Sunteți pe pagina 1din 155

UNIVERSITY OF CINCINNATI

3/1/2007
Date:___________________
Joseph Meiners
I, _________________________________________________________,
hereby submit this work as part of the requirements for the degree of:

Master of Science
in:

Electrical Engineering
It is entitled:
Simulation and Design of a Submicron Gate-Length AlGaN/GaN

HEMT

This work and its defense approved by:


Dr. Kenneth P. Roenker
Chair: _______________________________
Dr. Marc Cahay
_______________________________
Dr. David J. Klotzkin
_______________________________

_______________________________
_______________________________

Simulation and Design of a


Submicron Gate-Length AlGaN/GaN HEMT
A Thesis submitted to the
Division of Research and Advanced Studies of
the University of Cincinnati
in partial fulfillment of the
requirement for the degree of
MASTER OF SCIENCE
in the Department of Electrical and Computer Engineering
of the College of Engineering
March 2007
by
Joseph Meiners
Bachelor of Science (Electrical Engineering), College of Engineering,
University of Cincinnati, Cincinnati, Ohio, 2003
Thesis Advisor and Committee Chair: Dr. Kenneth P. Roenker
Committee Members: Dr. Marc Cahay, Dr. David J. Klotzkin

Abstract
Solid state semiconductor devices have received much attention in recent years
for microwave power amplification where vacuum tube devices have dominated the
market for high power, high-frequency operations. AlGaN/GaN High Electron Mobility
Transistors (HEMTs) are a promising candidate for high frequency, high power
microwave power amplification as the device offers high breakdown voltages (> 200 V)
and high frequency operation (> 100 GHz). Recent devices have been able to achieve
power densities as high as 10.5 W/mm at 40 GHz, cutoff frequencies (fT) of 163 GHz and
maximum oscillation frequencies (fmax) of 230 GHz. The AlGaN/GaN HEMT is able to
operate at high power densities due to the high breakdown voltages while the high current
densities of the device are attributed to the polarization charge created at the AlGaN/GaN
heterointerface.
This device modeling study examines the dc and microwave performance through
small-signal analysis of the AlGaN/GaN HEMT by employing the commercially
available numerical simulator DESSIS from Synopsis. Polarization charges present in
AlGaN/GaN HEMTs have been incorporated as sheet charges at the AlGaN/GaN
heterointerface and AlGaN surface. Donor traps were placed at the AlGaN surface as the
source of electrons for the two-dimensional electron gas (2DEG) and are thought to be
responsible for the current collapse observed in experimental devices. This modeling
study also investigates the devices performance when the device structure and
polarization charges are varied. Included in this investigation is the examination of the
AlGaN/GaN HEMTs microwave performance, specifically the peak transconductance,
cutoff frequency and maximum oscillation frequency.

Acknowledgments
I would like to thank the faculty, staff and students of the Department of
Electrical and Computer Engineering at the University of Cincinnati for the help I
received in completing my thesis. I would like to make particular mention of my advisor,
Dr. Kenneth P. Roenker who I thank for introducing my thesis topic and guiding me
through my work. His availability for advice and input has allowed me to learn about
and complete this work. I would also like to thank him for his time and effort in helping
me with my thesis work. Thanks are also due to Dr. Marc Cahay and Dr. David J.
Klotzkin for agreeing and taking the time to serve on my thesis defense committee as
well as for answering any technical questions I had during my stay here.
I would also like to thank Dr. Punit Boolchand for talks we had on the
polarization in the AlGaN and GaN heterostructure. I would also like to thank my former
and present colleagues in the Semiconductor Devices Laboratory and the Device
Simulation Laboratory particularly Yajun An, Aravind Appaswamy, Martin Arnold,
Aniket Breed, Saumitra Mehrotra, Ramanujan Sampathkumaran, Yuba Shrestha, and
Subramaniam Venkatesan. I have learned much from them and have enjoyed my time
working with them. I would also like to thank Kalyan Garre and Junjun Wan for their
various help. I would like to express my gratitude to John Phillips who I worked with
during my stay and also for the discussions we had. Thanks are also due to Chris Isbell
and Rob Montjoy for their time in helping me out with software usage and network
issues. Lastly, I would like to thank my parents and brothers for their support while I was
working on my thesis.

Table of Contents
Chapter 1

Introduction

1.1

General Introduction

1.2

AlGaN/GaN High Electron Mobility Transistor (HEMT)

1.3

Reported Results for AlGaN/GaN HEMTs

13

1.4

Focus of This Thesis

33

1.5

Description of Following Chapters

34

Chapter 2

Introduction

46

2.1

Device Simulation Software

47

2.1.1

Device Structure 2D Structure Editor

47

2.1.2

Device Simulator

48

2.2

2.3

2.1.2.1

Poissons Equation

49

2.1.2.2

Continuity Equation

49

Device Simulation Models

50

2.2.1

Carrier Statistics Model

50

2.2.2

Transport Models

51

2.2.3

Recombination Models

53

2.2.4

Band Structure

55

2.2.5

Mobility Models

57

2.2.5.1

Low-Field Mobility Model

58

2.2.5.2

High-Field Saturation Model

59

Polarization Incorporated into Device


i

61

2.4

Material Parameters Used in Study

66

Chapter 3

Device Modeling of 0.14 m Gate-Length AlGaN/GaN

70

Power HEMT
3.1

Device Structure

70

3.2

DC Device Characteristics

80

3.2.1

Drain Current Gate Voltage Characteristics

80

3.2.2

Transistor Output Characteristics

83

Microwave/Small-Signal Characteristics

88

3.3.1

Small-Signal Transconductance

88

3.3.2

Cutoff Frequency (fT)

90

3.3.3

Maximum Oscillation Frequency (fmax)

93

3.3

3.4

Summary and Conclusion

Chapter 4

93

Device Optimization

96

4.1

Polarization Charge Density

96

4.2

Donor Trap Density at the AlGaN Surface

103

4.3

Contact Resistance

109

4.4

AlGaN Thickness

114

4.5

Source-to-Gate Spacing

119

4.6

Discussion and Conclusions

125

ii

Chapter 5

Conclusions and Future Work

130

5.1

Conclusions

130

5.2

Future Work

132

iii

List of Figures
Figure 1.1

RF vacuum device and solid state device performance showing output


power versus frequency [5].

Figure 1.2

Schematic cross-section of an AlGaN/GaN High Electron Mobility


Transistor (HEMT).

Figure 1.3

Energy band diagram under the gate of an undoped AlGaN/GaN HEMT.

Figure 1.4

Crystal structure of wurtzite Ga-face and N-face GaN grown along in the
c-direction [22].

Figure 1.5

Diagrams showing the polarization induced sheet charge density and the
direction of spontaneous and piezoelectric polarization in Ga-face and Nface for strained and relaxed AlGaN/GaN heterostructures. Case (b) is the
typical growth structure in most AlGaN/GaN HEMTs [22].

Figure 1.6

Power sweep on a MOCVD Si3N4 passivated AlGaN/GaN HEMT where


record power density of 10.5 W/mm was reported at 40 GHz [24].

Figure 1.7

Comparison of gate leakage current for fluorine-plasma surface treatment


and different epitaxial structures showing improved reduction in gate
leakage for unpassivated GaN HEMTs where record results in power
density were reported without field plates [25].

Figure 1.8

Power characteristics of a MOCVD-grown unpassivated GaN HEMT with


a record power density of 17.8 W/mm with a PAE of 50% at 4 GHz [25].

Figure 1.9

IDS v. VDS output characteristics of an enhanced-mode AlGaN/GaN


HEMT with record reported current for enhanced-mode [26].

iv

Figure 1.10

a) Electron mobility, b) sheet electron density, and c) sheet resistance for


varying AlGaN thickness and with/without SiN passivation for the
AlGaN/GaN HEMT with reported 163 GHz cutoff frequency [29].

Figure 1.11

a) Cutoff frequency of 163 GHz extracted from current gain versus


frequency for an AlGaN/GaN HEMT with a gate length of 0.06 m and b)
maximum stable gain/maximum available gain (MSG/MAG) and
unilateral gain versus frequency for an AlGaN/GaN HEMT with a gate
length of 0.08 m [29].

Figure 1.12

Conduction band schematic of AlGaN/GaN HEMT with introduction of


InGaN back-barrier to improve fmax performance [30].

Figure 1.13

Small-signal characteristics of AlGaN/GaN HEMT with InGaN backbarrier with bias set to achieve for maximum fT and fmax [30].

Figure 1.14

Conduction band profile for MODFET structure in the [0001] direction for
VG = 0 V with and without polarization incorporated into the selfconsistent solution of the Schrdinger and Poisson equations coupled to a
quasi-2-D model for the current flow [49].

Figure 1.15

Calculated electron density distribution in the channel with several Al


concentrations for the AlxGa1-xN layer of the MODFET [49].

Figure 1.16

Calculated IDS v. VDS for an AlGaN/GaN MODFET with an Al


concentration of

Figure 1.17

a) 0.2 b) 0.4 [49].

IDS v. VDS characteristics for simulated and measured AlGaN/InGaN/GaN


HEMT. The simulation results using the hydrodynamic transport model
(solid lines) and the drift-diffusion transport model (dashed lines) for VG =

-2, 0 V are compared with the measured results (square markers) for VG =
-8, -2, -1, 0 V [50].
Figure 1.18

Cross-section of AlGaN/InGaN/GaN HEMT showing the electron


concentration at VD = 10 V and VG = 0 V when incorporating the a)
hydrodynamic transport model where the electrons are spreading away
from the channel and into the AlGaN and GaN layers and b) driftdiffusion model where the electrons are mostly confined in the channel
[50].

Figure 1.19

Distribution of trapped electrons by using the hydrodynamic model to


predict

the

occupied

bulk

traps

present

in

the

simulated

AlGaN/InGaN/GaN HEMT at VD = 10 V and VG = 0 V [50].


Figure 1.20

Cross-sections of the AlGaN/InGaN/GaN HEMT in the region of the gate


edge near the drain side of the device that show the conduction band
contours at a) VD = 2.4 V and b) VD = 6 V which are highlighted in c) as
the peak and valley of the ID v. VD curve. d) shows the 3-D surface plot of
the conduction band with the barrier located at the gate edge near the drain
highlighted.

This increased barrier is associated with the negative

differential conductance in the ID v. VD curve [50].


Figure 1.21

The effect of the drain current collapse on the experimental power output
of regular and field-plated III-Nitride HFETs and comparison to
theoretical output [56].

Figure 1.22

Measured drain current with varying rf input drive, the decreasing current
is the observed drain current collapse. The measurements were performed

vi

at a frequency of 4 GHz and VD = 20 V for ZL = 0 and VG = -3 V (circle)


and ZL = 38 + j39 and VG = -3 V (triangles) [52].
Figure 1.23

Peak normalized drain current transients under various drain voltage and
with and without surface passivation. The large reduction of the current
collapse in the passivated devices compared to the unpassivated devices
proves that the current collapse is related to surface states [53].

Figure 1.24

XPS measurements of plasma-treated AlGaN surfaces for a) Ga 3d and b)


N 1s. The groups results conclude that the surface states are 0.37 eVdeep donor N-vacancies [54].

Figure 1.25

a) IDS v. VDS representation of the drain current collapse and models for
current collapse b) under drain stress and c) under gate stress [54].

Figure 2.1

Conduction band diagram of an AlGaN/GaN HEMT showing the various


space charge components modeled into the device simulation from the
modified charge balance equation in [10]. The electrons in the 2DEG are
generated by the surface state charge and polarization charges as in (2.36).

Figure 3.1

AlGaN/GaN HEMT device structure used in this modeling following that


reported by Jessen et al. [1].

Figure 3.2

Device structure used for simulation of 0.14 m gate-length AlGaN/GaN


HEMT [1].

Figure 3.3

Device mesh used in dc simulations of AlGaN/GaN HEMT [1].

Figure 3.4

Detailed look at the spacing of the mesh under the gate region for the
simulated AlGaN/GaN HEMT [1]. The vertical spacing in the channel is
4.69 nm.

vii

Figure 3.5

Simulated band energy structure under the gate for a bias of 0 V on the
gate and 0 V on the drain, with a) showing the whole structure and b)
showing an expanded view of the band structure around the
heterostructure. The Si3N4 thickness is 100 Angstroms and starts at 0 m
on the plot and the AlGaN thickness is 260 Angstroms, with the
AlGaN/GaN interface located at 0.036 m.

Figure 3.6

Simulated conduction band profiles under the gate at various gate voltages
for no drain bias.

Figure 3.7

Simulated conduction band profiles under the gate at various gate voltages
for a drain bias of 10 V.

Figure 3.8

Simulated electron concentration profile under the gate for a bias of 0 V at


the gate for 0 V at the drain.

Figure 3.9

Simulated electron concentration profile under the gate for various gate
biases at a drain bias of 0 V.

Figure 3.10

Simulated electron concentration profiles under the gate for various gate
biases at a drain bias of 10 V.

Figure 3.11

Simulation results for the drain current plotted against voltage on linear
and logarithmic scales where the drain voltage was +10 V.

Figure 3.12

Experimental transfer characteristics of 0.14 m gate-length AlGaN/GaN


HEMTfor a drain bias of 10 V as reported by Jessen et al. [1].

Figure 3.13

Comparison of simulated and experimental drain current plotted against


gate voltage for drain biased at +10 V using the experimental results of
Jessen et al. [1].

viii

Figure 3.14

Simulated dc drain current versus drain voltage results where the gate
voltage was swept from -7 V to +1 V in 2 V increments.

Figure 3.15

Experimental drain current versus drain-source voltage of 0.14 um gatelength AlGaN/GaN HEMT as reported by Jessen et al. [1] where the gate
voltage was swept from -7 V to +1 V in 2 V steps.

Figure 3.16

Simulation results for the drain current plotted with the experimental
results from Jessen et al. [1] where the gate voltage was swept from -7 to
+1 V in 2 V increments.

Figure 3.17

Simulated small-signal transconductance (gm) and small-signal output


conductance (go) of the AlGaN/GaN HEMT. The frequency of the small
signal was set to 100 MHz.

Figure 3.18

Simulated current gain and unilateral power gain plotted against frequency
where the gate is biased at -3.5 V and the drain is biased at 10 V.

Figure 3.19

Simulated cutoff frequency (fT) and maximum oscillation frequency (fmax)


plotted as functions of the gate voltage where the drain voltage was kept at
10 V.

Figure 4.1

Threshold Voltage (VT, ) and the maximum drain current () at a bias of


VGS = +1 V, VDS = +10 V plotted against various polarization charge
densities at the AlGaN/GaN heterointerface and AlGaN surface. The
initial polarization charge density of 1.14x1013 cm-2 from the simulation
modeling in Chapter 3 is also plotted here and is identified by the arrow.

Figure 4.2

Plots of a) transconductance (gm) versus gate voltage for different


polarization charge densities and b) peak gm as a function of polarization

ix

charge density. The data for the polarization charge density of 1.14x1013
cm-2 is from the modeling results reported in Chapter 3 and identified by
the arrow in (b).
Figure 4.3

Cutoff frequency (fT) plotted as a function of the gate voltage for various
polarization charge densities at a drain bias of +10 V. The data for the
polarization charge density of 1.14x1013 cm-2 is from the modeling results
reported in Chapter 3.

Figure 4.4

Maximum oscillation frequency (fmax) plotted as a function of the gate


voltage for various polarization charge densities at a drain bias of +10 V.
The data for the polarization charge density of 1.14x1013 cm-2 is from the
modeling results reported in Chapter 3.

Figure 4.5

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax)
plotted as a function of polarization charge density for various densities.
The data for the polarization charge density of 1.14x1013 cm-2 (marked by
the arrow) is from the modeling results reported in Chapter 3.

Figure 4.6

Threshold Voltage (VT, ) and the maximum drain current () at a bias of


VGS = +1 V, VDS = +10 V plotted against various donor trap densities at
the AlGaN surface in contact with the Si3N4 layer. The data for the donor
trap density of 1.14x1013 cm-2 (identified by arrow) is from the modeling
results reported in Chapter 3 and is included in the plot.

Figure 4.7

Plots of a) transconductance (gm) versus gate voltage for different donor


trap densities and b) peak gm as a function of donor trap density. The data

for the donor trap density of 1.14x1013 cm-2 (identified by arrow) is from
the modeling results reported in Chapter 3 and is included in the plot.
Figure 4.8

Cutoff frequency (fT) plotted as a function of the gate voltage for various
donor trap densities at a drain bias of +10 V. The data for the donor trap
density of 1.14x1013 cm-2 is from the modeling results reported in Chapter
3 and is included in the plot.

Figure 4.9

Maximum oscillation frequency (fmax) plotted as a function of the gate


voltage for various donor trap densities at a drain bias of +10 V. The data
for the donor trap density of 1.14x1013 cm-2 is from the modeling results
reported in Chapter 3 and is included in the plot.

Figure 4.10

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax)
plotted as a function of the donor trap density for various densities. The
data for the donor trap density of 1.14x1013 cm-2 (indicated by arrow) is
from the modeling results reported in Chapter 3 and is included in the plot.

Figure 4.11

Threshold Voltage (VT, ) and the maximum drain current () at a bias of


VGS = +1 V, VDS = +10 V plotted as a function of the contact resistance.
The original device has a contact resistance of 1.46 -mm (indicated by
arrow); the data for this contact resistance is included in the plot.

Figure 4.12

Plots of a) transconductance (gm) versus gate voltage and b) peak gm as a


function of the device contact resistance.

The original device has a

contact resistance of 1.46 -mm (indicated by arrow); the data for this
contact resistance is included in the plot.

xi

Figure 4.13

Cutoff frequency (fT) plotted as a function of the gate voltage for various
contact resistances. The original device structure has a contact resistance
of 1.46 -mm whose cutoff frequency is included in the plot.

Figure 4.14

Maximum oscillation frequency (fmax) plotted as a function of the gate


voltage for various contact resistances. The original device structure has a
contact resistance of 1.46 -mm whose maximum oscillation frequency is
included in the plot.

Figure 4.15

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax)
plotted as a function of the device contact resistance. The original device
has a contact resistance of 1.46 -mm (indicated by arrow); the data for
this contact resistance is included in the plot.

Figure 4.16

Threshold Voltage (VT, ) and the maximum drain current () at a bias of


VGS = +1 V, VDS = +10 V plotted as a function of the AlGaN thickness.
The original device structure has an AlGaN thickness of 260 Angstroms
(indicated by arrow); the data for this device structure is included in the
plot.

Figure 4.17

Plots of a) transconductance (gm) versus gate voltage for various AlGaN


thickness and b) peak gm as a function of the AlGaN thickness. The
original device structure has an AlGaN thickness of 260 Angstroms
(indicated by arrow); the data for this device is included in the plot.

Figure 4.18

Peak electron concentration (ns) measured at the peak gm voltage and


plotted as a function of the AlGaN thickness.

xii

The original device

structure has an AlGaN thickness of 260 Angstroms (indicated by arrow);


the data for this device is included in the plot.
Figure 4.19

Cutoff frequency (fT) plotted as a function of the gate voltage for various
AlGaN thickness. The original device structure has an AlGaN thickness
of 260 Angstroms whose cutoff frequency is included in the plot.

Figure 4.20

Maximum oscillation frequency (fmax) plotted as a function of the gate


voltage for various AlGaN thickness. The original device structure has an
AlGaN thickness of 260 Angstroms whose maximum oscillation
frequency is included in the plot.

Figure 4.21

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax)
plotted as a function of the AlGaN thickness.

The original device

structure has an AlGaN thickness of 260 Angstroms (indicated by arrow);


the data for this device structure is included in the plot.
Figure 4.22

Threshold Voltage (VT, ) and the maximum drain current () at a bias of


VGS = +1 V, VDS = +10 V plotted as a function of the source-to-gate
spacing of the device structure. The original device structure has a sourceto-gate spacing of 1.3 microns (indicated by arrow); the data for this
device structure is included in the plot.

Figure 4.23

Plots of a) transconductance (gm) versus gate voltage for different sourceto-gate spacing and b) peak gm as a function of the source-to-gate spacing
of the device structure. The original device structure has a source-to-gate
spacing of 1.3 microns (indicated by arrow); the data for this device is
included in the plot.

xiii

Figure 4.24

Cutoff frequency (fT) plotted as a function of the gate voltage for various
source-to-gate spacing. The original device structure has a source-to-gate
spacing of 1.3 microns whose cutoff frequency is included in the plot.

Figure 4.25

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax)
plotted as a function of the source-to-gate spacing. The original device
structure has a source-to-gate spacing of 1.3 microns (indicated by arrow);
the data for this device structure is included in the plot.

Figure 4.26

Maximum oscillation frequency (fmax) plotted as a function of the gate


voltage for various source-to-gate spacing. The original device structure
has a source-to-gate spacing of 1.3 microns whose maximum oscillation
frequency is included in the plot.

xiv

List of Tables
Table 1.1

Comparison of Competing Materials used as Microwave Devices for


Power Amplification [7]

Table 1.2

Table of Material Properties and Figures of Merit of Competing Materials


for Consideration in Microwave Applications [9]

Table 1.3

Table of Competitive Advantages of GaN Devices for Power


Amplification [9]

Table 2.1

Shockley-Read-Hall Parameters for AlN and GaN [2]

Table 2.2

Parameters for Bandstructure [2]

Table 2.3

Effective Densities of States at 300 K for AlN and GaN [2, 3]

Table 2.4

AlN and GaN Parameters for Constant Mobility Model at Low Fields [2]

Table 2.5

AlN and GaN Values for Parameter

Table 2.6

Velocity Saturation Model Parameters for AlN and GaN [2]

Table 2.7

Description of Polarization and Surface State Charge Components in


Simulated AlGaN/GaN HEMT

Table 2.8

Parameter Values for AlN, GaN, Al0.34Ga0.66N at 300 K used in this study

Table 3.1

Important Lateral Dimensions of Modeled AlGaN/GaN HEMT following


Jessen et al. [1]

Table 3.2

Specifics of the Epitaxial Layers of the Modeled AlGaN/GaN HEMT [1]


after Jessen et al [1]

Table 3.3

Trap Specifications Used in Modeling of AlGaN/GaN HEMT after Jessen


et al. [1]

xv

Chapter 1

Introduction

1.1 General Introduction


Millimeter and microwave power amplifiers have come into focus in recent years
as solid state devices are trying to find a place in the market, where the market for these
amplifiers includes the military, commercial and consumer sectors.

These devices

include MESFETs, HEMTs and HBTs in GaAs, InP, SiC and GaN. In the military
sector, millimeter and microwave power amplifiers are used in radar systems,
communications, smart weapon systems, as well as in transmitters for small targets and
target imaging.

The commercial sector uses these amplifiers in high-data-rate

communication systems while the consumer market is looking at applications such as


automobile collision radar [1].
Vacuum tube devices have long dominated the market for high power amplifiers
at millimeter-wave frequencies.

However solid-state devices are a more attractive

alternative if better performance can be achieved that vacuum tube devices provide at a
comparable cost. Solid state devices are ultimately expected to be smaller, cheaper, more
robust, and reliable. Solid state devices could also lower system cost and possibly
increase the functionality of the system cost. To be considered as a replacement for
vacuum tube devices in power amplifiers, solid state devices must be able match the
performance of vacuum tube devices and ideally outperform them, and be cost
competitive.
Some of the device performance requirements for microwave power amplification
are high unity power gain frequency (fmax), minimized low and high frequency noise
1

(broad-band), a high signal-to-noise ratio (SNR), and the ability to operate over a wide
temperature range [2]. Also important in power amplification devices is high linearity
(low signal distortion), the specifications of which are increasing to avoid spillover into
adjacent channels of the spectrum in communications applications [3].

Another

important requirement is high efficiency which allows the transistor to convert DC


electrical power into high frequency output power.

As much as 90% of power

consumption is lost to heat in current amplifiers and designers would like to improve
upon the current 10% efficiency as a way to cut costs in cooling solutions [3].
Solid state devices have been able to make gains in recent years in microwave
systems. However, one area solid state devices have yet to replace vacuum tubes is high
frequency, high power applications (see Figure 1.1).

This area requires devices to

produce high radio-frequency (RF) power at frequencies over 100 GHz at temperatures
greater than 250 C.

Currently, this area is dominated by systems designed with

microwave tubes, such as klystrons and gyrotrons, and is able to deliver power levels in
the kilo- to megawatt range [4].
Current state of the art solid-state devices produce rf power levels below 100 W
and can reach frequencies around 100 GHz as shown in Figure 1.1 [5]. Vacuum tubes
devices currently offer better performance in high power (>100 W) and frequencies
above 100 GHz as illustrated in the figure.

Solid state power amplifiers can be

constructed with devices such as GaAs and InP HEMTs, but they can only operate at low
power and low voltage due in part to the small energy bandgap in these semiconductors,
1.42 eV and 1.35 eV, respectively.

Silicon Carbide based MESFETs have been

developed in recent years and show some promise. In 2004, Henry et al. [6] reported a

SiC MESFET with 20 W (4.4 W/mm) output power and 60% PAE for operation in the Sband. These power amplifiers can be combined to produce the power levels required as a
way to overcome the limitations of current solid state devices. However, by combining
multiple power amplifiers, the efficiency of the system is lowered due to the complexity.
In the last five years, another semiconductor Gallium Nitride (GaN) has been developed
that shows considerable promise due to its large energy bandgap (3.5 eV) and other
attractive properties. This thesis examines the promise and the potential of GaN HEMTs
that promise high power at high voltages with very good power efficiency.

Figure 1.1

RF vacuum device and solid state device performance showing output power versus
frequency [5].

1.2 AlGaN/GaN High Electron Mobility Transistor (HEMT)


The advantages of GaN in comparison to other materials for use in high
frequency, and high power applications are summarized in Table 1.1 [7].

Table 1.1

Comparison of Competing Materials used as Microwave Devices for Power


Amplification [7]

Characteristics

Silicon

(Units in

GaAs

InP

(AlGaAs/InGaAs)

(InAlAs/InGaAs)

SiC

GaN
(AlGaN/GaN)

Parentheses)
Bandgap (eV)

1.1

1.42

1.35

3.26

3.49

Electron

1500

8500

5400

700

1000-2000

Vsat (x107 cm/s)

1.0

1.3

1.0

2.0

1.3

Breakdown Field

0.3

0.4

0.5

3.0

3.0

1.5

0.5

0.7

4.5

> 1.5

11.8

12.8

12.5

10.0

9.0

Mobility (cm2/Vs)

(MV/cm)
Thermal
Conductivity
(W/cm-K)
Relative
Dielectric
Constant

The large bandgap of GaN allows it withstand higher electric fields which
translates into the material supporting a larger breakdown field, which is desirable for
high power operation. GaN has a large thermal conductivity, only out performed by SiC,
which allows much lower temperature rise in power HEMTs due to self heating. The
high saturation velocity of GaN is useful in allowing high frequency device operation,
which compensates for the lower electron mobility of GaN in comparison to competing

materials. It should be noted that although GaN has lower electron mobility, the material
presents very high electron mobility in an AlGaN/GaN HEMT design [8].
Key material advantages of GaN are shown in Table 1.2 [9] in comparison with
other semiconductors, along with several figures of merit used to determine the
usefulness of the material for high frequency, high power applications.
Table 1.2

Table of Material Properties and Figures of Merit of Competing Materials for


Consideration in Microwave Applications [9]

Material

(cm2/V-s)

Bandgap

BFOM

(eV)

Ratioa

JFM Ratiob

Tmax (C)

Si

1300

11.4

1.1

1.0

1.0

300

GaAs

5000

13.1

1.4

9.6

3.5

300

SiC

260

9.7

2.9

3.1

60

600

GaN

1500

9.5

3.4

24.6

80

700

a. Baliga Figure of Merit (s**ECR3)


b. Johnson Figure of Merit (ECR*vsat/2*)
The electron mobility of GaN is seen to be lower than some competing materials,
but is compensated by a higher saturation velocity that is useful in device design
considerations for AlGaN/GaN HEMTs. A higher maximum operating temperature is
possible in GaN based devices, which is an advantage where cooling equipment can be
eliminated or requirements reduced, such as when replacing a device based on another
material with a lower maximum operating temperature. The Johnson and Baliga figures
of merit listed in the table are combinations of important material parameters that are
used to rate the potential performance of materials in high frequency, high power
applications. The larger the number, the more useful the material is for microwave

applications. For high frequency, high power applications, GaN offers far better figures
of merit than Si, GaAs and even SiC.
The performance advantages of GaN HEMT devices for high frequency power
amplification and the feature that enables that performance advantage are summarized in
Table 1.3 [9].
Table 1.3

Table of Competitive Advantages of GaN Devices for Power Amplification [9]

Performance Need

Enabling Feature

Performance Advantage

High Power/Unit Width

Wide Bandgap, High Field

Compact, Ease of Matching

High Voltage Operation

High Breakdown Field

Eliminate/Reduce Step Down

High Linearity

HEMT Topology

Optimum Band Allocation

High Frequency

High Electron Velocity

Bandwidth, -Wave/mm-Wave

High Efficiency

High Operating Voltage

Power Saving, Reduced Cooling

Low Noise

High Gain, High Velocity

High dynamic range receivers

High Temperature Operation

Wide Bandgap

Rugged, Reliable, Reduced


Cooling

Thermal Management

SiC Substrate

High Power Devices with


Reduced Cooling Needs

Technology Leverage

Direct Bandgap, Enabler for

Driving Force for Technology,

Lighting

Low Cost

A high power per unit channel width allows for smaller devices with higher
impedance, which offers better matching of the device to a system. Higher voltage
operation can eliminate or reduce voltage step-down required. If an application is using
low-voltage devices, but is powered by a voltage outside the device range, then step
down circuitry is needed to drop the supply voltage. GaN offers the ability to use the
6

high voltage or at the very least reduce the circuitry needed to step down the voltage.
The high linearity is something GaN can take advantage of in the HEMT topology. High
linearity across a wide frequency range allows for better frequency band utilization.
Dingle [10] is credited with the early development of the AlGaAs/GaAs HEMT
due to his work on studying the AlGaAs/GaAs heterostructure.

The discovery of

modulation-doping in AlGaAs/GaAs heterostructures [11] propelled other groups to work


on the heterostructure as the basis for a field-effect device. At the same time Dingle was
working on developing the HEMT, a group at Fujitsu [12] working independently
reported on the first AlGaAs/GaAs HEMT.

The importance of this early work on

modulation doping using heterostructures resulted in Klaus von Klitzing [13] receiving
the Nobel Prize in 1985 due to his work in discovering the quantized Hall Effect in twodimensional electron gases (2DEGs). In recent years, the development of AlGaN/GaN
material systems resulted in the first AlGaN/GaN HEMT being developed by Asif Khan
et al. [14] in 1994 out of previous work done modeling the electron transport in GaN by
Monte Carlo simulations [15, 16].
A typical AlGaN/GaN HEMT device structure is shown in Figure 1.2 where the
HEMT structure and operation is similar to a field effect transistor (FET) in that the
current flows from the source to the drain and is controlled by a gate. A two-dimensional
electron gas (2DEG) is formed at the AlGaN interface in the GaN to form the channel for
the device using the discontinuity of the conduction band. The device structure is very
similar to that of the well established AlGaAs/GaAs and InP/InGaAs HEMTs. The
superior performance at high frequency, high power of the AlGaN/GaN HEMT arises
from the better material properties, e.g. large bandgap and breakdown field, of GaN

compared to GaAs and InGaAs. The common substrates used for an AlGaN/GaN HEMT
are Sapphire, SiC and recently Silicon [17, 18, 19]. Researchers have yet to find a way to
grow a pure enough boule of GaN for it to be considered for use as a substrate. However,
some groups have been working on ways to make pure GaN wafers [20]. For the GaN
HEMT, ohmic contacts are formed for the source and drain and a Schottky metal gate is
formed at the AlGaN layer surface to form the devices gate electrode.

Source

Gate

Drain

AlGaN

2DEG

GaN
Substrate

Figure 1.2

Schematic cross-section of an AlGaN/GaN High Electron Mobility Transistor


(HEMT).

The low field mobility of a material in a FET is limited by the scattering of


electrons in the channel by doping impurities and lattice vibrations.

This mobility

limitation can be avoided by using a structure that is termed a modulation-doped


heterostructure. In the HEMT, the heterostructure physically allows separating of the
dopant impurities, which lead to carrier scattering, from the channel. To do this in an
AlGaN/GaN HEMT, an n-type, silicon doped AlGaN layer is grown over an undoped
8

GaN layer. The silicon impurities present in the AlGaN donate electrons which are
attracted to a region of lower potential energy in the GaN. Since the silicon doped
AlGaN has a higher band gap energy than the undoped GaN, the electrons accumulate in
a region with lower potential called a quantum well which is just below the AlGaN/GaN
heterointerface. The energy band diagram under the gate of an AlGaN/GaN HEMT is
shown in Figure 1.3.

2DEG

EC
EF
AlGaN

GaN
EV

Figure 1.3

Energy band diagram under the gate of an undoped AlGaN/GaN HEMT.

The collection of electrons occupying this quantum well is called a twodimensional electron gas (2DEG) due to a very small thickness of the quantum well of
approximately 20 to 30 Angstroms [4] and is essentially two-dimensional as the length
and width of the channel are much larger. The electrons in the 2DEG are able to have a
higher mobility than in the bulk semiconductor due to the GaN being undoped, absent of
impurity atoms which cause scattering. The resulting high carrier velocities and low

resistance of the conducting channel enables the HEMT device to achieve highfrequencies and low-noise performance.
A unique feature of the AlGaN/GaN HEMT is that both the AlGaN and GaN can
be left undoped as the 2DEG is created even without intentional doping. This is due to
the spontaneous and piezoelectric polarizations that exist in the device. The polarization
present in the AlGaN/GaN is significant and polarization induced channel charges can
reach 1x1013 electrons/cm2, more than 4 times the total channel charges found in
AlGaAs/GaAs HEMTs [3]. Doping of the AlGaN layer does not contribute a significant
portion to the 2DEG as it has been demonstrated that the formation of the 2DEG in the
AlGaN/GaN HEMT is dominated by polarization induced effects [21].
The polarization of the GaN and AlGaN crystal structures are determined by the
growth process which can result in Gallium (Aluminum) faced crystals or Nitrogen faced
crystals [22]. For AlGaN/GaN HEMTs grown by MOCVD the crystal results in a Ga
face. MBE grown AlGaN/GaN HEMTs on sapphire result in N faced crystals unless an
AlN nucleation layer is deposited on the sapphire substrate which results in Ga faced
crystals [21]. The crystal structures for wurtzite Ga-face and N-face GaN grown in the cdirection are shown in Figure 1.4. For a Ga-face crystal structure, the gallium atom is on
top of a nitrogen atom and has [0001] polarity and the vector is pointing from a gallium
atom to a nearest-neighbor nitrogen atom. The direction vector of the polarization is
opposite for N-faced GaN, where the vector is pointing from a nitrogen atom to a nearestneighbor gallium atom.

10

Figure 1.4

Crystal structure of wurtzite Ga-face and N-face GaN grown along in the
c-direction [22].

The 2DEG is influenced by the amount of polarization present in the AlGaN/GaN


heterostructure, which depends on the Al content in the AlGaN. The negative charge in
the 2DEG is equal but opposite in magnitude to the positive polarization sheet charge
density at the AlGaN/GaN interface. Shown in Figure 1.5, the polarization induced sheet
charge density is a sum associated with the directions of the spontaneous and
piezoelectric polarization in the structure.

The net spontaneous and piezoelectric

polarizations will influence the collection of electrons or holes equal to the polarization
sheet charge density based on the growth of the AlGaN and GaN layers. In the case of
Ga-face heterostructures, the spontaneous polarization is negative meaning it is pointing
towards the substrate.

Under tensile strain, the piezoelectric and spontaneous

polarizations are parallel and the polarization point in the same direction.

Under

compressive strain, the piezoelectric and spontaneous polarizations are antiparallel. For
the tensile strained AlGaN grown over GaN commonly employed in the HEMT, the
11

polarization induced charge density is positive (+) and electrons will compensate the
charge

by

Figure 1.5

forming

2DEG

below

the

AlGaN/GaN

heterointerface.

Diagrams showing the polarization induced sheet charge density and the direction
of spontaneous and piezoelectric polarization in Ga-face and N-face for strained and
relaxed AlGaN/GaN heterostructures. Case (b) is the typical growth structure in
most AlGaN/GaN HEMTs [22].

The atoms in GaN and AlGaN are electronically charged and are arranged
irregularly due to the difference in size. The gallium atom is larger than the nitrogen
atom. This combination of ionization and irregularity leads to a spontaneous electrical
polarization within the crystal, or a separation of charge into countless, regularly spaced
12

negative and positive atoms [7].

During the growth process of AlGaN, the positive

atoms line up so that they that aligned towards the GaN surface. As the layer thickness
increases during growth, the atomic layers continue to align, creating an electric field
internal to the AlGaN layer. The positive side of the dipole faces the GaN layer and the
negative side of the dipole faces the growth surface. The electric field due to dipole is
enough to ionize some of the covalent electrons as well as impurities that happen to be
available in the AlGaN layer [4]. A large reservoir of electrons is available from the
loosely bound surface electrons. The major sources of electrons in the 2DEG come from
donor-like surface states or from a high density of both donor- and acceptor-like surface
states [23]. These ionized electrons eventually find their way to the lower potential in the
region below the AlGaN/GaN interface and form the 2DEG. As more electrons move
from the AlGaN into the 2DEG, the electric field reduces and when enough electrons are
in the 2DEG the electric field is reduced to the point where the electron transfer process
stops. In combination with the spontaneous polarization is the piezoelectric polarization
which is caused by the strain associated with the lattice mismatch of the AlGaN and GaN
crystals [7].

1.3 Reported Results for AlGaN/GaN HEMTs


This section will discuss experimental and theoretical work done with
AlGaN/GaN HEMTs and the improvements made in the device. The promise of high
performance AlGaN/GaN HEMTs for high frequency, high power applications has been
an incentive for a large amount of work being done in recent years to demonstrate and
improve the devices performance.

AlGaN/GaN HEMTs are being developed


13

commercially as well as researched in many universities; the work of which is discussed


below.
Palacios et al. [24] at the University of California at Santa Barbara (UCSB),
reported at the time of publication, record power density of 10.5 W/mm with 34% power
added efficiency (PAE) at 40 GHz for a Si3N4 passivated AlGaN/GaN HEMT. The
reported results for the power density, gain and PAE at 40 GHz are illustrated in Figure
1.6.

Figure 1.6

Power sweep on a MOCVD Si3N4 passivated AlGaN/GaN HEMT where record


power density of 10.5 W/mm was reported at 40 GHz [24].

The reported maximum fT and fmax were 70 GHz and 100 GHz, respectively, in
the passivated device with 9 V on the drain. Record results in power density were also
reported for unpassivated GaN HEMTs without field plates using a recessed design and a
new fluorine-plasma surface treatment [25]. By using this new recessed design and new
surface treatment they were able to increase the breakdown voltage to about 200 V and
14

reduce the gate leakage current. The reported gate leakage current for the GaN HEMT
with and without fluorine-plasma surface treatment are shown in Figure 1.7.

Figure 1.7

Comparison of gate leakage current for fluorine-plasma surface treatment and


different epitaxial structures showing improved reduction in gate leakage for
unpassivated GaN HEMTs where record results in power density were reported
without field plates [25].

The results for the MOCVD grown GaN HEMT were slightly better than the
MBE grown, with a power density of 17.8 W/mm and PAE of 50% at 4 GHz as shown in
Figure 1.8.

Palacios et al. [26] at UCSB have also fabricated enhanced-mode

AlGaN/GaN HEMTs with moderate results compared to depletion-mode AlGaN/GaN


HEMTs. They reported record current of 1.2 A/mm for enhanced-mode AlGaN/GaN
HEMTs with a threshold voltage of 0.1 V. The IDS v. VDS output characteristics are
shown in Figure 1.9. Research on AlGaN/GaN HEMTs has also been done at Cornell
University [27].

15

Figure 1.8

Power characteristics of a MOCVD-grown unpassivated GaN HEMT with a record


power density of 17.8 W/mm with a PAE of 50% at 4 GHz [25].

Figure 1.9

IDS v. VDS output characteristics of an enhanced-mode AlGaN/GaN HEMT with


record reported current for enhanced-mode [26].

16

Research efforts have also focused on extending the operating frequencies of the
AlGaN/GaN HEMT beyond 50 GHz into the V- and W-bands. In recent years, groups
working on AlGaN/GaN HEMTs have achieved improved high-frequency performance
in AlGaN/GaN HEMTs [28, 18]. In 2002, a 0.12 m gate-length AlGaN/GaN HEMT
grown on SiC demonstrated a unity current gain cutoff frequency (fT) of 121 GHz and
maximum frequency of oscillation (fmax) of 162 GHz with over 1 A/mm current density
and over 300 mS/mm[18]. In 2006, Higashiwaki et al. [29] reported a cutoff frequency
of 163 GHz in an AlGaN/GaN HEMT with an AlN spacer between the AlGaN and GaN
layers, which is a record as of the time of the reporting.

A maximum oscillating

frequency of 192 GHz was also achieved using an AlGaN/GaN HEMT with a slightly
large gate length. They were able to achieve better results by using a catalytic chemical
vapor deposition (cat-CVD) for the SiN passivation layer and by varying the gate length
and AlGaN thickness to optimize the device performance. Figure 1.10 shows their
results for the electron mobility, sheet electron density and sheet resistance with varying
AlGaN thickness and with/without SiN passivation.

Figure 1.10

a) Electron mobility, b) sheet electron density, and c) sheet resistance for varying
AlGaN thickness and with/without SiN passivation for the AlGaN/GaN HEMT with
reported 163 GHz cutoff frequency [29].

17

Figure 1.11 shows the extracted cutoff frequency from the current gain profile and
the maximum oscillation frequency from the maximum stable gain/maximum available
gain (MSG/MAG) and unilateral gain profiles. The authors expect that power densities
larger than 0.1-1 W/mm can be achieved in the V- and W-bands [29].

Figure 1.11

a) Cutoff frequency of 163 GHz extracted from current gain versus frequency for an
AlGaN/GaN HEMT with a gate length of 0.06 m and b) maximum stable
gain/maximum available gain (MSG/MAG) and unilateral gain versus frequency for
an AlGaN/GaN HEMT with a gate length of 0.08 m [29].

Record results have also been achieved in improving the maximum oscillation
frequency (fmax) in AlGaN/GaN HEMTs, with a reported fmax of 230 GHz being the best
thus far [30, 31]. Palacios et al. [30] at UCSB were able to achieve an fmax of 230 GHz
by introducing a thin InGaN layer between the GaN channel and GaN buffer. This
InGaN layer allows for better confinement of the electrons in the channel by increasing
the barrier of the GaN buffer and GaN channel. The conduction band diagram of this
AlGaN/GaN HEMT with the introduction of the InGaN back-barrier is illustrated in

18

Figure 1.12. They were able to improve the fmax performance of the AlGaN/GaN HEMT
by introducing the InGaN back-barrier, but the fT performance of the HEMT remained
relatively unchanged from the normal design.

Figure 1.12

Conduction band schematic of AlGaN/GaN HEMT with introduction of InGaN


back-barrier to improve fmax performance [30].

Figure 1.13 shows the recorded small-signal characteristics when biased for the
optimum fT and fmax of the HEMT. In another paper, Palacios et al. [31] reported on the
optimization of this HEMT design with the InGaN backlayer. They were able to achieve
a 20% improvement in fmax when adding the InGaN back-barrier to the AlGaN/GaN
HEMT design, by using -shaped gates to decrease the gate-to-drain capacitance (CGD),
and by reducing the gate width which reduces the gate resistance.

19

Figure 1.13

Small-signal characteristics of AlGaN/GaN HEMT with InGaN back-barrier with


bias set to achieve for maximum fT and fmax [30].

Recent commercial efforts have brought AlGaN/GaN HEMTs into the market.
Nitronex Corp. has recently released three RF power transistors in the 3.3-3.8 GHz and
2.3-2.7 GHz bands for the WiMAX market [32, 33, 34]. The RF power transistors are
AlGaN/GaN HEMTs grown on Si wafers. Some of the research done at Nitronex with
AlGaN/GaN HEMTs on Si is reported in [19, 35, 36, 37]. Cree Inc. have also released
two GaN HEMTs for WiMAX applications [38, 39]. In the past few years, Cree have
reported results of high-power transistors [17, 40, 41].

Another company who has

released GaN HEMTs commercially is RF Micro Devices Inc., with their line of highpower transistors targeting the WiMAX and WCDMA market [42].
Khan et al. [43, 44] at the University of South Carolina have in recent years
focused on developing metal oxide semiconductor heterostructure field effect transistors
(MOSHFETs), which should reduce gate leakage by several magnitudes due to the gate
oxide present and allow the device to operate with higher forward voltage. In 2000,
Khan et al. demonstrated MOSHFETs grown on sapphire [43] and SiC substrates [44]

20

which had a gate leakage reduction of several magnitudes compared to conventional


AlGaN/GaN HEMTs.
Efforts have also been made in recent years to enable higher voltage and more
reliable HEMT operation by incorporating a field plate which reduces the electric field at
the gate edge on the drain side of the device [45]. This reduction in the electric field
allows higher voltages to be applied due to a reduction of tunneling current from the gate
to the GaN semiconductor, permitting higher power generation. Field-plate designs also
reduce the current collapse which has been a common reliability problem that continues
to plague the device. In 2003, Ando et al [46] reported a power density of 10 W/mm at 2
GHz with the drain voltage operating at 65 V as well as reducing current collapse using a
field plate.

Since then, researchers have been able to further improve the device

performance with modified field plate design [47, 48, 40]. In 2005, Cree reported that it
had been able to achieve over 8 W/mm at 40 GHz using a field-plate design [41].
While experimental development of AlGaN/GaN HEMTs has been ongoing, there
have also been efforts at device modeling. However, much work still needs to be done in
the modeling of AlGaN/GaN HEMTs. Currently no software modeling software has the
ability to incorporate the polarization physics into the software. That is, based on the
AlGaN composition and known structure of the device, predict the magnitude of
polarization. The modeling work done so far has used various methods to incorporate the
effects of polarization present in AlGaN/GaN HEMTs to simulate the polarization
charges present near the 2DEG. Typically, this has been done by inserting sheets of
positive and negative fixed charge at the opposite faces of the AlGaN layer. The positive

21

charge at the AlGaN/GaN interface induces the negative charge of the 2DEG in the GaN
at the heterointerface.
Several groups have reported some success in modeling these AlGaN/GaN
HEMTs. Sacconni et al. [49] have modeled the device based on the self-consistent
solution of the Schrdinger and Poisson equations when coupled to a quasi-2D model for
current flow.

Their results demonstrate that without incorporating the piezoelectric

polarization into the model, the quantum well at the heterointerface is not well formed
and band bending at the interface will only occur when piezoelectric polarization is
incorporated into the model.

With both spontaneous and piezoelectric polarization

incorporated into the model, the quantum well is better formed. The formation of the
quantum well is demonstrated in Figure 1.14 where the conduction band of a modulation
doped-field-effect transistor (MODFET) structure along the [0001] direction is shown
with and without polarization fields being incorporated into the model. Their model also
showed that the electron concentration in the channel is formed and increases as the
aluminum concentration in AlGaN increases from 0.1 to 0.4 as shown in Figure 1.15.
Current-Voltage (I-V) characteristics were also calculated using the sheet charge density
obtained from a self-consistent Schrdinger-Poisson solution. The results of the I-V
calculations are shown in Figure 1.16 for AlGaN/GaN MODFET with an Al
concentration of 0.2 and 0.4.

22

Figure 1.14

Conduction band profile for MODFET structure in the [0001] direction for VG = 0
V with and without polarization incorporated into the self-consistent solution of the
Schrdinger and Poisson equations coupled to a quasi-2-D model for the current
flow [49].

Figure 1.15

Calculated electron density distribution in the channel with several Al


concentrations for the AlxGa1-xN layer of the MODFET [49].

23

Figure 1.16

Calculated IDS v. VDS for an AlGaN/GaN MODFET with an Al concentration of


a) 0.2 b) 0.4 [49].

Braga et al. [50] demonstrated the importance of incorporating the hydrodynamic


transport model into the device simulations to account for the negative differential
conductance in the output characteristics that is observed experimentally.
24

The I-V

characteristics are shown in Figure 1.17, with the simulation results of the
AlGaN/InGaN/GaN HEMT using the hydrodynamic transport model and the drift
diffusion model compared with the measured results of the real device.

Figure 1.17

IDS v. VDS characteristics for simulated and measured AlGaN/InGaN/GaN HEMT.


The simulation results using the hydrodynamic transport model (solid lines) and the
drift-diffusion transport model (dashed lines) for VG = -2, 0 V are compared with
the measured results (square markers) for VG = -8, -2, -1, 0 V [50].

The negative differential conductance is due to hot electrons present in the device
spreading from the 2DEG to the barrier and bulk where they are captured by bulk traps.
The spreading of the hot electrons is illustrated in Figure 1.18, where the electron
concentration in the device is shown to spread much more into the GaN and AlGaN
layers when using the hydrodynamic model.

25

Figure 1.18

Cross-section of AlGaN/InGaN/GaN HEMT showing the electron concentration at


VD = 10 V and VG = 0 V when incorporating the a) hydrodynamic transport model
where the electrons are spreading away from the channel and into the AlGaN and
GaN layers and b) drift-diffusion model where the electrons are mostly confined in
the channel [50].

Figure 1.19 shows the distribution of trapped electrons when incorporating the
hydrodynamic model.

Figure 1.19

Distribution of trapped electrons by using the hydrodynamic model to predict the


occupied bulk traps present in the simulated AlGaN/InGaN/GaN HEMT at VD = 10
V and VG = 0 V [50].

26

The large amount of occupied traps deep into the device where there is less
electron density causes extra negative charge which translates to an increased barrier in
the conduction band in the region of the gate edge on the drain side of the device [50].
This is shown in Figure 1.20 where Braga et al. [50] have proposed that the barrier in the
conduction band is associated with the negative differential conductance that is measured
in the real device by simulating the AlGaN/InGaN/GaN HEMT using the hydrodynamic
transport model to account for hot electrons.

Figure 1.20

Cross-sections of the AlGaN/InGaN/GaN HEMT in the region of the gate edge near
the drain side of the device that show the conduction band contours at a) VD = 2.4 V
and b) VD = 6 V which are highlighted in c) as the peak and valley of the ID v. VD
curve. d) shows the 3-D surface plot of the conduction band with the barrier located
at the gate edge near the drain highlighted. This increased barrier is associated
with the negative differential conductance in the ID v. VD curve [50].

In recent years, modeling studies have focused on explaining the current collapse
observed experimentally in AlGaN/GaN HEMTs [51, 52, 53, 54, 55]. This effect occurs
under microwave operation when large bias voltages are present on the device. Surface
27

states act as a virtual gate between the gate and drain and trap electrons which cause a
reduction in drain current output. This drain current collapse has become an issue in
achieving expected microwave/rf performance and is attributed to lower than expected
power levels achieved in devices as shown in Figure 1.21 [56].

Figure 1.21

The effect of the drain current collapse on the experimental power output of regular
and field-plated III-Nitride HFETs and comparison to theoretical output [56].

Nguyen et al. [52] investigated this current collapse phenomenon in AlGaN/GaN


HEMTs in 1999 and found that the drain current decreases under increasing rf input
drive.

A similar current collapse measurement without load impedance in the

experimental setup eliminates the possibility of the self-heating as a cause of the effect.
This measurement is shown in Figure 1.22.

28

Figure 1.22

Measured drain current with varying rf input drive, the decreasing current is the
observed drain current collapse. The measurements were performed at a frequency
of 4 GHz and VD = 20 V for ZL = 0 and VG = -3 V (circle) and ZL = 38 + j39 and VG =
-3 V (triangles) [52].

Vetury et al. [53] used a virtual gate model due to surface states as a way to
describe the observed current collapse in AlGaN/GaN HEMTs. The negative charge of
the surface states acts a reversed biased virtual gate whose extent and charge depend on
the quiescent bias condition of the device. Figure 1.23 shows the effect of the drain
current collapse under various drain bias conditions and with and without surface
passivation. Surface passivation of the device largely reduces the current collapse of the
device compared to the unpassivated device. Surface states are the apparent cause of the
current collapse as it is reduced in the passivated device.

29

Figure 1.23

Peak normalized drain current transients under various drain voltage and with and
without surface passivation. The large reduction of the current collapse in the
passivated devices compared to the unpassivated devices proves that the current
collapse is related to surface states [53].

Hasegawa et al. [54] measured gateless and Schottky-gate AlGaN/GaN HFETs


with and without various surface processing to investigate the current collapse
phenomenon in AlGaN/GaN HEMTs. DC I-V measurements were nonlinear due to the
virtual gating effect by surface states which were observed by x-ray photoelectron
spectroscopy (XPS) to be 0.37 eV-deep donors attributed to nitrogen vacancies. The
results of the XPS measurements are shown in Figure 1.24 and their models of the current
collapse are illustrated in Figure 1.25.

30

Figure 1.24

XPS measurements of plasma-treated AlGaN surfaces for a) Ga 3d and b) N 1s.


The groups results conclude that the surface states are 0.37 eV-deep donor Nvacancies [54].

31

Figure 1.25

a) IDS v. VDS representation of the drain current collapse and models for current
collapse b) under drain stress and c) under gate stress [54].

Analytical modeling has also been done in the past few years, where Ambacher et
al. [21, 22] have developed an analytical model for the spontaneous and piezoelectric
polarization, and the sheet charge density as a function of the polarization induced sheet
charge density. Green et al. [57] have developed a large signal model for AlGaN/GaN
HEMTs on SiC substrates by extracting the small-signal parameters of a built device.
Karmalkar et al. [58] have developed models for optimizing the field-plate design to
achieve the lowest electric field along the channel or along the surface.
Braga et al. [59, 60, 61] have done modeling that links current collapse to traps
present under the gate edges using DESSIS, a commercial software. Their work shows
that current collapse does not occur for a uniform trap distribution under the gate but is

32

observed for distribution of traps under the gate edges. They also link the current
collapse with hot electrons spreading into the buffer layer and propose a double
heterostructure field-effect transistor as a means to reduce current collapse.
Verzellesi et al. [62] and Meneghesso et al. [63] used DESSIS to model surface
traps at the ungated surface as the source for the drain current dispersion present in
AlGaN/GaN HEMTs. Meneghesso et al. [63] have shown that surface donor-like traps at
an energy located at ET-EV = 0.3 eV behave as hole traps and can account for the current
collapse if modeled with the polarization charges. However, simulations can not account
for the differences between their hole trap theory and the more conventional electron trap
theory as simulations can not account for nonideal surface conduction mechanisms due
to electron tunneling and electron hopping [63].

1.4 Focus of This Thesis


The focus of this thesis is to investigate the dc and small-signal microwave
performance of the AlGaN/GaN HEMT. This work will incorporate the polarization in
the device by placing interface sheet charge densities at the heterointerface and the
AlGaN surface. The polarization charge will be varied at the heterointerface and AlGaN
surface to study the effects on the device performance.

Further discussion of the

polarization incorporated into the device modeling can be found in the following chapter.
The source of electrons in the 2DEG will come from surface states as proposed by
Ibbetson et al. [23]. The surface state donors will be investigated in regards to their effect
on the device performance when varying the density.

Furthermore, this work will

examine the device performance when varying the device structure. This work will be
33

compared to the experimental results by Jessen et al. [64] as a means of verifying the
simulated results by using their device structure.

1.5 Description of Following Chapters


This initial chapter has introduced the reader to the basic operation of the
AlGaN/GaN HEMT as well as the potential applications for the device especially for
high frequency, high power applications such as power amplifiers for GHz frequency.
Discussed briefly were the material properties of GaN that makes it attractive for
microwave applications as well as the advantages and disadvantages of the material over
other semiconductors.

Previous experimental work on improving the AlGaN/GaN

HEMT performance is discussed as well as important modeling work done analytically


and through device software.
The second chapter describes the software used in modeling the device as well as
explaining the underlying physics that were modeled. The material properties of AlGaN
and GaN and how they are used in the physical models are discussed. The method of
incorporating the polarization charges in the model and its role in forming the 2DEG are
also discussed, the importance of which has been described in the first chapter.
The third chapter reports the results of the device modeling work as well as the
comparison of the results with previous reported experimental results. Included in this
chapter are the parameters of the dc and small-signal results. Chapter four discusses
results of optimizing and altering the original device design and structure with the results
in this chapter compared to the modeling work reported in chapter three. Chapter five
summarizes the results of the device modeling study of the AlGaN/GaN HEMT and
34

proposes future modeling work to be done to improve the model and characterization of
the device.

35

References
[1]

Y. Wu and P. Parikh, High-power GaN HEMTs battle for vacuum-tube


territory, Compound Semiconductor,
http://compoundsemiconductor.net/articles/magazine/12/1/4/1, January 2006.

[2]

J. D. Cressler, SiGe HBT Technology: A New Contender for Si-Based RF and


Microwave Circuit Applications, IEEE Transactions on Microwave Theory and
Techniques, vol. 46, no. 5, pp. 572-589, May 1998.

[3]

R. Borges, Gallium nitride electronic devices for high-power wireless


applications, RF Design, pp. 72-82, September 2001.

[4]

R. J. Trew, G. L. Bilbro, W. Kuang, Y. Liu, and H. Yin, Microwave AlGaN/GaN


HFETs, IEEE Microwave Magazine, vol. 6, no. 1, pp. 56-66, March 2005.

[5]

V. L. Granatstein, P. K. Parker, and C. M. Armstrong, Scanning the Technology:


Vacuum Electronics at the Dawn of the Twenty-First Century, Proceedings of
the IEEE, vol. 87, no. 5, pp. 702-716, May 1999.

[6]

H. G. Henry, G. Augustine, G. C. DeSalvo, R. C. Brooks, R. R. Barron, J. D.


Oliver, Jr., A. W. Morse, B. W. Veasel, P. M. Esker, and R. C. Clarke, S-Band
Operation of SiC Power MESFET With 20 W (4.4 W/mm) Output Power and
60% PAE, IEEE Transactions on Electron Devices, vol. 51, no. 6, pp. 839-845 ,
June 2004.

[7]

L. F. Eastman and U. K. Mishra, The Toughest Transistor Yet, IEEE Spectrum,


vol. 39, no. 5, pp. 28-33, May 2002.

36

[8]

M. Shur, B. Gelmont, and M. Asif Khan, Electron Mobility in Two-Dimensional


Electron Gas in AlGaN/GaN Heterostructures and in Bulk GaN, Journal of
Electronic Materials, vol. 25, no. 5, pp. 777-785, May 1996.

[9]

U. K. Mishra, P. Parikh, and Y. Wu, AlGaN/GaN HEMTs-An Overview of


Device Operation and Applications, Proceedings of the IEEE, vol. 90, no. 6, pp.
1022-1031, June 2002.

[10]

R. Dingle, New High-Speed III-V Devices for Integrated Circuits, IEEE


Transactions on Electron Devices, vol. 31, no. 11, pp. 1662-1667, November
1984.

[11]

R. Dingle, H. L. Strmer, A. C. Gossard, and W. Wiegmann, Electron mobilities


in modulation-doped semiconductor heterojunction superlattices, Applied
Physics Letters, vol. 33, no. 7, pp. 665-667, October 1978.

[12]

S. Hiyamizu, T. Mimura, T. Fugii, K. Nanbu, and H. Hashimoto, Extremely high


mobility of two-dimensional electron gas in selectively doped GaAs/N-AlGaAs
heterojunction structures grown by MBE, Japanese Journal of Applied Physics,
vol. 20, no. 4, pp. L245-L248, April 1981.

[13]

The Nobel Prize in Physics 1985, Press Release, The Royal Swedish Academy
of Sciences, http://nobelprize.org/nobel_prizes/physics/laureates/1985/press.html

[14]

M. Asif Khan, J. N. Kuznia, D. T. Olson, W. J. Schaff, J. W. Burm, and M. S.


Shur, Microwave performance of a 0.25 m gate AlGaN/GaN heterostructure
field effect transistor, Applied Physics Letters, vol. 65, no. 9, pp. 1121-1123,
August 1994.

37

[15]

M. A. Littlejohn, J. R. Hauser, and T. H. Glisson, Monte Carlo calculation of the


velocity-field relationship for gallium nitride, Applied Physics Letters, vol. 26,
no. 11, pp. 625-627, June 1975.

[16]

B. Gelmont, K. Kim, and M. Shur, Monte Carlo simulation of electron transport


in gallium nitride, Journal of Applied Physics, vol. 74, no. 3, pp. 1818-1821,
August 1993.

[17]

Y. F. Wu, D. Kapolnek, J. P. Ibbetson, P. Parikh, B. P. Keller, and U. K. Mishra,


Very-high power density AlGaN/GaN HEMTs, IEEE Transactions on Electron
Devices, vol. 48, no. 3, pp. 586-590, March 2001.

[18]

V. Kumar, W. Lu, R. Schwindt, A. Kuliev, G. Simin, J. Yang, M. Asif Khan, and


Ilesanmi Adesida, AlGaN/GaN HEMTs on SiC with fT of over 120 GHz, IEEE
Electron Device Letters, vol. 23, no. 8, pp. 455-457, August 2002.

[19]

J. W. Johnson, E. L. Piner, A. Vescan, R. Therrien, P. Rajagopal, J. C. Roberts, J.


D. Brown, S. Singhal, K. J. Linthicum, 12 W/mm AlGaN-GaN HFETs on silicon
substrates, IEEE Electron Device Letters, vol. 25, no. 7, pp. 459-461, July 2004.

[20]

L. Geppert, The Gallium Nitride Gamble, IEEE Spectrum, vol. 41, no. 1, pp.
52-55, 57-59, January 2004.

[21]

O. Ambacher, B. Foutz, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M.


Murphy, A. J. Sierakowski, W. J. Schaff, L. F. Eastman, R. Dimitrov, A.
Mitchell, and M. Stutzmann, Two-dimensional electron gases induced by
spontaneous and piezoelectric polarization in undoped and doped AlGaN/GaN
heterostructures, Journal of Applied Physics, vol. 87, no. 1, pp. 334-344, January
2000.

38

[22]

O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy, W. J.


Schaff, L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, and J.
Hilsenbeck, Two-dimensional electron gases induced by spontaneous and
piezoelectric

polarization

charges

in

N-

and

Ga-face

AlGaN/GaN

heterostructures, Journal of Applied Physics, vol. 85, no. 6, pp. 3222-3233,


March 1999.
[23]

J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.


Mishra, Polarization effects, surface states, and the source of electrons in
AlGaN/GaN heterostructure field effect transistors, Applied Physics Letters, vol.
77, no. 2, pp. 250-252, July 2000.

[24]

T. Palacios, A. Chakraborty, S. Rajan, C. Poblenz, S. Keller, S. P. DenBaars, J. S.


Speck, and U. K. Mishra, High-power AlGaN/GaN HEMTs for Ka-band
applications, IEEE Electron Device Letters, vol. 26, no. 11, pp. 781-783,
November 2005.

[25]

L. Shen, T. Palacios, C. Poblenz, A. Corrion, A. Chakraborty, N. Fichtenbaum, S.


Keller, S. P. DenBaars, J. S. Speck, and U. K. Mishra, Unpassivated high power
deeply recessed GaN HEMTs with fluorine-plasma surface treatment, IEEE
Electron Device Letters, vol. 27, no. 4, pp. 214-216, April 2006.

[26]

T. Palacios, C. S. Suh, A. Chakraborty, S. Keller, S. P. DenBaars, and U. K.


Mishra, High-performance E-mode AlGaN/GaN HEMTs, IEEE Electron
Device Letters, vol. 27, no. 6, pp. 428-430, June 2006.

[27]

L. F. Eastman, V. Tilak, J. Smart, B. M. Green, E. M. Chumbes, R. Dimitrov, H.


Kim, O. S. Ambacher, N. Weimann, T. Prunty, M. Murphy, W. J. Schaff, and J.

39

R. Shealy, Undoped AlGaN/GaN HEMTs for Microwave Power Amplification,


IEEE Transactions on Electron Devices, vol. 48, no. 3, pp. 479-485, March 2001.
[28]

M. Micovic, N. X. Nguyen, P. Janke, W. S. Wong, P. Hashimoto, L. M. McCray,


and C. Nguyen, GaN/AlGaN high electron mobility transistors with fT of 110
GHz, Electronics Letters, vol. 36, no. 4, pp. 358-359, February 2000.

[29]

M. Higashiwaki, N. Onojima, T. Matsui, and T. Mimura, High fT and fmax


AlGaN/GaN HFETs achieved by using thin and high-Al-composition AlGaN
barrier layers and Cat-CVD SiN passivation, Physica Status Solida A, vol. 203,
no. 7, pp. 1851-1855, May 2006.

[30]

T. Palacios, A. Chakraborty, S. Heikman, S. Keller, S. P. DenBaars, and U. K.


Mishra, AlGaN/GaN High Electron Mobility Transistors With InGaN BackBarriers, IEEE Electron Device Letters, vol. 27, no. 1, pp. 13-15, January 2006.

[31]

T. Palacios, Y. Dora, A. Chakraborty, C. Sanabria, S. Keller, S. P. DenBaars, and


U. K. Mishra, Optimization of AlGaN/GaN HEMTs for high frequency
operation, Physica Status Solida A, vol. 203, no. 7, pp. 1845-1850, May 2006.

[32]

A. Bindra, M. Valentine, K. Vick, Next-generation wireless propels GaN power


transistors, CMOS RFICs and passives, RF Design, pp. 16-21, January 2006.

[33]

M. Hatcher, WiMAX and WiBro emerge as major targets for GaN, Compound
Semiconductor, http://compoundsemiconductor.net/articles/magazine/11/12/3/1,
December 2005.

[34]

M. LePedus, Nitronex rolls GaN parts for WiMAX market, EE Times,


http://www.eetimes.com/news/design/showArticle.jhtml;jsessionid=IXJW25R3T
SBDQQSNDLRCKHSCJUNN2JVN?articleID=193104773, October, 2006.

40

[35]

R. Therrien, S. Singhal, J. W. Johnson, W. Nagy, R. Borges, A. Chaudhari, A. W.


Hanson, A. Edwards, J. Marquart, P. Rajagopal, C. Park, I. C. Kizilyalli, and K. J.
Linthicum, A 36mm GaN-on-Si HFET Producing 368W at 60V with 70% Drain
Efficiency, 2005 IEEE International Electron Devices Meeting (IEDM),
Washington, DC, pp. 568-571, December 2005.

[36]

W. Nagy, S. Singhal, R. Borges, J. W. Johnson, J. D. Brown, R. Therrien, A.


Chaudhari, A. W. Hanson, J. Riddle, S. Booth, P. Rajagopal, E. L. Piner, and K. J.
Linithicum, 150 W GaN-on-Si RF Power Transistor, 2005 IEEE MTT-S
International Microwave Symposium, Long Beach, CA, June 2005.

[37]

S. Singhal, T. Li, A. Chaudhari, A. W. Hanson, R. Therrien, J. W. Johnson, W.


Nagy, J. Marquart, P. Rajagopal, J. C. Roberts, E. L. Piner, I. C. Kizilyalli, and K.
J.

Linthicum,

Reliability

of

large

periphery

GaN-on-Si

HFETs,

Microelectronics Reliability, vol. 46, no. 8, pp. 1247-1253, August 2006.


[38]

PRNewswire, New Cree GaN HEMT Optimized for North American WiMAX
Applications, EE Times,
http://www.eetimes.com/press_releases/prnewswire/showPressRelease.jhtml%3bj
sessionid=RWIHFP4R1TULKQSNDLRCKHSCJUNN2JVN?articleID=X538024
&CompanyId=1, October, 2006.

[39]

Cree, Inc., http://www.cree.com.

[40]

Y. -F. Wu, A. Saxler, M. Moore, R. P. Smith, S. Sheppard, P. M. Chavarkar, T.


Wisleder, U. K. Mishra, and P. Parikh, 30-W/mm GaN HEMTs by Field Plate
Optimization, IEEE Electron Device Letters, vol. 25, no. 3, pp. 117-119, March
2004.

41

[41]

Y. -F. Wu, M. Moore, A. Saxler, T. Wisleder, U. K. Mishra, and P. Parikh, 8Watt GaN HEMTs at Millimeter-Wave Frequencies, IEEE IEDM, 2005.

[42]

RF Micro Devices, Inc., http://www.rfmd.com/.

[43]

M. Asif Khan, X. Hu, G. Sumin, A. Lunev, J. Yang, R. Gaska, and M. S. Shur,


AlGaN/GaN Metal Oxide Semiconductor Heterostructure Field Effect
Transistor, IEEE Electron Device Letters, vol. 21, no. 2, pp. 63-65, February
2000.

[44]

M. Asif Khan, X. Hu, A. Tarakji, G. Simin, and J. Yang, R. Gaska, and M. S.


Shur, AlGaN/GaN metal-oxide-semiconductor heterostructure field-effect
transistors on SiC substrates, Applied Physics Letters, vol. 77, no. 9, pp. 13391341, August 2000.

[45]

N. Q. Zhang, S. Keller, G. Parish, S. Heikman, S. P. DenBaars, and U. K.


Mishra, High Breakdown GaN HEMT with Overlapping Gate Structure, IEEE
Electron Device Letters, vol. 21, no. 9, pp. 421-423, September 2000.

[46]

Y. Ando, Y. Okamoto, H. Miyamoto, T. Nakayama, T. Inoue, and M. Kuzuhara,


10-W/mm AlGaN-GaN HFET With a Field Modulating Plate, IEEE Electron
Device Letters, vol. 24, no. 5, pp. 289-291, May 2003.

[47]

C. Lee, P. Saunier, J. Yang, and M. A. Khan, AlGaN-GaN HEMTs on SiC with


CW power performance >4 W/mm and 23% PAE at 35 GHz, IEEE Electron
Device Letters, vol. 24, no. 10, pp. 616-618, October 2003.

[48]

R. Thompson, T. Prunty, V. Kaper, and J. R. Shealy, Performance of the AlGaN


HEMT Structure With a Gate Extension, IEEE Transactions on Electron
Devices, vol. 51, no. 2, pp. 292-295, February 2004.

42

[49]

F. Sacconi, A. Di Carlo, P. Lugli, H. Morko, Spontaneous and Piezoelectric


Polarization Effects on the Output Characteristics of AlGaN/GaN Heterojunction
Modulation Doped FETs, IEEE Transactions on Electron Devices, vol. 48, no. 3,
pp. 450-457, March 2001.

[50]

N. Braga, R. Mickevicius, R. Gaska, X. Hu, M. S. Shur, M. Asif Khan, G. Simin,


and J. Yang, Simulation of hot electron and quantum effects in AlGaN/GaN
heterostructure field effect transistors, Journal of Applied Physics, vol. 95, no.
11, pp. 6409-6413, June 2004.

[51]

R. Vetury, Y. F. Wu, P. T. Fini, G. Parish, S. Keller, S. P. DenBaars, and U. K.


Mishra, Direct Measurement of Gate Depletion in High Breakdown (405 V)
AlGaN/GaN Heterostructure Field Effect Transistors, IEDM 1998 Technical
Digest, pp. 55-58, December 1998.

[52]

C. Nguyen, N. X. Nguyen, and D. E. Grider, Drain current compression in GaN


MODFETs under large-signal modulation at microwave frequencies, Electronics
Letters, vol. 35, no. 16, pp. 1380-1382, August 1999.

[53]

R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, The Impact of Surface


States on the DC and RF Characteristics of AlGaN/GaN HFETs, IEEE
Transactions on Electron Devices, vol. 48, no. 3, pp. 560-566, March 2001.

[54]

H. Hasegawa, T. Inagaki, S. Ootomo, and T. Hashizume, Mechanisms of current


collapse and gate leakage currents in AlGaN/GaN heterostructure field effect
transistors, Journal of Vacuum Science and Technology B, vol. 21, no. 4, pp.
1844-1855, August 2003.

43

[55]

A. Koudymov, G. Simin, M. Asif Khan, A. Tarakji, R. Gaska, and M. S. Shur,


Dynamic Current-Voltage Characteristics of III-N HFETs, IEEE Electron
Device Letters, vol. 24, no. 11, pp. 680-682, November 2003.

[56]

M. A. Khan, G. Simin, S. G. Pytel, A. Monti, E. Santi, and J. L. Hudgins, New


Developments in Gallium Nitride and the Impact on Power Electronics, IEEE
Conference on Power Electronics Specialists, pp. 15-26, June 12, 2005.

[57]

B. M. Green, H. Kim, V. Tilak, J. R. Shealy, J. A. Smart, and L. F. Eastman,


Validation of an Analytical Large Signal Model for AlGaN/GaN HEMTs on
SiC Substrates, 2000 IEEE/Cornell Conference on High Performance Devices,
vol. 2, pp. 237-241, August 2000.

[58]

S. Karmalkar, M. S. Shur, G. Simin, and M. Asif Khan, Field-Plate Engineering


for HFETs, IEEE Transactions on Electron Devices, vol. 52, no. 12, pp. 25342540, December 2005.

[59]

N. Braga, R. Mickevicius, R. Gaska, M. S. Shur, M. Asif Khan, and G. Simin,


Simulation of gate lag and current collapse in gallium nitride field-effect
transistors, Applied Physics Letters, vol. 85, no. 20, pp. 4780-4782, November
2004.

[60]

N. Braga, R. Mickevicius, R. Gaska, M. S. Shur, M. Asif Khan, and G. Simin,


Edge Trapping Mechanism of Current Collapse in III-N FETs, IEEE
International Electron Devices Meeting, pp. 815-818, December 13-15, 2004.

[61]

N. Braga, R. Mickevicius, R. Gaska, M. S. Shur, M. Asif Khan, and G. Simin,


Simulation of Gate Lag and Current Collapse in GaN Heterojunction Field

44

Effect Transistors, IEEE Compound Semiconductor Integrated Circuit


Symposium, pp. 287-290, October 24-27, 2004.
[62]

G. Verzellesi, R. Pierobon, F. Rampazzo, G. Meneghesso, A. Chini, U. K.


Mishra, C. Canali, and E. Zanoni, Experimental/Numerical Investigation on
Current Collapse in AlGaN/GaN HEMTs, IEEE International Electron Devices
Meeting, pp. 689-692, December 8-11, 2002.

[63]

G. Meneghesso, G. Verzellesi, R. Pierobon, F. Rampazzo, A. Chini, U. K.


Mishra, C. Canali, E. Zanoni, Surface-Related Drain Current Dispersion Effects
in AlGaN-GaN HEMTs, IEEE Transactions on Electron Devices, vol. 51, no.
10, pp. 1554-1561, October 2004.

[64]

G.H. Jessen, R. C. Fitch, J. K. Gillespie, G. D. Via, N. A. Moser, M. J. Yannuzzi,


A. Crespo, J. S. Sewell, R. W. Dettmer, T. J. Jenkins, R. F. Davis, J. Yang, M.
Asif Khan, and S. C. Binari, High Performance 0.14 m Gate-Length
AlGaN/GaN Power HEMTs on SiC, IEEE Electron Device Letters, vol. 24, no.
11, pp. 677-679, Nov. 2003.

45

Chapter 2

Introduction

The advantage of semiconductor device simulation is that the device can be


studied and optimized using commercially available software and then fabricated and its
performance verified experimentally.

This saves considerably on the cost of

development and the time needed to find the best device performance as compared to
fabricating many devices to experimentally determine the best performance and device
design. However, this necessitates accurate device modeling. To simulate the device
accurately, the material properties must be known as well as the incorporation of the
important relevant material physics. Since the use of AlGaN and GaN semiconductors is
relatively new compared to other semiconductors such as silicon, some of their material
properties are not well known or not measured. These material properties, when known,
are generally improving as researchers are developing new ways to grow higher quality
AlGaN and GaN with fewer defects and new experimental results. Since the use of
AlGaN/GaN is relatively new, the ISE software used in this study does not offer a well
established library of the material parameters so the material library had to be created by
incorporating the best available, known material properties into the softwares supported
physical models (NOTE: newer versions of the software do incorporate the AlGaN/GaN
material library). When no material properties were found for AlGaN or GaN or those
that could not be approximated by linear interpolation, between those for AlN and GaN,
silicon properties were used as default values instead. The first part of this chapter will
discuss the commercially available software used in this study, ISE TCAD; the second
part of this chapter will focus on the material properties and the incorporation of these
into the device physics models. The third section discusses the method of incorporating
46

the polarization physics into the device simulation, and the fourth section summarizes the
set of material properties used in this modeling study of AlGaN/GaN HEMTs.

2.1 Device Simulation Software


The commercial available software used for this study of the AlGaN/GaN HEMT
is ISE TCAD, which is capable of performing two- or three-dimensional device
simulations. The ISE TCAD package supports a set of tools in the simulation process
where GENESISe is employed as the user-interface for design, organization, and control
of all the simulation projects. GENESISe also serves as the framework for the design of
experiments (DOE) and device design optimization. Available and able to run under the
GENESISe framework are a fabrication process simulator, a device simulator, and
interfaces used to analyze and display the output characteristics of the device simulator.

2.1.1 Device Structure 2D Structure Editor


The device structure can initially be created using DIOS, a fabrication process
simulator in the ISE TCAD package, and then ported for use in the device simulator. An
alternative method for creating the device structure, which is used in this study, is also
available using MDRAW to draw the two-dimensional device structure and to specify the
material associated with each region, which is useful for distinguishing the electrodes,
substrate and the various semiconductors in the device structure. MDRAW can also be
used to define the mesh and the doping profiles present in the device. The mesh is the set
of vertical and horizontal lines that criss-cross in the device structure, the intersections of
which designate the grid points, where the program calculates the semiconductor devices
47

properties, such as carrier concentrations and electric potentials.

The device is

represented in the device simulator as a device whose physical properties are discretized
onto this non-uniform grid of points. Device physics equations are solved at each grid
point and the results are then used to calculate the devices output characteristics, such as
the terminal currents. Great care must be taken in defining the grid so that the device
simulator is able to get an accurate analysis of the simulated device performance. The
grid should be sufficiently dense enough so that the simulation accurately models the
device operation, yet not so dense that the simulation time is excessive. The grid should
be denser in areas of rapid variation in electron concentration, electric potential and
charge generation for good approximation of the device physics and operation. The
devices structure is input into the device simulator DESSIS as a geometry file and a data
file, which are the output generated by MDRAW. The geometry file contains the device
boundaries, material type for each region, the locations of electrical contacts, and the
locations of all the discrete nodes and their connectivity to adjacent nodes. The data file
contains the device doping profiles in the form of discrete nodes where the doping
between node points can be interpolated by the device simulator.

2.1.2 Device Simulator


Numerical simulations of the AlGaN/GaN HEMTs were performed using a
commercial device simulator DESSIS from Synopsis (formerly Integrated Systems
Engineering) [1]. Simulations done in this software model the physics of electron and
hole transport, and generation and recombination by solving Poissons equation along
with the electron and hole continuity equations at each node point, which results in the
electrostatic potential and carrier densities at each node. From these results, the flow of
48

carriers can be determined at each point in the device and the terminal characteristics of
the device are able to be calculated. The software is able to perform simulations for dc,
small-signal, and microwave operation of the device or multiple devices in a circuit using
a built-in SPICE model. The basic set of differential equations used to solve for the
electrical characteristics are described in the following sections. These equations are
used for all device simulations; an individual device is modeled by incorporating the
physical structure, composition and biasing for the particular device being investigated.

2.1.2.1 Poissons Equation


Poissons Equation is one of Maxwells equations which relates the electric field
to the charge distribution and is given by [1]

= q( p n + N D + N A )

(2.1)

where is the electrostatic potential, is the electrical permittivity, q is the magnitude of


the elementary electrical charge, n and p are the electron and hole densities, and ND+ is
the number of ionized donors and NA- is the number of ionized acceptors per unit volume.
Poissons equation is solved first to acquire the initial guess on the potential profile
throughout the device.

2.1.2.2 Continuity Equations


The continuity equations are based on the conservation of mobile charge with the
change in carrier concentration as it relates to the current flowing through a specific point
and the process of recombination and generation. The two continuity equations for
electrons and holes are written as [4]

49

r
n 1
= J n Rn + G n
t q

(2.2)

r
p
1
= J p Rp + Gp
t
q

(2.3)

where Jn is the total electron current density, Jp is the total hole current density , G is the
generation rate per unit volume, and R is the net electron-hole recombination rate per unit
volume.

2.2 Device Simulation Models


Discussed in the following sections are the physical models used in this study.
The version of DESSIS used in this work does not contain a built in material library for
GaN, AlN and AlGaN, so material libraries had to be created for each. The material
properties used in the library came from two primary sources [2, 3]. The majority of the
material properties for AlGaN were obtained by linear interpolation between the
properties for GaN and AlN by the DESSIS program using the general formula
PAl x Ga1 x N = (1 x )PGaN + xPAlN

(2.4)

where P is the material property, x is the aluminum mole fraction, where PAlGaN is the
resulting parameter value of AlxGa1-xN, PGaN is the parameter value for GaN, and PAlN is
the parameter value for AlN.

2.2.1 Carrier Statistics Model


By default, DESSIS uses Boltzmann statistics for electron and hole
concentrations. In heterostructure devices, such as the HEMT it is important to use
50

Fermi-Dirac statistics for the electron and hole concentrations. Fermi-Dirac statistics are
critical in areas where the carrier densities can reach values greater than 1x1019 cm-3,
which are reached in the HEMT channel or when the AlGaN buffer layer is doped. The
Fermi-Dirac distribution function gives the probability that an electronic state at energy E
is occupied by an electron and is given by [4]

f (E ) =

EE f

1+ e

(2.5)

kT

where Ef is the Fermi level, k is Boltzmanns constant, and T is the lattice temperature.

2.2.2 Transport Models


The electron and hole transport in semiconductors is modeled by the driftdiffusion model which characterizes the transport as separate drift and diffusion
components. The carrier transport for electrons and holes are written as [1]

r
J n = nq n n

(2.6)

r
J p = pq p p

(2.7)

where n and p are the electron and hole densities, q is the electrical charge, n and p are
the low field electron and hole mobilities, and n and p are the electron and hole quasiFermi potentials, respectively. The gradients of the quasi-Fermi potentials can be written
in one dimension as [4].
n =

d kT dn

dx qn dx

(2.8)

p =

d kT dp
+
dx qp dx

(2.9)

51

so that equations (2.6) and (2.7) can also be expressed as the sum of the drift and
diffusion current densities [4]
J n = qn n E + qDn

dn
dx

(2.10)

J p = qn p E qD p

dp
dx

(2.11)

where E is the electric field.


The drift-diffusion model is not the most accurate transport model available to
simulate AlGaN/GaN HEMTs, because the drift-diffusion model does not consider the
energy transport of the carriers and is not well suited for small active regions. Generally,
the drift-diffusion model is adequate for long channel devices, however, it does not
account for the hot electrons in AlGaN/GaN HEMTs. The hydrodynamic model offers a
better solution to account for the hot electrons in the device.

The use of the

hydrodynamic model in AlGaN/GaN HEMTs has been demonstrated by Braga et al. [5]
and is able in the DESSIS software to model the spreading of hot electrons into the bulk,
buffer layer, and traps present in the device. The negative differential conductance in the
output conductance is observed in simulations using the hydrodynamic models just like
the negative differential conductance is observed experimentally. The electron current
density for the more general, hydrodynamic model is written as [1]

r
J n = n nEC + k B Tn n + f ntd k B nTn 1.5nk B Tn ln me

(2.12)

where the first term includes the contribution due to the spatial variation of the
electrostatic potential, electron affinity, and the band gap where EC is the conduction
band energy. The second term is the contribution due to the gradient of concentration
(related to nonuniform doping).

The third term is the contribution of the carrier


52

temperature gradients, and the fourth term is the spatial variation of the electron effective
mass due to composition variations across regions of the device.

Although the

hydrodynamic model was used in the simulations, the computation of the electron
temperature was omitted due to a substantially longer simulation time that affected
getting results in a reasonable amount of time. Because of the omission of computing the
electron temperature, an electron temperature of 300 Kelvins was used in the
hydrodynamic model.

A more accurate, but more computationally time intensive

simulation would couple the electron temperature with the Poissons equation and
continuity equations to acquire a solution.

2.2.3 Recombination Models


The phenomenon of electron-hole recombination is modeled by the ShockleyRead-Hall recombination model. Although the model is used in the simulations, it has
relatively little effect on the output characteristics of the HEMT since it is a majority
carrier device. The Shockley-Read-Hall recombination model is given by [1]
R

SRH
net

np ni2,eff

p (n + n1 ) + n ( p + p1 )

(2.13)

where n and p are the electron and hole concentrations, and


Etrap

n1 = ni ,eff e

kT

(2.14)

and
Etrap

p1 = ni ,eff e

53

kT

(2.15)

where Etrap is the difference between the defect level and intrinsic level. In (2.13) the
minority lifetimes n and p are modeled as the product of the doping-dependent, fielddependent g(F), and temperature dependent f(T) factors expressed as [1]

n = dop

f (T )
1 + g n (F )

(2.16)

p = dop

f (T )
1 + g p (F )

(2.17)

The parameter values used for AlN and GaN in the Shockley-Read Hall model
are listed in Table 2.1.
Table 2.1

Shockley-Read-Hall Parameters for AlN and GaN [2].

Parameter

AlN

GaN

Unit

Etrap

eV

1x10-5 b

1x10-9

3x10-6 b

7x10-9

a. Etrap is determined by each individual trap level. See the following chapter for
specific Etrap levels for traps used in simulation.
b. DESSIS default values for silicon are used for this parameter
For AlN, the minority lifetimes of silicon were used since results are varying and
more research concerning this material needs to be published. The software does not
support mole fraction dependencies for Shockley-Read-Hall recombination parameters so
silicon values are used for AlGaN. It should also be noted that Shockley-Read-Hall
recombination is important in studying the transient effects of the AlGaN/GaN HEMT.
This is associated with the current collapse observed experimentally [6].

Transient

signals are used to study the trapped charge in the HEMT and their effects on the
54

behavior in microwave operation. Studying the traps in the AlGaN/GaN HEMT was the
initial goal in this study.

2.2.4 Band Structure


GaN and AlN are direct bandgap semiconductors with bandgaps of 3.497 eV [2]
and 6.242 eV [2], respectively. Due to their widebandgaps they can withstand high
electric fields and thus have high breakdown voltages making them suitable to high
power devices, such as the power HEMT.
The intrinsic bandgaps of AlN and GaN are dependent on temperature and are
described empirically using the function [1]

E g (T ) = E g (0)

T 2
(T + )

(2.18)

where Eg(0) is the band gap energy at 0 K, T is the lattice temperature in degrees Kelvin,
and and are fitting parameters, where the parameters used for GaN and AlN are listed
in Table 2.2 [2].
Table 2.2

Parameters for Bandstructure [2].

Parameters

AlN

GaN

Units

Eg(0)

6.242

3.497

eV

7.2x10-4

10-3

eV/K

5x102

103

1.9

3.1

eV

55

The bandgap of AlxGa1-xN is calculated by linear interpolation as a function of the


aluminum mole fraction x [1].
E g ( x ) = (1 x) * E g GaN + x * E g AlN

(2.19)

The intrinsic carrier density of the semiconductor for AlGaN and GaN is
calculated using the equation [1]
ni (T ) = N C (T )N V (T )e

E g (T )

(2.20)

2 kT

where NC is the conduction band effective density of states, NV is the valence band
effective density of states and Eg is the energy bandgap as a function of temperature
calculated from (2.18). The conduction and valence band effective density of states at
elevated temperatures are calculated from [1]
3

T 2
N C (Te ) = N C 300 * e cm 3
300

(2.21)

T 2
N V (Te ) = N V 300 * h cm 3
300

(2.22)

where NC300 and NV300 for GaN are 2.272x1018 cm-3 [2] and 4.6x1019 cm-3 [3], at T = 300
K, respectively. For AlN, the conduction and valence band effective density of states at
300 K are 4.174x1018 cm-3 [2] and 4.8x1020 cm-3 [3], respectively. These effective
densities of states for GaN and AlN are listed in Table 2.3.
Table 2.3

Effective Densities of States at 300 K for AlN and GaN [2, 3].

Parameters

AlN

GaN

Units

NC300

4.174x1018

2.272x1018

cm-3

NV300

4.8x1020

4.6x1019

cm-3

56

For AlGaN, the conduction and valence band effective density of states are
obtained by linear interpolation using [1]
N C 300 = (1 x ) * N C 300GaN + x * N C 300 AlN

(2.23)

N V 300 = (1 x ) * N V 300GaN + x * N V 300 AlN

(2.24)

where x is the composition of aluminum present in AlxGa1-xN. These effective densities


of states for AlGaN are dependent on the temperature and are calculated from their room
temperature values using (2.21) and (2.22). The effective carrier masses for GaN and
AlGaN are calculated from the effective densities of states at 300 K using [1]
2

me N C 300 3
=

mo 2.540 x1019

(2.25)

mh N V 300 3
=

mo 2.540 x1019

(2.26)

where me/mo is the effective electron mass and mh/mo is the effective hole mass
normalized to the free electron mass mo.

2.2.5 Mobility Models


The mobility is modeled in DESSIS by combining several components of the
mobility at low electric fields together using Mathiessens rule [1]. For this simulation,
two field components need to be considered, the low electric field mobility, which is
valid for electric fields where the carrier velocity is proportional to the electric field, and
the high electric field mobility where the carrier velocity reaches its saturation velocity at
high electric fields. The mobility at low electric fields is determined first and then the

57

final mobility is calculated from a formula that is a function of a driving force, i.e. the
electric field along the channel.

2.2.5.1 Low-Field Mobility Model


The low-field mobility is calculated from the constant mobility model which
assumes that the mobility is only affected by phonon scattering and thus is only
dependent on the lattice temperature for a given material. It is assumed that mobility
models incorporating impurity scattering are not needed in the simulation of AlGaN
HEMTs, as the device channel is located in the undoped (< 1015 cm-3) GaN layer where
very little scattering occurs due to scarcity of impurities. This is one of the strengths of
the HEMT device structure as there is little mobility degradation due to the lack of
impurities in the channel region. The low-field mobility model dependent on the lattice
temperature is given by [1]

const

T
= L
T0

(2.27)

where L is the mobility due to bulk phonon scattering, T is the lattice temperature, T0 =
300 K, and the exponent is a fitting parameter. The mobility models temperature
parameter values for GaN and AlN are listed in Table 2.4. The constant mobility of
AlGaN is obtained by linear interpolation based on the concentration of aluminum
present using [1]

const = (1 x ) const
where x is the aluminum concentration.

58

GaN

+ x const AlN

(2.28)

Table 2.4

AlN and GaN Parameters for Constant Mobility Model at Low Fields [2].

Parameter

AlN

GaN

Unit

Ln

300

1245

cm2/V-s

Lp

14

370

cm2/V-s

1.395a

1.107a

2.2a

1.652a

a. Parameter values calculated using Eq. 2.27 from data provided by [2].

2.2.5.2 High-Field Saturation Model


At high electric fields ( 104 V/cm), the carrier drift velocity saturates at a certain
velocity which is called the saturation velocity. The carrier drift velocity is no longer
proportional to the electric field at high fields and the mobility model must incorporate
this saturation of the velocity. Since the HEMT is a majority carrier device and the
electron saturation effect in the devices channel region is important, the hydrodynamic
Canali model is used for the electron mobility at high fields with the gradient of the
Fermi potential as the driving force. The Canali model used for the high-field mobility is
written as [1]

low

(F ) =

F
1 + low
v sat

(2.29)

and originates from the Caughey-Thomas model. In the equation low is the low field
mobility described above, F is the driving field, vsat is the saturation velocity and is a
fitting parameter. The fitting parameter is temperature dependent; the values used in
the simulation are those at 300 K which are given by Table 2.5.
59

Table 2.5

AlN and GaN Values for Parameter .

Parameter

AlN

GaN

Unit

81.79b

3.769b

1.213a

56b

a. DESSIS default values for silicon are used for this parameter
b. Parameter values calculated using Eq. 2.29 from data provided by [2].

The saturation velocity model is dependent on temperature and is expressed as [1]

v sat

T
= v sat , 0 0
T

v sat . exp

(2.30)

where T is the lattice temperature, T0 = 300 K. The velocity saturation model parameters
are listed in Table 2.6 for AlN and GaN.
Table 2.6

Velocity Saturation Model Parameters for AlN and GaN [2].

Parameter

AlN

GaN

Unit

vsat,0 n

1.5x107

1.9x107

cm/s

vsat.exp n

0.87a

0.87a

vsat,0 p

8.37x106 a

4.8x106

cm/s

vsat.exp p

0.52a

0.52a

a. DESSIS default values for silicon are used for this parameter
In the high-field hydrodynamic Canali model for the electron mobility, the
ordinary driving force (electric field) is replaced by a driving force that considers the
carrier thermal energy wc and is written as [1]

60

Fc =

wc w0
e ,c q

(2.31)

where wc = 3kBTc/2 is the average carrier thermal energy, w0 = 3kBTL/2 is the equilibrium
thermal energy, e,c is the energy relaxation time, Tc is the carrier temperature, and TL is
the lattice temperature. The Canali model then is written as [1]

low

(2.32)

2
2
1 + (wc w0 ) + (wc w0 )

when substituting the new driving force into the model. The fitting parameter is given
by [1].

1 low
=
2
2 q e,c v sat

(2.33)

In device simulations, the effect of the elevated temperature of the electrons in high
electric fields is incorporated self consistently within the calculations. To determine the
electron and hole mobilities at high electric fields in AlGaN, linear interpolation is
performed on and vsat as a function of the aluminum concentration.

2.3 Polarization Incorporated into Device


There are various methods used to incorporate the polarization effects of the
nitride materials in the AlGaN/GaN HEMT device simulation. Currently, the device
simulation software lacks the ability to model the piezoelectric and spontaneous
polarization physics from the physical description of the devices structure entered in
MDRAW. The effect on the output characteristics due to polarization physics in the
61

AlGaN/GaN HEMT has been discussed in Chapter 1 and is important to model in a


device simulation study of the devices performance.

Ambacher et al. [7, 8] have

developed analytical equations to describe the piezoelectric and spontaneous polarization


in an AlGaN/GaN heterostructure based on the lattice constants of the two layers for the
piezoelectric polarization and the composition of the aluminum in the AlxGa1-xN barrier
for the spontaneous polarization. They have also developed an analytical model to
determine the net sheet charge at the heterointerface due to the difference in polarization
between the AlGaN and GaN layers. From this net sheet charge equation, Ambacher et
al. [7, 8] have modeled an equation for the carrier concentration in the 2DEG. This
analytical model for the carrier concentration can be incorporated into the device
simulation software as the source of the 2DEG. This was done by another student in our
research group [9] who doped a small area of the AlGaN layer near the heterointerface
with the same concentration calculated by Ambachers carrier concentration equation [7,
8].
Another way to characterize the source of the 2DEG is to utilize a simple charge
balance equation developed by Ibbetson et al. [10], which is the approach that was taken
for this study. The charge balance equation proposed by Ibbetson et al. takes the form
[10]

surface pz + pz + AlGaN qn s buffer = 0

(2.34)

and includes the negative charge (-qns) due to the sheet carrier concentration in the
2DEG, the polarization-induced charges at the AlGaN/GaN interface (+pz) and the
AlGaN surface (pz), the integrated sheet charge due to ionized donors in the AlGaN
(+AlGaN), the charge due to ionized surface states (surface), and the buffer charge (buffer).

62

A few assumptions are made concerning the buffer charge; first the charge should be
negative otherwise the 2DEG would not be attractive to the +pz at the AlGaN/GaN
interface and would not be confined to the interface region, and second the buffer charge
should be relatively small in good design practice so it can be assumed that buffer 0
[10]. The charge balance equation can then be simplified to the form

surface pz + pz + AlGaN qn s = 0

(2.35)

where the sheet charge concentration present in the 2DEG is due to the charge associated
with ionized surface states, the ionized donors in the AlGaN buffer layer, along with the
polarization charges. The 0.14 m gate length AlGaN/GaN HEMT built by G. H. Jessen
et al. [11] incorporates an AlGaN buffer layer that is unintentionally doped which
approximates the charge associated with ionized donors in the AlGaN buffer layer to
zero. The charge balance equation for this study simplifies to

surface pz + pz qn s = 0

(2.36)

where the sheet charge concentration in the 2DEG is due to the charge associated with
the surface states and polarization charges.

The location of these charges in the

AlGaN/GaN HEMT is shown in the conduction band diagram provided in Figure 2.1.
The surface state charge and polarization charges in (2.36) are incorporated in the
simulations to generate the 2DEG sheet concentration, neglecting some bulk trap charges
that are included in the simulation.

The charge balance equation can be further

simplified, if the dipole of the polarization is considered, the polarizations charges cancel
each other. The simplified equation is given by
qn s surface

63

(2.37)

where the source of the electrons in the 2DEG can be largely attributed to the ionized
surface states.

Charge Balance Equation incorporated into Simulation


surface pz + pz qns = 0

2DEG (qns)
Surface
States (surface)

AlGaN

EC
EF
GaN

Figure 2.1

Polarization Charge

Polarization Charge

(-pz)

(+pz)

Conduction band diagram of an AlGaN/GaN HEMT showing the various space


charge components modeled into the device simulation from the modified charge
balance equation in [10]. The electrons in the 2DEG are generated by the surface
state charge and polarization charges as in (2.36).

The charge due to the surface states for this study were modeled into DESSIS by
placing surface donor traps at the AlGaN surface with a concentration of 1.14x1013 cm-2
[11] and at an energy of 1.65 eV [10] below the conduction band with electron and hole
cross-sections of 1x10-15 cm2. In DESSIS, charges and traps can not be placed at surfaces
but can be placed at interfaces. So to place the donor traps at the top AlGaN surface, an
interface was created by placing 100 Angstroms of Si3N4 on top of the AlGaN gate layer.
64

The negative polarization charge which is the combined piezoelectric and spontaneous
polarization was placed at the AlGaN surface by using the same Si3N4 and AlGaN
interface with a charge density of 1.14x1013 cm-2 [11]. The positive polarization charge
was placed at the AlGaN/GaN interface with a charge density of 1.14x1013 cm-2 [11].
The positive polarization charge is what attracts the 2DEG to the AlGaN/GaN interface.
Table 2.7 lists the concentrations of these charges as well as their locations for the
simulated AlGaN/GaN HEMT.
Table 2.7

Description of Polarization and Surface State Charge Components in Simulated


AlGaN/GaN HEMT.

Symbol

Description

Location

Density

surface

Donor Surface

At Si3N4/AlGaN

1.14x1013 cm-2

States Charge

interface, 1.65 eV
below conduction
band, electron and
hole concentrations
of 1x1015 cm2

-pz

+pz

Negative

At Si3N4/AlGaN

Polarization Charge

interface

Positive Polarization At AlGaN/GaN


Charge

interface

65

1.14x1013 cm-2

1.14x1013 cm-2

2.4 Material Parameters Used in Study


The material parameters used for AlN, GaN, and Al0.34Ga0.66N used in this study
are summarized in Table 2.8. For the majority of the parameter values for Al0.34Ga0.66N,
they were obtained by linear interpolation between those parameter values for GaN and
AlN by the simulation software. All parameter values are for simulations done at 300 K.
Table 2.8

Parameter Values for AlN, GaN, Al0.34Ga0.66N at 300 K used in this study.

Parameter

AlN

GaN

Al0.34Ga0.66N

Unit

8.3

9.04

8.79

Eg

6.161

3.4277

4.357

eV

1.9

3.1

2.69

eV

ni

1.672x10-33

4.35286x10-11

1.46532x10-18

cm-3

NC300

4.174x1018

2.272x1018

2.919x1018

cm-3

NV300

4.8x1020

4.6x1019

1.936x1020

cm-3

me/mo

0.3

0.2

0.236

mh/mo

7.09

1.486

3.873

const, n

300

1245

923.7

cm2/V-s

const, p

14

370

248.96

cm2/V-s

81.79

3.769

30.2961

1.213

56

37.3724

vsat,n

1.5x107

1.9x107

1.76x107

cm/s

vsat,p

8.37x106

4.8x106

6.01x106

cm/s

1x10-5

1x10-9

9x10-6 a

3x10-6

7x10-9

2.6x10-6 a

a. Silicon values were used for this parameter.

66

The following chapter will discuss the results of the device simulations and
compare them to the experimental work previously reported [11]. A subsequent chapter
will describe the optimization of the basic device structure.

67

References
[1]

DESSIS ISE TCAD Manual, Release 8.0 (ISE Integrated Systems Engineering
AG, Zurich 2002).

[2]

S. Adachi, Handbook on Physical Properties of Semiconductors: Volume 2, II-V


Compound Semiconductors, Kluwer Academic, Boston, 2004.

[3]

Properties of Advanced Semiconductor Materials: GaN, AlN, InN, BN, SiC, and
SiGe, edited by M. E. Levinshtein, S. L. Rumyantsev, and M. S. Shur, John Wiley
& Sons, New York, 2001.

[4]

Y. Taur and T. H. Ning, Fundamentals of Modern VLSI Devices, Cambridge


University Press, Cambridge, 1998.

[5]

N. Braga, R. Mickevicius, R. Gaska, X. Hu, M. S. Shur, M. Asif Khan, G. Simin,


and J. Yang, Simulation of hot electron and quantum effects in AlGaN/GaN
heterostructure field effect transistors, Journal of Applied Physics, vol. 95, no.
11, pp. 6409-6413, June 2004.

[6]

R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, The Impact of Surface


States on the DC and RF Characteristics of AlGaN/GaN HFETs, IEEE
Transactions on Electron Devices, vol. 48, no. 3, pp. 560-566, March 2001.

[7]

O. Ambacher, B. Foutz, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M.


Murphy, A. J. Sierakowski, W. J. Schaff, L. F. Eastman, R. Dimitrov, A.
Mitchell, and M. Stutzmann, Two-dimensional electron gases induced by
spontaneous and piezoelectric polarization in undoped and doped AlGaN/GaN
heterostructures, Journal of Applied Physics, vol. 87, no. 1, pp. 334-344, January
2000.
68

[8]

O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy, W. J.


Schaff, L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, and J.
Hilsenbeck, Two-dimensional electron gases induced by spontaneous and
piezoelectric

polarization

charges

in

N-

and

Ga-face

AlGaN/GaN

heterostructures, Journal of Applied Physics, vol. 85, no. 6, pp. 3222-3233,


March 1999.
[9]

A. C. Appaswamy, Simulation of Short Channel AlGaN/GaN HEMTs,


University of Cincinnati, February 2005.

[10]

J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.


Mishra, Polarization effects, surface states, and the source of electrons in
AlGaN/GaN heterostructure field effect transistors, Applied Physics Letters, vol.
77, no. 2, pp. 250-252, July 2000.

[11]

G.H. Jessen, R. C. Fitch, J. K. Gillespie, G. D. Via, N. A. Moser, M. J. Yannuzzi,


A. Crespo, J. S. Sewell, R. W. Dettmer, T. J. Jenkins, R. F. Davis, J. Yang, M.
Asif Khan, and S. C. Binari, High Performance 0.14 m Gate-Length
AlGaN/GaN Power HEMTs on SiC, IEEE Electron Device Letters, vol. 24, no.
11, pp. 677-679, Nov. 2003.

69

Chapter 3

Device Modeling of 0.14 m Gate-Length


AlGaN/GaN Power HEMT

This chapter will discuss the experimental device used as the starting point in this
simulation study of the effects of the device structure on the performance of the
AlGaN/GaN HEMT. Various profiles will be discussed with respect to the devices
cross-sectional structure for parameters such as the band gap and electron concentration.
The second part of this chapter will discuss the dc characteristics of the simulated device
and compare them to the reported experimental results of Jessen et al. [1]. Small-signal
simulations and the devices high frequency performance will also be discussed in the
last section and similarly compared to published experimental results.

3.1 Device Structure


The AlGaN/GaN HEMT simulated in this study was chosen to follow the basic
structure of the experimental device reported by Jessen et al. [1]. For the simulations, the
SiC substrate was neglected for the 0.14 m gate-length AlGaN/GaN HEMT since it does
not influence the devices performance and provides only a mechanical substrate for the
GaN growth and device handling. The device structure used in the simulations is shown
in Figure 3.1 where the devices lateral dimensions are given in Table 3.1.

70

Si3N4
LG

LS
LGS

Source

Gate

LD
LGD

Drain

AlGaN

GaN

Figure 3.1

AlGaN/GaN HEMT device structure used in this modeling following that reported
by Jessen et al. [1].

Table 3.1

Important Lateral Dimensions of Modeled AlGaN/GaN HEMT following Jessen et


al. [1]

Source Contact Length (LS)

0.2 m

Gate-to-Source Spacing (LGS)

1.23 m

Gate Contact Length (LG)

0.14 m

Gate-to-Drain Spacing (LGD)

1.23 m

Drain Contact Length (LD)

0.2 m

71

Table 3.2 provides the epitaxial layer details including the thickness, type and
doping concentration for each layer.
Table 3.2

Specifics of the Epitaxial Layers of the Modeled AlGaN/GaN HEMT [1] after Jessen
et al [1]

Layer

Thickness

Silicon Doping, n-type

Si3N4

100 Angstroms

Al0.34Ga0.66N

260 Angstroms

1x1014 cm-3

GaN

1.2 m

1x1014 cm-3

The specifics of how the polarization and surface state charges were incorporated
into the device structure used in the simulations are listed in Table 3.3. Also included in
the table, are the trap specifics for the bulk traps in the Al0.34Ga0.66N and GaN layers
which were employed in the simulation. A more detailed discussion of the modeled
polarization and surface charge which generate the two-dimensional gas (2DEG) in the
HEMT can be found in the previous chapter. The polarization charges are introduced as
sheet charges at the Al0.34GaN0.66/GaN heterointerface and Si3N4/Al0.34Ga0.66N interface,
while the bulk traps in the Al0.34Ga0.66N and GaN layers are introduced as traps in the
volume. The surface state charge is introduced as donor traps at the Si3N4/Al0.34Ga0.66N
interface with an energy level of 1.65 eV below the conduction band [2]. The electron
and hole cross-sections for the traps modeled into the device were assumed to be 1x10-15
cm2 [3, 4]. The Schottky barrier height for the gate was set to 1 eV for the simulations.
Parasitic series resistance at the drain and source were considered and a contact resistance
of 1.46 -mm was chosen for the drain and source contacts. Jessen et al. [1] reported

72

variations in the contact resistance from 0.24 to 4.41 -mm with an average value being
2.68 0.98 -mm. The contact resistance used in the simulations was determined by
trying several contact resistances in the specified range and choosing the contact
resistance whose simulated drain current versus drain voltage slope in the linear range
matched closely with the experimentally reported drain current versus drain voltage
characteristics reported by Jessen et al. [1].

Table 3.3

Location

AlGaN

GaN

AlGaN/GaN

Trap Specifications Used in Modeling of AlGaN/GaN HEMT after Jessen et al. [1]

Type

Concentration

Trap

Electron

Hole

Energy

Cross-

Cross-

Level

Section

Section

Acceptor

5x1015 cm-3

Ei

1x10-15

1x10-15

Donor

1x1015 cm-3

Ei

1x10-15

1x10-15

Acceptor

5x1015 cm-3

Ei

1x10-15

1x10-15

Donor

1x1015 cm-3

Ei

1x10-15

1x10-15

Sheet

1.14x1013 cm-2

-1.14x1013 cm-2

1.14x1013 cm-2

EC - 1.65

1x10-15

1x10-15

Heterointerface Charge
Si3N4/AlGaN

Sheet

Interface

Charge
Donor

eV

73

The simulated device structure and mesh for dc analysis are shown in Figure 3.2
and Figure 3.3, respectively.

Figure 3.2

Device structure used for simulation of 0.14 m gate-length AlGaN/GaN HEMT [1].

Figure 3.3

Device mesh used in dc simulations of AlGaN/GaN HEMT [1].

74

The vertical spacing of the mesh in the channel region in Figure 3.3 is 4.69 nm, a
smaller vertical spacing was prevented due to convergence problems. A more detailed
view of the mesh in the channel under the gate region is illustrated in Figure 3.4.

Figure 3.4

Detailed look at the spacing of the mesh under the gate region for the simulated
AlGaN/GaN HEMT [1]. The vertical spacing in the channel is 4.69 nm.

Denser vertical spacing of the mesh was used in the AlGaN layer and the upper
portion of the GaN layer near the heterointerface where the band bending is the most
rapid, with the densest area being these layers under the gate contact.
The simulated band structure under the gate is shown in Figure 3.5 for a bias of 0
V at the gate and 0 V at the drain. In the diagram the gate contact is located at 0 m on
the x-axis and the AlGaN layer thickness under the gate is 26 nm or 0.026 m, so that the
bottom of the notch is located at the AlGaN/GaN heterojunction interface. Figure 3.5 (b)
shows an expanded view of the energy band structure near the interface.

75

Figure 3.5

Simulated band energy structure under the gate for a bias of 0 V on the gate and 0
V on the drain, with a) showing the whole structure and b) showing an expanded
view of the band structure around the heterostructure. The Si3N4 thickness is 100
Angstroms and starts at 0 m on the plot and the AlGaN thickness is 260
Angstroms, with the AlGaN/GaN interface located at 0.036 m.

76

Conduction band profiles under the gate at various gate biases are plotted in
Figure 3.6 for zero drain bias.

Figure 3.6

Simulated conduction band profiles under the gate at various gate voltages for no
drain bias.

At zero drain bias, the two-dimensional electron gas (2DEG) is formed at the gate
biases of -3 V and +1 V (The Fermi level is located at 0); however the 2DEG does not
exist at a gate bias of -7 V. In contrast, Figure 3.7 shows the conduction band profiles at
high drain biases (+10 V), which allows for the 2DEG to exist and drain current
conduction at a gate bias of -7 V (see Figure 3.7). It should be noted that if the lateral
mesh spacing along the x-direction in Figure 3.4 is reduced under the gate, no current
conduction is observed at a gate bias of -7 V.

77

Figure 3.7

Simulated conduction band profiles under the gate at various gate voltages for a
drain bias of 10 V.

Figure 3.8 shows the electron concentration profile under the gate for a bias of 0
V on the gate and 0 V on the drain illustrating the two dimensional nature of the
distribution, i.e. its localization of the electrons at the AlGaN/GaN heterointerface. The
peak electron concentration located at the heterointerface in the 2DEG is approximately
1.3x1019 cm-3. Electron concentration profiles at various gate biases and with the drain
bias kept at 0 V are plotted in Figure 3.9. The electron concentration at a gate bias of -7
V is too small to be seen as the 2DEG has yet to form (see Figure 3.6). At a gate voltage
of +1 V, the electron concentration reaches 1.5x1019 cm-3 in the channel. When a drain
bias is applied as in Figure 3.10, the peak electron concentrations in the channel are lower
than when no drain bias is applied. The electrons also spread into the bulk GaN as can be
especially seen for the gate bias of -3 V.

78

Figure 3.8

Simulated electron concentration profile under the gate for a bias of 0 V at the gate
for 0 V at the drain.

Figure 3.9

Simulated electron concentration profile under the gate for various gate biases at a
drain bias of 0 V.

79

Figure 3.10

Simulated electron concentration profiles under the gate for various gate biases at a
drain bias of 10 V.

3.2 DC Device Characteristics


3.2.1 Drain Current Gate Voltage Characteristics
The drain current (Ids) versus gate voltage (Vgs) characteristics of the simulated
AlGaN/GaN HEMT are shown in Figure 3.11 for both the linear and logarithmic current
scales. The peak current density for the simulated device is 1.977 A/mm which is
achieved with a gate voltage of 1 V and the drain bias kept at 10 V. The simulated
threshold voltage of the device was determined by extrapolating the drain current to zero
for the linear scale using the slope at its maximum, yielding a threshold voltage of -6.6 V.

80

Also included in Figure 3.11, is the drain current plotted on a logarithmic scale against
gate voltage. The peak transconductance (gm) was obtained at a gate voltage of -2.17 V
and found to be 276 mS/mm by taking the difference in the drain current for a small
difference in the gate voltage.

Figure 3.11

Simulation results for the drain current plotted against voltage on linear and
logarithmic scales where the drain voltage was +10 V.

For comparison, Figure 3.12 shows the experimental results of the drain current
(Ids) versus the gate voltage (Vgs) reported by Jessen et al. [1]. The reported maximum
current density for the device was 1.352 A/mm and the reported maximum
transconductance was 295 mS/mm. The simulated peak drain current of 1.98 A/mm is
46% larger than the experimental peak drain current of 1.35 A/mm. The simulated peak
gm is 6% smaller than the experimental peak gm reported by Jessen et al. [1] for the
81

standard device structure. The threshold voltage extracted from the experimental linear
drain current versus gate voltage plot in Figure 3.12 was taken to be ~ -6.0 V, which is
approximately 10% smaller than the -6.6 V obtained from our simulation results. For an
offset device, with the gate shifted 0.75 m towards the source, a maximum current
density of 1.396 A/mm and maximum transconductance of 320 mS/mm were reported.

Figure 3.12

Experimental transfer characteristics of 0.14 m gate-length AlGaN/GaN HEMTfor


a drain bias of 10 V as reported by Jessen et al. [1].

A direct comparison of the simulated and experimental drain current (Ids) versus
gate voltage (Vgs) results is shown in Figure 3.13. The simulation results show higher
current density at each gate bias with a maximum current density of 1.98 A/mm
compared to the maximum experimental current density of 1.35 A/mm. The origin of the
difference between the simulation and experimental curves will be discussed in more
detail subsequently.

82

Figure 3.13

Comparison of simulated and experimental drain current plotted against gate


voltage for drain biased at +10 V using the experimental results of Jessen et al. [1].

3.2.2 Transistor Output Characteristics


The simulated drain current (Ids) versus drain voltage (Vds) characteristics at
various gate biases is shown in Figure 3.14 where the gate voltage is stepped up from -7
V to +1 V in increments of 2 V and the drain voltage is swept from 0 V to 10 V for each
incremental gate bias.

83

Figure 3.14

Simulated dc drain current versus drain voltage results where the gate voltage was
swept from -7 V to +1 V in 2 V increments.

The simulated Idss value of the device at zero volts on the gate and 10 V on the
drain is 1.80 A/mm. At a gate bias of -7 V, there is some current at high drain voltages,
from Vds = 6 to 10 V. This observed conductance in the simulation can be attributed to
the conduction band bending below the Fermi level as seen in Figure 3.7.

This

conduction band bending at Vgs = -7 V is insignificant at zero drain bias where no 2DEG
is formed as seen in Figure 3.6.
The experimental output characteristics for the same transistor are shown in
Figure 3.15 where the gate voltage is again stepped from -7 V to +1 V in 2 V increments
and the drain voltage is swept from 0 to 10 V.

84

Figure 3.15

Experimental drain current versus drain-source voltage of 0.14 um gate-length


AlGaN/GaN HEMT as reported by Jessen et al. [1] where the gate voltage was
swept from -7 V to +1 V in 2 V steps.

The maximum current density is approximately 1.35 A/mm at a gate bias of +1 V


and a drain bias of 10 V as compared to the 1.98 A/mm obtained from the simulations, a
difference of 46%. The drain current density at Vgs = -7 V shows very little conductance
until around 8 V on the drain, although the current density is much less than the
simulated current density at the same gate bias as shown in Figure 3.14.
A more direct comparison of the output characteristics for the simulated and
experimental results is shown in Figure 3.16.

85

Figure 3.16

Simulation results for the drain current plotted with the experimental results from
Jessen et al. [1] where the gate voltage was swept from -7 to +1 V in 2 V increments.

As can be seen, the simulation results have significantly higher values than the
experimental results at each gate and drain voltage, the most significant of which is the
simulation results for Vg = +1 V, which is 0.625 A/mm or 46% larger than the
experimental results at Vds = 10 V. The knee voltage of the simulated device is slightly
higher and the simulated HEMT requires a higher drain voltage to drive the device into
saturation. When the device is biased at a gate voltage of -3 V, the experimental knee
voltage is around 3 V and the simulated knee voltage is around 4 V, a 33% difference.
With 1 V biased on the gate, the knee voltage is harder to define. For the experimental
device, the knee voltage is around 6 V, while for the simulated device, it is debatable if
the device has reached saturation.
86

The possible origins of the discrepancies between the simulation and experimental
results can be attributed to several possible sources. First, the series resistance of the
source and drain contacts were varied in the simulation until a close match of the
transistor characteristics in the experimental device was found for the linear region. A
simulated series resistance of 1.46 -mm was chosen for the best match in the reported
experimental range of 0.24 to 4.41 -mm. For the experimental results in Figure 3.15,
no specific contact resistance was reported by the authors [1] so it is unknown whether
the simulated contact resistance is the same as the experimental contact resistance. It
should be noted, that a higher contact resistance will result in the lowering of the drain
current. Second, the experimental device may have some self-heating effects which
degrade the mobility and the velocity saturation, both of which reduce the current
achievable in the device. In this work, self-heating effects were not modeled, which
could explain our overestimation of the currents at all biases. Third, the method of
incorporating the polarization and charges in the device may also be a cause of
differences in the output conductance. The device simulator, DESSIS does not currently
offer a way to directly incorporate the polarization physics. As discussed in the previous
chapter, to incorporate the effects of polarization in the device, sheet charges were placed
at the AlGaN/GaN interface and Si3N4/AlGaN interface. The polarization sheet charge
densities chosen for the simulation were the same as that reported in the experimental
device for the 2DEG sheet charge density, 1.14x1013 cm-2 [1]. This is not necessarily the
most accurate way to model the polarization, because the polarization charges may not be
equal to the 2DEG sheet charge density in the real device. Also, the main source of the
electrons in the simulated device comes from the donor traps at the Si3N4/AlGaN surface.

87

The charge density of these traps were taken to be the same as the 2DEG sheet charge
density reported experimentally [1] and the energy location of these traps was set at 1.65
eV from the conduction band following the report of Ibbetson et al. [2]. In practice, there
may be multiple traps in the devices AlGaN layer so using only one trap over simplifies
the case. The last thing is that the experimental device may have defects introduced in
the fabrication process, which can not be accounted for in the simulations.

3.3 Microwave/Small-Signal Characteristics


The small-signal simulations of the AlGaN/GaN HEMT used a slightly different
mesh than the mesh used for the dc simulations. This was due to convergence issues
when using the same mesh for small-signal simulations. The small-signal mesh features
better vertical refinement in the channel region, but less mesh refinement in the bulk.
Thus the accuracy in dc simulations using this second mesh is not quite as refined as the
dc results discussed above, specifically for the transfer characteristics of the drain current
versus gate voltage.

3.3.1 Small-Signal Transconductance


As shown in Figure 3.17, the simulated small-signal transconductance peaks at
Vg = -2.15 V with a maximum transconductance of 289 mS/mm. These results should be
compared to the experimental transconductance seen in Figure 3.12, where the
experimental transconductance has a sharp peak at 295 mS/mm that is observed at Vg =
-4.75 V.

88

Figure 3.17

Simulated small-signal transconductance (gm) and small-signal output conductance


(go) of the AlGaN/GaN HEMT. The frequency of the small signal was set to 100
MHz.

The authors [1] reported the average peak experimental gm was 295 mS/mm in the
standard HEMT design and 320 mS/mm in the offset HEMT design. These values are
comparable but larger than the observed peak gm in simulations of 289 mS/mm by 2%
and 10%, respectively. While, the simulated and experimental peak transconductances
are very similar, the location of the peak in the transconductance curves is not as
agreeable. The simulated gm curve is smoother and broader than the experimental curve
and does not experience a smaller, secondary peak. The peak transconductance of the
simulated device is also at a higher voltage (-2.17 V) than the experimental device (-4.8
V). This dip and secondary peak in the experimental transconductance was proposed to
89

be due to the device being unpassivated and the presence of trapped charge located
between the gate and drain [1].
Also included in Figure 3.17 is the simulated output conductance of the
AlGaN/GaN HEMT plotted against the gate voltage with the drain voltage at 10 V. An
experimental output conductance of 25 mS/mm was extracted from Figure 3.15 for a gate
bias of 1 V. In comparison at this same bias, the simulated output conductance is 145
mS/mm, 480% larger than the experimental value.

At a gate bias of -3 V, the

experimental output conductance as extracted from Figure 3.15 to be 41 mS/mm. By


comparison, the simulated output conductance was found to be 29 mS/mm, 29% smaller
than the experimental value.

3.3.2 Cutoff Frequency (fT)


The cutoff frequency of a transistor is the frequency where the small-signal
current gain reaches unity. The cutoff frequency takes the general form in the HEMT [5]

fT =

gm
2Ci

(3.1)

where gm is the transconductance and Ci is the gate capacitance. The fT can also be
expressed as [5]
fT =

1
2 c

(3.2)

where c is the electron transit time through the channel and is expressed as [5]

c =

L
vs

(3.3)

where L is the gate length and vs is the electron drift velocity. From (3.3), the cutoff
frequency is determined by the gate length of the device; the shorter the gate length, the
90

more the saturation velocity is maximized for improved frequency performance. The
cutoff frequency is also dependent on the gm as seen from (3.1). Thus the bias of the
device affects the gm (see Figures 3.12 and 3.17) and so the cutoff frequency.
The cutoff frequency of the simulated HEMT is extracted from the current gain
plot versus frequency as shown in Figure 3.18.

Figure 3.18

Simulated current gain and unilateral power gain plotted against frequency where
the gate is biased at -3.5 V and the drain is biased at 10 V.

In this case the cutoff frequency is extracted to be approximately 266 GHz. The
fT is a function of the gate bias and Figure 3.19 shows the simulated fT plotted as a
function of the gate bias.

91

Figure 3.19

Simulated cutoff frequency (fT) and maximum oscillation frequency (fmax) plotted as
functions of the gate voltage where the drain voltage was kept at 10 V.

The peak cutoff frequency of 266 GHz is reached around -3.5 V on the gate,
which corresponds to 1.69 A/mm drain current where the drain was biased at 10 V in the
simulations. The experimental results report an average fT of 82 GHz for the standard
device and 87 GHz for the offset device at the same gate biasing as the peak gm [1]. The
simulated fT is much larger than the reported experimental value, 224% larger for the
standard device structure and 206% larger for the offset device structure. One source of
the difference in simulated and experimental cutoff frequencies is that the simulations do
not include all of the parasitics, specifically the parasitic gate capacitance that is
important as seen in (3.1). The only parasitics included in the simulations are the source
and drain contact resistances. The lack of parasitics in the simulations will also cause a
92

higher maximum oscillation frequency than would be expected if parasitics were


included.

3.3.3 Maximum Oscillation Frequency (fmax)


The maximum oscillation frequency (fmax) was also simulated for the AlGaN/GaN
HEMT. No fmax was reported for the experimental device so there are no comparisons to
be made between the experimental and simulated results. The unilateral power gain as a
function of frequency is plotted in Figure 3.18 along with the current gain.

The

maximum oscillation frequency of the device was extracted from this plot to be around
293 GHz at a gate bias of -3.5 V and 10 V on the drain. Figure 3.19 shows the simulated
fmax plotted as a function of the gate voltage. The fmax peaks at 298 GHz at a gate voltage
of -4.4 V. The simulated fmax is significantly larger than the recently reported maximum
experimentally reported fmax of 230 GHz which was achieved by introducing an InGaN
back-barrier into the design of the AlGaN/GaN HEMT [6].

3.4 Summary and Conclusion


In this chapter, the performance of the AlGaN/GaN HEMT was reported for the
simulated device structure following that used the experimental device by Jessen et al.
[1]. Simulations of this device were performed for dc and small-signal analysis and
compared to experimental results reported in [1]. The simulation results are in reasonable
agreement with the experimental results, with a few exceptions, which include the cutoff
frequency and output characteristics at high gate (> 0 V) and high drain bias (> 8 V).
In the next chapter, the results of device simulations are reported when changing
the polarization charges at the AlGaN surface and at the heterointerface and compared to
93

the simulation results reported in this chapter. This analysis is important as it shows the
role of the polarization in the device when changing the magnitude of the charge by a
slight amount. Similarly, the concentration of the surface traps will also be examined in
the next chapter. Finally, in the following chapter, structural parameters are altered to
demonstrate the effect of these parameters on the device performance. For comparison,
the results of these simulations will be compared to the simulation results reported in this
chapter. The goal is to show which parameters can be changed to improve the device
performance.

94

References
[1]

G.H. Jessen, R. C. Fitch, J. K. Gillespie, G. D. Via, N. A. Moser, M. J. Yannuzzi,


A. Crespo, J. S. Sewell, R. W. Dettmer, T. J. Jenkins, R. F. Davis, J. Yang, M.
Asif Khan, and S. C. Binari, High Performance 0.14 m Gate-Length
AlGaN/GaN Power HEMTs on SiC, IEEE Electron Device Letters, vol. 24, no.
11, pp. 677-679, Nov. 2003.

[2]

J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.


Mishra, Polarization effects, surface states, and the source of electrons in
AlGaN/GaN heterostructure field effect transistors, Applied Physics Letters, vol.
77, no. 2, pp. 250-252, July 2000.

[3]

G. Meneghesso, G. Verzellesi, R. Pierobon, F. Rampazzo, A. Chini, U. K.


Mishra, C. Canali, E. Zanoni, Surface-Related Drain Current Dispersion Effects
in AlGaN-GaN HEMTs, IEEE Transactions on Electron Devices, vol. 51, no.
10, pp. 1554-1561, October 2004.

[4]

N. Braga, R. Mickevicius, R. Gaska, X. Hu, M. S. Shur, M. Asif Khan, G. Simin,


and J. Yang, Simulation of hot electron and quantum effects in AlGaN/GaN
heterostructure field effect transistors, Journal of Applied Physics, vol. 95, no.
11, pp. 6409-6413, June 2004.

[5]

G. Gonzalez, Microwave Transistor Amplifiers: Analysis and Design, Prentice


Hall Inc., Englewood Cliffs, N. J., 1984.

[6]

T. Palacios, A. Chakraborty, S. Heikman, S. Keller, S. P. DenBaars, and U. K.


Mishra, AlGaN/GaN High Electron Mobility Transistors With InGaN BackBarriers, IEEE Electron Device Letters, vol. 27, no. 1, pp. 13-15, January 2006.
95

Chapter 4

Device Optimization

The simulation results of an AlGaN/GaN HEMT were discussed in Chapter 3


using the structure given by Jessen et al. [1] as a starting point for comparison to their
experimental results. In this chapter, various transistor structural parameter values are
altered from the values used in Chapter 3 as a means to pursue optimizing the device.
Comparisons are made of the simulations presented in this chapter with those using the
baseline structure in the previous chapter. For each device parameter that is altered, the
effect on the transistors transconductance, dc performance, cutoff frequency, and
maximum oscillation frequency are discussed.
The first parameter that is investigated is the polarization charge density
representing the sum of the spontaneous and piezoelectric polarization charge at the
AlGaN/GaN interface, but also its complementary charge at the top AlGaN surface.
Second, along with polarization charge density, the donor trap density at the
AlGaN/Si3N4 interface is varied to observe its effects on the device performance. Third,
the source and drain contact resistances are varied in a range reported in [1] for different
experimental devices. Fourth, the AlGaN thickness is varied and finally, the source-gate
and source-drain separations are varied for the device.

4.1 Polarization Charge Density


As discussed in Chapter 1, the formation of the two-dimensional electron gas
(2DEG) in the AlGaN/GaN HEMT is dominated by polarization effects [2]. In this
section, the charge densities of these polarizations are varied at the AlGaN/GaN

96

heterointerface and its complementary charge at the AlGaN surface in contact with a
Si3N4 layer in order to understand how the device performance is affected. Recalling the
charge balance equation from Chapter 2 that was introduced by Ibbetson et al. [3]

surface pz + pz + AlGaN qn s buffer = 0

(4.1)

where the negative charge due to the sheet carrier density in the 2DEG is represented by qns, the polarization-induced charges at the AlGaN/GaN interface and the AlGaN surface
are represented as +pz and pz, respectively, the integrated sheet charge due to ionized
donors in the AlGaN is represented by +AlGaN, the charge due to ionized surface states is
represented by surface, and the buffer charge is represented by buffer. Using this equation,
it would be expected that the polarization charges should negate each other and not have
much effect on the device performance when varied. However, this charge balance
equation is relatively simple and does not consider other effects going on in the device.
As shown in the results presented below, the variation of this polarization charge does
change the devices performance.
An increase in the magnitude of the polarization charge density at the
AlGaN/GaN heterointerface results in an increase in the maximum drain current and a
decrease in the threshold voltage as seen in Figure 4.1. Included in the plot is the data for
the polarization charge density of 1.14x1013 cm-2 used in the initial simulation results
from Chapter 3. As expected, the threshold voltage decreases as the amount of positive
polarization charge pz increases at the AlGaN/GaN interface. Correspondingly, for a
maximum VDS = 10 V and VGS = +1 V, the reduction in VT provides a correlated increase
in the maximum IDS.

97

Figure 4.1

Threshold Voltage (VT, ) and the maximum drain current () at a bias of VGS = +1
V, VDS = +10 V plotted against various polarization charge densities at the
AlGaN/GaN heterointerface and AlGaN surface. The initial polarization charge
density of 1.14x1013 cm-2 from the simulation modeling in Chapter 3 is also plotted
here and is identified by the arrow.

The transconductance as a function of gate voltage is plotted for different


polarization charge densities in Figure 4.2a. As seen in the plot, the transconductance
increases with increasing polarization charge density at low gate biases, then peaks and
subsequently decreases. The fall off at high VGS is typical for FETs and is associated
with reduced electron mobility due to the larger gate field and increased surface
scattering at high gate bias. Also noticeable in the plot, is that the transconductance peak
is enhanced and shifted towards a more negative gate bias as the polarization charge
density is increased. The devices gm is enhanced, because gm is proportional to the
product of the electron concentration (ns) and the mobility () and, as pz is increased, a
98

corresponding increase occurs for ns. However, n decreases due to the reasons specified
above, so the net effect is that peak gm goes thru a maximum as pz is varied. In Figure
4.2b, the peak transconductance is plotted against the polarization charge density. As
seen in the plot, the peak transconductance of the device increases with an increasing
polarization charge density. Since the baseline device had a pz = 1.14x1013 cm-2, a better
device (~ 10% higher peak gm) could be built employing a 50% larger pz.

Figure 4.2

Plots of a) transconductance (gm) versus gate voltage for different polarization


charge densities and b) peak gm as a function of polarization charge density. The
data for the polarization charge density of 1.14x1013 cm-2 is from the modeling
results reported in Chapter 3 and identified by the arrow in (b).

Since the AlGaN/GaN HEMT is of interest for high frequency, high power
applications, the speed of the device was examined (cutoff frequency and maximum
frequency of oscillation). The cutoff frequency is plotted in Figure 4.3 as a function of
the gate voltage for various polarization charge densities.
99

Figure 4.3

Cutoff frequency (fT) plotted as a function of the gate voltage for various
polarization charge densities at a drain bias of +10 V. The data for the polarization
charge density of 1.14x1013 cm-2 is from the modeling results reported in Chapter 3.

Similar to the gm, fT goes thru a peak as VGS is varied. A noticeable increase in
the peak cutoff frequency is observed for an increasing polarization charge density, with
the peak cutoff frequency shifting to a more negative voltage. This is consistent with a
noticeable shift towards a more negative voltage observed for the threshold voltage and
peak transconductance as is expected since the cutoff frequency typically scales linearly
with the transconductance. As expected, these high values (> 250 GHz) for the cutoff
frequency point to the devices promise for high frequency applications.
While the cutoff frequency defines where the current gain goes to unity, the
maximum frequency of oscillation defines where the power gain goes to unity for the

100

transistor. The maximum oscillation frequency (fmax) is plotted as a function of the gate
voltage for various polarization charge densities in Figure 4.4.

Figure 4.4

Maximum oscillation frequency (fmax) plotted as a function of the gate voltage for
various polarization charge densities at a drain bias of +10 V. The data for the
polarization charge density of 1.14x1013 cm-2 is from the modeling results reported
in Chapter 3.

Also included in the plot is the data from Chapter 3 when using the polarization
charge density of 1.14x1013 cm-2. An increase in the peak fmax is observed above 250
GHz when the polarization charge density is increased, as is observed in Figure 4.4.
To quantify the effects of the polarization charge on fT and fmax, Figure 4.5, shows
the plot of both peak fT and peak fmax vs. pz, which shows a steady increase with
increasing pz with an exception that the peak fmax for the density of 1.14x1013 cm-2 is
slightly lower than that for the lower polarization charge densities of 1x1013 cm-2 and
1.05x1013 cm-2.
101

Figure 4.5

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax) plotted as
a function of polarization charge density for various densities. The data for the
polarization charge density of 1.14x1013 cm-2 (marked by the arrow) is from the
modeling results reported in Chapter 3.

The larger peak fmax observed for lower polarization charge densities is largely
attributed to the smaller number of gate bias points used for the fmax extraction. If more
gate bias points were used, the peaks should be about the same or slightly higher for the
polarization density of 1.14x1013 cm-2. As seen in Figure 4.4, the maximum oscillation
frequency is larger for larger polarization charge densities before the peak, but smaller at
gate biases once the peak has been reached. Also plotted in Figure 4.5 is the peak cutoff
frequency as a function of the polarization charge density, which shows that the peak
cutoff frequency increases ~ 12% as the polarization charge density increases from
1x1013 to 1.45x1013 cm-2. As noted earlier, these results are consistent with the increase
102

in peak gm as pz increases since fT is proportional to the transconductance and fmax


typically scales up with fT. For the range of polarization charge density shown in Figure
4.5, fmax increases by ~ 7%, which is comparable but less than that seen for fT.
In summation, an increase in the magnitude of the polarization charge density
results in a negative shift in the threshold voltage and a shift in the peak transconductance
to a slightly more negative gate bias. More importantly, increasing pz produces a 10%
increase in the peak gm and a corresponding and comparable increase in peak fT and peak
fmax. As can be seen from these tests, the polarization charge present in the AlGaN/GaN
HEMTs plays a key role in the devices performance and if the magnitude of the
polarization charge density can be engineered, the devices performance can be improved
as seen in Figures 4.2 and 4.5. The mechanism by which the polarization charge can be
engineered is thru the growth of different AlGaN/GaN heterostructures and by selecting
the aluminum mole fraction present in the AlGaN layer, as first described by Ambacher
et al. [2, 4].

4.2 Donor Trap Density at the AlGaN Surface


The donor surface traps at the AlGaN/Si3N4 interface provide the source of the
electrons for the 2DEG as pointed out in Chapter 2. In this section, the effect on the
device performance is investigated when increasing or decreasing the donor trap density
from the baseline value of 1.14x1013 cm-2 used in the previous chapter. These donor
surface traps are also believed to be linked to the experimentally observed drain current
collapse phenomenon in AlGaN/GaN HEMTs [5]. For this part of the study, only the
density of traps is changed as the transient behavior of the device was not investigated.
103

Figure 4.6 shows the threshold voltage and maximum dc drain current plotted as
functions of the donor surface trap density.

Figure 4.6

Threshold Voltage (VT, ) and the maximum drain current () at a bias of VGS = +1
V, VDS = +10 V plotted against various donor trap densities at the AlGaN surface in
contact with the Si3N4 layer. The data for the donor trap density of 1.14x1013 cm-2
(identified by arrow) is from the modeling results reported in Chapter 3 and is
included in the plot.

When increasing the surface trap density, the threshold experiences a negative
shift, but one that is very small (about 140 mV difference for a density range of 7x1012
cm-2 to 3x1013 cm-2). In contrast, the threshold voltage changes discussed above for
simulations when the polarization charge density was varied were more substantial (~
few volts).

The maximum dc drain current of the device increases sharply with

104

increasing donor surface trap density, initially. However, once the trap density reaches
1.14x1013 cm-2, the maximum dc drain current of the devices rapidly saturates.
Seen in Figure 4.7a is the simulated transconductance of the device versus gate
voltage. The transconductance initially increases rapidly by ~ 20%, but then only slightly
when the donor trap density is increased above 8.5x1012 cm-2. As previously seen, the gm
goes through a peak as VGS is varied where the peak transconductance is plotted against
various donor trap densities in Figure 4.7b.

Figure 4.7

Plots of a) transconductance (gm) versus gate voltage for different donor trap
densities and b) peak gm as a function of donor trap density. The data for the donor
trap density of 1.14x1013 cm-2 (identified by arrow) is from the modeling results
reported in Chapter 3 and is included in the plot.

The results show that the peak transconductance experiences very little increase in
performance after 8.5x1012 cm-2 for the donor trap density, similar to the behavior seen
for the maximum current. This behavior is related to the density of the polarization
charge, pz at the AlGaN/GaN heterointerface and its control of the electron density in the
channel.

The donor traps must supply enough electrons to form a sheet carrier
105

concentration in the 2DEG to compensate pz, any donor trap density in excess of the
polarization charge density will produce no increase in gm or in the maximum drain
current.
Shown in Figure 4.8 is the cutoff frequency of the device plotted as a function of
the gate voltage as the surface donor trap density was raised.

Figure 4.8

Cutoff frequency (fT) plotted as a function of the gate voltage for various donor trap
densities at a drain bias of +10 V. The data for the donor trap density of 1.14x1013
cm-2 is from the modeling results reported in Chapter 3 and is included in the plot.

As in the case of pz, the peak gm rises as the trap density rises. However, the
increase in the cutoff frequency performance is not as substantial once the trap density
reaches close to 1x1013 cm-2. The increased performance in fT is much more noticeable

106

from 4.8x1012 cm-2 to 8.5x1012 cm-2, but is less rapid from 8.5x1012 cm-2 to 1x1014 cm-2
for the donor trap density.
The maximum oscillation frequency of the simulated device is plotted as a
function of gate voltage for various surface donor trap densities in Figure 4.9. Similar, to
the results for the fT, fmax shows an increase from 4.8x1012 cm-2 to 8.5x1012 cm-2 for the
donor trap density, but the simulated fmax is about the same in the donor trap density
range of 8.5x1012 cm-2 to 1x1014 cm-2.

Figure 4.9

Maximum oscillation frequency (fmax) plotted as a function of the gate voltage for
various donor trap densities at a drain bias of +10 V. The data for the donor trap
density of 1.14x1013 cm-2 is from the modeling results reported in Chapter 3 and is
included in the plot.

The peak fT and fmax versus donor trap density are shown in Figure 4.10. The plot
confirms that, in general, the peak fT and fmax rise with donor trap density although the
peak fmax is lower than expected for 1.14x1013 cm-2. The fmax at 1.14x1013 cm-2 should be
107

at the same value as that for 8.5x1013 cm-2 or slightly higher, but is not. This is believed
to be due to the number of bias points used to extract the fmax in the simulations since the
fT and fmax generally follow the gm trend, which is seen in Figure 4.7b.

Figure 4.10

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax) plotted as
a function of the donor trap density for various densities. The data for the donor
trap density of 1.14x1013 cm-2 (indicated by arrow) is from the modeling results
reported in Chapter 3 and is included in the plot.

To get the best performance out of the device, the surface donor trap density in
the device should be around 1x1013 cm-2. Any value higher than this and the additional
performance gain is minimal, but values lower than this have results that are much lower
than the devices potential. However, the presence of a high density of surface donor
traps may also contribute to more severe transient effects, e.g. current collapse [6, 7].
Finally, the donor trap density of 1.14x1013 cm-2 that should be targeted is the same
108

density for the polarization charge at the AlGaN/GaN heterointerface and AlGaN surface
used in this simulation. A lower donor trap density than 1.14x1013 cm-2 may not have
enough electrons free to form the 2DEG so that the magnitude of the sheet charge of the
2DEG can match the polarization charge located on the other side of the AlGaN/GaN
heterointerface. While a higher density of surface traps can contribute more electrons
than necessary needed for the 2DEG, the performance gains are minimal.

4.3 Contact Resistance


Also investigated in this study was the effect of the source/drain contact
resistance.

The contact resistance of the device was varied with specific contact

resistances of 0.24 -mm, 2.68 -mm, and 4.41 -mm simulated, which are within the
range measured as reported by Jessen et al. [1]. The results of the simulations with a
contact resistance of 1.46 -mm discussed in the previous chapter are also included for
comparison. Shown in Figure 4.11 are the threshold voltage and maximum drain current
as functions of the contact resistance. The threshold voltage shows some small (~ few
percent) variation as the threshold voltage experiences a negative shift when the contact
resistance of the device increases. By contrast, the simulated maximum drain current is
3.85 A/mm when the contact resistance is 0.24 -mm, but much smaller (0.91 A/mm)
when the contact resistance is increased to 4.41 -mm, a 323% difference. The variation
in the contact resistance significantly impacts the output current of the device due to a
high series resistance. A high contact resistance causes a large voltage drop across the
metal-semiconductor contacts at the source and drain, which lowers the electric field in
the channel and so the electron velocity.
109

Figure 4.11

Threshold Voltage (VT, ) and the maximum drain current () at a bias of VGS = +1
V, VDS = +10 V plotted as a function of the contact resistance. The original device
has a contact resistance of 1.46 -mm (indicated by arrow); the data for this contact
resistance is included in the plot.

Figure 4.12a shows the simulated transconductance as a function of the gate


voltage for various contact resistances. Consistent with the maximum drain current
results, the transconductance value is much higher at low contact resistance and decreases
rapidly as the contact resistance of the device is increased.

The decrease in

transconductance is related to an increase in contact resistance according to [8]

gm =

gi
1 + Rs g i

(4.2)

where gi is the intrinsic maximum transconductance and Rs is the series resistance at the
source. The simulated peak transconductance of the device decreases markedly with

110

increasing contact resistance as shown in Figure 4.12b, illustrating the importance of high
quality ohmic contacts to the AlGaN/GaN HEMTs performance.

Figure 4.12

Plots of a) transconductance (gm) versus gate voltage and b) peak gm as a function of


the device contact resistance. The original device has a contact resistance of 1.46 mm (indicated by arrow); the data for this contact resistance is included in the plot.

Figures 4.13 and 4.14 show the cutoff frequency and maximum frequency of
oscillation, respectively, plotted as a function of gate voltage for various contact
resistances.

111

Figure 4.13

Cutoff frequency (fT) plotted as a function of the gate voltage for various contact
resistances. The original device structure has a contact resistance of 1.46 -mm
whose cutoff frequency is included in the plot.

Figure 4.14

Maximum oscillation frequency (fmax) plotted as a function of the gate voltage for
various contact resistances. The original device structure has a contact resistance of
1.46 -mm whose maximum oscillation frequency is included in the plot.

112

The fT and fmax frequencies for the device are highest for a low contact resistance
of 0.24 -mm and decreases substantially (more than a factor of four) as the contact
resistance is increased similar to the results seen for the transconductance. The fT and
fmax frequencies have their highest values at a contact resistance of 0.24 -mm. As the
contact resistance is increased, the simulated fT and fmax decrease to ~ 25% of their peak
values as seen in Figure 4.15.

These trends are consistent with that seen for the

transconductance and arise for the same reason since both fT and fmax scale with the gm.

Figure 4.15

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax) plotted as
a function of the device contact resistance.

The original device has a contact

resistance of 1.46 -mm (indicated by arrow); the data for this contact resistance is
included in the plot.

In summary, an increase in contact resistance causes the drain current of the


device to decrease dramatically, the threshold voltage to shift slightly towards a negative
voltage, and the cutoff frequency and the maximum oscillation frequency of the device to
113

decrease substantially. Clearly, the quality of the ohmic contacts at the source and drain
are critical to the devices performance and to get the most power out of the device, the
contact resistance should be as small as possible.

4.4 AlGaN Thickness


Simulations were also performed to examine the effects on device performance of
variations in the thickness of the AlGaN barrier layer from 160 to 410 Angstroms. The
data from these simulations are compared to the simulations from the original device
structure which had an AlGaN thickness of 260 Angstroms. Figure 4.16 shows the
threshold voltage and maximum drain current of the device when varying the thickness of
the AlGaN layer.

Figure 4.16

Threshold Voltage (VT, ) and the maximum drain current () at a bias of VGS = +1
V, VDS = +10 V plotted as a function of the AlGaN thickness. The original device
structure has an AlGaN thickness of 260 Angstroms (indicated by arrow); the data
for this device structure is included in the plot.

114

The threshold voltage shifts rapidly (by volts) towards a more negative voltage as
the thickness of the layer is increased. The maximum simulated drain current at the
biases of 10 V on the drain and +1 V on the gate increases by nearly a factor of two as the
thickness of the AlGaN layer is increased. Since the maximum IDS is measured at a fixed
VGS, this result is consistent with the decrease in VT.
In Figure 4.17a, the simulated transconductance is plotted as a function of the gate
voltage for various AlGaN thicknesses. As previously seen, the gm goes through a peak
as VGS is varied where the peak of the transconductance shifts towards a more negative
gate voltage with increasing AlGaN thickness. The peak of the transconductance also
decreases by ~ 20% with an increase in the thickness of the AlGaN layer as can be seen
in Figure 4.17b.

Figure 4.17

Plots of a) transconductance (gm) versus gate voltage for various AlGaN thickness
and b) peak gm as a function of the AlGaN thickness. The original device structure
has an AlGaN thickness of 260 Angstroms (indicated by arrow); the data for this
device is included in the plot.

115

This is due to the reduction of the peak electron concentration in the channel as
the thickness of the AlGaN barrier is increased as shown in Figure 4.18.

Figure 4.18

Peak electron concentration (ns) measured at the peak gm voltage and plotted as a
function of the AlGaN thickness. The original device structure has an AlGaN
thickness of 260 Angstroms (indicated by arrow); the data for this device is included
in the plot.

Because gm is proportional to the product of the electron concentration (ns) and


the electron mobility (n), a thinner AlGaN thickness would result in a higher ns, but a
lower n since the transverse field in the channel is increased. The net result of this is a
higher gm. To optimize the transconductance for high frequency performance, the
thickness of the AlGaN layer should be reduced as this decreases the gate to channel
distance, where Figure 4.17b shows the trend of the transconductance is inversely
proportional to the gate-to-channel distance of the device.

116

Figures 4.19 and 4.20 show the cutoff frequency and maximum oscillation
frequency, respectively of the simulated AlGaN/GaN HEMT plotted as a function of the
gate voltage for various AlGaN thicknesses. The peaks of the cutoff and maximum
oscillation frequencies experience a shift towards negative gate voltage similar to the
transconductance and threshold voltage when the thickness of the AlGaN layer is
increased, but with little change in the peak fT and fmax.

The peaks of the cutoff

frequency and maximum oscillation frequency are shown in Figure 4.21 as functions of
the thickness of the AlGaN layer.

Figure 4.19

Cutoff frequency (fT) plotted as a function of the gate voltage for various AlGaN
thickness. The original device structure has an AlGaN thickness of 260 Angstroms
whose cutoff frequency is included in the plot.

117

Figure 4.20

Maximum oscillation frequency (fmax) plotted as a function of the gate voltage for
various AlGaN thickness. The original device structure has an AlGaN thickness of
260 Angstroms whose maximum oscillation frequency is included in the plot.

Figure 4.21

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax) plotted as
a function of the AlGaN thickness. The original device structure has an AlGaN
thickness of 260 Angstroms (indicated by arrow); the data for this device structure
is included in the plot.

118

The data show no strong dependence on the AlGaN thickness, although the data
may support a somewhat decreasing trend in the peaks of the cutoff frequency and
maximum oscillation frequency when the thickness of the AlGaN layer is increased. This
is somewhat surprising since the peak gm decreases ~ 20% with increasing AlGaN
thickness as seen in Figure 4.17. The inconclusive results seen here may be limited by
the uncertainty in the fT and fmax values due to their dependence on the number of nodes
used in the simulations. For the most accurate results, simulations with a larger number
of nodes are desirable. However, in practice this lends to an unacceptable large time for
simulations so a smaller number of nodes (~ 4000 to 5000), especially in a few regions of
the device were used in practice to get the results within an acceptable time frame. In this
case, more accurate results would need to be obtained using more nodes in the simulation
to clarify this issue. In summary, it is not clear that to maximize the transconductance of
the device, the AlGaN barrier thickness should be reduced for high frequency operation.

4.5 Source-to-Gate Spacing


The source-to-gate separation and drain-to-gate separation were also varied in
these simulations while keeping the device length (source to drain spacing) fixed at 3 m.
Simulations for source-to-gate spacings of 0.55 m, 0.95 m, 1.65 m, and 2.05 m were
done and compared to the devices original source-to-gate separations of 1.3 m. The
source-to-gate spacing of 0.55 m was specifically chosen as this is the spacing used for
the aggressive device reported in [1]. The simulations of these different source-to-gate
separations show substantial performance variation and are discussed below.
The simulation results for the threshold voltage and maximum simulated drain
current as functions of the source-to-gate spacing are shown in Figure 4.22.
119

Figure 4.22

Threshold Voltage (VT, ) and the maximum drain current () at a bias of VGS = +1
V, VDS = +10 V plotted as a function of the source-to-gate spacing of the device
structure. The original device structure has a source-to-gate spacing of 1.3 microns
(indicated by arrow); the data for this device structure is included in the plot.

The threshold voltage shows very little variation over the range of source-to-gate
spacing, although surprisingly the original device contact spacing of 1.3 m results in a
higher threshold voltage.

The maximum simulated drain current varies somewhat

substantially over the same source-to-gate separation range. The drain current decreases
~ 10% as the source-to-gate spacing increases from 0.55 to 2.05 m due to increased
source series resistance.
Figure 4.23a shows the simulated transconductance of the device as a function of
the gate voltage for various source-to-gate spacing and Figure 4.23b shows the simulated
peak transconductance plotted as a function of the source-to-gate spacing. The simulated
transconductance shows a decrease of ~ 30% as the source-to-gate spacing is increased
120

from 0.55 m to 2.05 m following the same trend seen for the drain current. The
location of peak transconductance also shifts to a less negative gate voltage as the sourceto-gate spacing is increased.

Figure 4.23

Plots of a) transconductance (gm) versus gate voltage for different source-to-gate


spacing and b) peak gm as a function of the source-to-gate spacing of the device
structure. The original device structure has a source-to-gate spacing of 1.3 microns
(indicated by arrow); the data for this device is included in the plot.

The cutoff frequency of the simulated device is plotted as a function of the gate
voltage for various source-to-gate spacing in Figure 4.24.

121

Figure 4.24

Cutoff frequency (fT) plotted as a function of the gate voltage for various source-togate spacing. The original device structure has a source-to-gate spacing of 1.3
microns whose cutoff frequency is included in the plot.

Unlike the gm trend, an increase of 10% in the cutoff frequency is observed for an
increase in the source-to-gate spacing. This may be due to a reduction in the gate source
capacitance as the source-to-gate spacing increases. Since the cutoff frequency (fT) is a
function of the transconductance (gm) and is expressed by [9]

fT =

gm
2C GS

(4.3)

where CGS is the gate-source capacitance. A decrease in fT might be expected for a


decrease in gm observed in Figure 4.23b for an increase in source-to-gate spacing.
However, an increase in source-to-gate spacing results in a substantial decrease in the
122

gate-source capacitance which according to (4.3) would allow the fT to increase even
with a decrease in gm, as is observed in Figure 4.25.

Figure 4.25

Peak cutoff frequency (fT) and peak maximum oscillation frequency (fmax) plotted as
a function of the source-to-gate spacing. The original device structure has a sourceto-gate spacing of 1.3 microns (indicated by arrow); the data for this device
structure is included in the plot.

The maximum oscillation frequency is plotted as a function of the gate voltage for
various source-to-gate spacing in Figure 4.26.

123

Figure 4.26

Maximum oscillation frequency (fmax) plotted as a function of the gate voltage for
various source-to-gate spacing. The original device structure has a source-to-gate
spacing of 1.3 microns whose maximum oscillation frequency is included in the plot.

Compared to the cutoff frequency, the maximum oscillation frequency


experiences an even more significant increase (~ 60%) as the source-to-gate spacing is
increased.

Figure 4.25 summarizes the increase in the simulated peak cutoff and

maximum oscillation frequencies as the source-to-gate spacing is increased.

The

maximum oscillation frequency (fmax) scales with the cutoff frequency (fT), but also
depends on various geometries as seen in [10]
f max

gm
2 (C gs + C gd )

C gd
4 g ds (Ri + RS + RG ) + 4 g m RG

C gs + C gd

124

1
2

(4.4)

where gm is the transconductance, Cgs and Cgd are the gate-source and gate-drain
capacitances, respectively, gds is the drain conductance, and Ri, RS and RG are the
intrinsic resistance of the gate, the extrinsic source resistance, and the extrinsic gate
resistance, respectively. The first term is basically another form of fT, while the second
term is the small-signal behavior that factors in with fT to get an fmax. Since the fT
increases with an increase in the source-to-gate spacing, a corresponding increase in fmax
is expected. This expected increase in fmax due to an increase in source-to-gate spacing is
a 60% increase for the simulated source-to-gate spacing range from 0.55 to 2.05 m,
while the fT had a smaller increase of 10% over the same range.
The aggressive experimental device reported in [1] had a best maximum drain
current of 1.48 A/mm, peak transconductance of 338 mS/mm, and fT at the peak gm of
91.4 GHz. In comparison, the simulated device with a contact spacing of 0.55 m had
maximum drain current of 2.06 A/mm, a 39% difference; a peak transconductance of 336
mS/mm, a difference under 1%; and a peak fT of 278, which is a 204% higher than the
experimental device.

4.6 Discussion and Conclusions


In summary, various parameters of the device were analyzed for optimization,
including the cutoff frequency, transconductance and maximum oscillation frequency by
changing the devices structure or incorporated polarization.

The results of these

simulations were compared to the results from the baseline structure reported in Chapter
3. An increase in the gate-to-source spacing of the device from 0.55 to 2.05 m will
experience a decrease in the transconductance and drain current, of about 30% and 10%,
125

respectively. At the same time, the cutoff frequency and maximum oscillation frequency
will experience an increase with a reduction of the gate-to-source spacing of 10% and
60%, respectively. The cutoff frequency of the device increases with increasing sourceto-gate spacing even though the transconductance decreases due to the reduction of the
gate-source capacitance when increasing the gate-to-source spacing.

The maximum

oscillation frequency increases with a large source-to-gate spacing as it scales with fT.
The polarization charge density (pz) at the AlGaN/GaN heterointerface was
varied from 1x1013 cm-2 to 1.45x1013 cm-2 in simulations. Over this density range, fT and
fmax were reported to be over 250 GHz with an increase in peak values of 13% and 6%,
respectively as the density was increased. The peak transconductance also increased ~
15% over this density range.
To get the most out of the device, the polarization charge should be as significant
as possible in order to achieve a high charge density at the AlGaN/GaN heterointerface.
This charge density should be significant enough to allow a higher 2-DEG. This higher
2-DEG is realized by a surface donor trap density that is large enough to match or dwarf
the magnitude of the polarization charge density at the AlGaN surface and AlGaN/GaN
heterointerface. If the surface donor trap density is not significant enough, the device
performance will perform poorly compared to its potential. The gm, fT, and fmax increased
with increasing donor trap concentration, but the values saturated as the donor trap
density neared 8.5x1012 cm-2. The increase in performance for gm, fT, and fmax levels off
as the donor trap density reaches 1.14x1013 cm-2, which is the value of the polarization
charge density at the AlGaN/GaN heterointerface.

126

There is little to suggest that

increasing the donor trap density above the value for pz does anything to improve device
performance.
The contact resistance of the device should be as small as possible to achieve
higher transconductance, higher fT, and higher fmax. As the source and drain contact
resistances were increased from 0.24 -mm to 4.41 -mm, the gm, fT and fmax decreased
by approximately 75% of their peak values. Increasing the contact resistance for the
source and drain contacts results in a voltage drop which in effect lowers the electric field
in the channel and electron velocity. The reductions in the electric field in the channel
and electron velocity are responsible for the reduced values of the gm, fT, and fmax as the
contact resistance is increased. From the simulations, it was not clear whether the AlGaN
barrier thickness should be reduced to maximize the transconductance for high frequency
operation.

While the peak transconductance decreases 20% as the AlGaN barrier

thickness is increased from 160 to 410 Angstroms, the data for fT and fmax is inconclusive
as whether decreasing the thickness improves the parameter values. If the device is to be
optimized for high power, and high frequency operation with the highest possible value
for fmax, then a polarization charge density of 1.45x1013 cm-2 or higher is recommended
with an equal density of donor traps, a contact resistance of 0.24 -mm, a source-to-gate
spacing of 2.05 m so the gate is offset closer to the source, and an AlGaN barrier
thickness that is the same as the baseline structure with a thickness of 260 Angstroms.

127

References
[1]

G.H. Jessen, R. C. Fitch, J. K. Gillespie, G. D. Via, N. A. Moser, M. J. Yannuzzi,


A. Crespo, J. S. Sewell, R. W. Dettmer, T. J. Jenkins, R. F. Davis, J. Yang, M.
Asif Khan, and S. C. Binari, High Performance 0.14 m Gate-Length
AlGaN/GaN Power HEMTs on SiC, IEEE Electron Device Letters, vol. 24, no.
11, pp. 677-679, Nov. 2003.

[2]

O. Ambacher, B. Foutz, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M.


Murphy, A. J. Sierakowski, W. J. Schaff, L. F. Eastman, R. Dimitrov, A.
Mitchell, and M. Stutzmann, Two-dimensional electron gases induced by
spontaneous and piezoelectric polarization in undoped and doped AlGaN/GaN
heterostructures, Journal of Applied Physics, vol. 87, no. 1, pp. 334-344, January
2000.

[3]

J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.


Mishra, Polarization effects, surface states, and the source of electrons in
AlGaN/GaN heterostructure field effect transistors, Applied Physics Letters, vol.
77, no. 2, pp. 250-252, July 2000.

[4]

O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy, W. J.


Schaff, L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, and J.
Hilsenbeck, Two-dimensional electron gases induced by spontaneous and
piezoelectric

polarization

charges

in

N-

and

Ga-face

AlGaN/GaN

heterostructures, Journal of Applied Physics, vol. 85, no. 6, pp. 3222-3233,


March 1999.

128

[5]

R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, The Impact of Surface


States on the DC and RF Characteristics of AlGaN/GaN HFETs, IEEE
Transactions on Electron Devices, vol. 48, no. 3, pp. 560-566, March 2001.

[6]

H. Hasegawa, T. Inagaki, S. Ootomo, and T. Hashizume, Mechanisms of current


collapse and gate leakage currents in AlGaN/GaN heterostructure field effect
transistors, Journal of Vacuum Science and Technology B, vol. 21, no. 4, pp.
1844-1855, August 2003.

[7]

G. Meneghesso, G. Verzellesi, R. Pierobon, F. Rampazzo, A. Chini, U. K.


Mishra, C. Canali, E. Zanoni, Surface-Related Drain Current Dispersion Effects
in AlGaN-GaN HEMTs, IEEE Transactions on Electron Devices, vol. 51, no.
10, pp. 1554-1561, October 2004.

[8]

F. Ali, and A. Gupta, HEMTs & HBTs: Devices, Fabrication, and Circuits,
Artech House, Boston, MA, 1991.

[9]

K. Chang, Handbook of RF/Microwave Components and Engineering, John


Wiley and Sons, Hoboken, N.J, 2003.

[10]

F. Schwierz and Juin J. Liou, Modern Microwave Transistors: Theory, Design,


and Performance, John Wiley and Sons, Hoboken, N.J., 2003.

129

Chapter 5

Conclusions and Future Work

5.1 Conclusions
The work presented in this thesis is the result of simulations of a 0.14 m gatelength AlGaN/GaN HEMT using the base structure reported by Jessen et al. [1] as the
starting point for the modeling. Discussed were the importance of the polarization effects
that occur in the AlGaN/GaN heterostructure with pioneering work done by Ambacher
[2, 3] and Ibbetson et al. [4] to describe and model the polarization in the structure
analytically. The polarization in the simulated device was incorporated by placing sheet
charges at their respective locations, i.e. the AlGaN/GaN heterointerface and the AlGaN
surface. Donor traps were also placed at the AlGaN/Si3N4 interface to provide the source
of electrons needed for the two-dimensional electron gas (2DEG). Modeling of hot
electrons was incorporated into the modeling by using the hydrodynamic model [5]. This
approach was used to simulate the device; the initial results were compared to the
experimental results [1].
The simulations of the device structure outlined above are in reasonable agreement
with the experimental results with a few exceptions. These exceptions include the cutoff
frequency (fT) and operation at high gate (> 0 V) and high drain biases (> 8 V). The
simulation peak fT was 266 GHz compared to the experimental peak fT of 82 GHz, a
224% difference, which was partially accounted for by the lack of parasitics included in
the simulations. Other output parameters were in better agreement, including the smallsignal transconductance (gm) whose peak simulation value was 289 mS/mm compared to
the experimental peak gm of 295 mS/mm, a 2% difference.
130

Variations of the baseline structure of experimental device were made and then
compared to the simulation results outlined above as a means to optimize the device
design. Increasing the polarization charge from 1x1013 cm-2 to 1.45x1013 cm-2 at the
AlGaN/GaN heterointerface and AlGaN surface resulted in higher gm, fT and maximum
oscillation frequency (fmax), and a negative shift in threshold voltage. Both fT and fmax
show promise for high frequency, high power applications as they achieved frequencies
over 250 GHz in the stated polarization charge density range. The density of the donor
surface traps were also investigated, the simulation results of which showed that a higher
trap density than the polarization charge density discussed above does not produce any
improved performance in the device, but a density lower than the polarization charge
density results in significantly lower gm, fT, and fmax. In order to maximize the device
performance, the donor trap density should thus equal or exceed the magnitude of the
polarization charge density so that the 2DEG can be filled with enough electrons.
Altering the AlGaN thickness showed inconclusive data that the thickness should be
smaller than the original AlGaN thickness in order to maximize the transconductance.
While the transconductance did decrease by 20% as the thickness of the AlGaN barrier
was increased from 160 to 360 Angstroms, the fT and fmax showed no general trend as to
increasing or decreasing with the AlGaN thickness. A small contact resistance is also
exceedingly important to maximize current, gm, fT, and fmax. Increasing the source and
drain contact resistances from 0.24 -mm to 4.41 -mm resulted in a 25% decrease in
the peak values of gm, fT, and fmax. Decreasing the source-to-gate spacing in the device
resulted in a 30% increase in peak transconductance while keeping the device length
fixed at 3 m. While the transconductance experienced an increase, the fT and fmax

131

decreased by 10% and 60%, respectively. The cutoff frequency decreased even with an
increase in peak gm due to a larger gate-source capacitance as the gate-to-source spacing
is decreased. The maximum oscillation frequency decreases over 60% with a decrease in
gate-to-source spacing as it scales with the cutoff frequency.
In conclusion, this simulation study showed the importance of incorporating the
polarization effects into the device structure as well as the small-signal and microwave
capabilities of the AlGaN/GaN HEMT. The study also shows the small-signal and
microwave performance gain when varying the device structure and polarization
incorporated in the device.

5.2 Future Work


A more accurate account of the polarization needs to be incorporated which should
include an analytical model such as the one proposed by Ambacher et al. [2]. While the
method used for incorporating the polarization at the AlGaN/GaN interface and AlGaN
surface is a step in the right direction, the densities used were the same as the sheet
charge density reported experimentally in [1]. Other work [2, 3] has shown that the
polarization may depend on other factors such as the mole fraction of aluminum in the
AlGaN layer. Incorporating a polarization charge model in the device structure, would
be more accurate especially when alterations are made to the device structure to optimize
the microwave performance.
Other additions that can be made to improve the model used in this work is the
incorporation of computing the electron temperature to account for hot electrons in the
hydrodynamic model, as well including a quantum confinement model such as the
132

density gradient model included in DESSIS [6]. These were considered for this work, but
later dropped due to issues related to simulation time and convergence problems. Selfheating effects should also be considered in future work as this would lower simulated
performance more in line with experimental results.

Other effects that should be

investigated in future work are the breakdown voltage due to the high power, high
frequency nature of the devices operation as well as the gate leakage current. Another
feature that is lacking in this work is the analysis of the transient behavior of the device.
Recent work [7] has shown that surface traps in the device are linked to the
experimentally observed current collapse. Transient analysis of the device can help
characterize this current collapse due to traps in the device. Early transient simulations
were done, but work stopped after difficulties were encountered in obtaining results.

133

References
[1]

G.H. Jessen, R. C. Fitch, J. K. Gillespie, G. D. Via, N. A. Moser, M. J. Yannuzzi,


A. Crespo, J. S. Sewell, R. W. Dettmer, T. J. Jenkins, R. F. Davis, J. Yang, M.
Asif Khan, and S. C. Binari, High Performance 0.14 m Gate-Length
AlGaN/GaN Power HEMTs on SiC, IEEE Electron Device Letters, vol. 24, no.
11, pp. 677-679, Nov. 2003.

[2]

O. Ambacher, B. Foutz, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M.


Murphy, A. J. Sierakowski, W. J. Schaff, L. F. Eastman, R. Dimitrov, A.
Mitchell, and M. Stutzmann, Two-dimensional electron gases induced by
spontaneous and piezoelectric polarization in undoped and doped AlGaN/GaN
heterostructures, Journal of Applied Physics, vol. 87, no. 1, pp. 334-344, January
2000.

[3]

O. Ambacher, J. Smart, J. R. Shealy, N. G. Weimann, K. Chu, M. Murphy, W. J.


Schaff, L. F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, and J.
Hilsenbeck, Two-dimensional electron gases induced by spontaneous and
piezoelectric

polarization

charges

in

N-

and

Ga-face

AlGaN/GaN

heterostructures, Journal of Applied Physics, vol. 85, no. 6, pp. 3222-3233,


March 1999.
[4]

J. P. Ibbetson, P. T. Fini, K. D. Ness, S. P. DenBaars, J. S. Speck, and U. K.


Mishra, Polarization effects, surface states, and the source of electrons in
AlGaN/GaN heterostructure field effect transistors, Applied Physics Letters, vol.
77, no. 2, pp. 250-252, July 2000.

134

[5]

R. Stratton, Diffusion of hot and cold electrons in semiconductor barriers,


Physical Review, vol. 126, no. 6, pp. 2002-2014, June 1962.

[6]

DESSIS ISE TCAD Manual, Release 8.0 (ISE Integrated Systems Engineering
AG, Zurich 2002).

[7]

R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, The Impact of Surface


States on the DC and RF Characteristics of AlGaN/GaN HFETs, IEEE
Transactions on Electron Devices, vol. 48, no. 3, pp. 560-566, March 2001.

135

S-ar putea să vă placă și