passivation reduces the interface states between the gate dielectric and the
AlGaN layer [32]. Thus instead of having a small impact on the threshold voltage as
seen in section 3.3, with less interface states, there should be a shift towards the
positive region in threshold voltage as seen in the simulation due to a change in 2DEG
density.
This work postulates that there are both acceptor-like traps as well as donor-like traps
at the interface of the gate dielectric and AlGaN layer and that the passivation process
is more likely to reduce the acceptor-like traps.
Although there is scant literature that shows the presence of acceptor-like traps at the
surface, it would be possible to simulate the effect of having both kinds of trap at this
interface, and assume that the surface passivation process reduces the density of the
acceptor-like traps. For the purpose of simplifying the simulation process, it is
assumed that the donor-like traps are not affected by the surface passivation process in
the simulation. The configurations of the simulations are thus done with different
77
acceptor-like trap concentration set at: a) No acceptor-like traps, b) 1.0×1012 cm-2 eV-1,
c) 5.0×1012 cm-2 eV-1 and d) 1.0×1013 cm-2 eV-1. The donor-like traps at this interface
is fixed at 3.0×1013 cm-2 eV-1 with and energy level of E0 = 0.2 eV while the energy
level of these acceptor-like traps were specified at E0 = -0.2 eV (i.e. 0.2 eV below
mid-gap) . In order for these acceptor-like traps to be ionised they must exist below
the Fermi level. The ID-VGS plot (Figure 4.17) gives an indication that the postulate
might be plausible.
Figure 4.18 shows that when the concentration of acceptor-like traps is reduced (e.g.
reduced by the surface passivation process), the drain current increases with little
change in threshold voltage. This is in better agreement with the results of the
Drain Current ID (uA/um)
experimental devices.
100
80
60
40
No Acceptors
12
-2
-1
1.0×10 cm eV
12
-2
-1
5.0×10 cm eV
13
-2
-1
1.0×10 cm eV
20
0
-5
-4
-3
-2
-1
Gate Voltage VGS (V)
0
Figure 4.18. The ID-VGS plot of devices with different concentration of acceptor-like trap.
78
An ID-VDS sweep was done with the same simulated devices with a gate overdrive of
VG-VTH = 3.5 V. The plot is shown in Figure 4.19.
Drain Current ID (uA/um)
600
500
400
300
No Acceptors
12
-2
-1
1.0×10 cm eV
12
-2
-1
5.0×10 cm eV
13
-2
-1
1.0×10 cm eV
200
100
0
0
2
4
6
8
Drain Voltage VDS (V)
10
Figure 4.19. The ID-VDS plot of devices with different concentration of acceptor-like traps.
The ID-VDS characteristics show good agreement with the results of the experimental
devices shown in section 3.3. In order to obtain drive current enhancement of >50%
like those seen in section 3.3, it is possible that the acceptor-like interface state
concentration underwent a reduction of ~ 5 times. A reduction of 50% (1.0×1013 cm-2
eV-1 - 5.0×1012 cm-2 eV-1) of traps resulted in a 75% increase in drive current
(200uA/um to 350uA/um) as demonstrated by the simulation results shown in Figure
4.19. However, it should be noted that this is based on an ideal condition where the
reduction of acceptor-like traps would unconditionally contribute to an increase in
drive current.
79
Chapter 5 Simulation of the Effect of Stress on AlGaN/GaN
MOSHEMT
In an attempt to achieve an enhancement mode device, a compressive stress liner was
deposited to reduce the intrinsic tensile stress that is present in the AlGaN/GaN
heterostructure (Figure 5.1). X. K. Liu has demonstrated in [31], that such an approach
is feasible experimentally. In addition, Liu explained that a coupled tensile stress is
present in the device in the region between the gate and the source/drain. This tensile
stress helped to enhance the device performance in a AlGaN/GaN MOSHEMT with
DLC stress liner.
In this part of the thesis, simulation is used to study the effect of compressive stress on
the device performance of AlGaN/GaN MOSHEMTs. Two simulation softwares were
used to simulate the effect of stress on the device. Taurus was used to simulate the
resulting stress on the device due to the DLC stress liner while Sentaurus TCAD was
used to simulate the device performance under stress.
80
Figure 5.1. The schematic view of the AlGaN/GaN MOSHEMT with a compressive DiamondLike Carbon (DLC) liner. The 2-DEG density decreases with application of compressive stress.
5.1
Device Structure and Stress Model
The device structure used in this part of the project is similar to the devices simulated
in Chapter 4. However, as the simulation suite has not been optimised for III-nitride
simulations, the effect of the stress on the AlGaN/GaN heterostructure is not
effectively modelled to portray the resulting piezoelectric effect that might result
because of this stress. In order to model the polarisation charges effectively, the
resulting polarisation induced charges were calculated and input into the simulation
code independently. This polarisation charge calculated is based on the piezoelectric
and spontaneous polarisations experienced at the interfaces.
81
To effectively simulate the stress that is induced on the device and the resulting effect
of this stress, mechanical stress simulation was first performed using the Abaqus
simulation suite. This was done by B. Liu whose work on DLC liner as a compressive
stress liner provided much of the expertise needed for this experiment and simulation
[53- 55]. The Abaqus simulation suite is able to simulate the mechanical stress that is
experienced in the entire device due to a stress liner being deposited. The schematic of
the structure being simulated in the Abaqus simulation software is shown in Figure 5.2.
As the DLC liner is only deposited over the gate region and the regions between the
gate and source/drain, only data from those regions were extracted after the simulation
in Abaqus. The X- and Y-axes denote the reference frame in which this device’s
geometry will be discussed.
Figure 5.2. The schematic diagram of the device being simulated in Sentarus TCAD and Abaqus
for the effect of compressive stress liner. The X and Y-axes denote the device geometry that was
used in this simulation.
82
Uniaxial stress was taken into consideration for this simulation as it was shown in [36]
that stress parallel to the gate width had little impact on the performance of the device.
The strain experienced along the x-axis due to the uniaxial stress induced on the
device (Figure 5.2) is known as stressXX. This is the dominant stress that affects the
in-plane piezoelectric polarisation in the AlGaN/GaN heterostructure.
Each region of the structure was configured with the appropriate material and its
properties. After the mechanical stress simulation was completed, the non-critical
regions (where little or no stress was experienced) were taken away to reduce data
extraction time. The resulting stress (stressXX) distribution in the AlGaN and GaN
layers due to the 40 nm of DLC compressive stress liner is shown in Figure 5.3.
High-K Dielectric (Al2O3) Layer
AlGaN Layer
GaN Layer
Figure 5.3. A map depicting the StressXX distribution that is experienced in a simulated
AlGaN/GaN device with LG = 300 nm due to 40nm of DLC compressive stress liner.
83
In the previous simulations, the GaN layer is assumed to be totally relaxed as it is
much thicker than the AlGaN layer. In order to maintain the consistency with previous
simulations, the GaN layer will also be assumed to be totally relaxed for this part of
the simulation; thus only the stress that is experienced in the AlGaN layer of the
device is taken into consideration in the simulation. The average stress over the 20 nm
of the AlGaN layer was computed and is shown in Figure 5.4. A negative value in
stress would denote a compressive stress while a positive stress value would denote a
tensile stress. It is observed that the stress experienced in the AlGaN layer under the
gate region is generally compressive in nature while the AlGaN regions beyond the
gate experience a weak tensile stress. This coupling of stress is due to the fact that the
DLC liner is deposited over the metal gate stack. In this manner, the compressive
stress has been concentrated on the gate edge as seen in Figure 5.3 while the regions
beyond the metal gate stack would result in a coupled tensile stress.
An empirical method was used in the translation of this stress experienced in the gate
region to the polarisation charge induced. Strain in a material [56] can be represented
by
(5.1),
where a is the unstrained bulk lattice constant,
is the strained bulk lattice constant,
σ is the stress induced and E is the Young’s modulus of the material under stress. The
Young’s Modulus of Al0.25Ga0.75N has been documented in literature to be between
the values of 320GPa – 350GPa [52, 56-59]. Thus an average value of 335GPa was
chosen for this empirical calculation.
84
400
StressXX (MPa)
200
0
-200
-400
-600
-800
-1000
-600 -400 -200
0
Gate Region
200 400 600
X (nm)
Figure 5.4. A plot of the average StressXX distribution that is experienced in the AlGaN layer of
the device structure shown in Figure 5.2.
It is observed that in Figure 5.4, the stress in the simulated region ranges from about
200 to -1000MPa. Thus by making use of results shown in Figure 5.4 and equation
(5.1), we will be able to obtain the effective lattice constant at each discreet point of
the device due to the stress at that point. This series of effective lattice constants is
then substituted into equations (4.1) – (4.11) in section 4.1.1. An effective nonuniform polarisation induced charge distribution at the interface of the AlGaN and
GaN layer is obtained and shown in Figure 5.5. The dashed horizontal line denotes the
uniform polarisation charge that would exist in a device without the DLC liner. The
compressive stress under the gate region gives rise to a polarisation charge value that
is lower than that of the uniform charge distribution while the tensile stress beyond the
gate region gives rise to a higher value of the polarisation charge.
85
2
1.5
13
Polarization Charge (x10 C/cm )
a)
1.4
1.3
1.2
1.1
1.0
-600 -400 -200
0
200
X (nm)
400
600
Y (nm)
b)
X (nm)
Figure 5.5. a) Polarisation induced charges at the interface of AlGaN/GaN due to compressive
stress by the DLC stress liner, dashed horizontal line denotes the uniform polarisation charge in a
non-stressed device. b) A corresponding device schematic illustrating the device geometry.
This polarisation charge was input into the simulation at discreet points in 10 nm
intervals at the AlGaN/GaN interface. Due to a change in piezoelectric polarisation in
the AlGaN layer, a similar change in polarisation induced charge distribution is also
experienced at the High-k dielectric/AlGaN interface. The corresponding charge
distribution at that interface is also calculated and put at discreet points at 10 nm
intervals. The simulation was then performed with these changes in polarisation
86
charge distribution due to stress from the DLC liner. Interface states were also
included between the AlGaN layer and the high-k dielectric layer. These interface
traps were set to be donor-like, at a concentration and energy level of 3.0×1013 cm-2
eV-1 and 0.2 eV above mid-gap respectively. No acceptor-like interface states were
included in this simulation.
5.2
Effect of Stress on Electrical Characteristics
The ID-VGS characteristics is first analysed. The control ID-VGS plot was taken from the
ID-VGS plot in section 4.2 where the effect of the thickness of the AlGaN layer was
studied. The ID-VGS plot of the device with DLC stress liner, as well as that of the
control device is shown in Figure 5.6.
Figure 5.6. The ID-VGS plot of the MOSHEMT device with and without the DLC stress liner. The
gate length of the devices, Lg = 300nm.
87
Figure 5.7. The effect of compressive stress from DLC stress liner on the ID-VDS plot of simulated
device with Lg = 300nm.
A shift in the threshold voltage from -4.20 V (no stress) to -3.50 V (with compressive
stress) is observed in the ID-VGS plot in Figure 5.6. The trend observed is consistent
with the results of the experimental devices discussed in section 3.3.2 where devices
with DLC compressive stress liner were characterised.
In addition, in section 3.3.3, an ID-VDS measurement was also performed on these
devices where a 36% enhancement in drive current was observed. An ID-VDS sweep
was also performed for the simulated device and the result is shown in Figure 5.7.
The simulation result did not show a similar amount of enhancement compared to the
experimental devices. However, it does show a similar trend of a higher drive current
(2.5% enhancement). This could be attributed to the fact that the effect of stress was
88
not fully simulated in this process. Instead, only the change in polarisation induced
charge was taken into account. The enhancement of drive current could also be due to
a few factors such as a change in the effective mass of the electrons due to a change in
band structure [2],[3] or a change in trap concentration [60]. Some of these factors
such as change of effective mass is not taken into considerations as it would require
one to include the calculations due to change in band structure which is not in the
scope of this thesis.
5.3
Effect of Stress on Carrier Concentration
Although some factors that could lead to drive current enhancement is not taken into
consideration in this simulation, it is still meaningful to look at how the other
parameters in this simulated device affect the electrical results. In the previous
sections, it is observed that a change in 2DEG density affects the threshold voltage of
a device. When stress is induced in the device, a non-uniform polarisation charge is
observed. That would give rise to a non-uniform distribution of the carriers in the
2DEG. In this case the 2DEG density of the device at 1 nm below the AlGaN/GaN
interface was extracted from the device simulation. The 2DEG density distribution is
shown in Figure 5.8.
89
6
2.4
~ 7% Increase
2.2
2.0
-500
4
-450
-400
-350
-300
-250
X axis (nm)
2
-1000
19
No Stress
With Stress
Electron Density ( x10 )
19
Electron Density ( x10 )
8
~15-35%
Decrease
-500
0
500
X Axis
X (nm)
1000
Figure 5.8. A plot of the simulated 2DEG density in the x-direction of the device at a depth of 1
nm below the AlGaN/GaN interface. A reduction of electron density is observed in the channel
region under the gate while a slight increase in electron density is observed outside the gate region.
Figure 5.8 shows that the distribution of the electron density follows the distribution
pattern of the polarisation charge shown in Figure 5.5a. It is observed that the region
under the gate has about 15-35% less carriers as compared to the device without the
DLC stress liner. This reduction in 2DEG density under the gate would thus require a
less negative gate voltage to deplete the 2DEG in order to turn off the device. This
explains the positive shift in the threshold voltage of the device with the DLC liner, as
observed in the ID-VGS plot in Figure 5.6.
On the other hand, the tensile stress in the regions beyond the gate has resulted in an
average increase of about 3-7% in the 2DEG density in those regions. This slight
increase in electron concentrations could have provided a lower external resistance in
90
the LS-G and LD-G region. This higher conductivity in these regions is one of the factors
that enhanced the drive current in the ID-VDS plot shown in Figure 5.7.
5.4
Effect of Stress on Carrier Mobility
The non-uniform distribution of the polarisation charge could have had other effects
than the 2DEG distribution. Thus a plot of the mobility of the electrons in the 2DEG
(1 nm below the AlGaN/GaN interface) is extracted from the simulation and is shown
in Figure 5.9.
2
Electron Mobility (cm /Vs)
1200
With Stress
No Stress
1000
800
6 - 17%
Increase
600
400
200
-1000
-500
0
500
1000
X (nm)
Figure 5.9. A plot of the simulated electron mobility value along the x-axis of the MOSHEMT
with and without stress, at a depth of 1 nm below the AlGaN/GaN interface.
91
The mobility plot in Figure 5.9 indicates a 6-17% increase in mobility of the 2DEG in
the region under the gate. This increase in mobility could have been a result of a
change in the vertical (y-direction) electric field in the device due to the non-uniform
distribution of the polarisation charges. The reduction in 2DEG density, which reduces
carrier-carrier scattering, could also be a factor in the increase in electron mobility in
the channel. The vertical electric field is extracted from the simulation and is shown in
Figure 5.10. Only the vertical electric field is taken into consideration here as the
mobility degradation effect would be caused predominantly by the vertical electric
field as presented by the Lombardi Model [43],[45]. The electric field shown here is
the vertical electric field 1 nm below the AlGaN/GaN interface.
By looking at equations (4.20) (Lombardi model) in the previous section (section
4.1.3), it is observed that this model is directly affected by the vertical electric field
present in the device. Thus it is reasonable to attribute this change in mobility to the
change in electric field shown in Figure 5.10. Equation (4.20) is re-produced here for
ease of reference,
(
)
(
(4.20).
)
92
-1
Vertical Electric Field (MV/cm )
1.0
With Stress
Without Stress
0.9
0.8
0.7
0.6
0.5
-600
.
-400
-200
0
200
X (nm)
400
600
Figure 5.10. A plot of the simulated vertical electric field along the x-direction of the MOSHEMT
with and without stress, at a depth of 1nm below the AlGaN/GaN interface.
However, it should be noted that this change in mobility is not due to the change in
effective mass or change in band structure as the models that describe the mobility
change due to these factors are not used in this simulation. The increase in mobility in
this region could be one of the factors that resulted in the increase in drive current in
the device. However, the values of the fitting parameters such as B and C are not
known for GaN materials, thus the model may only give a qualitative evaluation and
not a quantitative one.
93
Chapter 6 Conclusion
In this thesis, characterisation work had been carried out on experimental AlGaN/GaN
MOSHEMT devices. Before that, some preparatory work needs to be done in order to
achieve a successful fabrication of the required devices. This thesis had given an
overview of the mask design and some of the various designs of test structures that
were designed and included in the mask layout. In addition, a brief description of the
mechanism and physics of how each test structure works is also presented.
After these preparatory works were presented, the fabrication process was described
with the various splits in the experiment included. Different electrical characterisation
techniques were presented and the results and analysis of each characterisation were
also discussed in a succinct manner.
The observations in the characterisation work of experimental devices shows that
devices with surface passivation (in situ VA and SiH4 treatment) gave a 53%
enhancement in saturation drain current over the control device without any surface
passivation. However, it had little impact on the threshold voltage which was
determined using the ID-VGS measurement. The electrical characterisations of devices
with DLC stress liners were also presented. These devices had a +0.7 V threshold
voltage shift over the control devices without DLC liners. This change in threshold
voltage could be due to the change in 2DEG density under the gate region. In addition
to the change in threshold voltage, the devices with DLC liners also saw a 36%
enhancement in saturation drain current over devices without DLC liners. This could
94
be a result of a change in mobility of the channel electrons due to the stress induced by
the DLC liner.
Numerical simulations were performed to complement the experimental work.
Various simulation models and the physics used in the simulation were described and
explained. The different relevant models were integrated to perform several device
simulations. Firstly, the effect of the thickness of the AlGaN barrier layer on the
threshold voltage of the AlGaN/GaN MOSHEMT was studied. This part of the work
showed consistency with the literature which showed that a thicker AlGaN layer
would generally gave rise to a higher 2DEG density before stress relaxation sets in.
The effect of interface states were also simulated with different types of traps put into
place at the AlGaN/high-k dielectric interface to study their effect on device
performance. A postulation was also made with respect to the mechanism of the
surface passivation work where it was hypothesized based on simulation result that
during the surface passivation, acceptor-like traps are the traps that are being
passivated thus improving the device performance.
Finally, an AlGaN/GaN MOSHEMT with stress induced by a stress liner was also
simulated in order to better understand the effect of the compressive stress that the
DLC liner induced on the experimental devices. It is shown in the simulations that the
stress induced in the AlGaN layer resulted in a different distribution of polarisation
induced charges. This non-uniform distribution of polarisation charges would thus
give a non-uniform electron distribution in the 2DEG. As a result of the difference of
95
stress in different parts of the device, a shift in threshold voltage is observed while an
enhancement in drive current is also observed due to change in electric field
experienced in the device.
6.1
Possible Future Work
Future projects can be done as extensions to the simulation work done in this thesis.
One of these would be to make use of the stress model that might be available in
future versions of Sentaurus TCAD to simulate the effect of stress on device
performance. In this simulation work, the effect of stress is configured such that it
only affects the polarisation charges and that alone has shown that a shift in threshold
voltage and enhancement in drive current is possible. If the other effects of stress such
as the effect of stress on the effective mass of the carriers or the energy band structure
of the GaN and AlGaN can be incorporated, a more accurate study of the stress effect
could be performed.
Temperature effect on device performance can also be studied on this simulation suite
as the thermodynamic and hydrodynamic models would give accurate simulation on
temperature effects. This is especially important as GaN-based devices are largely
used for high power devices and temperature effect could play a large effect on device
performance.
96
References
[1] J. Ibbetson, P. Fini, K. Ness, S. DenBaars, J. Speck, U. Mishra, "Polarization
Effects, Surface States, and The Source of Electrons in AlGaN/GaN Heterostructure
Field Effect Transistors," Applied Physics Letters, vol. 77, pp. 250-252, 2000.
[2] Y. Yeo, T. Chong, M. Li, "Uniaxial Strain Effect on the Electronic and Optical
Properties of Wurtzite GaN-AlGaN Quantum-well Lasers," IEEE Journal of Quantum
Electronics, vol. 34, no. 11, pp. 2224-2224, 1998.
[3] M. Suzuki, T. Uenoyama, "Strain Effect on Electronic and Optical Properties of
GaN/AlGaN Quantum-well Lasers," Journal of Applied Physics, vol. 80, no. 12, pp.
6868-6868, 1996.
[4] S. Pearton, F. Ren, A. Zhang, K. Lee, "Fabrication and Performance of GaN
Electronic Devices," Materials Science and Engineering: R: Reports, vol. 30, no. 3-6,
pp. 55-55, 2000.
97
[5] U. Mishra, P. Parikh, Y. Wu, "AlGaN/GaN HEMTs - An Overview of Device
Operation and Applications," Proceedings of the IEEE, vol. 90, no. 6, pp. 1022-1022,
2002.
[6] F. Bernardini, V. Fiorentini, D. Vanderbilt, "Spontaneous Polarization and
Piezoelectric Constants of III-V Nitrides," Physical Review B, vol. 56, no. 16, pp.
10024-10024, 1997.
[7] Synopsys Inc., "Sentaurus Device User Guide (Version D-2010.03, March
2010),":187-191.
[8] R. Szweda, Gallium Nitride & Related Wide Bandgap Materials & Devices,
Elsevier Science, 2000.
[9] J. Piprek, Nitride Semiconductor Devices, LibreDigital, 2007.
[10] C. Wood, Polarization Effects in Semiconductors, Springer Verlag, pp.380-383,
2008.
98
[11] F. Marino, N. Faralli, T. Palacios, D. Ferry, S. Goodnick, M. Saraniti, "Effects of
Threading Dislocations on AlGaN/GaN High-Electron Mobility Transistors," IEEE
Transactions on Electron Devices, vol. 57, no. 1, pp. 353-353, 2010.
[12] G. Jessen, R. Fitch, J. Gillespie, G. Via, B. White, S. Bradley et al, "Effects of
Deep-level Defects on Ohmic Contact and Frequency Performance of AlGaN/GaN
High-Electron Mobility Transistors," Applied Physics Letters, vol. 83, no. 3, pp. 485485, 2003.
[13] W. Saito, M. Kuraguchi, Y. Takada, K. Tsuda, I. Omura, T. Ogura, "Influence of
Surface Defect Charge at AlGaN-GaN-HEMT upon Schottky Gate Leakage Current
and Breakdown Voltage," IEEE Transactions on Electron Devices, vol. 52, no. 2, pp.
159-159, 2005.
[14] W. Saito, Y. Takada, M. Kuraguchi, K. Tsuda, I. Omura, T. Ogura, "600V
AlGaN/GaN Power-HEMT: Design, Fabrication and Demonstration on High Voltage
DC-DC Converter," IEEE International Electron Devices Meeting, pp. 23.7.1-23.7.1,
2003.
99
[15] W. Saito, M. Kuraguchi, Y. Takada, K. Tsuda, I. Omura, T. Ogura, "High
Breakdown Voltage Undoped AlGaN-GaN Power HEMT on Sapphire Substrate and
its Demonstration for DC-DC Converter Application," IEEE Transactions on Electron
Devices, vol. 51, no. 11, pp. 1913-1913, 2004.
[16] R. Quay, Gallium Nitride Electronics, Springer Verlag, 2008.
[17] W. Lanford, T. Tanaka, Y. Otoki, I. Adesida, "Recessed-Gate EnhancementMode GaN HEMT with High Threshold Voltage," Electronics Letters, vol. 41, no. 7,
pp. 449, 2005.
[18] S. Huang, H. Chen, K. Chen, "Effects of the Fluorine Plasma Treatment on the
Surface Potential and Schottky Barrier Height of AlxGa1-xN/GaN Heterostructures,"
Applied Physics Letters, vol. 96, no. 23, 2010.
[19] T. Mizutani, M. Ito, S. Kishimoto, F. Nakamura, "AlGaN/GaN HEMTs With
Thin InGaN Cap Layer for Normally Off Operation," IEEE Electron Device Letters,
100
no. 7, pp. 549-551, 2007.
[20] "Inlustra Technologies : Technology : The Nonpolar Advantage," 2011.[Online].
Available: http://www.inlustra.com/technology/nonpolar.html.
[21] R. King-Smith, D. Vanderbilt, "Theory of Polarization of Crystalline Solids,"
Physical Review B (Condensed Matter), vol. 47, no. 3, pp. 1651-1654, 1993.
[22] O. Ambacher, B. Foutz, J. Smart, J. Shealy, N. Weimann, K. Chu et al, "Two
Dimensional Electron Gases Induced by Spontaneous and Piezoelectric Polarization in
Undoped and Doped AlGaN/GaN Heterostructures," Journal of Applied Physics, vol.
87, pp. 334, 2000.
[23] H. Morkoç, R. Cingolani, B. Gil, "Polarization Effects in Nitride Semiconductor
Device Structures and Performance of Modulation Doped Field Effect Transistors,"
Solid-State Electronics, vol. 43, no. 10, pp. 1909-1927, 1999.
101
[24] R. Resta, "Macroscopic Electrical Polarization of Materials (Disordered and
Crystalline)," in Journal of Physics, vol. 117, pp. 012024 (8), 2008.
[25] F. Bernardini, V. Fiorentini, "Macroscopic Polarization and Band Offsets at
Nitride Heterojunctions," Physical Review B (Condensed Matter), vol. 57, no. 16, pp.
R9427-R9430, 1998.
[26] F. Bernardini, V. Fiorentini, "Spontaneous Versus Piezoelectric Polarization in
III-V Nitrides: Conceptual Aspects and Practical Consequences," Physica Status
Solidi B, vol. 216, no. 1, pp. 391-391, 1999.
[27] T. Mizutani, Y. Ohno, M. Akita, S. Kishimoto, K. Maezawa, "Current Collapse in
AlGaN/GaN HEMTs Investigated by Electrical and Optical Characterizations,"
Physica Status Solidi A, vol. 194, no. 2, pp. 447-451, 2002.
[28] E. H. Rhoderick, "The Hall Effect — An Important Diagnostic Tool," III-Vs
Review, vol. 13, no. 3, pp. 46-51, 2000.
102
[29] S. Cohen, "Contact Resistance and Methods for its Determination," Thin Solid
Films, vol. 104, no. 3-4, pp. 361-361, 1983.
[30] D. K. Schroder, Semiconductor Material and Device Characterization, WileyIEEE Press, 2006.
[31] X. K. Liu, B. Liu, E. Low, H. C. Chin, W. Liu and M. C. Yang et al, "Diamondlike Carbon (DLC) Liner with Highly Compressive Stress Formed on AlGaN/GaN
MOSHEMTs with In Situ Silane Surface Passivation for Performance Enhancement,"
in IEEE Electron Devices Meeting, pp. 11.3.1-11.3.4, 2010.
[32] X. K. Liu, H. C. Chin, L. S. Tan, Y. C. Yeo, "In Situ Surface Passivation of
Gallium Nitride for Metal-Organic Chemical Vapor Deposition of High-Permittivity
Gate Dielectric," IEEE Transactions on Electron Devices, vol. 58, no. 1, pp. 95-102,
2011.
[33] S. Wolf and R. N. Tauber, Silicon processing for the VLSI era, 1995.
103
[34] H. Chin, M. Zhu, X. Liu, H. Lee, L. Shi, L. Tan et al, "Silane-Ammonia Surface
Passivation for Gallium Arsenide Surface-Channel n-MOSFETs," IEEE Electron
Device Letters, vol. 30, no. 2, pp. 110-110, 2009.
[35] B. Kang, S. Kim, J. Kim, F. Ren, K. Baik, S. Pearton et al, "Effect of External
Strain on the Conductivity of AlGaN/GaN High-Electron Mobility Transistors,"
Applied Physics Letters, vol. 83, no. 23, pp. 4845-4845, 2003.
[36] C. Chang, S. Hsiao, E. Chang, C. Lu, J. Huang, C. Lee, "Changes of Electrical
Characteristics for AlGaN/GaN HEMTs Under Uniaxial Tensile Strain," IEEE
Electron Device Letters, vol. 30, no. 3, pp. 213-213, 2009.
[37] K. M. Tan, M. Zhu, W. W. Fang, M. Yang, T. Y. Liow and R. T. P. Lee et al, "A
New Liner Stressor with Very High Intrinsic Stress (> 6 GPa) and Low
Permittivity Comprising Diamond-Like Carbon (DLC) for Strained p-Channel
Transistors," in IEEE International Electron Devices Meeting, pp. 127-130, 2007.
[38] X. Liu, B. Liu, E. Low, W. Liu, M. Yang, L. Tan et al, "Local Stress Induced by
Diamond-like Carbon Liner in AlGaN/GaN Metal-Oxide-Semiconductor High-
104
Electron Mobility Transistors and Impact on Electrical Characteristics," Applied
Physics Letters, vol. 98, pp. 183502, 2011.
[39] D. Neamen, Semiconductor Physics And Devices, McGraw-Hill
Science/Engineering/Math, 2011.
[40] K. Shimada, T. Sota, K. Suzuki, H. Okumura, "First-Principles Study on
Piezoelectric Constants in Strained BN, AlN, and GaN," Japanese Journal of Applied
Physics, Part 2: Letters, vol. 37, no. 12 A, pp. L1421-L1423, 1998.
[41] D. Look, Z. Q. Fang, B. Claflin, "Identification of Donors, Acceptors, and Traps
in Bulk-like HVPE GaN," Journal of Crystal Growth, vol. 281, no. 1, pp. 143-150,
2005.
[42] E. Miller, X. Dang, H. Wieder, P. Asbeck, E. Yu, G. Sullivan et al, "Trap
Characterization by Gate-Drain Conductance and Capacitance Dispersion Studies of
an AlGaN/GaN Heterostructure Field-Effect Transistor," Journal of Applied Physics,
vol. 87, no. 11, pp. 8070-8073, 2000.
105
[43] S. Xie, J. Yin, S. Zhang, B. Liu, W. Zhou, Z. Feng, "Trap Behaviors in AlGaNGaN Heterostructures by C-V Characterization," Solid-State Electronics, vol. 53, no.
11, pp. 1183-1185, 2009.
[44] A. Brannick, N. Zakhleniuk, B. Ridley, L. Eastman, J. Shealy, W. Schaff,
"Hydrodynamic Simulation of Surface Traps in The AlGaN/GaN HEMT,"
Microelectronics Journal, vol. 40, no. 3, pp. 410-410, 2009.
[45] C. Lombardi, S. Manzini, A. Saporito, M. Vanzi, "A Physically Based Mobility
Model for Numerical Simulation of Nonplanar Devices," IEEE Transactions on
Computer-Aided Design of Integrated Circuits and Systems, vol. 7, no. 11, pp. 11641164, 1988.
[46] M. O. Manasreh, H. X. Jiang and J. H. Edgar, Three-Nitride Semiconductors,
Elsevier Science, 2000.
106
[47] G. Masetti, M. Severi, S. Solmi, "Modeling of Carrier Mobility Against Carrier
Concentration in Arsenic-, Phosphorus-, and Boron-Doped Silicon," IEEE
Transactions on Electron Devices, vol. 30, no. 7, pp. 764-764, 1983.
[48] C. Canali, G. Majni, R. Minder, G. Ottaviani, "Electron and Hole Drift Velocity
Measurements in Silicon and Their Empirical Relation to Electric Field and
Temperature," IEEE Transactions on Electron Devices, vol. 22, no. 11, pp. 1045-1045,
1975.
[49] D. Caughey, R. Thomas, "Carrier Mobilities in Silicon Empirically Related to
Doping and Field," Proceedings of the IEEE, vol. 55, no. 12, pp. 2192-2192, 1967.
[50] M. Darwish, J. Lentz, M. Pinto, P. Zeitzoff, T. Krutsick, H. Vuong, "An
Improved Electron and Hole Mobility Model for General Purpose Device Simulation,"
IEEE Transactions on Electron Devices, vol. 44, no. 9, pp. 1529-1529, 1997.
[51] B. Van Daele, G. Van Tendeloo, M. Germain, M. Leys, Z. Bougrioua, I.
Moerman, "Relation Between Microstructure and 2DEG Properties of AlGaN/GaN
Structures," Physica Status Solidi B, vol. 234, no. 3, pp. 830-830, 2002.
107
[52] M. Fieger, M. Eickelkamp, L. Koshroo, Y. Dikme, A. Noculak, H. Kalisch et al,
"MOVPE, Processing and Characterization of AlGaN/GaN HEMTs with Different Al
Concentrations on Silicon Substrates," Journal of Crystal Growth, vol. 298, pp. 843847, 2007.
[53] R. Cheng, B. Liu, Y. Yeo, "Carrier Transport in Strained p-Channel Field-effect
Transistors with Diamondlike Carbon Liner Stressor," Applied Physics Letters, vol. 96,
no. 9, 2010.
[54] B. Liu, K. Tan, M. Yang, Y. Yeo, "NBTI Reliability of P-Channel Transistors
With Diamond-Like Carbon Liner Having Ultrahigh Compressive Stress," IEEE
Electron Device Letters, vol. 30, no. 8, pp. 867-867, 2009.
[55] B. Liu, H. Wong, M. Yangyang, Y. Yeo, "Strained Silicon Nanowire p-Channel
FETs With Diamond-Like Carbon Liner Stressor," IEEE Electron Device Letters, vol.
31, no. 12, pp. 1371-1371, 2010.
108
[56] A. Wright, "Elastic Properties of Zinc-blende and Wurtzite AlN, GaN, and InN,"
Journal of Applied Physics, vol. 82, no. 6, pp. 2833-2839, 1997.
[57] Y. Duan, J. Li, S. Li, J. Xia, "Elasticity, Band-gap Bowing, and Polarization of
AlxGa1-xN Alloys," Journal of Applied Physics, vol. 103, no. 2, pp. 023705-023701,
2008.
[58] F. Xu, B. Shen, M. Wang, J. Xu, L. Lu, Z. Miao et al, "Mechanical Properties of
AlxGa1-xN Films with High Al Composition Grown on AlN/Sapphire Templates,"
Applied Physics Letters, vol. 91, no. 9, pp. 1-3, 2007.
[59] D. Caceres, I. Vergara, R. Gonzalez, E. Monroy, F. Calle, E. Munoz et al,
"Nanoindentation on AlGaN Thin Films," Journal of Applied Physics, vol. 86, no. 12,
pp. 6773-6778, 1999.
[60] M. Mastro, J. LaRoche, N. Bassim, C. Eddy, "Simulation on the Effect of Nonuniform Strain from the Passivation Layer on AlGaN/GaN HEMT," Microelectronics
Journal, vol. 36, no. 8, pp. 705-705, 2005.
109
[61] B. Luo, R. Mehandru, J. Kim, F. Ren, B. Gila, A. Onstine et al, "Comparison of
surface passivation films for reduction of current collapse in AlGaN/GaN high
electron mobility transistors," Journal of the Electrochemical Society, vol. 149, no. 11,
pp. G613-G619, 2002.
110
Appendix A
The baseline simulation code is given here. The simulation code consists mainly of the
structure file (which describes the structure of the device) and the simulation file
(which describes the necessary models to be used and the other electrical and material
specifications required).
Structure File:
;------------------------------------------------------------;
; vertical dimensions
(define hAlN
0.100)
(define hGaN
2.000)
(define hAlGaN
0.020)
(define hPass
0.050)
(define hDielectric 0.007)
; horizontal dimensions
(define Xmin
-5.650)
(define SrcLngth 5.000)
(define DrnLngth 5.000)
(define GtLngth
0.300)
(define SrcSep
0.500)
(define DrnSep
0.500)
; Ohmic contact definitions
(define Sep
0.00)
(define Dpng
1E20)
; Molefraction definition
(define x_AlGaN 0.25)
;----------------------------------------------------------------------;
; Derived quantities
;----------------------------------------------------------------------;
(define Xmax
(+ Xmin SrcLngth SrcSep GtLngth DrnSep DrnLngth))
(define Ymax
(+ hGaN))
(define Ymin
(- 0 hAlGaN hPass))
(define Ysrfc
(- 0 hAlGaN))
(define Yjnctn (- 0 Sep))
(define
(define
(define
(define
Xsrc
XgtLft
XgtRght
Xdrn
(+
(+
(+
(+
Xmin SrcLngth))
Xsrc SrcSep))
XgtLft GtLngth))
XgtRght DrnSep))
111
;----------------------------------------------------------------------;
; Create structure
;----------------------------------------------------------------------;
(sdegeo:create-rectangle (position Xsrc Ymin 0) (position XgtLft (Ysrfc hDielectric) 0) "Nitride" "LeftPassivation" )
(sdegeo:create-rectangle (position Xsrc (- Ysrfc hDielectric) 0)
(position Xdrn Ysrfc 0) "Al2O3" "Dielectric")
(sdegeo:create-rectangle (position XgtRght Ymin 0) (position Xdrn (Ysrfc hDielectric) 0) "Nitride" "RightPassivation" )
(sdegeo:create-rectangle (position Xmin Ysrfc 0) (position Xmax 0 0)
"AlGaN" "AlGaN_barrier" )
(sdegeo:create-rectangle (position Xmin 0 0) (position Xmax Ymax 0)
"GaN" "GaN_bulk" )
; ----------------------------------------------------------------------; Place AlGaN mole fraction
; ----------------------------------------------------------------------(sdedr:define-constant-profile "CP.xMole" "xMoleFraction" x_AlGaN)
(sdedr:define-constant-profile-material "CP.xMole" "CP.xMole" "AlGaN" 0
"Replace")
; ----------------------------------------------------------------------; Place doping profiles to emulate metal spikes
; ---------------------------------------------------------------------(sdedr:define-refinement-window "Pl.Source" "Rectangle" (position Xmin
Ysrfc 0) (position Xsrc Yjnctn 0))
(sdedr:define-constant-profile "P.source" "PhosphorusActiveConcentration"
Dpng)
(sdedr:define-constant-profile-placement "P.source" "P.source"
"Pl.Source")
(sdedr:define-refinement-window "Pl.Drain" "Rectangle" (position Xdrn
Ysrfc 0) (position Xmax Yjnctn 0))
(sdedr:define-constant-profile "P.drain" "PhosphorusActiveConcentration"
Dpng)
(sdedr:define-constant-profile-placement "P.drain" "P.drain" "Pl.Drain")
; ----------------------------------------------------------------------; Place doping profiles to emulate background doping of GaN
; ----------------------------------------------------------------------(sdedr:define-refinement-window "Pl.GaN" "Rectangle" (position Xmin 0 0)
(position Xmax Ymax 0))
(sdedr:define-constant-profile "P.gan" "NDopantActiveConcentration" 1E16)
(sdedr:define-constant-profile-placement "P.gan" "P.gan" "Pl.GaN")
; ----------------------------------------------------------------------; Create and place all electrodes
; ----------------------------------------------------------------------(sdegeo:define-contact-set "source")
(sdegeo:set-current-contact-set "source")
(sdegeo:set-contact-edges (find-edge-id (position (+ Xmin 0.001) Ysrfc
0)))
(sdegeo:define-contact-set "gate")
(sdegeo:set-current-contact-set "gate")
(sdegeo:set-contact-edges (find-edge-id (position (+ XgtLft 0.001) (Ysrfc hDielectric) 0)))
(sdegeo:set-contact-edges (find-edge-id (position (* 0.5 (+ XgtLft
XgtRght)) (- Ysrfc hDielectric) 0)))
(sdegeo:set-contact-edges (find-edge-id (position (- XgtRght 0.001) (Ysrfc hDielectric) 0)))
112
(sdegeo:define-contact-set "drain")
(sdegeo:set-current-contact-set "drain")
(sdegeo:set-contact-edges (find-edge-id (position (+ Xdrn 0.001) Ysrfc
0)))
; ------------------------------------------------------------------------------; Grid refinement definitions
; ------------------------------------------------------------------------------(sdedr:define-refinement-window "Pl.Default" "Rectangle" (position Xmin
Ymin 0) (position Xmax Ymax 0))
(sdedr:define-refinement-size "Ref.Default" 0.4 0.2 99 0.01 0.005 66 )
(sdedr:define-refinement-placement "Ref.Default" "Ref.Default"
"Pl.Default" )
(sdedr:define-refinement-function "Ref.Default" "DopingConcentration"
"MaxTransDiff" 1)
(sdedr:define-refinement-window "Pl.eDensity" "Rectangle"
(position (- XgtRght 0.05) Ymin 0)
(position (+ XgtRght 0.3)
0.1 0))
(sdedr:define-refinement-size "Ref.eDensity" 0.005 99 99 0.005 66 66 )
(sdedr:define-refinement-placement "Ref.eDensity" "Ref.eDensity"
"Pl.eDensity" )
(sdedr:define-refinement-window "Pl.channel_h" "Rectangle"
(position (- XgtLft 0.1) Ymin 0)
(position (+ XgtRght 1)
0.1 0))
(sdedr:define-refinement-size "Ref.channel_h" 0.1 99 99 0.001 66 66 )
(sdedr:define-refinement-placement "Ref.channel_h" "Ref.channel_h"
"Pl.channel_h" )
(sdedr:define-refinement-window "Pl.drain_c" "Rectangle"
(position (- XgtRght 0.2) Ymin 0)
(position (+ XgtRght 0.7)
0 0))
(sdedr:define-refinement-size "Ref.drain_c" 0.04 99 99 0.005 66 66 )
(sdedr:define-refinement-placement "Ref.drain_c" "Ref.drain_c"
"Pl.drain_c" )
(sdedr:define-refinement-window "Pl.contact_r" "Rectangle"
(position (- XgtRght 0.005) Ymin 0)
(position (+ XgtRght 0.05)
0.1 0))
(sdedr:define-refinement-size "Ref.contact_r" 0.005 99 99 0.001 66 66 )
(sdedr:define-refinement-placement "Ref.contact_r" "Ref.contact_r"
"Pl.contact_r" )
(sdedr:define-refinement-window "Pl.contact_l" "Rectangle"
(position (- XgtLft 0.005) Ymin 0)
(position (+ XgtLft 0.005) 0.1 0))
(sdedr:define-refinement-size "Ref.contact_l" 0.005 99 99 0.001 66 66 )
(sdedr:define-refinement-placement "Ref.contact_l" "Ref.contact_l"
"Pl.contact_l" )
(sdedr:define-refinement-window "Pl.surface" "Rectangle"
(position Xmin Ysrfc 0)
(position Xmax 0 0))
(sdedr:define-multibox-size "MB.surface" 99 0.05 99 66 0.001 66 1 2 1 )
(sdedr:define-multibox-placement "MB.surface" "MB.surface" "Pl.surface" )
113
(sdedr:define-multibox-size "MB.barrier" 99 0.05 99 66 0.0004 66 1 -2 1 )
(sdedr:define-multibox-placement "MB.barrier" "MB.barrier" "Pl.surface" )
(sdedr:define-refinement-window "Pl.surface_Pol" "Rectangle" (position
Xmin Ysrfc 0) (position Xmax (+ Ysrfc 0.0001) 0))
(sdedr:define-multibox-size "MB.surface_Pol" 99 0.05 99 66 0.0002 66 1 4
1 )
(sdedr:define-multibox-placement "MB.surface_Pol" "MB.surface_Pol"
"Pl.surface_Pol" )
(sdedr:define-refinement-window "Pl.channel" "Rectangle"
(position Xmin
0 0)
(position Xmax hGaN 0))
(sdedr:define-multibox-size "MB.channel" 99 0.1 99 66 0.0004 66 1 1.5 1 )
(sdedr:define-multibox-placement "MB.channel" "MB.channel" "Pl.channel" )
(sdeaxisaligned:set-parameters "yCuts"
(list -0.025 -0.0249 -5e-5 0 5e-5)
)
;--- Generate and save the mesh using Mesh
(sde:build-mesh "snmesh" "" "n2_dielec")
Simulation file
Electrode {
{ Name="gate"
Voltage= -0.8 Workfunction = 4.4 }
{ Name="source" Voltage= 0 Resist = 500 }
{ Name="drain" Voltage= 0 Resist = 500 }
}
File {
* Input files
Grid= "n2_dielec_msh.tdr"
Parameter= "pp4_des.par"
* Output files
Current= "n4_des.plt"
Plot= "n4_des.tdr"
Output= "n4_des.log"
}
Physics {
Mobility(
Enormal(Lombardi)
DopingDependence
eHighfieldsaturation
)
EffectiveIntrinsicDensity (Nobandgapnarrowing)
Fermi
Recombination(SRH)
RecGenHeat
Aniso(Poisson)
Thermionic
}
114
Physics (Material="GaN") {
Traps (
(Acceptor Level Conc= 5e17 EnergyMid= 1.0 EnergySig= 0
FromMidBandGap
eXSection= 1e-15 hXSection= 1e-15)
)
}
Physics (Material="AlGaN") {
Traps (
(Acceptor Level Conc= 5e17 EnergyMid= 1.0 EnergySig= 0
FromMidBandGap
eXSection= 1e-15 hXSection= 1e-15)
)
}
Physics (MaterialInterface="AlGaN/GaN") {
Charge(Uniform Conc=1.328115e+13
)
}
Physics (MaterialInterface="AlGaN/Al2O3") {
Charge(Uniform Conc=-3.13841e+13
)
Traps (
(Donor Level Conc= 3.0e13 EnergyMid= 0.2 FromMidBandGap)
)
}
Plot {
Potential Electricfield/Vector
eDensity hDensity
eCurrent/Vector hCurrent/Vector
TotalCurrent/Vector
SRH Auger Avalanche
eMobility hMobility
eQuasiFermi hQuasiFermi
eGradQuasiFermi hGradQuasiFermi
eEparallel hEparallel
eMobility hMobility
eVelocity hVelocity
DonorConcentration Acceptorconcentration
Doping SpaceCharge
ConductionBand ValenceBand
BandGap Affinity
xMoleFraction
eTemperature hTemperature
eTrappedCharge hTrappedCharge
eInterfaceTrappedCharge hInterfaceTrappedCharge
}
Math {
Extrapolate
Iterations= 16
Digits= 6
ErrRef(electron) = 1E5
ErrRef(hole) = 1E3
RHSmin= 1e-10
RHSmax= 1e30
CDensityMin= 1e-20
115
DirectCurrentComputation
RelTermMinDensity= 1e5
eMobilityAveraging= ElementEdge
}
Solve {
Coupled (Iterations= 100000 LinesearchDamping= 0.001)
Coupled (Iterations= 100) {Poisson Electron Hole}
{Poisson}
****************************************************************
* Zero bias plot
****************************************************************
Plot(FilePrefix="300nm_NS_n4_Zero_Bias")
****************************************************************
* IdVd curve with Vg=0 V
****************************************************************
NewCurrentFile="300nm_NS_IdVd_Vg0_"
Quasistationary (
InitialStep= 5e-3 Minstep= 1e-7 MaxStep= 0.05 Increment=
1.55
Goal {Name="drain" Voltage= 10}
) {
Coupled {Poisson Electron Hole}
}
Plot(FilePrefix="300nm_NS_Vg0_Vd10")
****************************************************************
* IdVg curve with Vd=0.5 V
****************************************************************
*NewCurrentFile="NS_IdVg_Vd10_"
*Quasistationary (
*
InitialStep= 0.025 Minstep= 1e-7 MaxStep= 0.05 Increment=
1.5
*
Goal {Name="gate" Voltage= -5}
*) {
*
Coupled {Poisson Electron Hole}
*}
*
*Plot(FilePrefix="NS_Vg-5_Vd10")
}
116
Appendix B
The Taurus Abaqus simulation for simulating the stress on the device is shown here.
The author would like to thank Liu Bin for his help in providing the simulation source
code for this part of the project.
TaurusProcess
# 1/2 gate length
Define (Lg = 150nm)
Define (sp = 700nm)
# Define the initial simulation domain and the initial grid. Also
# initialize the Boron concentration in the silicon substrate.
DefineDevice
(
xSize = expr(($Lg)+ $sp)
ySize = 0.5um
#zSize = 0.5um
Initialize(
Name = phosphorus
Value = 1.0e16
)
AmbientHeight = 1um
RefinementsFile = refinements_data
)
#
#
#
#
Add a set of regrid parameters to the current list of refinements.
This specifies a maximum mesh spacing to 0.3um overall, and 5nm
at region boundaries. The refinements are used later when mesh
adaptation and boundary fitting is done during the process flow.
Physics(
Material=diamond
YoungsModulus=760e+9
PoissonRatio=0.20
)
Physics(
Material=Silicon
C11= 3.9e+11
C12= 1.45e+11
C44= 1.05e+11
)
Physics(
Material=Germanium
YoungsModulus=3.97e11
)
117
Refinements(
Regrid(
MeshSpacing = 0.3um
MergeRegionsOfMaterial = oxide
MergeRegionsOfMaterial = diamond
CriticalFeatureSize = 0.1nm
ThinLayer = 1nm
Criterion(
AllInterfaces
MeshSpacing = 5nm
)
)
Regrid(
MeshSpacing = 0.003um
MinX = expr(($sp)-0.1) MaxX = expr(($Lg)+ ($sp))
MinY = 0um MaxY = 0.01um
)
Regrid(
MeshSpacing = 0.005um
MinX = 0.2 MaxX = expr(($Lg)+ ($sp))
MinY = 0um MaxY = 0.2um
)
)
# Enable automatic stress history simulation during process flow.
Physics(
KeepStressHistory
Material = silicon
Anisotropic = true
Bandgap (StressInducedBGNActive = true)
)
Physics
(
Material=Silicon
Equation=Germanium
Diffusion(
StrainFactor(
StrainDependency = true
StrainCoefficient = 40.0)
)
Equation=Boron
Diffusion(
StrainFactor(
StrainDependency=true
StrainCoefficient= -17.0)
)
)
# Modify the linear solver and associated solution parameters.
Numerics(
Linearsolver = direct
NewtonResid = 1e-10
)
Numerics (ImbalanceLimit = 10.0)
Numerics(
Iterations=20
maxDivergenceCount=10
)
118
# Deposit thin planar oxide and nitride layers, then etch a 0.5 um
# deep vertical trench using an etch mask at X = 0.2 um.
Deposit(
Material = Germanium
Thickness = 0.02um
)
#Deposit(
#
Material = oxide
#
Thickness = 0.01um
#)
#Deposit(
#
Material = nitride
#
Thickness = 0.1um
#)
#Etch
#(
#
#
#
#
#
#
#
#
#
#)
Thickness = 0.5um
EtchType = dry
MaskPolygon
(
Point(x=0.2um z=-0.3um)
Point(x=expr(($Lg)+ ($sp)) z=-0.3um)
Point(x=expr(($Lg)+ ($sp)) z= 0.3um)
Point(x=0.2um z= 0.3um)
)
# Grow liner oxide.
#Diffuse
#(
#
Time = 5min
#
Temperature = 900C
#
DryO2
#
NativeLayerThickness = 50A
#
InitialTimeStep = 0.5min
#)
#Deposit
#(
#Layer(
#
Material = oxide
#
thickness = 5nm
#
Onto(Material = silicon)
#)
#)
Save(MeshFile = SiGe_50nm_nosti01.tdf)
# Deposit TEOS infill to form the show trench isolation region,
# which has a planar surface within the trench. Then strip the
# nitride layer away.
#Deposit
#(
#
Material = teos
#
SurfacePosition = -0.02um
#)
#Etch
#(
#
Material = nitride
#
EtchType = all
#)
#Save(MeshFile = SiGe_50nm_nosti02.tdf)
119
# Perform channel doping.
#Implant
#(
#
Name = boron
#
Energy = 500.0
#
Dose = 1e13
#
Tilt = 0
#
Rotation = 0
#
NoRegrid
#)
#Implant
#(
#
Name = boron
#
Energy = 140.0
#
Dose = 1e13
#
Tilt = 0
#
Rotation = 0
#)
#Implant
#(
#
Name = boron
#
Energy = 50.0
#
Dose = 1e12
#
Tilt = 0
#
Rotation = 0
#)
#Etch
#(
#
Material = oxide
#
Thickness = 150A
#
EtchType = dry
#)
#Save(MeshFile = SiGe_50nm_nosti03.tdf)
# Create the gate oxide and the Poly gate layer.
Deposit
(
Material = hafniumoxide
thickness = 7nm
)
Deposit
(
Material = Tantalum
Thickness = 0.1um
)
#Save(MeshFile = SiGe_50nm_nosti04.tdf)
# Define the poly gate.
Etch
(
Material = Tantalum
EtchType = dry
MaskPolygon
(
Point(z=-1um
Point(z=-1um
Point(z= 1um
Point(z= 1um
)
)
x=expr($sp))
x=expr(($Lg)+ ($sp)))
x=expr(($Lg)+ ($sp)))
x=expr((($sp))))
120
#recess etch
#Etch
#(
#
#
#
#
#
#
#
#
#
#
#
Material = oxide
Material = silicon
EtchType = dry
Thickness = 700A
MaskPolygon
(
Point(z=-1um
Point(z=-1um
Point(z= 1um
Point(z= 1um
)
x=expr($sp))
x=expr(($Lg)+ ($sp)))
x=expr(($Lg)+ ($sp)))
x=expr((($sp))))
#)
#Save(MeshFile = SiGe_50nm_nosti05.tdf)
#Deposit(
#
AnisotropyFactor = 5
#
MismatchStrain = True
#
Material=silicon
#
SurfacePosition = -0.006um
#
#25% Ge concentration
#
Initialize (
#
Name=Germanium
#
Value=1.25e22
#
)
#)
# silicon atomic density=5e22
#Save(MeshFile = SiGe_50nm_nosti06.tdf)
# Do the LDD implant.
#Implant
#(
#
Name = Boron
#
Energy = 5
#
Dose = 2.5e14
#
Tilt = 0
#
Rotation = 0
#)
# Liner Oxide Deposition (10nm)
#Deposit (
#
Material=oxide
#
Thickness=10nm
#
)
# Create the nitride spacer.
#Deposit
#(
#
Material = nitride
#
Thickness = 0.04um
#)
121
#Save(MeshFile = unstrained_400nm_07.tdf)
#Etch
#(
#
Material = nitride
#
Thickness = 0.08um
#
EtchType = dry
#)
#Etch (
#
Material=oxide
#
EtchType=dry
#
Thickness= 10nm
#
)
#Save(MeshFile = SiGe_50nm_nosti08.tdf)
# Perform the high dose arsenic implant.
#Implant
#(
#
Name = Boron
#
Energy = 4.5
#
Dose = 1.0e15
#
Tilt = 0
#
Rotation = 0
#
NoRegrid
#)
#Save(MeshFile = SiGe_50nm_nosti09.tdf)
#Do the RTA.
#Diffuse
#(
#
Time = 5s
#
Temperature = 1000C
#
NoRegrid
#)
#Save (MeshFile="SiGe_50nm_nosti10.tdf")
#Reflect in the symmetry plane to form the full structure.
define(a=0)
while ($a[...]... LIST OF SYMBOLS AND ABBREVIATIONS ABBREVIATIONS DESCRIPTION/ EXPANSION GaN Gallium Nitride AlGaN Aluminium Gallium Nitride TaN Tantalum Nitride HFET Heterojunction Field Effect Transistor MODFET Modulation Doped Field Effect Transistor HEMT High Electron Mobility Transistor MOSFET Metal- Oxide- Semiconductor Field Effect Transistor MESFET Metal- Semiconductor Field Effect Transistor 2-DEG 2-Dimensional Electron. .. there is a substantial gate leakage current present in such a structure Unlike the MOSFET structure, the absence of a gate dielectric would be detrimental to the gate current leakage levels Comparing this gate leakage current to an ideal Schottky gate device shows that in the AlGaN/GaN HEMT device, the Schottky gate leakage current is very much higher than that of an ideal Schottky gate reverse current... of the 2DEG without any doping involved The polarisation effect and the operation of the HEMT device will be further elaborated in the next chapter 1.2 Scope and Purpose Having discussed the advantages of GaN semiconductor material as being a suitable candidate for high power and high temperature purposes, a GaN -based HEMT would thus be the choice candidate for electronic devices capable of withstanding... (W/cm K) Dielectric constant ε CFOM = (χ ε µ vsat EB) / ( χ ε µ vsat EB)Si CFOM: Combined figure of merit for high temperature /high power /high frequency applications GaN with a large bandgap (Eg = 3.40 eV), large critical breakdown field of 4.0 MV cm-1, coupled with good electron transport properties (theoretical electron mobility, µe, of up to 2000 cm2 V-1s-1 [9] and a peak saturation velocity, vsat, of. .. performance enhancement of GaN -based HEMTs for applications in highpower electronics Simulation work was performed to understand the effect of different parameters on the electrical performance of the device Effect of various parameters such as barrier layer thickness, concentration and energy level of interface traps as well as thermal effects were investigated The simulation of the effect of stress due to... Effect Transistor (MOSFET) and Metal- Semiconductor Field Effect Transistor (MESFET) However, due to the immaturity of the technology of material growth [8], it is still a challenge to make high quality epitaxial layer on various substrates Thus GaN layers often come with high density of defects, primarily due to the difference in lattice constant of GaN and the substrate it is grown on This makes surface... slight increase in electron density is observed outside the gate region 90 XIII Figure 5.9 A plot of the simulated electron mobility value along the x-axis of the MOSHEMT with and without stress, at a depth of 1 nm below the AlGaN/GaN interface 91 Figure 5.10 A plot of the simulated vertical electric field along the x-direction of the MOSHEMT with and without stress, at a depth of 1nm below the AlGaN/GaN... MOSHEMT at gate region extracted from the simulation 76 Figure 4.18 The ID-VGS plot of devices with different concentration of acceptor-like trap 78 Figure 4.19 The ID-VDS plot of devices with different concentration of acceptor-like traps 79 XII Figure 5.1 The schematic view of the AlGaN/GaN MOSHEMT with a compressive Diamond-Like Carbon (DLC) liner The 2-DEG density decreases with application of compressive... Matching Eliminate/Reduce need for High Voltage High Breakdown Field voltage conversion Operation Bandwidth, µ-wave/mmHigh Frequency High Electron Velocity wave High dynamic range Low Noise High Gain, High Velocity receivers High Temperature Wide Bandgap Rugged, Reliable Operations Direct Bandgap Technology Leverage Enabler for Lighting carrier transport or a high breakdown voltage Table II presents the... effect transistor (HFET) as a general class of devices Also known as the High Electron Mobility Transistor (HEMT), the ability to induce a conductive 2 DEG in intrinsic GaN heterostructure layers of the device has ignited interest of many people In the rest of this thesis, we will be working on the AlGaN/GaN HEMT structure that is currently the de facto norm for GaN -based HEMTs The High Electron Mobility ... Metal-Oxide-Semiconductor High Electron Mobility Transistor 20 Figure 3.1 a) A typical transistor design for the purpose of DC characterisation W, L and LD-G referred to the gate width, gate length and drain -gate. .. suitable candidate for high power and high temperature purposes, a GaN-based HEMT would thus be the choice candidate for electronic devices capable of withstanding high power and high temperature However,... purpose of DC characterisation W, L and LD-G referred to the gate width, gate length and drain -gate separation respectively b) The width of a mesa island would dictate the gate width of a transistor