Citation
Advanced Three-Dimensional Packaging Schemes for Microelectronic and Microsystem Applications

Material Information

Title:
Advanced Three-Dimensional Packaging Schemes for Microelectronic and Microsystem Applications
Creator:
Anderson, Travis James
Place of Publication:
[Gainesville, Fla.]
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (219 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Chemical Engineering
Committee Chair:
Ren, Fan
Committee Members:
Lin, Jenshan
Ziegler, Kirk
Pearton, Stephen J.
Graduation Date:
5/1/2008

Subjects

Subjects / Keywords:
Electric current ( jstor )
Electric potential ( jstor )
Electrons ( jstor )
Etching ( jstor )
Hydrogen ( jstor )
Lasers ( jstor )
Semiconductors ( jstor )
Sensors ( jstor )
Silicon ( jstor )
Wavelengths ( jstor )
Chemical Engineering -- Dissertations, Academic -- UF
Genre:
Electronic Thesis or Dissertation
bibliography ( marcgt )
theses ( marcgt )
Chemical Engineering thesis, Ph.D.

Notes

Abstract:
The emergence of GaN-based devices promises a revolution in areas requiring high performance electronics, such as high speed earth and space-based communication systems, advanced radar, integrated sensors, high temperature electronics, and utility power switching. The properties of this system make it ideally suited for operation at elevated temperatures and at voltage and current levels well beyond that accessible by Si. Recent improvements in material quality and device performance are rapidly opening the door to commercialization, and III-N technologies are demonstrating exciting developments of late. Though devices are entering commercialization, there is still some work to be studied. In particular, GaN high electron mobility transistors (HEMTs) show potential as gas sensors, which will be relevant to the emerging hydrogen fuel cell vehicle market. Devices were fabricated and show a very low detection limit of < 10 ppm, and have been integrated in a wireless network that is currently undergoing field testing. Devices have also been fabricated on different substrates such as Si (low-cost), SiC (high performance), and SopSiC (novel material ? potentially combines advantages of Si and SiC) to study the effects on device performance and their compatibility with processing for 3-D integration. Reliability testing is a major area of interest. A 32-channel stress test system has been designed and is currently being built. A 3-D integration project is being undertaken to achieve 3-D bonding of integrated circuits and other components fabricated from dissimilar materials. The thermal design of a vertically integrated multi-chip-module (MCM) based on GaN High Electron Mobility Transistor (HEMT) power amplifiers (PA) on SiC substrates with a backside heat sink/antenna and Si modulator, bonded to a common ground plane using polydimethylsiloxane (PDMS), was studied. Heat transfer was estimated using finite element modeling for different PA power densities, HEMT gate finger pitch, layer thickness, the presence or absence of the thermally insulating layers, and the thickness of dielectric isolation interlayers. Laser drilling is a promising method for through hole via formation in SiC, presenting several advantages over dry etching, the most important being considerably higher etch rates. Studies have been undertaken to minimize the surface contamination and semiconductor degradation due to this process and, ultimately, make this process competitive with conventional dry etch techniques. By using a UV excimer laser source (193 nm) for drilling, we can achieve considerably better smoothness inside the holes and minimize surface contamination, compared to the use of the more common Nd:YVO4 laser. The device characteristics of AlGaN/GaN HEMT layers grown on SiC substrates were similar after formation of vias by 193 nm laser drilling to those from an undrilled reference sample. By sharp contrast, 1064, 532, and 355 nm laser drilling produces significant redepostion of ablated material around the via and degrades the electrical properties of the HEMT layers. Flip-chip bonding is critical for 3-D integration is it is the method that actually forms the bonds. Studies have been performed to look at various materials systems, particularly In, Au, and SU-8 polymer, as candidate materials for flip-chip bonding applications. Bonding protocols have been developed to optimize the mechanical strength of the bond for a MEMS application. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2008.
Local:
Adviser: Ren, Fan.
Statement of Responsibility:
by Travis James Anderson

Record Information

Source Institution:
UFRGP
Rights Management:
Copyright by Travis James Anderson. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
7/11/2008
Classification:
LD1780 2008 ( lcc )

Downloads

This item has the following downloads:


Full Text































*.C I'- s.,.;u If M*r 1 1 ;


M*g'* .O*


Figure 8-9. SEM images of cross-sections of via holes formed in poly-SiC substrates using DPSS

laser (left) and excimer laser (right) at low magnification (top) and high magnification

(bottom)


Figure 8-10. Optical image of array of holes formed in GaAs


I .- ; Ii. .


,, 9i ,Iup rI ..T. ; Ir F i'n,7 T.T., II : -


r~~ ;r~~ ;rr~r~ ;~ ~rri- rals~
u~ u. ; I~*r~r,;


Yl 'W' 'Yl' "'~ "' '''~"'"
n~ Rry VU1'4~ '1F rl*lr4









wafer consists of the SiC substrate, followed by 0.1 [im of thermal Si02, and 0.2 [im of (111) Si.

The Molecular Beam Epitaxy (MBE) growth in the rf N2 plasma-assisted mode of the HEMT

structure began with deposition of an -0.5 rm thick GaN buffer, 1.4 rm additional buffer of

Al0.25Gao.75 N, and was followed by the active regions of a 100 A thick GaN channel, then a

GaN/AIN 10 period superlattice with total thickness 32 nm and average Al content of 32 at.%, a

Al0.25Gao.75N barrier (50 A) and undoped 20 A GaN cap layer. The sheet resistance of the HEMT

was 456 Q/ sq, with a sheet carrier density of 7 x 1012 cm 2 and electron mobility at 300 K of

1400 cm2 /V-s. A cross-sectional transmission electron microscopy (TEM) micrograph of the

entire structure is shown in Figure 4-6. The structure shows single crystal nature and clean, sharp

interfaces, while there is the usual density (-5 x 109 cm 2) of threading dislocations originating

from the lattice mismatch at the hetero-interface. The -500 .im thick polycrystalline SiC

provides electrical isolation, heat dissipation and mechanical strength. This is followed by -200

nm of Si02 which establishes the quality of the wafer bonding and finally -470 nm of high

resistivity (111) Si which provides a suitable surface for growth of hexagonal GaN and also

provides electrical isolation. Device fabrication followed the steps discussed in the previous

section for SiC HEMTs, with the exception being that the gate length was written as 0.5 |tm in

the e-beam lithography system [73].

Figure 4-7 shows a typical IDs-VDs characteristics from a HEMT. At top, the device 2x100

lm2 gates with 2 [im gate-drain spacing. All of the devices showed excellent gate-drain

breakdown voltages in the range 40-350V for gate-drain spacing of 2-125 rim, where the reverse

breakdown voltage was defined as the voltage at which the reverse current density was 1 LA-cm

2. These correspond to breakdown fields in the range 3.2-20 x 104 V/cm and are comparable to

HEMTs on other substrates with the same sheet carrier density. The maximum drain-source









7. Close the nitrogen valves, open the valve to the roughing pump and chamber
8. Wait for the chamber to pump down. Switch to the turbo pump by closing the roughing
pump valve opening the valve when the pressure = 100 |tm Hg. Turn off roughing
pump
9. Wait for the chamber to pump. Switch to the cryo pump by closing the turbo pump
valve and chamber valve and opening the gate valve when the pressure = 10 |tm Hg.
Turn off turbo pump.
10. Wait 4-6 hours for the system to reach high vacuum (pressure < Ix10-6 mTorr)
11. Turn on chiller, wait 10 minutes for warm up
12. Turn on power supply, wait 10 minutes for warm up. Turn on the deposition controller
and set the deposition parameters as per the log book while waiting.
13. Turn on the high voltage (key and push button) and the beam sweep (push button).
14. Verify that the beam position is near the center of the crucible. The beam should not
need adjusting.
15. Monitor time during deposition. Do not deposit for more than 5 minutes to avoid
damage to the system. Always wait 5 minutes between depositions.
16. When deposition is complete, wait 10 minutes and then turn off the power supply. Wait
another 10 minutes and turn off the chiller.
17. Vent the chamber, remove the sample, and pump the chamber again.

A.8 Thermal Evaporator

1. Turn on the control panel, roughing pump and turbo pump
2. Open the ball valves for nitrogen and house chilled water supply and return
3. Vent the chamber using the toggle valve at the system interface
4. Use the hoist toggle switch to raise the chamber top
5. Remove the glass chamber wall and clean with acetone if necessary
6. Load the boat with metal charges and replace the glass wall. The relationship between
metal mass loaded and final film thickness is linear, and a calibration curve should be
established with all new metals.
7. Lower the chamber top and begin roughing the chamber by turning the 3-way valve to
the rough position
8. When the chamber pressure reaches 100 |tm Hg, change the 3-way valve to the foreline
position, wait a minute to pump out the line, and then open the turbo pump valve
9. Turn on the high vacuum gauge. Wait for the chamber to pump down (<1x10-5 mTorr)
10. To deposit, turn on the current supply. Manually ramp up the current. Dwell for 1
minute at a low setting relative to the deposition current for initial heating.
11. Slowly ramp the current to an intermediate setting and watch the metal charge for
melting. Dwell until the charge is completely melted.
12. Ramp the current to the deposition level and open the shutter. Evaporate the entire
charge.
13. Slowly ramp down the current supply, turn it off and wait 5 minutes for cooling.
14. Turn off the turbo pump valve, turn the 3-way valve to off, and vent the chamber.
15. Clean the chamber wall and pump the chamber.
























I5' '.-


.0


0


0
4


U'


Figure 8-7. Optical microscope images of front and back of via holes formed in poly-SiC
substrates with a DPSS laser (left) and excimer laser (right) from the top
semiconductor surface (top) and bottom SiC surface (bottom)


Figure 8-8. SEM images of via holes formed in poly-SiC substrates with a DPSS laser (left) and
excimer laser (right) from the top semiconductor surface.


* ,









C.2.5 Image Reversal (5214)


1. spin 5214, 4000 RPM, 30s
2. soft bake, 100 C, 1.5 min (oven)
3. expose 10s with mask
4. bake, 110 C, 2 min (hot plate)
5. flood expose, 30s
6. develop 30s in AZ400K:H20 (1:4)
7. hard bake, 110 C, 30 min (oven)

C.2.6 Air Bridge

1. spin PMGI, 3000 RPM, 30s
2. soft bake, 180 C, 20 min
3. spin 1045, 3000 RPM, 30s
4. soft bake, 110 C, 5 min
5. expose 25s, develop 45s in MF-322
6. hardbake, 110 C, 15 min
7. Etch PMGI in 02 plasma
8. Remove photoresist in acetone
9. Bake at 180 OC for 1 hour to round the edges of the PMGI
10. Deposit seed Au layer in evaporator (1000 A)
11. Pattern air bridge level using 1045 resist (see previous section for processing)
12. Electroplate Au (-5 |tm) in opened areas
13. Remove photoresist with acetone
14. Remove seed layer by plasma etching

C.3 E-Beam Lithography

C.3.1 Etching (PMMA)

1. Spin PMMA to desired thickness (2% = -1000, 4% = -2000), 4000 RPM, 30s
2. Remove photoresist form one corer for conductivity in the SEM using acetone
3. Soft bake, 195 C, 10 min
4. Expose at 10 kV, 100 [tC/cm2
5. Develop 30s in MIBK:IPA (1:3), Rinse with IPA

C.3.2 Lift-off (PMMA/PMMA-MAA)

1. Spin PMMA-MAA copolymer to desired thickness (7% = -2000), 4000 RPM, 30s
2. Remove photoresist from one corner for conductivity in the SEM using acetone
3. Soft bake, 195 C, 10 min
4. Spin PMMA to desired thickness (2% = -1000, 4% = -2000), 4000 RPM, 30s
5. Remove photoresist from one corner for conductivity in the SEM using acetone
6. Soft bake, 195 C, 10 min
7. Expose at 10 kV, 100 [tC/cm2


206









then be stepped across a sample (i.e. a 4 x 4 array of squares can be stepped twice in each

direction to create an 8 x 8 array). The third, and most intriguing, method is to coordinate beam

scanning with stage translations to control the vertical dimension as a function of x and y

position. This can create stepped, sloped, or rounded walls [138].

The ultimate resolution of the beam will be determined by geometry, depth, optics, and

material properties. Patterning features to micron scale resolution is quite possible with

appropriate quality focusing optics, and 10 rtm is possible with a standard system. There will be

an aspect ratio limitation as well though, with limitations from about 10:1 to 20:1, depending on

the material. Hard materials such as SiC will require a large fluence, and there will be a

demagnification factor associated with the projection process to achieve said fluence. This will

thus limit the feature size and aspect ratio achievable. For polymers and thin films, it is much

easier to achieve high resolution patterns because the material couples with the laser very well

[138].

Upon irradiation with a UV laser beam, four processes can occur: photolysis, electron

ejection/ionization, electron-hole pair generation, and heating. Photolysis refers to molecular

bond breaking, i.e. the following reaction:

AB + hv A + B

The bond that is broken in the process can represent a bond between diatomic molecules,

resulting in individual atoms A and B; a bond within a molecule, creating fragments; a bond

between a molecule and a surface, resulting in desorption; or a bond within a molecular solid,

which is broken. This is the mechanism for laser ablation. This process occurs when the photon

energy from the laser beam excites a molecular bond, as typically the case with UV wavelength














Table 7-1. Effect of boundary conditions on package temperatures
Top Forced Free Forced Forced Free
Bottom Forced Free Free Fixed Fixed
GaN 124 227 141 57 57
Si 96 198 111 25 25
Antenna 92 196 109 25 25


High Resistivity Si

II I IIIII II II I IIIII I


3-D Antenna

Figure 7-1. First-generation prototype of integrated heat sink/antenna incorporating GaN PA and
Si modulator


Figure 7-2. Concept of the 3-D multi-layer structure combining heat-sink antenna, GaN RF and
silicon signal processing electronics



















































Figure 3-13. Comparison of pattern transfer via wet and dry etching


Wirebond TAB

EpO,, ACI.L,.' Irinnlr Lead Bond
Wire | lce Enr:c apulaiorr
-dhe I.., h e L
Pad




CHIP CARRIER


Two options: Two options:
Ball bonding Face up chip
Wedge bonding Face down chip


FIGURE 9.2 Chip to package or substrate interconnection techniques.


Figure 3-14. Schematic of packaging schemes


FipI



TAB Be in
e.a-.j Bu


Three <
* N
* N
a
* A


)









micromachining and scribing. There is current interest in developing laser drilling processes to

create through-wafer vias in hard compound semiconductor substrate materials, such as SiC, for

both electronic and micro-electro-mechanical systems (MEMS) applications. AlGaN/GaN

HEMTs grown on SiC are very promising for high power, high temperature devices due to the

superior thermal conductivity of SiC, compared to the original sapphire substrates. Monolithic

mm-wave integrated circuits based on these devices will require through wafer via connections

to provide a common ground and low inductance path from the source of the FET to the ground

plane in the back of the wafer [19-36].

1.2 Study Outline

There are four sections of this dissertation. In addition to providing the results of research

performed over the past four years, I also wish to share my background knowledge on the subject

to create a sort of condensed manuals for future students in the group. To that end, the next

couple of chapters will provide purely background information. Chapter two will provide a

cursory literature review of compound semiconductors, from initial research and development up

to the current state of the art. Chapter three will provide a short summary of device processing,

starting with substrate wafer fabrication and the processing steps for device fabrication, all the

way up through device packaging, circuit board fabrication, and systems assembly and testing.

These two chapters are intended serve as an introduction to the field, and suggest first-day

literature for new students to the group.

The next sections will cover research performed over the past four years in detail. This can

be divided into two main sections: discrete devices and packaging-level work. Chapters four,

five, and six will focus on the devices, and cover efforts to fabricate high performance devices

(Ch. 4), gas sensor devices (Ch. 6), and reliability and stress testing (Ch. 5), which is currently a

hot topic. Chapters seven, eight, nine, and ten will focus in 3-D packaging work. Topics to be









because it offered the lowest temperature to still make a solid bond that exceeded the strength of

the bond form the pad to the substrate. Real devices were then bonded using this method and

successfully released. An optical microscope image of the device die is shown on Figure 10-4,

and an image of a bonded die after release is shown in Figure 10-5 [146-148].

10.3 Device Demonstration

After device fabrication, the die were bonded to carriers and wire bonded for testing.

Reflectance was measured using a white light source directed at the reflector structure. A tuning

voltage range of 0-10V was investigated and a reference sample consisting of Au evaporated on

Si was used as a reference reflectance standard. The results are shown in Figure 10-6. A

wavelength tuning range of 930-990 nm was achieved. Though the design range matches the

actual wavelength range, the design actuation voltage vs wavelength curve does not correlate

well compared to the real measurement. The reason for this is mainly due to stress factors that

were not accounted for. The DBR was not planar, and actually formed a concave cavity. This

was verified using interferometry, with the data shown in Figure 10-7. When a simulation was

performed using a nonplanar cavity, the results correlated well with real device results, shown in

Figure 10-8. All design, simulation, and testing was performed at AFRL, while the bonding was

performed at UF [146-148].









diatomics, thus the proper term, though not widely used, would be exciplex. A table of excimer

molecules and their wavelength is listed in Table 8-2. Typically excimer lasers use a mixture of

<1% halogen (F2, C12) gas, 2-5% rare gas (Xe, Kr, Ar), and the remaining 94-95% a buffer gas

(Ne, He). The bulk of this discussion will refer to an ArF excimer gas at 193 nm. The primary

reason for this is because this is the wavelength of the system that will be purchased, and

therefore it is the most relevant wavelength. Note though that since an excimer laser is a gaseous

source, the source gas and thus wavelength of the system can be changed by simply purging and

filling the chamber with a different mixture. The optics must also be changed to be compatible

with the new wavlength as well, but the entire assembly can be purchased for -$5K [138].

The principle of operation is very complex. The buffer gas absorbs most of the initial

excitation energy, then an electron is transferred to the halogen, creating a negative ion that bond

rapidly with the rare-gas positive ion to produce an ion pair in an excited state. As the molecule

falls back to a dissociated ground state, it emits at a characteristic wavelength (bound-free

transition) [136]:

Kr + F-+ Kr Kr+F- + Kr Kr + F + Kr + hv (248 nm)

The gain in this system is inherently high (up to 2x104 in a single pass) since the density of

excited states exceeds the unbound ground state population in the laser cavity. Excimer systems

therefore do not require the high reflectivity cavity mirrors to provide optical feedback and can

operate with a rear reflector and an uncoated front window. Excimer lasers are pumped by a fast

electrical discharge. In this type of situation, the excimer action must be achieved before

breakdown and arcing occur, so pulses are short (10-50 ns) [136].

All systems will degrade with time as the halogen gas slowly reacts and absorbing

impurities form in the gas mixture. For this reason, a periodic maintenance issue with excimer












ctn


400-


300-



-.. 200-


100-


0-


ksubstrate (Wm-oC)


Figure 7-5. Effect of substrate thermal conductivity on the junction temperature of the HEMT


130-


120-


110-


S100-

I-
9 -


80-


70-


Power Level: 3 wlmm
h=30 WIm2z-C
GaN Thickness=5 pmn
No wire
No via



/
//
/


0 100 200 300 400

SiC Thickness (jrm)


500 600


Figure 7-6. Effect of silicon carbide substrate thickness on the junction temperature of the
HEMT


Sapphire
Substrate Power Level: 3 W/mm
Substrate h= 30 W/m

Substrate Thickness=500 prm
GaN Thickness=5 pm
No wire
No via



GaN .-----
Substrate SiC
Substrate









a particular concern when dealing with these systems due to the dangers associated with these

precursors (silane is highly flammable, arsine is toxic, etc). There are three main components of

an MOCVD system: gas delivery, reactor, and exhaust. Typically the organometallic sources are

enclosed in bubblers, with hydrogen as the carrier gas, and the group V source as a compressed

gas. Extreme care must be taken in the delivery system to avoid condensation or pyrolysis of the

organometallic precursors. A general issue is the gas flow pattern and introduction into the

chamber. To obtain atomically smooth interfaces, the valves must be designed so that there will

be no perturbations as flows start and stop. A schematic of an MOCVD reactor is shown in

Figure 3-6 [68].

3.3 Dopant

Impurity dopant is used to change the electrical properties of semiconductors. It is an

important step in every device fabrication process since it creates the p-n junctions that are the

key active parts of the devices. The doping process must be very tightly controlled and

reproducible. It is critical to control the process both in terms of introducing the impurity

concentration to design specifications and controlling the area where the impurity is introduced.

There are two methods of doping semiconductors: thermal diffusion, which is typically used to

create deep wells, and ion implantation, which is used to create precise shallow junctions or very

high impurity concentrations [69].

3.3.1 Diffusion

Diffusion is a fundamental transport phenomenon that involves the redistribution of free

material in response to a concentration gradient. The material flux is proportional to the

concentration, and movement will be away from the high concentration region towards the

region of lowest concentration. There are two basic atomic diffusion mechanisms in

semiconductor crystals: movement through vacancy lattice sites, and movement through









Convective heat transfer involves energy exchange between a surface and adjacent fluid.

Natural convection occurs when there are no external forces driving the fluid, just the natural

circulation due to density differences in the fluid from temperature variation. Forced convection

occurs when there is an external driving force such as a pump or fan. Convective heat transfer is

described by Newton's law of cooling, also known as Newton's rate equation [126].

q= hAT
A

In this equation, h is the convective heat transfer coefficient. It is not a constant and

generally is a function of system geometry, fluid, flow properties, the difference in temperature,

and the type of convection present.

Radiative heat transfer occurs between two surfaces and does not require a medium to

propagate. This is the only method of heat transfer in a vacuum. The energy emission by a

perfect radiator (black body) is given by the Stefan-Boltzmann law of thermal radiation [126].

q T oT4
A

This equation must then be corrected for deviations from black body behavior. It is

common for all three methods of heat transfer to be taking place simultaneously. For this work,

the effects of radiative heat transfer are minimal compared to conduction and convection.

Consider as an example a situation where a hot gas is contacting a 3-material composite

wall, with a cold gas on the other side. The heat transfer can be described as a series sum of

convective heat transfer between the hot gas and the first surface, conduction through the first

material of thermal conductivity kl, conduction through the second material of thermal

conductivity k2, conduction through the third material of thermal conductivity k3, and

convection again at the wall. This is shown schematically in Figure 7-3. Considering these walls









APPENDIX B
ELECTRICAL MEASUREMENTS

B.1 Purpose

This section is meant to provide information on the standard electrical measurements that

are performed to quantify HEMT performance. Background and system setup diagrams will be

given when applicable. A thorough knowledge of electrical testing is a useful tool. Just as much

diagnostic information, if not more, can be gathered from a round of electrical testing as from a

full battery of analytical equipment.

B.2 Ohmic Contacts

B.2.1 Overview

Transmission Line (TLM) testing is used to quantify the ohmic contacts. A typical

transmission line consists of a series of ohmic contacts with varying spacing on one mesa. By

measuring the current-voltage curve between each contact, one can calculate the resistance from

Ohm's Law (V=IR) as the simple slope of the line. The resistance is a linear function of the

spacing between contacts, so from here one can fit a curve if these values are plotted. The

resulting slope will be the sheet resistance (Rs), which is reported in ohms/sq. This is the

resistance purely of the semiconductor material. In an ideal situation, the intercept of this line

will be zero, but since the contacts themselves have some resistance, the slope will represent the

transfer resistance (RT), which is reported in ohm/mm. These factors are combined into a specific

contact resistance (Re), which is defined as Rs/RT, and is reported in ohm-cm2. A good ohmic

contact will have Re -1x10-6 ohm-cm or smaller.

B.2.2 Procedure

An optical image of a TLM pattern is shown in Figure B-l to demonstrate the terminals.

Use the parameter analyzer and probe station for this test, as shown in Figure B-2. Set up the



















2-dimensional
electron gas


GaN


L-x


Figure 2-1. Energy band structure of AlGaN/GaN heterojunction


Sapphire


Figure 2-2. Cross-section of an AlGaN/GaN HEMT


-Al


AIGaN









C.5 Plasma Enhanced Chemical Vapor Deposition


C.5.1 Silicon Nitride (SiNx)

Temp = 255 C
Pressure = 900 mTorr
RF Power = 30 W
DC Bias = 0
Ref Power = 0
Gas flows: SiH4


Dep Rate = -
Refractive in


100


N2 = 400
NH3 = 27
115 A/min
dex = -1.7-1.9


C.5.2 Silicon Dioxide (SiO2)

Temp = 255 C
Pressure = 900 mTorr
RF Power = 30 W
DC Bias = 0
Ref Power = 0
Gas flows: SiH4 =


Dep Rate = -
Refractive in


200


N20 = 200
300 A/min
dex = -1.38-1.46


C.6 Wet Etching

C.6.1 Buffered Oxide Etch (SiNx, SiO2)

SiNx etch rate = -1000 A/min
SiO2 etch rate = -6000 A/min
Aqua Regia (Ni)
Formulation = HN3:HCl:H3P04 (3:1:1)
Ni etch rate = -1 tm/min

C.7 Thermal Evaporation


C.7.1 Indium

Current = 200 A
Deposition Rate = 125 A/s
1 In charge = -1 tm film


208


































2008 Travis James Anderson









For the MIPS project, this allowed for formation of 8 |tm mesas in SiC (etching down to

the n-type layer). The Ni can then be stripped very rapidly and selectively using an aqua regia

etch (1:3 HNO3:HC1). Further processing involves the deposition of the n-type ohmic contact,

which is Ni, patterened by lift-off and annealed at 700 OC. The p-type ohmic contact is also

deposited by evaporation, and consists of Al/Ti/Au, annealed at 200-300 C. A passivation layer

of 1000A SiNx is deposited by PECVD. This represents the entire device structure, shown in

cross section in Figure 9-4. An optical image of a 10 finger device is shown in Figure 9-5 [145].









if it is moved in a direction off-axis. The bond is formed at a lower temperature (50-125 C), a

controlled wedge bonding force, and ultrasonic excitation. This method essentially applies a

force to pinch the wire and the ultrasonic excitation forms the bond [21].

The advantages of wire bonding are the extremely high reliability, ease of automation, and

high level of industry infrastructure support. Current automated equipment can bond an entire

chip in a fraction of a second. The disadvantages are that it is a serial process so it is an

inherently slower bonding process relative to flip-chip, long bond lengths due to the loop of the

wire can degrade chip performance, especially for RF packages, there is a larger chip footprint

required again due to the wire loop and routing outside of the chip, and there is potential for wire

movement and shorting during encapsulation [21].

3.7.2 Flip-Chip Bonding

Flip-chip bonding involves quite literally the face-down direct bonding of the chip pads to

the carrier. This method uses solder bump (sphere) interconnections to form both the mechanical

and electrical bond, in contrast to wire bonding where an epoxy is used to mount the chip to the

carrier and then the gold wires form the electrical connection, with no contribution to mechanical

strength. This bonding method is one of the most significant packaging developments to improve

cost, reliability, and productivity. There are considerably more processing issues with this

method, particularly compatibility between chip pad metallization and solder material and I/O

routing issues. Solder bumps are deposited using optical lithography to define the bump pattern

and electroplating or evaporation to deposit the metal. After patterning, the bumps are metal

posts, and are then reflowed to homogenize the structure, since phase separation can occur

during electroplating [21].












At350 "C
-- Regular
Boide Source,IEain
10 P/Au
M /Au
---Pt/BorideAu
0 +/Bcride/Au
\ --* ------
-to
-10 i .O- ----___--
A -20 -




-650


-70 -
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Time (Days)

Figure 5-1. Changes of IDS as a function of time at 3500C for HEMTs with different
combinations of Ohmic and Schottky contacts.


RT
B 15(P C



0E 10,
1 4!.. -




2 4 2 8 3 2 3.6
10 00'Tempermture
0 20 40 6& 80 100 120
Stre time, Hnurs

Figure 5-2. RF power output as a function of time









current, Imax, gm, and threshold voltage (Vth) high-temperature (>150 C) DC 3-terminal

aging test in the pinch-off state will be conducted and focused on gate leakage current equivalent

to high power operation. Third, we will study the effects of fabrication techniques on current

degradation mesa isolation vs. ion implantation isolation, different Ohmic and Schottky

metalliation, Schottky vs. metal oxide semiconductor (MOS) based gate contact, and different

passivation approaches will be included in the study. Fourth, we will study the effect of DC

operation conditions on the degradation different drain voltages, drain currents, and junction

temperatures will be used in the study. Finally, we will study accelerated life tests performed at

elevated temperatures.

Effectively our protocol will be a combination of Si industry-standard approaches and

areas specific to compound semiconductors (such as surface passivation and a higher defect

concentration). We will run sufficient samples to get decent statistics and determine the lifetime-

limiting factors at different temperatures. The usual Arrhenius approach to obtaining operating

lifetime from an extrapolation of high temperature data clearly is a problem since mechanisms

that occur at high temperature may not really be a factor at lower temperatures. We will use a

full range of device and materials characterization methods to better understand the factors

determining the reliability of GaN HEMT technology. Failed devices can be analyzed to

determine if macroscopic defects (dislocations) have played a role in the device break down. We

largely have these capabilities in house so we can provide quick turnaround on device

fabrication.

5.3 Stress Test System

A block diagram of a single frequency 32-channel test system is shown in Figure 5-3. The

system we are going to build consists of 4 RF driver boards with an output to 2-way splitters,













400

Vg = 0
350 -
Vstep = -1 V
300 -

250 -

200

150

100 -

50 -

0
0 1





Figure 4-8. DC I-V curve for SopSiC HEMT


Figure 4-9. DC I-V curve for SopSiC HEMT









CHAPTER 10
FLIP-CHIP BONDING

10.1 Overview

Flip-chip bonding is one of the biggest advances in packaging technology. It has been used

primarily as a method of packaging high performance electronic components to achieve shorter

bond lengths, smaller chip size, and a higher I/O density. A new trend that is emerging is the use

of flip-chip bonding for device integration, particularly for MCM electronic devices and MEMS

devices. This process involves individually fabricating and optimizing components (i.e. an

amplifier and a sensor) and then integrating them using flip-chip bonding. This can be viewed as

a system-on-package (SOP) approach, rather than system-on-chip (SOC) approach, which

involves directly fabricating all components on one chip. The disadvantage of the SOC approach

is that process integration is particularly difficult, such as integration of Si and compound

semiconductors on one chip, and typically performance of one or both components must be

sacrificed in order to achieve the desired integration.

Standard flip-chip processing has been discussed in Chapter 3. This chapter focuses on the

usage of flip-chip technology to integrate electronics components (in this case a VCSEL) and

MEMS (a DBR structure) to create a hybrid device. The goal of this work is to create a tuneable

filter. The DBR structure is a GaAs/AlGaAs superlattice and the emitter is a Si-based reflector to

demonstrate the concept. The ultimate goal would be to fabricate a tuneable filter where the DBR

structure is electrostatically actuated using metal arms. By raising and lowering the DBR, one

could change the resonant frequency of the device, thus creating a tuneable filter. This approach

is a very hot research topic and many such integration schemes have been proposed. Another

such idea involves bonding a photodetector directly to an amplifier chip [146-148].









B .4.1 O overview ................................................................ ....186
B .4 .2 P ro c ed u re ...................... .. ............. .. ................................................18 6
B.5 Leakage/Isolation ........................ .... ............................... .... 187
B .5.1 O overview ................................................................ ....187
B.5.2 Procedure ................................. .............................. ........ 187
B .6 Breakdow n/H igh Pow er ................................................................... ............... 188
B .6.1 O overview ................................................................ ....188
B.6.2 Procedure ........................................... ... .................. 189
B.7 Sm all Signal RF (S-param eters) ............................................................................. 189
B .7 .1 O v e rv iew ........................................................................................................ 1 8 9
B.7.2 Procedure .................. ......... .......................190
B .8 L large Signal R F (L oad-Pull)................................................. ............................. 191
B .8.1 O overview ................................................................ ....191
B .8 .2 P ro c e d u re ....................................................................................................19 2

C PR O C E SSIN G R E C IPE S ................................................................................. ..... ...... 205

C 1 P u rp o se ................... ...................2............................5
C.2 Photolithography ...................................................... ............ ....... ...... 205
C .2.1 M esa Etching (S-1045) ............................................... ............................. 205
C .2 .2 L ift-O ff (S-18 18) .............................................................. ...........................205
C .2.3 Lift-O ff (S-1808) .............. ...... .................................................................205
C.2.4 Lift-Off (LOR/1808) ..................................................................... 205
C .2 .5 Im age R eversal (52 14).............................................................. .....................206
C .2.6 A ir B ridge ..........................................................................................206
C .3 E -B eam L ithography .............................................................................. ............... 206
C .3.1 Etching (PM M A ) ....................................................... .............................. 206
C.3.2 Lift-off (PM M A/PM M A-M AA)................................... .................................... 206
C.3.3 T-gate (PMMA/PMMA-MAA/PMMA).............................................................207
C .4 P lasm a E tch in g ..................... ........ ......... ...................... .. ... .. ........ .. ................... 2 0 7
C.4.1 Inductively Coupled Plasma Etch With Cl2 (AlGaN/GaN and Au)..................207
C.4.2 Reactive Ion Etch W ith CF4 (Si02, SiN x)...........................................................207
C.4.3 Reactive Ion Etch W ith 02 (polymers) ................................... .................207
C.5 Plasma Enhanced Chemical Vapor Deposition .................................. ...............208
C.5.1 Silicon Nitride (SiNx)..................................... ............................ 208
C .5.2 Silicon D dioxide (Si0 2) ........................................................................... .... ... 208
C .6 W et E tch in g .............................................................................................................. 2 0 8
C.6.1 Buffered Oxide Etch (SiNx, SiO2) ............................... ... ....................... 208
C .7 T herm al E vaporation ........................................................................... .....................208
C .7 .1 In d iu m ...................................................................................2 0 8
C .8 Flip-C hip B ending .............................................. .. .. .... ................. 209
C .8.1 Indium -G old ............ ........ ................................ ...................209
C .8 .2 G o ld -G o ld ...................... .. ............. .. ...............................................2 0 9
C.8.3 Polym er (SU -8-SU -8) ........................... ..............................................209





9










9.1.2 A lpha-V oltaics ...................................... ...................................... ......... .. ...... ... 15 1
9.2 Electroless Plating Procedures ......................................................... ............... 152
9.2 .1 Solutions M ade In-H ou se ......................................................................... ... ... 152
9.2.2 Com m ercial Plating Solutions........................................................... ... .......... 155

10 F L IP -C H IP B O N D IN G .......................................................................... ........................163

1 0 .1 O v e rv iew ................................................................................................................. 1 6 3
10.2 Bonding Studies........... .... ................... ................. 164
10.2.1 M etal Bonding ............... ................. ........... ............ ......... 165
10.2.2 Polym er B ending .................................................... ...................... 166
10.3 D vice D em onstration ....................................................... ........ ............. 167

11 FUTURE WORK ............... ................. ........... ............................. 172

11.1 Processing Studies ....... ...... ... ...... ...... ................ 172
1 1 .2 R eliab ility ................................................................... 17 2
11.3 G as Sen sors ......... ..................................... ...........................173
1 1.4 S im u nation s ...............................................................17 3
11.5 Laser Drilling ................................................. ..... ..................... 174
11.6 E lectroless P lated M etal ................................................................. ........................ 175
11.7 Flip-Chip Bonding ............... ................. ........... ................. ....... 175

12 C O N CLU SIO N .......... ............................................................... ................. .. 177

APPENDIX

A E Q U IPM EN T O PER A TIO N ......................................................................................178

A 1 P u rp o se ................... ...................1.........................8
A .2 Spin-C eating ........................................................................... 178
A .3 M ask A ligner (M JB -3).......................................................................... ................... 178
A .4 R aith E-B eam Lithography .............................. .... .............. .................................. 179
A .5 Plasm a-Therm PECVD .............................................................. ................ 180
A .6 Evaporator (CH A)................................................................ .......... .... 181
A.7 Old Evaporator................................ .. .... ...... .... .......... 181
A 8 T h erm al E v ap orator ................. .................................................. .. ........ ........... .... 182

B ELECTRICAL MEASUREMENTS .............................................................................183

B 1 P u rp o se ................... ...................1............................3
B .2 O hm ic Contacts .............................................. 183
B .2.1 O overview ................................................................ ....183
B .2 .2 P procedure ......................................................................................... ....... 183
B .3 D ire ct C u rren t .......................................................................................................... 1 8 4
B .3.1 O overview ................................................................ ....184
B.3.2 Procedure .................. .................................... ........ ..185
B .4 P u lse (G ate L ag )............................................................................................. ..... 18 6


8






































Rath 150 30pm*
Mag= 276X ---|H


En I = 10u.uu V signal A = InLns Date :14 Apr 2{j0
WO= 12mm User Name = TRAINING Time :142:05


.4:.S -Z o 0



Rith 150 10 pm EHT =1000 kV Signal A = InLens Dae :14Apr2f05
Mag= 892X WO= 12mm USer Name = TRAINING Time :140605



Figure 8-2. SEM images of hole drilled from front at 1064 nm (above) and 532 nm (below)









CHAPTER 1
INTRODUCTION

1.1 Motivation

The Galluim Nitride (GaN) materials system is attracting much interest at an industrial

level in the late 2000s. Due to the wide-bandgap nature of the material, the material is very

thermally stable, and electronic devices can be operated at 500 OC. The material is also

chemically stable, with the only known wet etchant being molten NaOH or KOH, making it very

suitable for operation in chemically harsh environments or in radiation. This system also

demonstrates a high electron mobility, making it suitable for the high frequency communications

market, and high breakdown field, making it marketable to the high power industry as well.

While this technology has had a very long infancy as researchers strived to develop high quality

material on high performance substrates, and develop novel processing schemes to improve

device performance to the inherent material limits, it appears in the past couple years that the

technology is beginning to mature and move from the lab into production.

1.1.1 Electronic Devices

There is currently strong interest in developing AlGaN/GaN High Electron Mobility

Transistors (HEMTs) for use in high power microwave transmission systems, millimeter-wave

(MMW) military communications links, and X-band radar systems. Nitride-based HEMTs can

operate from very high frequency (VHF) through X-band frequencies with higher breakdown

voltage, better thermal conductivity, and wider transmission bandwidths than Si or GaAs

devices. GaN-based HEMTs can operate at significantly higher power densities and higher

impedance than currently used GaAs devices. The higher sheet carrier density in the two-

dimensional electron gas compared to the AlGaAs/GaAs system, and higher electron mobility

relative to Si, suggest that nitride-based HEMTs will also exhibit low on-state resistance, an









the material, while etching refers to the laser excitation of a chemical etchant for enhancement of

the etch. To this end, pure laser ablation has been studied to date in this work. Laser etching

could possibly be studied to selectively etch material, such as enhanced etching of GaN using

KOH [136].

Ablation can be broken down into 3 areas: highly absorbing materials, weakly absorbing

materials, and organic. Highly absorbing materials are capable of being intensely heated by the

laser pulse. Ablation then occurs primarily by sublimation. Metals fall into this category, and it is

for this reason that metals typically machine poorly in UV systems. There is usually an area

around the periphery of the pattern where there is beam-induced damage in the form of an area

that was melted during ablation. Another concern is thermal conductivity. Metals prove difficult

to machine because, though they are highly absorbing, they tend to conduct the energy away

very rapidly, thus requiring high fluence lasers to continuously provide sufficient energy on

target. This is epitomized in an example of a Cu film. Bulk Cu is very difficult to machine with a

UV laser and requires several J/cm2 for ablation to occur, however ablation will occur at 0.1

J/cm2 in a 1000 A Cu film on an insulator such as SiO2. This is because the low thermal

conductivity of the dielectric confines the laser energy to the Cu layer [136, 138].

In weak absorbers, such as semiconductors, several processes may occur. First, basic

photoablation can occur if the energy is larger than the bandgap, which will always be the case

for an ArF laser. Second, multiphoton absorption can lead to electron generation and ionization,

which will cause an avalanche breakdown process and generation of a plasma condition,

resulting in material ejection. This is typically the case for insulators, which have a bandgap

larger than the photon energy. Third, defects will play a major role by acting as localized

absorption centers, which can cause heating, and subsequent breakdown and ejection. Ablation









addition, there is less surface debris on the surface of the sample processed with the excimer

laser, where it is just a small radius of material.

To further examine the drilling processes, the samples were cross-sectioned with a dicing

saw. The SEM images are shown in Figure 8-9, with the DPSS hole on the left and excimer on

the right, and low magnification on the top and high magnification on the bottom. The excimer

hole shows some taper towards the bottom of the hole. This is characteristic of excimer laser

processing, and is actually beneficial for the purpose of the via hole process because the tapered

hole will facilitate metallization. The degree of taper can be controlled with the laser parameters,

and if perfectly straight holes are necessary, either a higher power or overablation can be used.

The disadvantage of higher power is that some control over the etch rate is lost. The excimer

holes show a smoother surface with less beam-induced damage. There are obvious molten

regions on the DPSS hole, visible at high magnification.

As another test, an array of holes in InP and GaAs was formed. The optical image is shown

below in Figure 8-10. An 8x8 array of holes were drilled through a 150 |tm thick wafer with an

entrance hole diameter of 110 rtm and exit hole diameter of 100 rtm, with a pitch of 200 |tm.

While these were processed serially, it would be possible to create a mask that could be stepped,

such as the one described above (a 4 x 4 array stepped twice in both the x and y direction). This

test showed very good feature control and reproducibility.

Ultimately it was decided that the JPSA IX-260 machining system was perfect for the

needs of the research group and the university. It offers the resolution and flexibility of materials

that are necessary. It will be configured initially for 193 nm, but can be altered to perform at any

excimer gas wavelength. In addition, high-resolution optics are available to achieve micron-level

resolution. It has been fitted with an air-bearing stage that provides 0.1 |tm resolution and 1 |tm









growth. The dislocations did not appear to be efficient non-radiative centers, so while optical

devices could be achieved, the defects do affect reliability [37, 47-50].

Rapid progress has been demonstrated in field effect transistor (FET) structures. This

places very strict demands on the growth and processing, but the technology has been developed

in the past decade, and present published results indicate that GaN will play a significant role in

the development of high temperature, high power, and high frequency electronic devices. Further

improvements in materials quality will enhance device operation, though processing technology

must also be improved for significant development. There are many processing issues, such as

poor p-type doping, low resistance and thermally stable ohmic contact development, high

temperature requirements for implant activation, poor wet etching processes, low dry etch rates,

and other issues. As advances are made, we can expect further improvements in device

performance and increasing commercialization [37, 43].

2.2 High Electron Mobility Transistors

High Electron Mobility Transistors (HEMTs) have emerged in the past twenty years as

promising candidates for microwave and high voltage applications. HEMTS were originally

based on AlGaAs/GaAs, AlGaAs/InGaAs, AlInAs/InGaAs, and related systems on GaAs or InP

substrates. In the 1990s, AlGaN/GaN HEMTs on sapphire, SiC, or even bulk GaN substrates

have demonstrated considerably larger power density (>12 W/mm), and have become contenders

for high power amplification and switching applications. The use of wide bandgap

semiconductors presents an advantage because, in addition to the increased output power, the

temperature tolerance and radiation hardness is also extended. Recently, GaN-based HEMTs

have been demonstrated operating at 750 C [43].









CHAPTER 3
SEMICONDUCTOR PROCESSING

3.1 Overview

This chapter is meant to give an overview of semiconductor material growth and device

processing, starting from raw materials all the way up to a functioning circuit board.

3.2 Crystal Growth

The first step towards microelectronic devices fabrication is the substrate wafer growth. In

a melt process, the raw materials (ultra high purity Si, for example) are simply melted and

resolidified in single crystal form. The process is much more complicated than this, as a

considerable amount of control is required to maintain crystal orientation and impurity

concentration over the large quantities of material grown. A related technology is the growth of

thin single crystal layers on the single crystal substrates. This process is called epitaxy. This

process is particularly important for high-performance devices because it allows for precise

control of layer thickness and impurity doping concentration. At the current technology level,

high quality Si and GaAs wafers grown from the melt are widely available, while GaN is grown

epitaxially (bulk GaN wafers are available, but they are still grown epitaxially and are of poor

quality) [69].

3.2.1 Growth from Melt

Silicon is typically grown using the Czochralski method (CZ growth). The starting material

is quartzite, which is a pure form of sand (Si02). It is refined to form ultra high purity

polycrystalline silicon (99.999999999% pure) through various treatments. The first is reduction

in a furnace using carbon, shown in reaction 3.1. This is followed by a HC1 treatment to dissolve

the Si, shown in reaction 3.2. The solution is distilled to remove impurities, followed by a









covered include thermal simulations of a packaged chip to optimize heat transfer (Ch. 7), laser

drilling for interconnect formation (Ch. 8), electroless metal plating for thick film deposition

(Ch. 9), and flip-chip bonding for assembly (Ch. 10).

The fourth section will conclude my research. This will contain a chapter on future work

(Ch. 11), which will outline the steps to be taken to wrap up the current work and suggestions for

where to take the work in the future. Chapter 12 will serve as the conclusion, and wrap

everything up and tie it together.

The fifth section will be operating procedures. Appendix A will cover operating

procedures for processing equipment in the lab, Appendix B will cover electrical testing, and

Appendix C will include tables of recipes used in processing. The intent of this section is to

consolidate information about the lab. I am a firm believer in leaving somewhere in better shape

than it was in when I arrived, and wish to make my contribution by clearly documenting and

arranging the necessary information about the lab. I hope that this section can again be

distributed to new students on the first day to aid in training, and serve as a master document that

is added to and modified by students each year to keep the information current.









Free convection is the only heat dissipation mechanism through the outer vertical-surface

of the MCM. The amount of heat dissipation through free convection is five orders smaller as

compared to the conduction through the integrated heat sink antenna. The thermal resistance at

the convective boundary is -85 K/W, compared to -2x 104 K/W for conduction through the

package. Therefore, the temperature rise at the center of the chip on the top surface was highest

(around 110 C), and there was a -40 C difference between the center and edge of the

structure, based on a power amplifier density of 3 W/mm. To investigate the temperature rise

while the device is turning on, the transient equation was used. The temperature profile reached

steady state in the range of -30 sec [92-93].

Other MCM design considerations that were also considered, but were not found to have a

significant effect on the operating temperature, were the GaN epi-layer thickness (-5 C

difference for 10 utm versus 3 [tm) and the wire thickness and length (-2-8 C difference for wire

lengths from 2-100 mm and thicknesses of 0.2-10 [tm) [92-93].

7.2.3 Generation II: Combined Amplifier and Silicon Chip

After initial study and optimization of heat transfer through the GaN PA on the heat sink, it

is logical to go back and consider the more complex problem of modeling the entire 3-D

package. The optimized GaN PA parameters were used as starting points for this work: 3 W/mm

dissipated power in the GaN power amplifier, a 2 |tm thick GaN layer, 100 |tm thick SiC layer,

and an effective convective heat transfer coefficient, of 30 W/m2-oC. The heat sink was, again,

assumed to be -20 mm2 area, with 11 fins, 10 mm fin height, and 0.94 mm fin width. The

additional layers added in a vertical stack to complete the 3-D structure were the PDMS layers

for bonding (4 jtm each), an Au common ground plane (5 [tm), and the Si modulator chip (500

[tm) [92-93].









8.2.2 Experiments: Frequency Effects

The early work on this project built on the postdoctoral work of Dr. Suku Kim at UF, who

explored via hole formation in SiC using laser processing [135]. The primary goal of this work

has been proof-of-concept for the via hole process and to evaluate systems available on the

market with the goal of purchasing one that will serve the purposes of research at UF. To that

end, samples were sent to several companies for simple processing on their system to evaluate

compatibility and limits. The primary methods used to quantify performance were feature size

and the degree of damage and surface debris, both determined using a combination of optical

microscopy and SEM.

The first studies evaluated the performance of systems at Lenox Laser and US Laser

Corporation. These were selected as they were highly reputable marking systems, however they

were typically used for metal processing. They both used a 1064 nm YAG laser, and US Laser

provided the capabilities to study both frequency doubled and frequency tripled modes. This

allowed us to effectively study the effects of decreasing the wavelength. Performance improved

significantly when the UV region was reached, as predicted by the literature and mentioned in

the discussion both above and below. In this set of experiments, a HEMT structure was used, and

devices were fabricated on the sample after laser drilling [71, 74].

An AlGaN/GaN HEMT structure was grown by MOCVD on a SiC substrate (4H-polytype,

N-doped, 1017cm-3, thickness was -400 tm). Laser drilling was employed in both directions,

i.e. from front-to-back and vice versa. Via holes with nominal diameters of 50-500 |m were

obtained by laser drilling with a Nd:YVO4 laser (k = 1064 nm), frequency doubled Nd:YVO4

laser (532nm) or frequency tripled Nd:YVO4 laser (355nm). The pulse frequency was up to 60

kHz at an average power of 11 W, leading to an average pulse energy of 183 PJ. Mesas were












C


B


E


To source


To drain


To gate


To source


Digital Multimeter


DC Power Supply


DC Power Supply


Local V
Control
Input


Resistor(1 MO)
Output
To drain


Figure B-15. Test setup for breakdown measurement <400 V (top) and >400 V (bottom)


Tektronix 370A

Curve Tracer


To gate


Glassman High Voltage
Power Supply


I


Local V
C ontrol
Input









ADVANCED THREE-DIMENSIONAL PACKAGING SCHEMES FOR
MICROELECTRONIC AND MICROSYSTEM APPLICATIONS




















By

TRAVIS JAMES ANDERSON


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2008















0.8


0.6
f(t)
0.4 -
0" ..,'rl. r.i:c..:,-, C,:' .l.i.:'r ,ni
9 r. pl al. ': I, ..

0.2
E Oul

0.0-'
0 80 160 240 320 400

Time (Hours)



Figure 3-17. Failure density function









DT
V kVT + q + =pc, D- + v 2v
Dt

In this equation, q refers to the total energy generation, A refers to the viscous work rate

per unit volume, and potential and kinetic energy terms are simplified in the right hand side of

the equation. From here, the equation can be further simplified for the purposes of the simulation

by taking the case where there is no fluid motion and all heat transfer is by conduction. This

equation, which is the one used in the thermal simulations, can be written as follows [126]:

DT
pc =kV2Tq
P Dt

In this case, T is the temperature, t is time, p is density, Cp is the heat capacity, k is thermal

conductivity (W/cm-K), and PD is the dissipated power (generated in the device). The latter can

be determined from the product of bias voltage and drain current through the HEMT, divided by

the HEMT layer volume [126].

To solve the differential equation, boundary conditions must be specified. The initial

conditions refer to the values at the start of the time period of interest. For this equation,

temperature is the only variable, and therefore the initial condition can be considered to be room

temperature. The boundary conditions refer to the values of the variable of interest at the system

boundaries. There are three types of boundary conditions for this type of problem: isothermal

boundary, which is a constant temperature (T=const), insulated boundary, where there is no heat

flow across the boundary (dT/dx = 0), and boundaries with a function describing the temperature.

This type of boundary is typically described by a change in the method of heat transfer, such as

conduction transferring heat to a surface and convection removing heat at the surface. This can

be written as follows, assuming a convective heat transfer coefficient and temperature far from

the surface as Tinf[126]:









CHAPTER 11
FUTURE WORK

11.1 Processing Studies

In pursuit of a functional prototype of the 3-D package concept described above, the goal

for HEMT fabrication is to design and fabricate high power devices compatible with the laser

drilling process. Work must be done to investigate fully the effect of laser drilling by examining

HEMT performance before and after laser drilling. On the packaging level, work will also

include investigation of novel oxide dielectrics for device passivation and packaging. To further

push the state-of-the-art of AlGaN/GaN HEMT technology, e-beam lithography based "T-gate"

or "mushroom gate" process needed to be used to achieve very small gate contact area, which

improves the RF characteristics and, at the same time, have the large trace on top, minimizing

gate resistance and further enhancing the RF performance.

11.2 Reliability

The topic of reliability is very critical for the realization of any practical uses of the nitride

HEMT technology. Typical rf stress test systems can be quite expensive ($250K for 8 channels,

up to $1M for >50 channels), so we are in the process of building one reliability test set at UF

using a design adapted from Sandia National Labs. After completion of this system, there is a

tremendous amount of interesting dc and rf reliability tests that can be done, as there is little

work performed to date. Primarily, there should be an investigation into the failure mechanisms

of AlGaN/GaN HEMTs, by stressing devices under a combination of DC, RF, and thermal stress.

FIB and TEM analysis can be performed to look at the semiconductor layers. Typically, for the

conventional III-V devices, failure rate can be linearly related to thermal stress, therefore

accelerated testing is performed at elevated temperature. This is not the case for rf nitride based









them at times, but I learned a lot and the experience gave me new ideas that I brought back to

UF.

I would not be where I am today if it were not for Dr. Cliff Henderson at Georgia Tech,

and former students Drs. Celesta White and Benita Comeau, who took a chance on hiring me as

an undergraduate assistant on a microsystem fabrication project. They took the time to teach me

many of the processing skills that made my transition to Dr. Ren's group easy. This project was

what convinced me to go to graduate school and do research for a living. It was Dr. Henderson

who pushed me to meet with Dr. Ren when I got accepted to UF.

The reason I was able to finish this work was the close support from my friends. In

particular, I would like to thank Justin Sellers, Alex Welsh, Mark Kozek, and Luke Neal. They

have been there for me for at least the past 4 years. Through good times and bad times, they have

been there to support me.

I thank my family for their love and support. I hope this work in some way repays them for

all of the love they have given me over the years. Finally, I want to thank my wife, Kristin. She

went above and beyond anything I could ever ask of her to stick with me and support me. She is

the love of my life, and best friend.









quality, morphology, and electrical properties. Atmospheric CVD is particularly advantageous in

that it allows for a continuous feed system since no pumping is required, and it has high

deposition rates. The reactant gas is simply injected to a showerhead above the wafers on a

heated chain track. The disadvantage to this system is the poor quality and the particle

contamination associated with atmospheric pressure processes. Low pressure CVD is very

popular in the IC business because, while a batch system, it allows for high throughput and

excellent uniformity. PECVD offers the advantage of low temperature processing and the ability

to integrate with a cluster tool, but uniformity can be poor [68].

Silicon nitride and silicon oxide are the dielectrics primarily used with compound

semiconductors. Silicon nitride in particular is important as a passivation and capping material. It

oxidizes very slowly, serves as a diffusion barrier to water and sodium, and has a slower etch

rate in buffered HF than silicon dioxide. Silicon dioxide can be deposited by thermal oxidation

on Si, or on any semiconductor by LPCVD or PECVD using silane, NO, and nitrogen.

Atmospheric CVD can be used to deposit silicon films containing 4-12% phosphorous or boron

(phosphosilicate or borosilicate glass PSG or BSG). Silicon nitride is very difficult to grow

thermally, so LPCVD or PECVD are typically used to react silane and ammonia. It deposits as

an amorphous dielectric, and can contain hydrogen. This is especially true in PECVD, where the

nitride is rarely stoichiometric (Si3N4), and instead is denoted as SiNx, with up to 25% hydrogen.

The quality of the film can be determined immediately by comparing the refractive index

measured by ellipsometry to that of an ideal stoichiometric film [69].

3.4.2 Metal Deposition

While metal can be deposited using CVD from metalorganic sources, most metal films are

formed by physical vapor deposition (PVD). Two common methods are evaporation and

sputtering. The mechanism is simple the metal atoms are ejected from the source into a




















125



100


L%=1 pm
W=1 mm
SIC=100 vm
Au5 prn
81=100 pm
POMS=4 um
h=30 Wim'-C


0 2 4 6 8 10

Power Level (WImm)



Figure 7-12. Effect of the removal of top PDMS layer and both PDMS layers on the maximum
temperature in both active layers at varying power levels


300 1-


250



200



150


1UU
0 1 2 3 4 5

SiNx Thickness (urn)



Figure 7-13. Effect of SiNx passivation thickness on the junction temperature of the HEMT


-- GaN, 1 layer PDMS
..... Si, 1 layer PDMS A
A GaN, no PDMS
-- Si, no PDMS
A




A- ,


A -- -- - -
I.a

.jjA
I Er


---- --






I LG=I p
/ W =1 mm

/ SiC=100 rn
/ Au=5 pm
/ Si=100 pm

h=30 Wlm2-C

U









anywhere from 3 to 9 heating zones to finely control temperature and usually a couple cooling

zones. The preheat step is to remove the solvent from the solder. The soak phase is to equalize

the temperature of all components. The reflow phase is to melt the solder and form the electrical

joint, and the cooling phase allows for the solder to solidify before exiting the oven. Control of

the temperature profile is critical because this determines the yield, mechanical and electrical

integrity, and reliability [21].

Through-hole soldering uses a wave-soldering machine. This type of system solders all

components in one pass by passing the board with all components on it across a wave of molten

solder where the bottom of the board barely touches the solder. The solder will wet areas of

exposed metal, but not areas covered by the top layer of the board, called a solder mask [21].

There is a final test after assembly to ensure functionality. There is actually a test step

before each package level. IC testing is performed by the chip manufacturer. This is a simple sort

that tests basic functionality. Good chips are packaged and tested again after packaging. PCBs

are tested for opens and shorts inside the board, then again after assembly for solder joint

continuity. At the system level, more extensive functional testing is performed to make sure the

board carries out it's intended operation [21].

3.9 Reliability

The reliability of a device is the probability that a system will operate within acceptable

limits for a given period of time. Since the design lifetime is usually on the order of years to tens

of years, accelerated testing must be performed to determine failure rates. Device failure can be

caused by thermomechanical, electrical, and chemical mechanisms. An ideal failure distribution

follows a "bathtub" curve, shown in Figure 3-17. This implies that at the beginning of the life

cycle, there will be a high failure rate, referred to as infant mortality, in which devices fail due to

defects introduced in manufacturing. After the initial failure period, there will be a long stable









B -18. T typical s-param eter data......... .. ............... ................. .................................................203

B-19. Typical H21 and U calculation for Ft and Fmax .................................... ............... 203

B-20. Test setup for load-pull measurement......................................... 204

B-21. Typical load-pull data ................................... .. .. ........ .. ............204
















































17









experimentally by machining a custom heat sink, putting a thermocouple on the top surface, a

thermocouple and heat source on the bottom surface, and therefore knowing Q, A, and AT, one

can back out h from the convective heat transfer equation [127].

7.2.2 First Generation Simulation: GaN Power Amplifier on Heat Sink

As an initial test of the model, a simplified system was studied. This system consisted of a

GaN HEMT on the heat sink. This structure resembles the schematic for the first generation

device. Figure 7-4 shows a schematic of the device centered on the integrated heat sink/antenna.

The assumed initial values were: 3 W/mm dissipated power in the GaN power amplifier, a 5 am

thick GaN layer, 500 [m thick SiC layer, and an effective convective heat transfer coefficient, h,

of 30 W/m2-oC. The heat sink was assumed to be -20 mm2 area, with 11 fins, 10 mm fin height,

and 0.94 mm fin width, which are the dimensions required for a 2.4 GHz antenna. The finned

heat sink acts as a large thermal mass for heat removal. From these starting values, each layer

was optimized by systematically varying the thickness to find the minimum steady state

temperature. The most realistic boundary condition is a convective boundary on all sides [92-93].

The first variable studied was thermal conductivity. This is manifested in the design in the

substrate selection. Figure 7-5 shows the expected maximum temperatures reached by the

integrated power amplifier and modulator at fixed power and layer thickness for different

substrates. Clearly the use of sapphire substrates is not an effective solution because of their poor

thermal conductivity. The resultant high device temperature would be a major detriment to the

reliability of the GaN power amplifier in particular, since it is very difficult to prevent reaction of

the gate metal at 400 C during extended operation. The surprising result from Figure 7-5 is that

use of a bulk GaN substrate, while still inferior to SiC, would still maintain the die temperature









92. T. J. Anderson, F. Ren, L. Covert, J. Lin, S. J. Pearton. Journal of Vacuum Science and
Technology B, 24 (1), Jan/Feb 2006.

93. T. J. Anderson, F. Ren, L. Covert, J. Lin, S. J. Pearton. Journal of the Electrochemical
Society, 153 (10), G906-G910 (2006)

94. P. Tobias, A. Baranzahi, A. L. Spetz, O. Kordina, E. Janzen and I. Lundstrom, IEEE
Electron. Dev. Lett. 18 287(1997).

95. A. Vasiliev, W. Moritz, V. Fillipov, L. Bartholomaus, A. Terentjev and T. Gabusjan, Sens.
Actuators B49 133(1998).

96. J. Kim, F. Ren, B. Gila, C. R. Abernathy, and S. J. Pearton, Appl. Phys. Lett. 82 739 (2003).

97. J. Kim, B. Gila, C. R. Abernathy, G. Y. Chung, F. Ren, and S. J. Pearton, Solid State
Electronics, 47 1487 (2003).

98. A. Lloyd Spetz, P. Tobias, L. Uneus, H. Svenningstorp, L.-G. Ekedahl and I. Lundstrom,
Sens. Actuators B70 67(2000).

99. E. J. Connolly, G. M. O'Halloran, H. T. M. Pham, P. M. Sarro and P. J. French, Sens.
Actuators A99 25(2002).

100. G. W. Hunter, P. G. Neudeck, R. S. Okojie, G. M. Beheim, V. Thomas, L. Chen, D. Lukco,
C. C. Liu, B. Ward and D. Makel, Proc. ECS Vol 01-02 212(2002)

101. R. Neuberger, G. Muller, O. Ambacher, M. Stutzmann, Phys Status Solidi A 185 85 (2001).

102. J. Schalwig, G. Muller, O. Ambacher, M. Stutzmann, Phys Status Solidi A 185 39 (2001).

103. G. Steinhoff, M. Hermann, W. J. Schaff, L. F. Eastmann, M. Stutzmann, M. Eickhoff, Appl.
Phys. Lett. 83 177 (2003).

104. M. Eickhoff, R. Neuberger, G. Steinhoff, O. Ambacher, G. Muller, M. Stutzmann. Phys.
Status Solidi B 228 519 (2001).

105. J. Schalwig, G. Muller, M. Eickhoff, O. Ambacher, M. Stutzmann. Sens Actuat B 81 425
(2002).

106. M. Eickhoff, J. Schalwig, G. Steinhoff, O.Weidmann, L.Gorgens, R. Neuberger, M.
Hermann, B.Baur, G.Muller, O. Ambacher, and M. Stutzmann, Phys. Stat. Solidi C 6 1908
(2003)

107. H. Svenningstorp, P. Tobias, I. Lundstrom, P. Salomonsson, P. Martensson, L.-G. Ekedahl
and A. L. Spetz, B57 159(1999).

108. L. Chen, G. W. Hunter, and P. G. Neudeck, J. Vac. Sci. Technol A 15, 1228 (1997); J. Vac.
Sci. Technol. A16 2890(1998).


215













Mass flow controllers


Exhaust


Figure 3-6. A MOCVD reactor


Figure 3-7. An ion implantor


FL-











Purifie
H,









lasers, and is the most relevant reaction for the purposes of this work. A schematic of the

photoablation process is shown in Figure 8-6 [136].

If the photon energy (6.7 eV for ArF) exceeds the work function (3-5 eV) of a metal

surface, a photoelectron will be ejected. This is relevant particularly with metal surfaces, and the

ejected electron can actually serve to enhance a chemical etch by assisting in desorption of

reaction products. Electron-hole pair generation occurs in semiconductors when the photon

energy exceeds the bandgap of the material. The effects can be the same as photoelectron

emission in metals, but it can also have potentially implications for reliability testing as the

additional electron-hole pairs on a biased device will cause a higher current level and more stress

on the device [136].

Sample heating is also a relevant process, and occurs when photon energy is converted into

excited electronic states, which decay into ground state atomic vibrations, thus raising the

temperature. Each pulse is able to raise the temperature very rapidly (up to 1010 K/s) because a

large amount of energy is imparted into a small volume in a short time. This is complementary to

the ablation process as it forces the sublimation or desorption of nonvolatile ablation products.

At UV wavelengths, most of the photon energy goes into exciting molecular bonds rather than

pure heating. The temperature can get very high on a per pulse basis (AC temperature rise), but

the repetition rate is relatively low that the sample always reaches equilibrium temperature

between pulses (DC temperature rise). For high pulse rates or long wavelengths, such as in a

DPSS laser, the AC temperature may never reach the ambient value, leading to a rise in the DC

temperature of the sample. This can cause damage to the material [136].

A distinction must be made between ablation and etching. In the context of laser

processing, ablation refers to material removal purely by the interaction of the laser beam with














800

700 Reference
V =OV
600 V =1 V
step
500
E
E 400

S300
Q
200 .

100 ,

0 4 i
0 2 4 6 8
VD (V)



800 I I
Hole drilled at 355 nm
V = OV
v = -IV --

600o Vst= -1 V e





E /

200 -



0
0 2 4 6 8

VD (V)

Figure 4-3. I-V Characteristics of reference HEMT (top) and drilled HEMT (bottom)










"' :1 .,i 1.-l ",,


o P as Ag'iilent 4156

b. X ^ *,^.,.. .. ,.
,* ,,^ wi




SParameter Arlyzer



,-Figure B-. et et o TLM measreent
FSi2et r TL m.easureme

BI !.-", rI..". ","'.: ."
I






Figure B-l. Optical image of TLM pattern.
SMU1

To TLM iads Agilent 4156
_s___3 Parameter Analyzer



Figure B-2. Test setup for TLM measurement














I I I


0.0

Voltage (V)


0.5 1.0


Figure B-3. Typical I-V curves for TLM measurement


Slope = 3.8091
ntercept = 28.21


0 10 20 30 40 50 60 70 80 90

Gap (pm)


Figure B-4. Calculation of Rs, Rt, and Rc from TLM measurement


--- 5 Aim
10 pm
S20 pm
S40 pm
80 pm


0.020

0.015

0.010

0.005

0.000

-0.005

-0.010

-0.015

-0.020


I I I I'


I


400

350

300

5 250

o 200

. 150

0 AA
AnnJ


' ' '


I I II I I I I









200 |tm. It also serves to encapsulate the chip to protect it from environmental and mechanical

stress, and it acts as a heat sink for the chip, providing some level of thermal management. There

are three main first-level packaging techniques in the industry: wirebond, tape automated bond

(TAB), and flip-chip bond. Wirebonding and flip-chip bonding will be discussed in detail. TAB

is used in the LCD industry and involves mounting and interconnecting IC on metallized flexible

polymer tape. A schematic of the three packaging schemes is shown in Figure 3-14 [21].

3.7.1 Wire Bonding

Wire bonding is an interconnection technique that involves attaching a fine wire between

the I/O pads on the chip and the associated pins on the package. It is widely used in industry and

in the early 2000s accounted for over 90% of all chip level interconnections. Typically a gold

wire 25 |tm thick is bonded ultrasonically between the chip pad and the carrier pin. There are

two options for forming the bond: ball bonding and wedge bonding [21].

Ball bonding is a thermosonic welding technique. The primary advantage of ball bonding

is that the bond formed is circular, so the wire can be pulled to the next bond in any direction.

The bond is formed under a controlled force (<100 g), modest temperatures (150-200 C), and

ultrasonic excitation (60-120 kHz). The wire is then paid out from a spool to form a loop, where

the next bond is made. This bond has to be a wedge bond or line fracture since no ball can be

formed. After wire fracture, an electronic flame off (EFO) fires to form the next ball bond.

Process times can be <20 ms per bond cycle [21].

Wedge bonding is advantageous in fine pitch situations, since the bond wire is only

deformed 20-30% beyond the original diameter, compared to 60-80% for ball bonds. It also

demonstrates a higher yield, but there is a major drawback: the first bond must be aligned

perfectly along the axis between the first and second bonds. Otherwise, the first bond can pull off




















60




40




E 20
0


-6 -5 -4 -3 -2 -1 0

VG (V)



Figure B-11. Typical pulsed Vg-Ids curve compared to DC


8.00E-009



4.00E-009



O.OOE+000



-4.00E-009



-8.00E-009



-1.20E-008
-2


0


0

Voltage (V)


Figure B-12. Typical isolation data


i
- U


I




- II



























) Super high pressure mercury lamp
SEllipsoidal concave mirror
O First mirror 9
( Collimator
NO filter Second mirror
( L-39 filter Condenser lens
Fly's eye lens Mask plane


Figure 3-11. An optical lithography system


A

tid emllssi o

erm lim01i'
pOTrtu [2mi9l1


1..ir ll2 Ialar r

lid anIe \500

i mi dalldiarli
,,I I X,,,F X t -
pilalr e ard1

P1,161 1, A11 4


plilus .lior l,
ss mbl y





Figure 3-12. An electron beam lithography system














10

---- Reference
103 500 ppm
v 50 ppm

102





10







3.5 -3.0 -2.5 -2.0 -15 -1.0 -0.5 0.0 0.5 1.0 1.5

Bias (V)



Figure 6-2. Forward and reverse bias plot of diode current in varying atmospheres


250



200



150



100



50



n


ft


-3.5 -3.0 -2.5 -2.0 0.5 1.0 1.5 2.0

Bias (V)



Figure 6-3. Percentage change in current as a function of bias at 500 ppm H2


2.5


- SOOppmH,











mil


* *


-














PDMS thickness=4 pm
Si thickness=25 pm
LG=1 Lm
W0=1 mm
P=3 W/mm
h=30 WIm -K





a r
.r- <


Ir


0.0100

x (m)


Figure 7-10. Typical temperature profile in a cross section
near the gate


350

300

250

200

150


of the structure zoomed in to the area


0 2 4 6 8 10

Power Level (Wlmm)


Figure 7-11. Effect of HEMT power dissipation on the maximum temperature in both active
layers


80 L
0.0075


0.0125


I I I I I
LG=1 lm
W0=1 mm
SiC=100 pm
Au=5 pm
Si=100 nm /
PDMS=4 um
h=30 W/mz-C ,



GaN
Si


-*- GaN PA
. Au
---Si









3.7.3 E ncapsulation ................. .................................... ...... ........ .. ............. 54
3.8 B oard-L evel O perations.......................................................................... ....................55
3.8.1 Printed W iring B oard Fabrication ........................................ ...................... 55
3.8.2 A ssem bly and Testing .................................................. .............................. 56
3 .9 R e lia b ility ................................................................................................................... 5 7

4 HIGH ELECTRON MOBILITY TRANSISTORS..................................... ...............69

4 .1 O v e rv ie w ..................................................................................................................... 6 9
4.2 Silicon Carbide Substrates ............................................................................... 69
4.3 Silicon On Polycrystalline Silicon Carbide Substrates................. .............................71
4.4 Silicon Substrates.......... ........ ..... .......... ................. ..74

5 RF RELIABILITY......................................................... ..............82

5 .1 O v erview ..................................................................................................................... 82
5.2 R research Plan Stress Testing ........................................... ........ ..... ............... .83
5.3 Stress Test System ................ ......................... ................ .............. 84

6 G A S S E N S IN G ................................................................................................................ 9 1

6 .1 O v e rv iew ......................................................................................... 9 1
6.2 H hydrogen Sensors .............................. ........... .. .. ........................ .......... 92
6.2.1 Gallium Nitride High Electron Mobility Transistor Sensors ..............................92
6.2.2 Wireless Hydrogen Sensor Network .......................... .................................... 95

7 THERMAL SIMULATIONS OF THREE-DIMENSIONAL PACKAGE ........................106

7 .1 O v erv iew ............. ...... ... ........................................................................................... 10 6
7.2 Finite-Elem ent M modeling ......................................... ........................ ............... 106
7.2.1 H eat Transfer .............. ................ .....................................107
7.2.2 First Generation Simulation: GaN Power Amplifier on Heat Sink......................113
7.2.3 Generation II: Combined Amplifier and Silicon Chip .............. ................115

8 UV LASER PROCESSING .......................................................... ... ...............126

8 .1 O v e rv iew ................................................................................................................... 12 6
8.2 Solid State L aser Processing.................................................... ................................. 129
8.2.1 Background: Diode Pumped Solid State Lasers................... ........................ 129
8.2.2 Experim ents: Frequency Effects................................................. ...... ......... 131
8.3 E xcim er L aser P rocessing............................................................................. ..............133
8.3.1 B background: Excim er Lasers ........................................ .......... ............... 133
8.3.2 Experim ents: L aser Type............................................. ............................. 139

9 ELECTROLESS METAL DEPOSITION................................................................. 149

9 .1 O v erv iew ................................................................
9.1.1 N ickel P lating ............................................................................................ ..... 149


7









Typically this measurement requires an RF source that has the capability to set a single

frequency with variable power, power meters at least for input and output (some systems also use

reflected power to measure the true power input to the system), tuners for impedance matching,

and a DC bias system. A considerable amount of time must be put into this measurement, as each

component must be measured and calibrated out so that only the characteristics of the device are

calculated.

B.8.2 Procedure

A schematic of the load-pull system is shown in Figure B-20. All components must be

calibrated first in order to make the measurement. This process can take several days. First, the

network analyzer must be started and calibrated as described in the previous section. The tuners

are then connected to the network analyzer for calibration. S-parameters are taken at specified

impedance points on the tuner. This procedure typically takes 8-10 hours per tuner as the tuner

must move to each position. The Maury load-pull software manages all of the data acquired. The

power meters must be set to the correct frequency, and it is best that they are zeroed as well. S-

parameters are taken for the directional coupler, power meters, bias tees, and probe tips. All of

this data is assembled by the Maury software for calibration.

When a device is placed on the probe station, a DC curve is taken first. The tuners must

then be moved to perform impedance matching. When the input and output impedance is

matched, the DC bias is then applied, and then RF power is applied. A sweep is performed to

measure input and output power as a function of the input power from the RF generator. Input

and output power and gain are then plotted as a function of input power for the device. A typical

curve is shown in Figure B-21. While most of this is performed by the software, it is important to

note that the dB scale is logarithmic, so while gain is defined as the difference between output

and input power, this cannot be calculated in dB. The power must be converted to mW,









LIST OF REFERENCES


1. A. P. Zhang, L. B. Rowland, E. B. Kaminsky, J. B. Tucker, J. W. Kretchmer, A. F. Allen, J.
Cook, and B. J. Edward, Electron. Lett. 39, 245(2003).

2. W. Saito, Y. Takada, M. Kuraguchi, K. Tsuda, I. Omura, T. Ogura and H. Ohashi, IEEE Trans
Electron Dev. 50, 2528 (2003 ).

3. A.P. Zhang, L.B. Rowland, E.B. Kaminsky, V. Tilak, J.C. Grande, J. Teetsov, A. Vertiatchikh
and L.F. Eastman, J. Electron. Mater.32 388(2003).

4. W. Lu, V. Kumar, E. L. Piner and I. Adesida, IEEE Trans. Electron Dev. 50, 1069 (2003).

5. P. Valizadeh and D. Pavlidis, IEEE Trans. Electron. Devices, 52, 1933 (2005).

6. S. J. Pearton, J. C. Zolper, R. J. Shul, F. Ren, J. Appl. Phys. 86, 1 (1999) and references
therein

7. M. Hikita, M. Yanagihara, K. Nakazawa, H. Ueno, Y. Hirose, T. Ueda, Y. Uemoto, T.
Tanaka, D. Ueda and T.Egawa, IEEE Trans. Electron Devices, 52, 1963 (2005).

8. S. Nakazawa, T. Ueda, K. Inoue, T. Tanaka, H. Ishikawa and T. Egawa, IEEE Trans. Electron
Devices, 52, 2124 (2005).

9. T. Palacios, S. Rajan, A. Chakraborty, S. Heikman, S. Keller, S.P. DenBaars and U.K. Mishra,
IEEE Trans. Electron Dev. 52, 2117 (2005)

10. U. K. Mishra, P. Parikh, and Y. F. Wu, Proc. IEEE 90, 1022 (2002).

11. 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, J. R. Shealy, IEEE Trans.
Electron Dev., 48 479(2001).

12. S. Keller, Y.-F. Wu, G. Parish, N. Ziang, J. J. Xu, B. P. Keller, S. P. DenBaars, U. K.
Mishra, IEEE Trans. Electron. Dev., 48 552 (2001).

13. V. Adivarahan, M.Gaevski, W.H. Sun, H.Fatima, A. Koudymov, S. Saygi, G. Simin, J. Yang,
M.A. Khan, A. Tarakji, M.S. Shur and R. Gaska, IEEE Electron Device Letters, 24, 514
(2003).

14. A. Tarakji, H.Fatima, X.Hu, J.P.Zhang, G. Simin, M.A. Khan, M.S. Shur and R. Gaska,
IEEE Electron Dev. Lett. 24, 369 (2003).

15. S. Iwakami, M. Yanagihara, O. Machida, E. Chino,N. Kaneko, H. Goto and K. Ohtsuka, Jpn.
J. Appl. Phys. 43, L831 (2004).

16. R. Mehandru, S. Kim, J. Kim, F. Ren, J. Lothian, S.J. Pearton, S.S. Park and Y.J. Park, Solid-
State Electron. 47, 1037 (2003).


210









CHAPTER 8
UV LASER PROCESSING

8.1 Overview

Excimer laser processing has emerged over the past 30 years as a powerful

micromachining technique. They are particularly flexible tools due to their short wavelength,

high efficiency, and high power. A sampling of the many processes that use excimer lasers are as

follows: photoablation, etching, micron-scale lithography, doping, texurizing surfaces,

planarizing and cleaning surfaces, and medical applications. The most notable development in

the medical field is laser keratectomy with a 193 nm laser, known today most commonly as

LASIK. It is also used in the electronics industry in several capacities, most notably being for via

hole formation in dielectric films, but it is also used to selectively remove metal to rework

circuits or masks where there are electrical shorts. Pulsed laser deposition is one of the most

studied applications, and offers several advantages over conventional CVD or sputtering, the

most notable being stoichiometric ablation of an alloy.

Pulsed long wavelength lasers such as the CO2 and YAG lasers, were first developed

commercially. They are used primarily for bulk micromachining, particularly engraving, of metal

or ceramic parts. These lasers are quite limited though in that they are incapable of producing

high-resolution features, reflection from the surface becomes more efficient at larger

wavelengths, thus requiring more input power, and longer wavelength photons have less energy,

so photochemical reactions cannot be initiated efficiently. For this reason the primary ablation is

heating and subsequent sublimation, which can create damage in the material and debris on the

surface [133].

UV laser processing is advantageous in that UV photons are capable of excitation of gas,

liquid, or solid molecules, which is a first step in the photoablation process. UV photons are









modulation. In a depletion mode device, the gate voltage is modulated from 0 in steps of-1 V

until pinch-off is reached. The current saturation level should decrease with increasing reverse

gate bias. Typically pinch-off occurs between -2 and -7 V, and if nothing is reached by -10 V,

there is most likely a problem with the gate of the device.

The second curve taken is the gate voltage vs drain current curve. Typically the gate

voltage is taken from beyond pinch-off (-10 V, for example) to 0.5 V. One should see the drain

current leveling off beyond Vg=0, so this is why it is taken to slightly beyond 0. The drain

current should linearly decrease with increasing reverse gate bias as pinch-off is approached, and

be nearly flat at a very low current level (nA) beyond pinch-off. A line can be extrapolated back

to the x-axis to determine the threshold voltage. The transconductance (gm, reported normalized

as mS/mm or S/mm) can also be determined from this curve. It is defined as dI/dV, so it amounts

effectively to the first derivative of this curve. It can be calculated manually by calculating dl

and dV between two consecutive points for the entire curve. A good maximum gm is above 200

mS/mm, which implies a very sharp transition from on to off states.

B.3.2 Procedure

An optical image of a HEMT is shown in Figure B-5 to demonstrate the terminals. Use the

parameter analyzer and probe station for this test, as shown in Figure B-6. Set up the parameter

analyzer first for FET Vd-Id measurement. The drain bias range depends on the source-drain

spacing, but 0-10 V is typically a good starting point. For a depletion mode device, the gate

should go from 0 to pinch-off in -1 V steps. If the pinch-off voltage is not known, -5 or -6 V is a

good starting point. A sample curve is shown in Figure B-7.

After performing this measurement, change the measurement mode to FET Vg-Id

measurement. The drain bias will be constant in this mode, and should be set to just beyond









creating 8 outputs at approximately 5 W each. A voltage-controlled oscillator is used to

synthesize the RF signal, then a series of amplifiers and attenuators is used to tune the output to

the required power level, shown schematically in Figure 5-4. The splitter boards have isolators,

directional couplers, and diode detectors to detect both forward and reflected power as an output

voltage, detailed in Figure 5-5.

The signal then goes into the device boards. These have the input and output matching

networks fabricated on the circuit board or integrated on the GaN chip, with a thick copper layer

on the back side for heat sinking and as a common ground plane. A hole is bored through the

boards to the copper for the device cavity. The device is mounted directly on the copper using

Diemat, and wire bonded, as shown in Figure 5-6.

This pre-matched circuit board is designed as disposable board. On the back side of the

board, thermoelectric heaters are clamped on with a thermostat for temperature sensing. This

allows for temperature control. The heaters will be controlled by PID controllers with a

computer interface. The device boards also have DC control and monitoring capabilities. The RF

output from the boards goes to a detector diode board. The diode detectors work by rectifying the

RF signal into a DC voltage output. This output must be calibrated since the response will be

different at different frequencies. This can be done by using a synthesized sweeper to input a

known power to the diode, and simply measuring the output with a digital multimeter. A curve

can then be fit to this data and an equation and constants extracted, as shown in Figure 5-7. We

have performed the thermal simulations of the HEMT for different power levels and device

configuration. The HEMT junction temperature can be easily higher than the physical device by

more than 100 C [92-93]. It is very important to determine the junction temperature during the









from the front, while some epitaxial layer cracking is visible on holes drilled from the back-side.

This is an indication of high thermally-induced stresses in the latter case and is clearly not

acceptable for device fabrication [71, 74].

Figure 8-4 shows SEM at an angle of holes drilled from the front at 1064 nm (top), from

the back at 355 nm (center), and from the front at 355 nm (bottom). The shorter wavelengths

produce significantly cleaner and rounder holes, along with minimal surface damage.

The left picture in Figure 8-5 shows the undrilled alloyed Ti/Al/Pt/Au-based metallization

in transmission line measurement (TLM) patterns. The dimension of the square metal pads is 100

x 100 im2. The center picture shows the same pattern on a substrate drilled with the 1064 nm

laser. In this case, there is significant debris caused by the drilling, contaminating the areas

surrounding the vias. The picture on the right shows a hole drilled with a 355 nm laser in the

TLM region, without damaging the TLM pattern. Table 8-1 shows the electrical data from the

TLM patterns before and after the laser drilling at two different wavelengths. The sheet and

contact resistance from a 355 nm laser drilled sample are very similar to the undrilled reference

sample [71, 74].

8.3 Excimer Laser Processing

Since the UV solid state laser performed much better than the IR solid state laser, it makes

sense to try an even shorter wavelength using an excimer laser. A brief background section on

excimer laser processing will be presented next, followed by results from a comparison of

excimer and solid state laser drilled holes.

8.3.1 Background: Excimer Lasers

An excimer laser is a gas-phase system that uses electronically excited molecules to emit

high intensity pulses of UV light. The term was originally meant as a shorthand for excited

dimmer, which implies two identical atoms. Most excimer gases are actually heteronuclear









to note the peak in the response at around 200 C. This is due to competing effects on the

surface. It is expected that with increasing temperature, the cracking of the hydrogen on the

surface will become more efficient. This will result in more molecular hydrogen, which will also

diffuse to the metal-semiconductor interface faster at increased temperature. This effect is

countered by basic kinetics. There will also be a decreased surface concentration of hydrogen,

since there will be more transport to and from the surface at elevated temperature.

With this in mind, studies were performed to see if even lower detection limits were

possible under elevated temperature conditions. A temperature of 150 C was chosen since it

demonstrated a large response without pushing the temperature so high that the device was

damaged. As shown in Figure 6-7, a repeatable detection limit of 1 ppm was observed. Given the

relatively large response even at 1 ppm, there is speculation that even lower detection limits can

be achieved, but 1 ppm was the limit of the testing system.

6.2.2 Wireless Hydrogen Sensor Network

We have demonstrated a wireless hydrogen sensing system using commercially available

wireless components and GaN Schottky diodes as the sensing devices. Our sensors have

achieved ppm level detection, with the added advantages of a very rapid response time within a

couple of seconds, and rapid recovery. The sensors have shown current stability for more than 8

months in an outdoor environment. Our wireless network sensing system enables wireless

monitoring of independent sensor nodes and transmits wireless signals. This is especially useful

in manufacturing plants and hydrogen-fuelled automobile dealerships, where a number of

sensors, possibly with each detecting different chemicals, would be required. We have also

developed an energy-efficient transmission protocol to reduce the power consumption of the

remote sensor nodes. This enables very long lifetime operation using batteries. Experimental




















Figure 9-4. Cross-section of SiC MIPS device









penetrate the wafer and are scattered, and can re-emerge from the wafer. This is in fact one

method of detection in an SEM. These backscattered electrons will affect the irradiation in

neighboring areas of the pattern. It is possible to simulate these effects and correct the dose on

regions that are close enough to experience this effect. A schematic of the electron beam system

is shown in Figure 3-12 [68].

The major disadvantage of the EBL systems is that the throughput is extremely low since it

is a serial direct-write process. Exposure of just one wafer can take an hour or more because the

stage has to step and expose every stitch field. For this reason, they are excellent for research

level fabrication where small features are required (minimum feature size can be <50 um), but

impractical for mass production. This represents one of the fundamental challenges for the

industry over the next 10 years how to push the limits of optical lithography to this resolution

to maintain the throughput [68].

3.6 Etching

Etching is used to remove material through a mask patterned by lithography. There are two

basic types of etched: wet and dry, both with their own advantages that will be discussed. The

differences between wet and dry etching in terms of pattern transfer are shown in Figure 3-13

[69].

3.6.1 Wet Etching

Wet etching is a purely chemical attack of the material. It is primarily used to etch

dielectrics such as SiO2 and SiNx. There are three basic steps in the reaction mechanism -

diffusion of etchant to the surface, chemical reaction, and diffusion away form the surface. That

being said, there are two types of wet etch diffusion limited and reaction limited. The majority

of wet etched are diffusion limited, so an acid spray can be used to continuously supply fresh

etchant to the surface to enhance the etch rate and the uniformity. Since both the reaction and









that any later diffusion or thermal oxidation steps will cause redistribution of the dopants that are

already in the material, so the combined effect of multiple temperature cycles must be

considered. It is therefore undesirable for shallow junctions as they cannot be controlled as well.

It is still used to form deep junctions and as an alternative to ion implantation when crystal

damage is an issue. For compound semiconductors, ion implantation almost exclusively used. As

mentioned in previous sections, high temperature processes for long time periods can cause

preferential loss of one component, so these materials are more compatible with low temperature

processing unless capping layers are used [69].

3.3.2 Ion Implantation

Ion implantation uses charged particles to introduce impurities into the material. The ion

dose can be controlled by monitoring the ion current, making it very attractive for shallow or

lightly doped junctions. The principle of operation is as follows: a heated filament breaks up

source gas into charged ions, which are extracted from the source. The specific ion to be

implanted is selected using a mass analyzer to filter by mass to charge ratio. The ions are then

accelerated toward the sample at 10-400 kV, where they strike the samples and penetrate to a

given depth, with a distribution determined by the ion dose, accelerating voltage, and type of ion.

A schematic of an ion implantor is shown in Figure 3-7 [68].

There are two stopping mechanisms: electronic stopping, where the attractive power of

electrons in the lattice eventually slows the ion and stops it, and nuclear stopping, where

collisions with nuclei in the lattice cause energy loss and ultimately stopping of the ion. It is

possible for ions to travel down channels created by the crystal planes, so most commercial

systems tilt the sample at 7-10 degrees relative to the major plane. As the ion travels into the

lattice, it will displace nuclei, causing lattice damage and in the case of high doses,

amorphization. When ion implantation is used, a post-implant anneal step is required to









of As is lower (610 C), so an overpressure of As will form, causing transport of As to the Ga

melt (1240 C), forming GaAs, then cooling the melt. From the raw material, ingots can be

formed by the Czochralski method and the Bridgman method. The Bridgman process is very

similar to the material synthesis process. An overpressure of As is created in a lateral furnace to

prevent decomposition, and the polycrystalline GaAs is placed with a seed in a high temperature

zone of the furnace. The furnace is moved so that the material will melt and then freeze

following the seed. This process can be compared to a horizontal float-zone process. A schematic

is shown in Figure 3-3 [69].

After growth, wafers are prepared from the ingots for fabrication. The seed and tail ends

are cut off, the ingot is ground to a uniform diameter, and flats are ground along the length of the

crystal to define crystal orientation and conductivity type. Standard flat orientations are shown in

Figure 3-4. A diamond saw is used to slice the crystal into wafers (thickness of 500-700 [tm),

which are then lapped to a flatness of around 2 itm, etched to remove contamination, and

polished on one or both sides for device fabrication [68].

3.2.2 Epitaxial Growth

Epitaxial growth methods are used to grow single-crystal semiconductor layers on single-

crystal substrates. Homoepitaxy is the term used to denote growth when the substrate and the

grown layer are the same, for example growth of a highly doped layer of GaAs on a single

crystal bulk GaAs substrate. Heteroepitaxy is used to denote growth of a semiconductor layer on

a different substrate, such as GaN on SiC, or AlGaAs on GaAs. The growth systems used must

be highly precise. The layers must be of nearly perfect quality and have atomically smooth

interfaces. The most common growth methods are chemical vapor deposition (CVD), metal

organic CVD (MOCVD), and molecular beam epitaxy (MBE). Since epitaxial growth involves a









7-14. Effect of gate pitch on the junction temperature of the HEMT in a multiple finger
co n fig u ratio n ...................................... ................................................. 12 5

8-1. SEM cross sections of holes drilled at 1064 nm (left) and 532 nm (center) ........................142

8-2. SEM images of hole drilled from front at 1064 nm (above) and 532 nm (below)............... 143

8-3. SEM image of holes drilled at 532 nm from front (above, left) and back (below, left)
and 355 nm from front (above, right) and back (below, right) ............. ................144

8-4. SEM at an angle of holes drilled from the front at 1064 nm (top), from the back at 355
nm (center), and from front at 355 nm (bottom) .....................................................145

8-5. Photographs of TLM patterns with Ti/Al/Pt/Au based metallization on an undrilled
sample (left), drilled with 1064 [m laser (center), drilled with 355 nm laser (right)......146

8-6. The U V photoablation process. ................................................ ................................ 146

8-7. Optical microscope images of front and back of via holes formed in poly-SiC substrates
with a DPSS laser (left) and excimer laser (right) from the top semiconductor surface
(top) and bottom SiC surface (bottom ) ................................................. ....... ........ 147

8-8. SEM images of via holes formed in poly-SiC substrates with a DPSS laser (left) and
excimer laser (right) from the top semiconductor surface....................... ...............147

8-9. SEM images of cross-sections of via holes formed in poly-SiC substrates using DPSS
laser (left) and excimer laser (right) at low magnification (top) and high
m agnification (bottom ) ...................... ...................... .. .. .... ......... .. .... .. 148

8-10. Optical image of array of holes formed in GaAs........... ..............................................148

9 -1 A n -v o ltaic b battery .................................................................................. ............... ..... 15 8

9-2. Optical microscope pictures of Ni plated on Si (4 |tm on top, 11 tm on the bottom),
with exposed holes............ ...... ....................... .........159

9-3. Photo (top) and optical microscope image (bottom) of Ni hard mask for SiC trench
etch in g ................... .......................................................... ................ 16 0

9-4. C ross-section of SiC M IP S device ........................................................................... .... 161

9-5. Photo (top) and optical microscope image (bottom) of MIPS device fabricated using an
electroless N i etch m ask ........................................................ .. ...... ...... 162

10-1. A tuneable filter device............................................................................. .................... 168

10-2. SEM cross-section of In film bonded to Au film ............................. 168













5 tO 25 50 00 250 500


-10 -

0 360 720 1080 1440 1800 2160 2520 2880 3240 3600

Time (s)



Figure 6-4. Time dependence of Schottky diode sensor when switching from pure nitrogen
atmosphere to hydrogen concentrations from 1-500 ppm


200


100


0 100 200 300 400 500

Hz Concentration (ppm)



Figure 6-5. Percentage change in current as a function of hydrogen concentration under both
forward and reverse bias.


---



Reverse Bias
-- Forward Bias

V =er =-3.5 V

UVForvd = 1.0 V




/
;'U
0 ...-"
U. ""


so









B.6.2 Procedure

Breakdown measurements below 400 V can be performed with the curve tracer, as shown

in top of Figure B-15. On this system, the compliance setting is a power level, so it is a

combination of current and voltage. Therefore the compliance can be reached at high current and

low voltage or high voltage and low current. While the compliance on this system will protect

the system, it can still create hazardous voltages or currents to the user. Set the sweep for a DC

Vd-Id measurement. It is best then to begin at the lowest settings and gradually work up. The

sweep is set to go from high to low power, and is represented at a percentage of maximum

power. Therefore the setting should be gradually increased, with single measurements, and then

when the desired point is reached, a sweep back can be used to obtain the full I-V curve. The

curve that is generated is shown in Figure B-16.

If the breakdown voltage is too high to measure with the curve tracer, the Glassman High

Voltage power supplies can be used. A DC power supply must be connected to the system to

control the voltage. An applied voltage of 0-10 V will correspond to 0-max power (1 kV, 6 kV,

20 kV) on the high voltage supply. The high voltage is connected to the drain terminal through a

large (1 MQ) resistor to limit the current for safety. The source terminal is connected to a digital

multimeter to measure the source-drain current. The gate voltage is controlled through a second

DC power supply to maintain the gate voltage near pinch-off. The breakdown is determined by

monitoring the current level and looking for a sudden sharp rise. A schematic of the test system

is shown in the bottom of Figure B-15.

B.7 Small Signal RF (S-parameters)

B.7.1 Overview

This test is used to look at the frequency response of a HEMT. It inputs a small RF signal

over a large frequency range (typically 10 MHz to 40 GHz or more) and measures what are









CHAPTER 9
ELECTROLESS METAL DEPOSITION

9.1 Overview

While the electroless metal deposition project was initially studied for metallization of via

holes in SiC, the most interesting device application that has emerged has been it's use for hard

mask deposition for dry etching. To this end, the following section will provide an overview of

the fundamentals of Ni deposition, followed by an overview of a photovoltaic device that has

been fabricated in SiC using electroless Ni as an etch mask.

9.1.1 Nickel Plating

Electroless (or catalytic) metal deposition is a subset of the general category of

nonelectrolytic metal coating processes. All metal depositions from solution follow the same

general mechanism, in which the metal in solution is reduced by a source of electrons to the solid

state on the surface to be coated, shown in the following reaction:

M+x (aq) + xe- M (s)

In a typical electrodeposition process, this reaction is quite straightforward, with the

electrons being supplied by an external power supply. In this case oxidation occurs at the anode,

producing the metal ions from a solid source into solution, and at the cathode the ions are

deposited onto the surface of interest by the reduction reaction given above. An electroless

plating method follows this same principle, but the reactions occur simultaneously in solution

with no external source required. This process is, therefore, considerably more complex and not

completely understood as of yet [139].

Nickel is a metal of interest both for via hole filling and as an etch mask for SiC during

dry etching. Since the etch rates are so slow, the mask material must be quite robust, since it will

be sputtered away during the etching process. This calls for a very thick mask, approximately 10









The next step is to repeat these results on SiC. The transition proved quite difficult. Initial

efforts to duplicate the process were failures in that the metal film always delaminated.

Ultimately, it was determined that surface conditions are absolutely critical to the process. The

SiC samples from one company resulted in delamination after less than one minute in the plating

bath, while samples from a different company, which will have a different surface condition, will

plate easily. It was determined that the "homemade" solution is too unstable for a reproducible

process. Commercial plating solutions are readily available and testing has produced stable and

reproducible results requiring considerably less solution and effort.

9.2.2 Commercial Plating Solutions

The first commercial plating solution investigated is known by the trade name Nickel-B

from Transene. This solution was chosen because it resembled the solution produced in the lab in

that it requires an activator (therefore is selective) and grows Ni films at a very high rate

(desirable to reduce process time). A water bath in a double beaker system was used to maintain

a constant temperature of 64-66 C, and a 31:1 solution of activator to Nickel-B was used. A

growth rate of 0.2 [tm/min was observed. This film actually grew too fast and resulted in a very

strained film, which peeled off after about 1 |tm was grown.

Based on the observations from the Nickel-B experiments, it was determined that many of

the problems were due to the fast growth rate. A different solution, known as Nickelex, also from

Transene, was recommended due to the slower growth rate. The primary disadvantage for this

solution is that it is not selective and a seed layer is required, thus some work is necessary to

make a pattern in the film. The seed layer was e-beam evaporated Ti/Ni (200/800 A). The Ti

layer is meant to improve adhesion. Since the film requires a clean seed layer, the samples must

be acid etched to remove any native oxide and create a clean Ni surface for plating. The etch









CHAPTER 4
HIGH ELECTRON MOBILITY TRANSISTORS

4.1 Overview

Current work has focused on fabrication of AlGaN/GaN HEMTs on various substrates. As

discussed previously, the SiC substrate is of vital importance when heat transfer and high power

operation is critical, such as in an application integrating Si-based electronics with GaN. A

particularly novel substrate has been demonstrated by Picogiga. It uses a bulk poly-SiC wafer,

with a Si layer bonded on the surface, and the GaN grown on the Si. This could potentially

integrate the low cost Si approach as a growth template and the high thermal conductivity of SiC.

Devices have been fabricated on both substrates, and performance results will be compared in

this chapter.

4.2 Silicon Carbide Substrates

SiC substrates are ideal for GaN device fabrication because they offer a thermal

conductivity that is more than an order of magnitude higher than sapphire (nearly 400 W/cm-K

vs 20 W/cm-K), and a much better lattice match to GaN (3.5% vs 13%). The major drawback is

cost. SiC wafers are by far the most expensive (hundreds of dollars per wafer compared to tens

of dollars per sapphire wafer), making the devices cost-prohibitive except in high performance

markets where extremely high thermal conductivity is a must. Devices fabricated in SiC typically

do produce the best performance, and to that end, all devices presented in this section will be

compared to SiC devices as a reference [71, 74].

The first section of this chapter will focus on HEMTs fabricated on SiC substrates. As

mentioned previously, laser drilling presents an interesting opportunity for novel processing

schemes and in particular 3-D integration. Therefore to study the effect of laser drilling on

devices, HEMTs were fabricated in a SiC reference sample and a laser drilled SiC sample. For









241 is typically used, which has a half-life of 432.7 years), the semiconductor is damaged within

a matter of hours, thus reducing the power output. To this end, SiC is particularly attractive due

to it's inherent resistance to radiation damage. A schematic of such a device is shown in Figure

9-1 [140].

9.2 Electroless Plating Procedures

Initial testing used a basic solution formulated in the lab. It operates on the principle of the

reactions described above. Most industrial solutions are based on this on a fundamental level,

with many proprietary chemicals added to improve stability and to tailor the solution to specific

applications (i.e. high rate, selective deposition, etc). While this is useful in determining the

fundamentals of the process, it is impractical for real processing because the solution will be

inherently unstable. Therefore when investigating a plating process for a real application, such as

the MIPS device described above, commercial solutions were used. Results and observations

from both are described.

9.2.1 Solutions Made In-House

Electroless Ni deposition on Si has been demonstrated using the following bath conditions:

NiSO4 (29 g/L), NaH2PO2 (17 g/L), sodium succinate (15 g/L), and succinic acid (3.2 g/L). A

500 mL solution was made of these components, as well as 200 mL solutions of the following

two pretreatment solutions: SnC12-HCl (0.1 g/L-0.1 mL/L) and PdC12-HCl (0.1 g/L-0.1 mL/L).

Initially, various preparation methods were used to determine the best method of initiating the

reaction. If the aqueous Pd pretreatment is used, then, of course, Pd will adsorb on all surfaces

contacting the solutions, causing Ni to plate everywhere. Since we are searching for a pattemable

plating process, this is clearly undesirable. Evaporated Au, Ni, and Pt (1000 A) were also

investigated, and it was found that only Ni functioned as a catalytic surface to initiate the

reaction. Using sputtered Ni as a seed metal, followed by activation in the pretreatment solutions













QuickTimeTM and a
decompressor
are needed to see this picture.





Figure 8-5. Photographs of TLM patterns with Ti/Al/Pt/Au based metallization on an undrilled
sample (left), drilled with 1064 [m laser (center), drilled with 355 nm laser (right).








Absorption




Electronic
Bond Breaking
& Ablation


Thin Layer Removed
by Single Pulse




Multiple Pulses
Provide Depth Control




Figure 8-6. The UV photoablation process.





























(a) (b) (c)


Figure 6-11. Photo of sensor system (a) Sensor with sensor device; (b) Sensor and base station;

(c) Computer interface with base station


't1*.t:c~9 Sirrc. I
-I*1 y: q... .'nr


Figure 6-12. Field test results of four sensors


I 1-0 E91 7- '-.-r"""""1, .


PI 9
i"/i I*'


Figure 6-13. Field test results for single diode sensors over a period of one week


''
''
-- ipliN I
Smii r









repeatability over the entire 6-inch range of movement. A system has been ordered and will be

installed in spring 2008.









6.2 Hydrogen Sensors

6.2.1 Gallium Nitride High Electron Mobility Transistor Sensors

AlGaN/GaN Schottky diodes with Pt sensing metal demonstrate improved sensitivity

under reverse bias and elevated temperature. A detection limit of 10 ppm was achieved under

reverse bias with a current increase of 14%. The detection limit was further decreased to 1 ppm

by operating the sensor at 150 OC, still demonstrating a current change of nearly 50%. This is a

considerable improvement over the forward bias detection limit of 100 ppm, with a current

change of 4% [117].

The device layer structures were grown on C-plane A1203 substrates by Metal Organic

Chemical Vapor Deposition (MOCVD). The layer structure included an initial 2[tm thick

undoped GaN buffer followed by a 35nm thick unintentionally doped A10.28Ga0.72N layer. Mesa

isolation was achieved by using an inductively coupled plasma system with Ar/Cl2 based

discharges. The Ohmic contacts were formed by lift-off of Ti (200A)/Al (1000A)/TiB2

(200A)/Ti (200A)/Au (800A). The metals were deposited by Ar plasma-assisted RF sputtering.

The contacts were annealed at 850 C for 45 sec under a flowing N2 ambient in a Heatpulse 610T

system. A 100 A thick Pt Schottky contact was deposited by e-beam evaporation for the

schottky metal. The final step was deposition of e-beam evaporated Ti/Au (300A/1200A)

interconnection contacts. The individual devices were diced and wirebonded to carriers. These

were then placed in an environmental test chamber and connected to an electrical feedthrough for

testing. Mass flow controllers were used to control the gas flow through the chamber, and the

devices were exposed to either 100% pure N2, or H2 concentrations of 500 ppm down to 1 ppm

in N2 and temperatures from 25 to 500 OC [117].









75. 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. R. Shealy, IEEE
Trans. Electron Dev., 48, 479(2001).

76. S. C. Binari, K. Ikossi, J. A. Roussos, W. Kruppa, D. Park, H. B. Dietrich, D. D. Koleske, A.
E. Wickenden and R. L. Henry, IEEE Trans. Electron Dev., 48, 465(2001).

77. R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, IEEE Trans. Electron. Dev. ED48,
560(2001).

78. B. M. Green, K. K. Chu, E. M. Chumbes, J. A. Smart, J. R. Shealy, and L. F. Eastman, IEEE
Electron. Dev. Lett., 21, 268(2000).

79. B. Luo, J. W. Johnson, J. Kim, R. M. Mehandru, F. Ren, B. P. Gila, A. H. Onstine, C. R.
Abernathy, S. J. Pearton, A. G. Baca, R. D. Briggs, R. J. Shul, C. Monier and J. Han, Appl.
Phys. Lett. 80,1661 (2002).

80. A. Vertiatchikh, L. F. Eastman, W. J. Schaff and I. Prunty, Electronics Lett. 38,. 388 (2002).

81. Binari, S.C., Klein, P.B., and Kazior, T.E., Proc. IEEE, 2002, 90, pp. 1048-1058.

82. T. Kikkawa, Japanese J. Appl. Phys., vol. 44, pp. 4896-4901, 2005.

83. D.C.Dumka, C. Lee, H.Q. Tserng and P. Saunier, 2004 Electronics Letters, vol. 40, no. 24.

84. T. Kikkawa, K. Imanishi, M. Kanamura, K. Joshin, 2006 CS-MANTECH Technical Digest.,
pp.171-174, 2006.

85. 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, K.J. Linthicum,
Microelectronics Reliability 46 (2006) pp.1247-1253.

86. H. Kim, V. Tilak, B. M. Green, H.Y. Cha, J. A. Smart, J. R. Shealy, and L. F. Eastman, 2001
IEEE 39th Annual International Reliability Physics Symposium, pp214-218.

87. R. Khanna, L. Stafford, S.J.Pearton, T.J.Anderson, F. Ren, I.I. Kravchko, A.Dabiran,
A.Osinsky, J.Y.Lee, K.Y.Lee and J.Kim, Electron. Mater. 36, 379(2007).

88. R. Khanna, S.J. Pearton, F. Ren, I.I. Kravchenko, Applied Surface Science 253, pp. 2340-
2344 (2006).

89. A. Koudymov, M.S. Shur, G. Simin, R. Gaska. Phys. Stat. Sol. (RRL) 1, No. 3, 116-118
(2007)

90. W. Kuang, R. Trew, G. Bilbro, Y. Liu. Mater. Res. Soc. Symp. Proc. Vol. 955 (2007)

91. A. Koudymov, M.S. Shur, G. Simin, C. Lee, K. Chu, P.C. Chao, J. Jimenez, T. Balistrerei.
International Semiconductor Device Research Symposium Proceesings (2007) 1-2


214









38. H. Morkoc, S. Strite, G.B. Gao, M.E. Lin, B. Sverdlov, M. Nurns, J. Appl. Phys. 76, 1363
(1994)

39. T. P. Chow, R. Tyagi, IEEE Trans. Electron. Dev. 41, 1481 (1994)

40. Z.Z. Bandic, P.M. Bridger, E.C. Piquette, T.C. McGill, R.P. Vaudo, V.M. Phanse, J.M.
Redwing, Appl. Phys. Lett. 74, 1266 (1998)

41. R. Gaska, J.W. Yang, A. Osinsky, Q. Chen, M.A. Khan, A.O. Orlov, G.L. Snider, M.S. Shur,
Appl. Phys. Lett. 72, 1266 (1998)

42. Y. Uzawa, Z. Wang, A. Kawakami, B. Komiyama, Appl. Phys. Lett. 66, 1992 (1995)

43. F. Ren, J.C. Zolper, Wide Energy Bandgap Electronic Devices (World Scientific Press,
Singapore, 2003)

44. H. Marchand, J.P. Ibbetson, P.T. Fini, P. Kozodoy, S. Keller, S. DenBaars, J.S. Speck, U.K.
Mishra, Mater. Res. Soc. Int. J. Nitride Semi. Res. 3, 3 (1998)

45. R. Radar, G. Jacob, J. Hallais, R. Frucgart, J. Cryst. Growth 31, 197 (1975)

46. J. Karpinski, S. Porowski, S. Mitokowska, J. Cryst. Growth 56, 77 (1982)

47. S. Nakamura, T. Mukai, M. Senoh, Appl. Phys. Lett. 64, 1687 (1994)

48. H. Amano, M. Kito, K. Hiramatsu, I. Akashi, Jpn. J. appl. Phys. 28, L139 (1989)

49. S. Nakamura, M. Senoh, S.I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku,
Y.S. Ugimoto, Jpn. J. Appl. Phys. 35, L217 (1996)

50. S. Nakamura, M. Senoh, S.I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku,
Y.S. Ugimoto, Appl. Phys. Lett. 68, 2105 (1996)

51. S. Li, Semiconductor Physical Electronics (Springer, 2006)

52. M.A. Khan, A.R. Bharrari, J.N. Kuznia, D.T. Olson, Appl. Phys. Lett. 63, 1214 (1993)

53. M.A. Khan, J.N. Kuznia, D.T. Olson, W. Schatt, J. Burm, M.S. Shur, Appl. Phys. Lett 65,
1121 (1994)

54. L. Esaki and R. Tsu, IBM Research, RC 2418 (1969)

55. R. Dingle, H.L. Stormer, A.C. Gossard, W. Weigmann. Appl. Plys. Lett. 33, 665 (1978)

56. T. Mimura, S. Hiyamizu, T. Fujii, K. Nanbu. Jap. J. Appl. Phys. 19, L225-227 (1980)

57. D. Delagebeaudeuf, P. Delescluse, P. Etienne, M. Laviron, J. Chaplart, N.T. Linh. Electron
Lett. 16, 667-668 (1980)


212









Figure 7-8 shows the expected maximum temperatures reached by the power amplifier for

various Si layer thickness and fixed power density of 3 W/mm. The effect is small, changing by

only approximately 10 C over an order of magnitude thickness range. Therefore, standard

thickness Si substrates can be used for the modulator device without incurring any substantial

losses in thermal management. It would be beneficial however, cost permitting, to use thin Si

membranes for the modulator, using PDMS as a chip carrier. This could prove essential

technology if this extra 10 OC drop will make the difference between the function and failure of

the Si devices, since these are the most temperature-sensitive component in the MCM [92-93].

The typical temperature distribution in both GaN and Si layers is shown in Figure 7-9. The

Si temperature distribution is considerably broader, due to lateral spreading from the high

thermal resistance PDMS layers. This effect is highlighted in Figure 7-10, which shows a typical

cross sectional temperature distribution in various layers in the device. Note from the scale that

this shows only a small area, 2.5 mm on either side of the gate. The temperature in all layers

reaches the same value within 2 mm of the gate, and is sharply peaked in the GaN layer,

becoming broader down through the Au and Si layers, with the PDMS layers spreading the heat.

The effects of both PDMS and Au thickness were also studied, and the effect over a range of 15

[tm was negligible (<3 C) [92-93].

After optimization of all layer thicknesses in the structure, the device power level was

studied. Previous work had shown that optimization of the GaN device design could drive down

the high power operating temperature considerably from a reference case. Now, due to the high

thermal resistance from the PDMS layers, the junction temperature of the HEMT rises again to

3500C at high power levels, even with all layers optimized, as shown in Figure 7-11. Again, the

GaN PA can operate at this temperature, but the Si modulator would be inoperable, even with









higher driving force is applied to the electrons to move across the depletion region. Thus the

dipole layer is amplified at the Pt/AlGaN interface for higher reverse bias voltage. Due to this

dipole layer amplification, the detection sensitivity is enhanced at higher reverse bias voltage.

The change in current upon exposure to hydrogen is directly related to the hydrogen

concentration, as demonstrated in Figure 6-4, where current was monitored as a function of time,

with the hydrogen concentration being increased from 1 ppm to 500 ppm, alternating between

hydrogen and nitrogen exposure [117].

It is also evident in Figure 6-5 showing the detection sensitivity as a function of

hydrogen concentrations, that the diodes are much more sensitive under reverse bias conditions.

A detection limit of 100 ppm is achieved under forward bias, but the reverse bias detection limit

is an order of magnitude lower, 10 ppm. The change in current at 10 ppm is 14% and over 200%

at 500 ppm under reverse bias conditions. This demonstrates again that the same sensor under

reverse bias has a much lower detection limit, as forward bias operation results in changes of 25-

75% over the 100-500 ppm range. This is consistent with published reports indicating improved

sensitivity under reverse bias [124]. The reliability of the hydrogen sensor may be quite different

under the two bias voltage polarities, since different degradation mechanisms in GaN devices are

accelerated by either the presence of high voltage depletion regions (reverse bias) or current

injection (forward bias in this experiment) [89].

Temperature effects were studied next. A hydrogen concentration of 25 ppm was chosen to

study the temperature effects since it demonstrated a large response at room temperature. The

diode current was measured in both 25 ppm H2 and 100% N2 atmospheres at temperatures from

25 C to 500 OC. The percentage change in current as a function of temperature is shown in

Figure 6-6. At temperatures above 550 C, the device was irreversibly damaged. It is interesting









important advantage for rf applications. Many communications systems for the military operate

in the frequency range from 7->44 GHz, and in some cases, the systems are space-borne, where

amplifier efficiency, reliability, and low weight are key requirements [1-16].

1.1.2 Three-Dimensional Packaging

As demands on portable devices increase, new packaging techniques must be developed to

meet the demands for smaller, lighter, and more functional devices. The early solutions, such as

ball grid arrays (BGA) and flip-chip technologies, have met the demands for decreasing package

size and interconnect distance and increasing I/O count, but in today's mobile landscape, the chip

footprint area is becoming a huge factor. 3-D packaging solutions meet both the size and

performance requirements. Vertical stacking of multiple die within a package, using specialized

substrates and interconnects, will reduce the number of chip-to-board connections and decrease

the area required for chips and inter-chip wire traces. These techniques are also advantageous

from a power consumption standpoint since 40% of power consumption comes from chip-to-chip

interconnects. The module-to-board solder connects account for almost 90% of board failures, so

reducing the number of connections, we can decrease board failures and attain an overall

increase in reliability and decrease in power consumption [17-18].

Laser drilling is a well established method for the machining of metals and hard materials.

The mechanism for this process is simple. Essentially, the laser supplies sufficient power to eject

the material. The actual ejection mechanism is dependent upon wavelength. At longer

wavelengths (532 and 1064 nm), the mechanism is sublimation of the sample and subsequent

ejection. As the wavelength decreases, moving into the UV region (355 nm), the mechanism

changes to a molecular level bond-breaking phenomenon. This, therefore, results in much

cleaner profiles and less redeposited ablation debris. Since the laser can be focused and

controlled for much smaller features than standard machining tools, it is an ideal method for









parameter analyzer for a diode (Vf-If) measurement. A typical test range is from -1 V to 1 V.

Probe the TLM pads, starting with the largest spacing. Determine the slope of each line (the

intercept should be zero). This can be done by either curve fitting in Excel after testing, as shown

in Figure B-3, or directly measuring the tangent on the parameter analyzer. The resistance is

equal to I/slope from Ohm's Law (V=IR, therefore R=V/I). Plot the resistance as a function of

the gap between pads, as shown in Figure B-4. Perform a linear fit to determine the slope and the

intercept of this line. The sheet resistance (Rs) is equal to 100*slope. The transfer resistance (RT)

is equal to intercept*0.1/2. The 0.1 comes from the pad length, in mm, and the 2 comes from the

number of pads.

B.3 Direct Current

B.3.1 Overview

DC testing is performed to check for basic function of HEMTs. This primarily means

checking for current saturation and gate modulation. To this end, two curves are typically taken -

source-drain voltage vs source-drain current at varying gate voltages and gate voltage vs source-

drain current at a constant drain voltage. The former measurement is used to check for saturation,

and the latter used to check the gate characteristics.

Ideally, the current will linearly increase at low voltages and then level off at saturation. In

sapphire or Si HEMTs, the current will actually decrease linearly with increasing drain voltage

past the saturation voltage. This is due to the self-heating of the device, and therefore is not seen

in SiC HEMTs. By extrapolating the linear region forward and the saturation level back, the knee

voltage can be found as the intersection of these lines. Knee voltage is a function of gate to drain

distance, and thus increases linearly with this gap. The current level is typically normalized and

reported as mA/mm or A/mm, where mm refers to the total gate width. A state-of-the-art GaN

HEMT approaches the theoretical limit of 1.2 A/mm. This test is also used to check for gate













IL~, l~L'V
P.~ *~
- v JA A


: .'. -

-. .
.. 1 n .i






I

F
""' i" ^






it





* p '- :
. *^ -J '


"I ,,' ,,V".,'
.t":l,, u,l e .%' 1

-' '.. *' *-". .r. "J" ',
,_ ..& . ._ '. '
t ~ Un)" .1 ~' : '
r *' t .

.. .... .....'"" t "'"*r '\ *< '* ";* '"" -" "'' ) "' '" i*. ,",* *' '*' : '1 *.)
,,._ '".&. ..* .... i. '*




^ / ; i ^ ^ **.. ; V t '.- >', *' .* -.: *" !
ii1 ,
,: r',




", "---'-,'"______:___,.^.* ,
.r: ,~~r. V

I oj.
;-"i A 'i -:"<.."' ,'.'.',' i. ., ,-<,." .'; ., .. *,.
.,, ,... A,: ,,. :,A ..L






an~ ~ ~ ~ eletros Ni etc mak
162, h '.
,* UZj.' -i,





Fi ur.95.Ph to '.. -. -"'-l-- (to )'nd ptcalm i ros op im ge (b tt m of M IP de icab i at d usn
an~~~rr. elcrls Nieth as

162
'.' ~ .... i. ., r,; I .. ."'
.. s ._ .... .. .... l i ., j1 ,... -" I ; .. .. ...

... '" I ;- ': "
.. "r 'L J' "".t1: "r,


Figure 9-5. Photo (top) and optical microscope image (bottom) of MIPS device fabricated using
an electroless Ni etch mask.



162


....* -

........ i

-A






1-HI




*-<& .h L"
A" I
~j4A4i
hit'
L;~


r;

;* ~:











subtracted, and then converted back to dB. The correlation is PdBm = 10log(Pmw), or reversed,

Pmw = 10PdBm/10
Pmw = 10









58. M.A. Khan, J.M. Van Hove, J.N. Kuznia, D.T. Olson. Appl. Phys. Lett. 58, 2408-2410
(1991)

59. M.A. Khan, D.T. Olson, J.N. Kuznia. Appl. Phys. Lett 65, 1121-1123 (1994)

60. Q. Chen, J.W. Yang, R. Gaska, M.A. Khan, M.S. Shur, G.J. Sullivan, A.L. Sailor, J.A.
Higgings, A.T. Ping, I. Adesida. IEEE Electron Device Lett. 19, 44-46 (1998)

61. Y.-F. Wu, B.P. Keller, P. Fini, J. Pusl, M. Le, N.X. Nguyen, C. Nguyen, D. Widman, S.
Keller, S.P. Denbaars, U.K. Mishra. Electron. Lett. 33, 1742-1743 (1997)

62. Wakejima, Matsunaga, Okamoto, Ota, Ando, Nakayama, Miyamoto. IEEE MTT-S
International Microwave Symposium Digest, 2006, 1360-1363

63. Nitronex Corporation, Available Products.

64. Choi, Pophristic, Peres, Cha, Spencer, Eastman. Semiconductor Science and Technology 22
(2007), 517-521

65. Higashiwaki, Mimura, Matsui. Phys. Stat. Sol. (A) 204, No. 6, 2042-2048 (2007)

66. Palacios, Dora, Chakraborty, Sanabria, Keller, DenBaars, Mishra. Phys. Stat. Sol (A) 203,
No. 7, 1845-1850 (2006)

67. Y.-F. Wu, A. Saxler, M. Moore, R.P. Smith, S. Sheppard, P.M. Chavarkar, T. Wisleder, U.K.
Mishra, P. Parikh, IEEE Electron Device Letters, v 25, n 3, March 2004, p 117-19

68. Campbell. The Science and Engineering of Microelectronic Fabrication. Oxford University
Press, 2001

69. Sze. Semiconductor Devices: Physics and Technology. Wiley & Sons, 2002

70. Pearton, Zolper, Shul, Ren, Appl. Phys. Rev. 86, #1, 1 July 1999

71. T.J. Anderson, F. Ren, L. Covert, J. Lin, S.J. Pearton, T. Dalrymple, C. Bozada, R. Fitch, N.
Moser, R. Bedford, M. Schimpf. Journal of Electronic Materials. Vol. 35, No. 4, 2006,
675-679

72. A.Minko,V.Hoel,E.Morvan,B.Grimbert, A.Soltani, E.Delos, D.Ducateau, C.Gaquiere,
D.Theron, J.C.Jaeger, H.Lahreche, L.Wedzikowski,R.Langer and P.Bove, IEEE Electron
Dev.Lett.25,453 (2004).

73. T. J.Anderson, F. Ren, L. Voss, M. Hlad, B. P. Gila, L. Covert, J. Lin, S. J. Pearton, P. Bove,
H. Lahreche, and J. Thuret, J. Vac. Sci. Technol. B 24, 2302 (2006).

74. Travis Anderson, F. Ren, S.J. Pearton, M. Mastro, R. Holm, R. Henry, C. Eddy, J.Y. Lee,
K.Y. Lee, J. Kim. J. Vac. Sci. Technol. B. 24 (5) Sept/Oct 2006


213









4-5. RF characteristics of reference HEMT (top) and drilled HEMT (bottom).............................78

4-6. Simulation results of junction temperature as a function of substrate............................... 79

4-7. TEM im age of SopSiC structure ................................................. ............................... 79

4-8. D C I-V curve for SopSiC H EM T ............................................. ..... .......................... 80

4-9. DC I-V curve for SopSiC HEM T ...................................................................... 80

4-10. Pulse curve for SopSiC H EM T ................................................. ............................... 81

5-1. Changes of IDS as a function of time at 3500C for HEMTs with different combinations
of Ohmic and Schottky contacts. ...... ........................... .......................................87

5-2. RF power output as a function of time ..................................................... ..................87

5-3. A single frequency 32 channel RF and DC reliability test stand.........................................88

5 -4 T h e R F so u rc e ................................................................................................................... 8 8

5-5. The splitter board ........................................................................................................ ......89

5-6. A pre-m watched circuit board .................................................................... ..... ....................89

5-7. Calibration and curve fitting of a diode detector........................................ ............... 90

5-8. D ata acquisition. ...............................................................................90

6-1. A Schottky diode hydrogen sensor shown in cross-section (top) and optical image
(b bottom ) ................... .......................................................... ................. 99

6-2. Forward and reverse bias plot of diode current in varying atmospheres..............................100

6-3. Percentage change in current as a function of bias at 500 ppm H2 ........... ................... 100

6-4. Time dependence of Schottky diode sensor when switching from pure nitrogen
atmosphere to hydrogen concentrations from 1-500 ppm ............................................101

6-5. Percentage change in current as a function of hydrogen concentration under both
forward and reverse bias. ....................................... .. .............. .... ............101

6-6. Percentage change in current as a function of temperature at a hydrogen concentration
of 50 0 p p m ........................................................................... 102

6-7. Time dependence of Schottky diode sensor when switching from pure nitrogen to 1
ppm hydrogen under reverse bias at 150 C ............................................................ 102

6-8. Block diagram of sensor module and wireless network server ................. ................103









operation period with few failures. Finally, there will be another period of high failure rate as the

devices reach the end of their design life cycle. This curve is referred to as the failure density

function, which is the time derivative of the cumulative failure function, which is the fraction of

a group of devices that has failed ay any given time [21].

The main chemical failure mechanism is corrosion. A physical failure mechanism may be

due to creep or other physical aging of polymer encapsulants. Electrical failure can be caused by

electromigration. Most failures on the system level are thermomechanical failures. They are

caused by packaging stress and thermal expansion coefficient mismatch. This can be manifested

in the form of chip cracking, solder joint failure, and delamination. Accelerated testing can be

performed to investigate all of these failure mechanisms. Some accelerated tests include: thermal

cycling or thermal shock (cycle from -20 C to 120 OC, for example), thermal soaking (baking for

long time), mechanical vibration or applied force (shear applied to solder joints), high voltage or

power cycling, and high humidity (environmental chamber at say 90% humidity and 90 C) [21].

A set of standards has been developed by several different groups, each with different acceptance

criteria. The most stringent reliability standards are the US military standards. A more thorough

discussion of electrical reliability will follow in Chapter 5.









10-3. Optical microscope of test die after separation at 165 C (top), 135 C (middle), and
105 C (b bottom ) .......................................................................... 169

10-4. Optical microscope images of actual MUMPS die (left) and DBR die (right) .................169

10-5. O optical im age of a bonded device ...................................................... .......................... 170

10-6. Device testing results using a white light source compared to design simulations using
a planar cavity ............................................................... ..... ..... ........ 170

10-7. Interferometry data showing the curvature of the DBR ............................................... 171

10-8. Device testing results using a white light source compared to calculations based on a
curved cavity ...................................... ................. ................ ........... 171

B- Optical im age of TLM pattern. ................................................ ................................ 194

B-2. Test setup for TLM m easurem ent............................................... ............................. 194

B-3. Typical I-V curves for TLM m easurem ent................................. ........................ .. ......... 195

B-4. Calculation of Rs, Rt, and Rc from TLM measurement............... ................195

B -5. O optical im age of G aN H EM T ................................................................... .....................196

B -6. Test setup for D C m easurem ent................................................. .............................. 196

B -7. T typical D C V ds-Ids curve ......... ..... ............ ................. ................................................197

B -8. T typical D C V g-Ids curve............................................................................ ................... 197

B-9. Test setup for gate lag m easurem ent......................................................... ............... 198

B-10. Typical pulsed Vds-Ids curve compared to DC................ ......... .................... 198

B-11. Typical pulsed Vg-Ids curve compared to DC ..................................... ..................199

B-12. Typical isolation data .............. ................. ............. ........... .. ............ 199

B-13. Schematic of test setup for gate leakage measurement....................... ............... 200

B-14. Typical leakage data ................................... .. ....... .. .... .............. .. 200

B-15. Test setup for breakdown measurement <400 V (top) and >400 V (bottom)....................201

B -16. T typical breakdown n data .................................................................................. ......... 202

B-17. Test setup for small signal RF measurement .............. ......... ...................202









109. A. Baranzahi, A. Lloyd Spetz and I. Lundstrom, Appl. Phys. Lett. 67 3203 (1995).

110. B.S. Kang, R. Mehandru, S. Kim, F. Ren, R. Fitch, J. Gillespie, N. Moser, G. Jessen, T.
Jenkins, R. Dettmer, D. Via, A. Crespo, B.P. Gila, C.R. Abernathy and S.J. Pearton, Appl.
Phys. Lett. 84 4635 (2004).

111. B.S. Kang, Y.W. Heo, L.C. Tien, D.P. Norton, F.Ren, B.P. Gila and S.J. Pearton, Appl.
Phys. A 80 1029 (2005).

112. A. El. Kouche, J. Lin, M.E. Law, S. Kim, B.S. Kim, F. Ren and S.J. Pearton, Sensors and
Actuators B: Chemical 105 329 (2005).

113. S.J. Pearton, B.S. Kang, S. Kim, F. Ren, B.P. Gila, C.R. Abernathy, J. Lin and S.N.G. Chu,
J. Phys: Condensed Matter 16 R961 (2004).

114. S. Kim, B. Kang, F. Ren, K. Ip, Y. Heo, D. Norton and S.J. Pearton, Appl. Phys. Lett. 84,
1698 (2004).

115. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z.N. Low, J. Lin, S.J. Pearton, F. Ren. NHA
Conference Proceedings 2007

116. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z.N. Low, J. Lin, S. J. Pearton, F. Ren, A.
Dabiran, A. Osinsky, J. Painter. Hydrogen Safety Report, July 2007


117. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z.N. Low, J. Lin, S. J. Pearton, F. Ren, A.
Dabiran, A. Osinsky, Journal of the Electrochemical Society (submitted October 2007)

118. Jun-Rui Huang, Wei-Chou Hsu, Yeong-Jia Chen, Tzong-Bin Wang, Kun-Wei Lin, Huey-
Ing Chen and Wen-Chau Liu, Sensors Actuators B 117, 151(2006).

119. B. S. Kang, S. Kim, F. Ren, B. P. Gila, C. R. Abernathy and S. J. Pearton, Sensors
Actuators B 104, 232(2005).

120. K. Matsuo, N. Negoro, J. Kotani, T. Hashizume and H. Hasegawa, Appl. Surf. Sci.244, 273
(2005).

121. J. Song, W. Lu, J. S. Flynn and G. R. Brandes, Solid-State Electronics, 49, 1330(2005).

122. L. Voss, B.P. Gila, S.J. Pearton, Hung-Ta Wang, F. Ren. J. Vac. Sci. Technol. B 23, 6
(2005)

123. H.-T. Wang, B.S. Kang, F. Ren, R.C. Fitch, J.K. Gillespie, N. Moser, G. Jessen, T. Jenkins,
R. Dettmer, D. Via, A. Crespo, B.P. Gila, C.R. Abernathy, S.J. Pearton. Appl. Phys. Lett.
87, 172105 (2005)

124. C.-W. Hung, H.-L. Lin, H.-I. Chen, Y.-Y. Tsai, P.-H. Lai, S.-I Fu, W.-C. Liu. Electron
Device Letters, 27, 12 (2006)


216









established. The other materials are high performance materials for specialized applications. The

encapsulation process is typically a simple transfer molding process for wire-bonded packages.

Encapsulation of flip-chip packages is accomplished by application around the edge of the chip,

and then a capillary force will pull the material under the chip to fill the void [21].

3.8 Board-Level Operations

A printed wiring board (PWB), also called printed circuit board (PCB), is typically a

composite of organic and inorganic materials. It contains internal and external wiring, power,

and heat sinking. It carries all components of an entire system, so it can also be called a system

board or motherboard. It is made of an insulating structure with copper wires both on the surface

and in internal layers. Electrical components are soldered on the board to form both the electrical

and mechanical connection. FR-4 is the most commonly used composite board material,

constructed of multiple plies of epoxy-resin impregnated onto a woven glass cloth. A cross

section of a typical PCB is shown in Figure 3-15 [21].

3.8.1 Printed Wiring Board Fabrication

There are two methods of board fabrication: sequential build-up and parallel build-up. The

sequential process follows the device chip fabrication process. A series of layers of dielectrics

and metals are sequentially deposited and patterned. The parallel build-up process involves

fabricating each think film layer separately, then laminating all layers together. The processes

used for PCB fabrication are primarily lithography and etching. Most processing is subtractive,

with Cu metal deposited on the board from 10-50 |tm thick, followed by lithography with a

negative resist and acid etching. An additive process can be used to build thick films, however

instead of a lift-off process, a combination of electroless and electrodeposition is used. The

electroless plated copper can be used to selectively grow a copper film on the PCB material

where there was no copper before. It will selectively grow on windows opened up in the






















Vf Aul





Figure 10-1. A tuneable filter device


Figure 10-2. SEM cross-section of In film bonded to Au film


|i|niG i| 161 JtPi




















170



165



160



155



150


U









U







U


U
* I I I I


L=I i
W -250 mn
4 fingers
SiC=100 Jm
Au=5 mn
Si=25 pm
PDMS=4 um
Power= 10 WImr
h=s3 Wim-C


Gate Pitch (um)


Figure 7-14. Effect of gate pitch on the junction temperature of the HEMT in a multiple finger
configuration


1111111111















120- DC
-1 kHz
100- 100 kHz

Vds=7V
80


60 -


40 -


20 -
0
0 '- ,, e -- -
-6 -5 -4 -3 -2 -1 0





Figure 4-10. Pulse curve for SopSiC HEMT









Acid solutions present many benefits, including higher deposition rates, greater stability,

and improved physical characteristics of the deposition. The general reaction is as follows:

Ni2 + NaH2PO2 + buffers, complexors, stabilizers -> (Ni + P) + H2 + NaHPO3

This reaction takes place on a catalytic surface, either activated using Sn/Pd or an acid-

activated seed metal surface. Iron has been discussed as a seed metal, but it is a deep level trap

for most semiconductors, making it of little use. Also discussed have been Pt and Ni. The most

common salts used for the Ni source are the chloride and sulfate, in concentrations of 4-10 g/L.

Sodium hypophosphite is virtually the only reducing agent for Ni. The reduction potential for Ni

is -0.2 V, so this reaction requires a strong reducing agent. The buffers and complexors are

typically single ingredients, and almost all are organic acids. These stabilize the pH of the bath

and mildly complex the Ni. Again, moderators and wetting agents can be added if necessary.

Temperature is the most important factor that influences deposition rate, with essentially no

deposition occurring below 50 C, and most baths operating between 80 and 100 C [139].

9.1.2 Alpha-Voltaics

An alpha-voltaic device is very similat to a solar cell, except it operates by producing

power from alpha-particle radiation rather than solar (visible) radiation. This is particularly

useful for portable power supply applications, in which a radiation source is coupled to the

devices, thus producing a completely self-contained long-lifetime power supply. Alpha-voltaics

are particularly attractive over beta-voltaics due to the less strict shelding requirements for alpha

particles. The device consists of an alpha radiation source coupled with a semiconductor p-n

junction, which will collect the electron-hole pairs created by the particle as it penetrates the

material. The limiting factor in the development of these devices has been the radiation damage

to the semiconductor. While the radiation source is stable over a long time period (Americium-


































Figure 4-6. Simulation results of junction temperature as a function of substrate.


Figure 4-7. TEM image of SopSiC structure


5 W/mm
A 10W/mm
* reference
A 15W/mm




Full Text

PAGE 1

1 ADVANCED THREE-DIMENSIONAL PACKAGING SCHEMES FOR MICROELECTRONIC AND MICROSYSTEM APPLICATIONS By TRAVIS JAMES ANDERSON A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008

PAGE 2

2 2008 Travis James Anderson

PAGE 3

3 To my family.

PAGE 4

4 ACKNOWLEDGMENTS First of all, I would like to thank m y s upervisory committee chair, Dr. Fan Ren. His technical and personal support have been priceless. He has always been there to give me advice and guide me when I had problems. He has treate d me like a family member, and I truly consider Dr. Ren and the members of my research group to be like a family to me. I am also indebted to the rest of my committee: Drs. Steve P earton, Jenshan Lin, and Kirk Ziegler. This work could not have been completed without the support and a ssistance of current and former group members Hung-Ta Wang, Yu-Lin Wang, Nimo Chen, Barrett Hicks, and Drs. Byoung Sam Kang and Soohwan Jang in chemical engineering; Lance Covert, Jaeshin Kim, Changzhi Li, Zhen-Ning Low, Xiaogang Yu, and Fu-Yi Han in electrical engineering; Lars Voss, Jon Wright, Wantae Lim, and Drs. Luc Stafford and Rohit Kh anna in materials science. I also wish to thank Dr. Brent Gila in materials sc ience for all of the help with equipment service and materials analysis. The support staff in the Chemical Engineering has been first-class: Deborah Aldrich in purchasing and Dennis Vince and Jim Hinnant in the machine shop. Jim has been a savior on the reliability project. The UF nanofab facility staff, Ivan Kravchenko and Bill Lewis, deserve recognition for keeping the equi pment running and assisting me with processing and troubleshooting. Our collaborators have been especially impor tant in providing me with samples. In particular, I thank Drs. Robert Bedford, Tom Nelson, and Eddie Ochoa at Air Force Research Labs for the flip-chip bonding work; Wayne J ohnson, Pradeep Rajagopal, and Chris Park at Nitronex Coropration, Jeff LaRoche at Raythe on, and Jihyun Kim at Korea University. I also want to thank Dr. Al bert Baca at Sandia National Labs for hosting me on a summer internship; and staff members Garth Kraus, Torben Fortune, and Carlos Sanchez for assisting me in learning the equipment and accomplishing my goa ls for the project. I know it was difficult for

PAGE 5

5 them at times, but I learned a lot and the experi ence gave me new ideas that I brought back to UF. I would not be where I am today if it were not for Dr. Cliff Hende rson at Georgia Tech, and former students Drs. Celesta White and Be nita Comeau, who took a chance on hiring me as an undergraduate assistant on a microsystem fabrication project. They took the time to teach me many of the processing skills that made my tran sition to Dr. Rens group easy. This project was what convinced me to go to graduate school an d do research for a livi ng. It was Dr. Henderson who pushed me to meet with Dr. Ren when I got accepted to UF. The reason I was able to finish this work was the close support from my friends. In particular, I would like to thank Justin Sellers Alex Welsh, Mark Kozek, and Luke Neal. They have been there for me for at least the past 4 years. Through good times and bad times, they have been there to support me. I thank my family for their love and support. I hope this work in some way repays them for all of the love they have given me over the year s. Finally, I want to th ank my wife, Kristin. She went above and beyond anything I coul d ever ask of her to stick w ith me and support me. She is the love of my life, and best friend.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ........11 LIST OF FIGURES.......................................................................................................................12 ABSTRACT...................................................................................................................................18 CHAP TER 1 INTRODUCTION................................................................................................................. .20 1.1 Motivation.........................................................................................................................20 1.1.1 Electronic Devices..................................................................................................20 1.1.2 Three-Dimensional Packaging...............................................................................21 1.2 Study Outline.............................................................................................................. ......22 92 COMPOUND SEMICONDUCT OR LITERATURE REVIEW............................................ 24 2.1 Wide Bandgap Electronic Devices................................................................................... 24 2.2 High Electron Mobility Transistors.................................................................................. 26 3 SEMICONDUCTOR PROCESSING.................................................................................... 32 3.1 Overview...........................................................................................................................32 3.2 Crystal Growth............................................................................................................. .....32 3.2.1 Growth from Melt................................................................................................... 32 3.2.2 Epitaxial Growth.................................................................................................... 34 3.3 Dopant...............................................................................................................................36 3.3.1 Diffusion.................................................................................................................36 3.3.2 Ion Implantation..................................................................................................... 38 3.4 Film Deposition............................................................................................................ ....39 3.4.1 Chemical Vapor Deposition................................................................................... 39 3.4.2 Metal Deposition.................................................................................................... 40 3.5 Lithography................................................................................................................ .......42 3.5.1 Photoresists.............................................................................................................42 3.5.2 Optical Lithography................................................................................................45 3.5.3 Electron Beam Lithography................................................................................... 47 3.6 Etching..............................................................................................................................48 3.6.1 Wet Etching............................................................................................................ 48 3.6.2 Dry Etching............................................................................................................50 3.7 Packaging..........................................................................................................................51 3.7.1 Wire Bonding......................................................................................................... 52 3.7.2 Flip-Chip Bonding.................................................................................................. 53

PAGE 7

7 3.7.3 Encapsulation.........................................................................................................54 3.8 Board-Level Operations....................................................................................................55 3.8.1 Printed Wiring Board Fabrication.......................................................................... 55 3.8.2 Assembly and Testing............................................................................................56 3.9 Reliability.........................................................................................................................57 4 HIGH ELECTRON MOBILITY TRANSISTORS................................................................ 69 4.1 Overview...........................................................................................................................69 4.2 Silicon Carbide Substrates................................................................................................ 69 4.3 Silicon On Polycrystalline Silicon Carbide Substrates..................................................... 71 4.4 Silicon Substrates......................................................................................................... .....74 5 RF RELIABILITY............................................................................................................... ...82 5.1 Overview...........................................................................................................................82 5.2 Research Plan Stress Testing.........................................................................................83 5.3 Stress Test System......................................................................................................... ...84 6 GAS SENSING.................................................................................................................. ....91 6.1 Overview...........................................................................................................................91 6.2 Hydrogen Sensors.............................................................................................................92 6.2.1 Gallium Nitride High Electron Mobility Transistor Sensors................................. 92 6.2.2 Wireless Hydrogen Sensor Network...................................................................... 95 7 THERMAL SIMULATIONS OF THREE-DIMENSIONAL PACKAGE .......................... 106 7.1 Overview.........................................................................................................................106 7.2 Finite-Element Modeling................................................................................................ 106 7.2.1 Heat Transfer........................................................................................................107 7.2.2 First Generation Simulation: GaN Power Amplifier on Heat Sink...................... 113 7.2.3 Generation II: Combined Am plifier and Silicon Chip ......................................... 115 8 UV LASER PROCESSING................................................................................................. 126 8.1 Overview.........................................................................................................................126 8.2 Solid State Laser Processing........................................................................................... 129 8.2.1 Background: Diode Pumped Solid State Lasers................................................... 129 8.2.2 Experiments: Frequency Effects........................................................................... 131 8.3 Excimer Laser Processing............................................................................................... 133 8.3.1 Background: Excimer Lasers............................................................................... 133 8.3.2 Experiments: Laser Type...................................................................................... 139 9 ELECTROLESS METAL DEPOSITION............................................................................ 149 9.1 Overview.........................................................................................................................149 9.1.1 Nickel Plating....................................................................................................... 149

PAGE 8

8 Alpha-Voltaics ......................................................................................................151 9.2 Electroless Plating Procedures........................................................................................ 152 9.2.1 Solutions Made In-House.....................................................................................152 9.2.2 Commercial Plating Solutions.............................................................................. 155 10 FLIP-CHIP BONDING........................................................................................................ 163 10.1 Overview.......................................................................................................................163 10.2 Bonding Studies............................................................................................................164 10.2.1 Metal Bonding....................................................................................................165 10.2.2 Polymer Bonding................................................................................................ 166 10.3 Device Demonstration.................................................................................................. 167 11 FUTURE WORK................................................................................................................. .172 11.1 Processing Studies........................................................................................................172 11.2 Reliability............................................................................................................... ......172 11.3 Gas Sensors...................................................................................................................173 11.4 Simulations............................................................................................................... ....173 11.5 Laser Drilling............................................................................................................ ....174 11.6 Electroless Plated Metal............................................................................................... 175 11.7 Flip-Chip Bonding........................................................................................................175 12 CONCLUSION.................................................................................................................. ...177 APPENDIX A EQUIPMENT OPERATION................................................................................................ 178 A.1 Purpose...........................................................................................................................178 A.2 Spin-Coating..................................................................................................................178 A.3 Mask Aligner (MJB-3)................................................................................................... 178 A.4 Raith E-Beam Lithography............................................................................................ 179 A.5 Plasma-Therm PECVD.................................................................................................. 180 A.6 Evaporator (CHA)..........................................................................................................181 A.7 Old Evaporator...............................................................................................................181 A.8 Thermal Evaporator.......................................................................................................182 B ELECTRICAL MEASUREMENTS.................................................................................... 183 B.1 Purpose...........................................................................................................................183 B.2 Ohmic Contacts............................................................................................................. .183 B.2.1 Overview..............................................................................................................183 B.2.2 Procedure.............................................................................................................183 B.3 Direct Current............................................................................................................. ...184 B.3.1 Overview..............................................................................................................184 B.3.2 Procedure.............................................................................................................185 B.4 Pulse (Gate Lag).............................................................................................................186

PAGE 9

9 B.4.1 Overview..............................................................................................................186 B.4.2 Procedure.............................................................................................................186 B.5 Leakage/Isolation...........................................................................................................187 B.5.1 Overview..............................................................................................................187 B.5.2 Procedure.............................................................................................................187 B.6 Breakdown/High Power.................................................................................................188 B.6.1 Overview..............................................................................................................188 B.6.2 Procedure.............................................................................................................189 B.7 Small Signal RF (S-parameters).................................................................................... 189 B.7.1 Overview..............................................................................................................189 B.7.2 Procedure.............................................................................................................190 B.8 Large Signal RF (Load-Pull)..........................................................................................191 B.8.1 Overview..............................................................................................................191 B.8.2 Procedure.............................................................................................................192 C PROCESSING RECIPES..................................................................................................... 205 C.1 Purpose...........................................................................................................................205 C.2 Photolithography............................................................................................................205 C.2.1 Mesa Etching (S-1045)........................................................................................ 205 C.2.2 Lift-Off (S-1818)................................................................................................. 205 C.2.3 Lift-Off (S-1808)................................................................................................. 205 C.2.4 Lift-Off (LOR/1808)............................................................................................ 205 C.2.5 Image Reversal (5214)......................................................................................... 206 C.2.6 Air Bridge............................................................................................................206 C.3 E-Beam Lithography......................................................................................................206 C.3.1 Etching (PMMA).................................................................................................206 C.3.2 Lift-off (PMMA/PMMA-MAA).......................................................................... 206 C.3.3 T-gate (PMMA/PMMA-MAA/PMMA)..............................................................207 C.4 Plasma Etching............................................................................................................. ..207 C.4.1 Inductively Coupled Plasma Etch With Cl2 (AlGaN/GaN and Au).................... 207 C.4.2 Reactive Ion Etch With CF4 (SiO2, SiNx)............................................................207 C.4.3 Reactive Ion Etch With O2 (polymers).............................................................. 207 C.5 Plasma Enhanced Chemical Vapor Deposition............................................................. 208 C.5.1 Silicon Nitride (SiNx)...........................................................................................208 C.5.2 Silicon Dioxide (SiO2).........................................................................................208 C.6 Wet Etching................................................................................................................ ....208 C.6.1 Buffered Oxide Etch (SiNx, SiO2).......................................................................208 C.7 Thermal Evaporation......................................................................................................208 C.7.1 Indium..................................................................................................................208 C.8 Flip-Chip Bonding.........................................................................................................209 C.8.1 Indium-Gold.........................................................................................................209 C.8.2 Gold-Gold............................................................................................................209 C.8.3 Polymer (SU-8-SU-8).......................................................................................... 209

PAGE 10

10 LIST OF REFERENCES.............................................................................................................210 BIOGRAPHICAL SKETCH.......................................................................................................219

PAGE 11

11 LIST OF TABLES Table page 7-1. Effect of boundary conditions on package temperatures...................................................... 118 8-1. Electrical data demonstr ating the effect of laser drilling on etching and contact properties..........................................................................................................................142 8-2. Excimer laser gases and wavelengths...................................................................................142

PAGE 12

12 LIST OF FIGURES Figure page 2-1. Energy band structure of AlGaN/GaN heterojunction ...........................................................31 2-2. Cross-section of an AlGaN/GaN HEMT ................................................................................31 3-1. A Czochralski growth system............................................................................................... ..59 3-2. A float-zone refining system..................................................................................................60 3-3. A Bridgman furnace........................................................................................................ .......60 3-4. Standard flat orientations for Silicon.................................................................................... ..61 3-5. A MBE reactor........................................................................................................................61 3-6. A MOCVD reactor.................................................................................................................62 3-7. An ion implantor.....................................................................................................................62 3-8. Positive and negative photoresist patterning..........................................................................63 3-9. Photolysis of DQ upon UV exposure.....................................................................................64 3-10. SEM image of a dual layer PMMA-MAA/PMMA resist structure for lift-off.................... 64 3-11. An optical lithography system............................................................................................ ..65 3-12. An electron beam lithography system.................................................................................. 65 3-13. Comparison of pattern transf er via wet and dry etching ...................................................... 66 3-14. Schematic of packaging schemes......................................................................................... 66 3-15. Cross section of a PCB................................................................................................... ......67 3-16. Demonstration of through-hole and surface mount techn ology........................................... 67 3-17. Failure density function........................................................................................................68 4-1. Cross-Section of an AlGaN/GaN HEMT on SiC ................................................................... 75 4-2. Image of reference FET (left) and FET adjace nt to hole drilled at 355 nm (center, right).... 75 4-3. I-V Characteristics of reference HEMT (top) and drilled HEMT (bottom )........................... 76 4-4. I-V characteristics of reference HE MT (top) and drilled HEMT (bottom )............................ 77

PAGE 13

13 4-5. RF characteristics of reference HE MT (top) and drilled H EMT (bottom)............................. 78 4-6. Simulation results of junction temp erature as a function of substrate. ................................... 79 4-7. TEM image of SopSiC structure............................................................................................79 4-8. DC I-V curve for SopSiC HEMT...........................................................................................80 4-9. DC I-V curve for SopSiC HEMT...........................................................................................80 4-10. Pulse curve for SopSiC HEMT............................................................................................ 81 5-1. Changes of IDS as a function of time at 350C for HEMTs with different combinations of Ohmic and Schottky contacts........................................................................................ 87 5-2. RF power output as a function of time................................................................................... 87 5-3. A single frequency 32 channel RF and DC reliability test stand............................................ 88 5-4. The RF source.........................................................................................................................88 5-5. The splitter board........................................................................................................ ............89 5-6. A pre-matched circuit board............................................................................................... ....89 5-7. Calibration and curve fitting of a diode detector....................................................................90 5-8. Data acquisition.......................................................................................................... ............90 6-1. A Schottky diode hydrogen sensor shown in cross-section (top) and optical im age (bottom)....................................................................................................................... .......99 6-2. Forward and reverse bias plot of diode current in varying atm ospheres.............................. 100 6-3. Percentage change in current as a function of bias at 500 ppm H2......................................100 6-4. Time dependence of Schottky diode sensor when switching from pure nitrogen atmosphere to hydrogen concen trations from 1-500 ppm............................................... 101 6-5. Percentage change in current as a function of hydrogen concentration under both forward and reverse bias. .................................................................................................101 6-6. Percentage change in curre nt as a function of tem peratur e at a hydrogen concentration of 500 ppm.......................................................................................................................102 6-7. Time dependence of Schottky diode sens or when switching from pure nitrogen to 1 ppm hydrogen under reverse bias at 150 C....................................................................102 6-8. Block diagram of sensor modul e and wireless network server ............................................ 103

PAGE 14

14 6-9. System timing of at wireless sensor node .............................................................................103 6-10. Interface of online hydrogen level m onitoring................................................................... 103 6-11. Photo of sensor system (a) Sensor with sensor device; (b) Sensor and base station; (c) com puter interface with base station................................................................................ 104 6-12. Field test result s of four sensors .........................................................................................104 6-13. Field test results for single diod e sensors over a period of one week ................................ 104 6-14. Differential diode test results using 1% H2 at room temperature and elevated temperature.................................................................................................................... ..105 6-15. Field test results over a m onth period for differential di ode sensors with boride-based ohmic contacts................................................................................................................. 105 7-1. First-generation prototype of integrated heat sink/antenna incorporating GaN PA and Si modulator .........................................................................................................................118 7-2. Concept of the 3-D multi-layer structure com bining heat-sink antenna, GaN RF and silicon signal pro cessing electronics................................................................................ 118 7-3. Example of heat transfer through a composite wall............................................................. 119 7-4. Mesh used for finite element modeling................................................................................ 119 7-5. Effect of substrate thermal conductivity on the junction tem perature of the HEMT........... 120 7-6. Effect of silicon carbide substrate thickn ess on the junction tem perature of the HEMT..... 120 7-7. Effect of power dissipation on th e junction temperature of the HEMT ............................... 121 7-8. Effect of Si thickness on the junction temperature of the HEMT........................................ 121 7-9. Typical temperature distri bution under base o perating cond itions in the GaN layer (top) and Si layer (bottom)....................................................................................................... 122 7-10. Typical temperature profile in a cross section of the stru cture zoom ed in to the area near the gate.................................................................................................................. ...123 7-11. Effect of HEMT power dissipation on the m aximum temperature in both active layers... 123 7-12. Effect of the removal of top PDMS layer and both PDMS layers on the m aximum temperature in both active laye rs at varying power levels............................................... 124 7-13. Effect of SiNx passivation thickness on the junc tion temperature of the HEMT............... 124

PAGE 15

15 7-14. Effect of gate pitch on the junction tem perature of the HEMT in a multiple finger configuration....................................................................................................................125 8-1. SEM cross sections of holes drille d at 1064 nm (left) and 532 nm (center)........................ 142 8-2. SEM images of hole drilled from fr ont at 1064 nm (above) and 532 nm (below)............... 143 8-3. SEM image of holes drilled at 532 nm from front (above, left) and back (below, left) and 355 nm from front (above, righ t) and back (below, right)........................................ 144 8-4. SEM at an angle of holes drilled from the front at 1064 nm (top), from the back at 355 nm (center), and from front at 355 nm (bottom)..............................................................145 8-5. Photographs of TLM patterns with T i/Al/Pt/Au based m etalliz ation on an undrilled sample (left), drilled with 1064 m laser ( center), drilled with 355 nm laser (right)...... 146 8-6. The UV photoablation process.............................................................................................146 8-7. Optical microscope images of front and back of via holes for med in poly-SiC substrates with a DPSS laser (left) a nd excimer laser (right) from the top semiconductor surface (top) and bottom SiC surface (bottom)............................................................................ 147 8-8. SEM images of via holes formed in polySiC substrates with a DPSS laser (left) and excim er laser (right) from the top semiconductor surface............................................... 147 8-9. SEM images of cross-sections of via hol es for med in poly-SiC substrates using DPSS laser (left) and excimer laser (right ) at low magnification (top) and high magnification (bottom)....................................................................................................148 8-10. Optical image of array of holes formed in GaAs................................................................ 148 9-1. An -voltaic battery .............................................................................................................158 9-2. Optical microscope pictures of Ni plated on Si (4 m on top, 11 m on the bottom), with exposed holes...........................................................................................................159 9-3. Photo (top) and optical microscope image (bottom ) of Ni hard mask for SiC trench etching........................................................................................................................ ......160 9-4. Cross-section of SiC MIPS device.......................................................................................161 9-5. Photo (top) and optical microscope image (bottom ) of MIPS device fabricated using an electroless Ni etch mask................................................................................................... 162 10-1. A tuneable filter device.......................................................................................................168 10-2. SEM cross-section of In film bonded to Au film............................................................... 168

PAGE 16

16 10-3. Optical microscope of te st die after separation at 165 C (top), 135 C ( middle), and 105 C (bottom)...............................................................................................................169 10-4. Optical microscope images of actual MUMPS die (lef t) and DBR die (right).................. 169 10-5. Optical image of a bonded device...................................................................................... 170 10-6. Device testing results using a white light source com pared to design simulations using a planar cavity..................................................................................................................170 10-7. Interferometry data showing the curvature of the DBR.....................................................171 10-8. Device testing results us ing a white light source com par ed to calculations based on a curved cavity....................................................................................................................171 B-1. Optical image of TLM pattern............................................................................................. 194 B-2. Test setup for TLM measurement........................................................................................ 194 B-3. Typical I-V curves for TLM measurement.......................................................................... 195 B-4. Calculation of Rs, Rt, a nd Rc from TLM m easurement...................................................... 195 B-5. Optical image of GaN HEMT.............................................................................................. 196 B-6. Test setup for DC measurement........................................................................................... 196 B-7. Typical DC Vds-Ids curve...................................................................................................197 B-8. Typical DC Vg-Ids curve.....................................................................................................197 B-9. Test setup for gate lag measurement.................................................................................... 198 B-10. Typical pulsed Vds-Ids curve compared to DC................................................................. 198 B-11. Typical pulsed Vg-Ids curve compared to DC.................................................................. 199 B-12. Typical isolation data................................................................................................... ......199 B-13. Schematic of test setup for gate leakage measurement...................................................... 200 B-14. Typical leakage data..................................................................................................... .....200 B-15. Test setup for breakdown measurem ent <400 V (top) and >400 V (bottom ).................... 201 B-16. Typical breakdown data.....................................................................................................202 B-17. Test setup for small signal RF m easurement..................................................................... 202

PAGE 17

17 B-18. Typical s-parameter data................................................................................................. ...203 B-19. Typical H21 and U calculation for Ft and Fmax............................................................... 203 B-20. Test setup for load-pull measurement................................................................................ 204 B-21. Typical load-pull data................................................................................................... .....204

PAGE 18

18 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ADVANCED THREE-DIMENSIONAL PACKAGING SCHEMES FOR MICROELECTRONIC AND MICROSYSTEM APPLICATIONS By Travis James Anderson May 2008 Chair: Fan Ren Major: Chemical Engineering The emergence of GaN-based devices promis es a revolution in areas requiring high performance electronics, such as high speed earth and space-based communication systems, advanced radar, integrated sensors, high temp erature electronics, and utility power switching. The properties of this system make it ideally suit ed for operation at elevated temperatures and at voltage and current levels well beyond that acces sible by Si. Recent improvements in material quality and device performance are rapidly ope ning the door to commerc ialization, and III-N technologies are demonstrating exciting developments of late. Though devices are entering commer cialization, there is still some work to be studied. In particular, GaN high electron mobility transistors (HEMTs) show potential as gas sensors, which will be relevant to the emerging hydrogen fuel ce ll vehicle market. Devices were fabricated and show a very low detection limit of <10 ppm, and have been integrat ed in a wireless network that is currently undergoing field testi ng. Devices have also been fabricated on different substrates such as Si (low-cost), SiC (high performan ce), and SopSiC (novel material potentially combines advantages of Si and SiC) to st udy the effects on device performance and their compatibility with processing for 3-D integration. Re liability testing is a major area of interest. A 32-channel stress test system has been de signed and is currently being built.

PAGE 19

19 A 3-D integration project is being undertaken to achieve 3-D bonding of integrated circuits and other components fabricated from dissimilar materials. The thermal design of a vertically integrated multi-chip-module (MCM) based on Ga N High Electron Mobility Transistor (HEMT) power amplifiers (PA) on SiC substrates with a backside heat sink/an tenna and Si modulator, bonded to a common ground plane using polydimet hylsiloxane (PDMS), was studied. Heat transfer was estimated using finite element m odeling for different PA power densities, HEMT gate finger pitch, layer thickness, the presence or absence of the thermally insulating layers, and the thickness of dielectric isolation interlayers. Laser drilling is a promising method for thr ough hole via formation in SiC, presenting several advantages over dry etch ing, the most important being c onsiderably higher etch rates. Studies have been undertaken to minimize the surface contamination and semiconductor degradation due to this process and, ultimately, ma ke this process compe titive with conventional dry etch techniques. By using a UV excimer lase r source (193 nm) for drilling, we can achieve considerably better smoothness inside the holes and minimize surface contamination, compared to the use of the more common Nd:YVO4 laser. The device characteristics of AlGaN/GaN HEMT layers grown on SiC substrates were si milar after formation of vias by 193 nm laser drilling to those from an undrilled reference sample. By sharp contrast, 1064, 532, and 355 nm laser drilling produces significan t redepostion of ablated materi al around the via and degrades the electrical properties of the HEMT layers. Flip-chip bonding is critical for 3-D integrati on is it is the method that actually forms the bonds. Studies have been performed to look at vari ous materials systems, particularly In, Au, and SU-8 polymer, as candidate materials for flip -chip bonding applications. Bonding protocols have been developed to optimize the mechanical st rength of the bond fo r a MEMS application.

PAGE 20

20 CHAPTER 1 INTRODUCTION 1.1 Motivation The Galluim Nitride (GaN) materials system is attracting much intere st at an industrial level in the late 2000s. Due to the wide-bandgap na ture of the material, the material is very thermally stable, and electronic devices can be operated at 500 C. The material is also chemically stable, with the only known wet et chant being molten NaOH or KOH, making it very suitable for operation in chemically harsh environments or in radiation. This system also demonstrates a high electron mobility, making it suitable for the high frequency communications market, and high breakdown field, making it marketable to the high power industry as well. While this technology has had a very long infancy as researchers st rived to develop high quality material on high performance substrates, and develop novel processing schemes to improve device performance to the inherent material limits it appears in the past couple years that the technology is beginning to mature and m ove from the lab into production. 1.1.1 Electronic Devices There is currently strong interest in de veloping A lGaN/GaN High Electron Mobility Transistors (HEMTs) for use in high power microwave transmission systems, millimeter-wave (MMW) military communications links, and X-band radar systems. Nitride-based HEMTs can operate from very high frquency (VHF) thr ough X-band frequencies with higher breakdown voltage, better thermal conductivity, and wider transmission bandwidths than Si or GaAs devices. GaN-based HEMTs can operate at signi ficantly higher power densities and higher impedance than currently used GaAs devices. The higher sheet carrier density in the twodimensional electron gas compared to the AlGa As/GaAs system, and higher electron mobility relative to Si, suggest that nitride-based HEMT s will also exhibit low on-state resistance, an

PAGE 21

21 important advantage for rf applications. Many communications systems for the military operate in the frequency range from 7->44 GHz, and in some cases, the systems are space-borne, where amplifier efficiency, reli ability, and low weight ar e key requirements [1-16]. 1.1.2 Three-Dimensional Packaging As de mands on portable devices increase, new packaging techniques must be developed to meet the demands for smaller, lighter, and more functional devices. The early solutions, such as ball grid arrays (BGA) and flip-chip technologie s, have met the demands for decreasing package size and interconnect distance and increasing I/O count, but in toda ys mobile landscape, the chip footprint area is becoming a huge factor. 3-D packaging solutions meet both the size and performance requirements. Vertical stacking of multiple die within a package, using specialized substrates and interconnects, wi ll reduce the number of chip-toboard connections and decrease the area required for chips and inter-chip wire traces. These techniques are also advantageous from a power consumption standpoint since 40% of power consumption comes from chip-to-chip interconnects. The module-to-board solder connects account for almost 90% of board failures, so reducing the number of connections, we can decrease board failures and attain an overall increase in reliability and decrea se in power consumption [17-18]. Laser drilling is a well established method for the machining of metals and hard materials. The mechanism for this process is simple. Essent ially, the laser supplies sufficient power to eject the material. The actual ejection mechanis m is dependent upon wavelength. At longer wavelengths (532 and 1064 nm), the mechanism is sublimation of the sample and subsequent ejection. As the wavelength decreases, moving into the UV region (355 nm), the mechanism changes to a molecular level bond-breaking phe nomenon. This, therefore, results in much cleaner profiles and less redeposited ablation debris. Since the laser can be focused and controlled for much smaller features than standard machining tools, it is an ideal method for

PAGE 22

22 micromachining and scribing. There is current inte rest in developing lase r drilling processes to create through-wafer vias in ha rd compound semiconductor substrate materials, such as SiC, for both electronic and micro-electro-mechanical systems (MEMS) applications. AlGaN/GaN HEMTs grown on SiC are very promising for high power, high temperature devices due to the superior thermal conductivity of SiC, compared to the original sapphire substrates. Monolithic mm-wave integrated circuits based on these de vices will require through wafer via connections to provide a common ground and low inductance path from the source of the FET to the ground plane in the back of the wafer [19-36]. 1.2 Study Outline There are four sections o f this dissertation. In addition to prov iding the results of research performed over the past four years, I also wi sh to share my background knowledge on the subject to create a sort of condensed manuals for futu re students in the group. To that end, the next couple of chapters will provide purely backgr ound information. Chapter two will provide a cursory literature review of compound semiconductors, from initial research and development up to the current state of the art. Chapter three will provide a short summary of device processing, starting with substrate wafer fabrication and th e processing steps for de vice fabrication, all the way up through device packaging, circuit board fa brication, and systems assembly and testing. These two chapters are intended serve as an introduc tion to the field, a nd suggest first-day literature for new students to the group. The next sections will cover re search performed over the past four years in detail. This can be divided into two main sections: discrete de vices and packaging-level work. Chapters four, five, and six will focus on the devices, and cove r efforts to fabricate high performance devices (Ch. 4), gas sensor devices (Ch. 6), and reliabilit y and stress testing (Ch. 5) which is currently a hot topic. Chapters seven, eight, nine, and ten will focus in 3-D packaging work. Topics to be

PAGE 23

23 covered include thermal simulations of a packaged chip to optimize heat transfer (Ch. 7), laser drilling for interconnect formation (Ch. 8), elec troless metal plating for thick film deposition (Ch. 9), and flip-chip bonding for assembly (Ch. 10). The fourth section will conclude my researc h. This will contain a chapter on future work (Ch. 11), which will outline the steps to be taken to wrap up the current work and suggestions for where to take the work in the future. Chap ter 12 will serve as th e conclusion, and wrap everything up and tie it together. The fifth section will be operating pro cedures. Appendix A will cover operating procedures for processing equipment in the lab, Appendix B will cover electrical testing, and Appendix C will include tables of recipes used in processing. The intent of this section is to consolidate information about the lab. I am a firm believer in leaving so mewhere in better shape than it was in when I arrived, and wish to make my contribution by clearly documenting and arranging the necessary information about the lab. I hope that this section can again be distributed to new students on the first day to aid in training, and serve as a master document that is added to and modified by students each ye ar to keep the information current.

PAGE 24

24 CHAPTER 2 COMPOUND SEMICONDUCTOR LITERATURE REVIEW 2.1 Wide Bandgap Electronic Devices Since the 1970s, III-nitride sem iconductors have been considered promising materials for device applications at blue and ultraviolet (UV) wavelengths, much in the same way that GaAsbased and InP-based materials have been commer cialized for infrared (IR), red, and yellow wavelengths. AlN, GaN, and InN are all wide bandgap materials and ca n crystallize in both wurtzite and zinc-blende structures. These material s, along with their alloys offer an essentially continuous range of direct band gap energies through the visible spectrum and into UV wavelengths. This materials system is, therefore, quite attractive for optoelectronic applications. Applications include light emitting diodes (LEDs) laser diodes (LDs), and detectors. GaN LEDs are essential for development of full-color LED displays, combining, for example, GaN-based blue and green LEDs with GaAs-based red LEDs to form a white LED [37-38]. Another area of interest for III-N materi als is the high temperature and high power electronics market. These material s are perfectly suited for such a pplications due to their wide bandgap nature and high breakdown fields. Due to the wide bandgap, these materials become intrinsic at much higher temperatures than Ge, Si, or GaAs, allowing for fewer high-cost processing steps and design issues to maximize cooling. The breakdown field scales with the square of the band gap, so, ag ain, the wide bandgap material s demonstrate higher breakdown fields, allowing for potential use as a replacement for vacuum tubes in the utility industry. GaN in particular shows excellent tr ansport properties, including high carrier mobility and high drift velocity, making it suitable for general electronics particularly microwave devices. This allows for the exploitation of three regi mes which have previously been unattainable: high temperature, high power, and high frequency. Perhaps the best feat ure of III-N materials is the heterostructure

PAGE 25

25 technology. These materials can support quantum well, modulation-doped heterointerface, and heterojunction structures. Also attractive is the exceptionally high thermal and mechanical stability [37, 39-42]. Research on III-N growth began in the 1960s One difficulty encountered is the lack of available substrates. For this reason, efforts we re essentially stopped on III-N growth during the 1970s and 1980s. Given the constraints, sapphire (Al2O3) and SiC are the most popular substrates currently used. Bulk GaN wafers would be idea l, but that technology has yet to be developed. SiC is a particularly interest ing substrate due to its high thermal conductivity (~350 W/m-K). Metal organic chemical vapor deposition (MOC VD) has emerged as the choice method for growing GaN due to its high degr ee of compositional control and uniformity, availability of high purity sources, and large scale manufacturing compatibility [37,43]. One key parameter for a GaN film is defect density. There has been considerable progress in the last decade on epitaxial growth techniqu es and heterostructures, which will minimize defect density and allow for device fabrication. Another issue with wide bandgap growth is the doping. Native defects, particularly nitrog en vacancies, make the MOCVD material predominantly n-type; however, progress has rece ntly been made allowing MOCVD growth of p-type material [37, 44-46]. The first p-n junction LED was demonstrated in 1989 by Amano et al. Nichia Chemical Industries was the first to commercially produc e blue LEDs following this breakthrough with very high efficiency and intensity. Nakamura et al first reported curren t-injection GaN LDs in 1996 and subsequently achieved continuous wave (CW) lasing at room temperature. It is interesting to note that these devices were achieved while still in the stage of low quality film

PAGE 26

26 growth. The dislocations did not appear to be efficient non-radiative centers, so while optical devices could be achieved, the defect s do affect reliability [37, 47-50]. Rapid progress has been demonstrated in fiel d effect transistor (FET) structures. This places very strict demands on the growth and processing, but the technology has been developed in the past decade, and present published results indicate that GaN will play a significant role in the development of high temperature, high power and high frequency electronic devices. Further improvements in materials quality will enha nce device operation, t hough processing technology must also be improved for significant developmen t. There are many processing issues, such as poor p-type doping, low resistance and thermally stable ohmic contact development, high temperature requirements for implant activation, poor wet etching processes, low dry etch rates, and other issues. As advances are made, we can expect further improvements in device performance and increasing commercialization [37, 43]. 2.2 High Electron Mobility Transistors High Electron Mobility Transist ors (HEMTs) have em erged in the past twenty years as promising candidates for microwave and high vo ltage applications. HEMTS were originally based on AlGaAs/GaAs, AlGaAs/InGaAs, AlInAs/I nGaAs, and related systems on GaAs or InP substrates. In the 1990s, AlGaN/GaN HEMTs on sa pphire, SiC, or even bulk GaN substrates have demonstrated considerably larger power density (>12 W/mm), and have become contenders for high power amplification and switching applications. The use of wide bandgap semiconductors presents an advantage because, in addition to the increased output power, the temperature tolerance and radiation hardness is also extended. Recently, GaN-based HEMTs have been demonstrated operating at 750 C [43].

PAGE 27

27 The HEMT is also known as MODFET (modulation-doped FET), TEGFET (twodimensional electron gas FET), SDHT (Selectivel y Doped Heterostructure Transistor), or HFET (Heterojunction FET). The defining feature of a HEMT is the channel formed by carriers accumulating along an asymmetric heterojunction. A schematic of a GaN/AlGaN heterostructure is shown in Figure 2-1. There is a large conduction band discontinui ty, causing the electrons to diffuse from the AlGaN (large bandgap) in to the GaN (small bandgap), forming a twodimensional electron gas (2DEG) in the triangular quantum well at the interface. The carrier density is further enhanced in GaN by a strong piezoelectric effect [37, 43, 51]. The conceptual physics of the HEMT were fi rst discussed as earl y as 1969 [54]. The development in the 1970s of MOCVD and MBE techniques allowed for the growth of heterojunctions. Mobility enhancement in an Al GaAs/GaAs heterojunction was demonstrated in 1979, which led to the demonstration of the fi rst HEMT in 1980 [55-57]. Although GaN growth was first studied in the 1960s, suitable substr ates and growth technology did not catch up until the late 1980s. Mobility enhancement in AlGaN/GaN heterojunctions was first demonstrated in 1991, with HEMTs for microwave applications being demonstrated in 1993 [58-59]. Considerable progress was made during the 1990s in the areas of material quality, heterostructure design, and ohmic contact form ation, and the first high power devices were demonstrated in 1997 [60-61]. Khan demonstrated the first AlGaN/GaN HEMT in 1993, reporting a gm of 23 mS/mm, fT of 11 GHz and fmax of 14 GHz [37, 43, 51-53]. To demonstrate the progress in the field since the early developments, some recent statistics will be reported. Due to high 2DEG sheet carrier concentrati on and saturation velocity, power density of nitride HEMT has been repo rted over 30 W/mm [67]. A research group from Japan has recently reported a power output of 370 W at 2 GHz. This demonstrates the superiority

PAGE 28

28 of GaN over GaAs and Si, which are very ma ture technologies, and have maximum power outputs reported in the range of 200-300 W and power density of 2W /mm [62]. This is ideal for W-CDMA base station applications. This type of performance is typically achieved with a large periphery device (in this case 120 fingers of 400 m each), which can be realized in a controlled mass production environment. (power matching issu e will reduce the efficiency) This particular device employed a high performance SiC substrate that was thinned and had thermal vias for improved heat transfer. This type of structure cannot be realized in a cost-effective manner for mass production. Commercially available devices are available though. Nitronex corporation, for example, offers devices operating up to 4 GHz with output powers up to 180 W at a production environment that can be scaled up to a high throughput level by using GaN on Si. The particular advantages of substrate choice will be discussed in Chapter 4 [63]. A high breakdown voltage is required for pr oposed For high power sw itching applications, devices with high breakdown voltage are preferre d, which can be operated at lower current level to reduce the diameter of the cable. A research group at Cornell has recently reported HEMTs with an off-state forward breakdown (gate comple tely pinched off) of 1350 V for a gate-drain distance of 16 m. This employed a high resistivity carbon-doped GaN layer for high breakdown and low leakage. The calculated breakdown field is about 1.6 MV/cm, which is approaching the theoretical limit for this material [64]. Simple calculations can then prove that devices with breakdown voltages in the tens of kV range can be fabricated easily by pushing the gate to drain distance larger. Some work in this area will be discussed in Chapter 4. The third realm that GaN is important for is the high frequency market. Maturation of processing techniques and of device structure has been used to eliminate the parasitic effects seen in early HEMTs that limited RF performan ce. On the processing side, a group in Japan has

PAGE 29

29 increased 2DEG carrier density by depositing a cat alytic CVD process for nitride passivation, a high aluminum content barrier layer, and a short ga te length. This group was able to achieve a 30 nm gate contact length using a Tgate architecture. The reported unity gain cut-off frequency, fT of over 180 GHz is almost double an y previously reported value, due primarily to the short gate length [65]. Other groups, in particular Palacios at UCSB, have reported structure optimization to improve small-signal gain [66]. This work focused in obtaining a high fmax by minimizing parasitic resistances and capacitanc es in the device structure base d on the small signal equivalent circuit (discussed in Appendix B). An InGaN layer was inserted below the AlGaN layer, with a thin InGaN layer between AlGaN and GaN. This serves to create a doub le heterojunction. The conduction band is raised at the InGaN/GaN in terface on the GaN side relative to the InGaN channel. This improves confinement and reduces short-channel effects, creating a higher output resistance. Parasitic capacitances were reduced using field plate technology (discussed below), resulting in a reported fT of near 130 GHz and a record fMax of 230 GHz [66]. The HEMT is a lateral current flow device wi th ohmic source and drain contacts and a Schottky gate contact. A schematic is shown in Figure 2-2. Device isolation is typically achieved by either etching through the AlGaN layer a nd partially into the GaN layer with a Cl2-based plasma to confine the 2DEG, or by implan ting the material with a heavy dose of He+ or N+ ions to create a thick layer with high resistivity. Oh mic metallization is accomplished using lift-off a metal stack that is typically Ti-Al based. Such formulations in clude Ti/Al/Pt/Au and Ti/Al/Ni/Au, with rapid thermal annealing used to alloy the metal at temperatures from 850-950 C for 30-60s. Specific contact resi stances are typically around 1 10-6 -cm2. Gate contacts are a stack of either Ti, Ni, or Pt and Au. The gate can be patterned by op tical lithography down to 1 m dimension and e-beam lithography can be used to pattern gate length down to 100 nm. A

PAGE 30

30 final metal layer is used to route out the de vice contacts to bond pads for probing and bonding to a carrier. This is typically a st ack of either Ti/Au or Ti/Pt/Au and is usually relatively thick, around 3000

PAGE 31

31 Figure 2-1. Energy band structure of AlGaN/GaN heterojunction Figure 2-2. Cross-section of an AlGaN/GaN HEMT

PAGE 32

32 CHAPTER 3 SEMICONDUCTOR PROCESSING 3.1 Overview This chapter is m eant to give an overview of semiconductor material growth and device processing, starting from raw materials all the way up to a functioning circuit board. 3.2 Crystal Growth The first step towards microelectronic devices fabrication is the substrate wafer growth. In a m elt process, the raw materials (ultra high purity Si, for example) are simply melted and resolidified in single crystal form. The process is much more complicated than this, as a considerable amount of control is required to maintain crystal orientation and impurity concentration over the larg e quantities of material grown. A related techno logy is the growth of thin single crystal layers on the single crystal substrates. This process is called epitaxy. This process is particularly important for high-perf ormance devices because it allows for precise control of layer thickness and impurity doping concentration. At th e current technology level, high quality Si and GaAs wafers grown from the melt are widely available, while GaN is grown epitaxially (bulk GaN wafers are available, but they are still grown epita xially and are of poor quality) [69]. 3.2.1 Growth from Melt Silicon is ty pically grown using the Czochralsk i method (CZ growth). The starting material is quartzite, which is a pure form of sand (SiO2). It is refined to form ultra high purity polycrystalline silicon (99.999999999% pure) through various treatments. The first is reduction in a furnace using carbon, shown in reaction 3.1. Th is is followed by a HCl treatment to dissolve the Si, shown in reaction 3.2. The solution is distilled to remove impurities, followed by a

PAGE 33

33 hydrogen reduction to recover the so lid electronic grade silicon. Th is reaction is essentially the reverse of reaction 3.2. The reac tions are given below: [69] SiC (s) + SiO2 (s) Si (s) + SiO (g) + CO (g) (3.1) Si (s) + 3HCl (g) SiHCl3 (g) + H2 (g) (3.2) This high quality poly-Si is then melted in a silica crucible in a vacuum chamber (melting point = 1412 C). A small seed crystal is lowered into the melt and slowly withdrawn. Freezing will occur at the interface, yielding a large single crystal. These are referred to as boules or ingots. Modern silicon ingots can be 400 mm (16 in ) diameter and 2 m long. The pull rates are very slow, on the order of mm/min, to balance the heat transfer. A schematic of a crystal pull system is shown in Figure 3-1 [68]. Control of impurity concentrati on is critical. Magnetic fields are typically added to crystal pullers to prevent the incorporation of ionized im purities. A float-zone process can also be used to refine the ingots and further remove impurities. In this process, RF heating is used to melt a small cross section of a polycrysta lline or single crystal rod. As this coil traverses the length of the rod, impurities will follow the molten region, u ltimately being concentrated at one end of the rod. A schematic of a float zone system is show n in Figure 3-2. This growth method is limited in that the melted region must be able to support the we ight of the entire crystal, so ingots of only a few kilograms can be refined [68]. The major difficulty in producing large co mpound semiconductor ingots from a melt is maintaining the stoichiometry from the melt to th e solid. In all methods, an overpressure of the more volatile component is requi red to prevent prefer ential loss. In th e case of GaAs, the stiochiometric solid is synthesized first by plac ing high purity solid Ga and high purity As in a sealed two-temperature furnace, heating each co mponent to the melting point. The melting point

PAGE 34

34 of As is lower (610 C), so an overpressure of As will form, causing transport of As to the Ga melt (1240 C), forming GaAs, then cooling the melt. From the raw material, ingots can be formed by the Czochralski method and the Brid gman method. The Bridgman process is very similar to the material synthesis process. An overp ressure of As is created in a lateral furnace to prevent decomposition, and the polycry stalline GaAs is placed with a seed in a high temperature zone of the furnace. The furnace is moved so that the material will melt and then freeze following the seed. This process can be compared to a horizontal float-zone process. A schematic is shown in Figure 3-3 [69]. After growth, wafers are prepared from the i ngots for fabrication. The seed and tail ends are cut off, the ingot is ground to a uniform diam eter, and flats are ground along the length of the crystal to define crystal orientat ion and conductivity type. Standard flat orientations are shown in Figure 3-4. A diamond saw is used to slice the crystal into wafers (thickness of 500-700 m), which are then lapped to a flatness of around 2 m, etched to remove contamination, and polished on one or both sides for device fabrication [68]. 3.2.2 Epitaxial Growth Epitaxial growth m ethods are used to grow single-crystal semic onductor layers on singlecrystal substrates. Homoepitaxy is the term used to denote grow th when the substrate and the grown layer are the same, for example growth of a highly doped layer of GaAs on a single crystal bulk GaAs substrate. Heteroepitaxy is used to denote growth of a semiconductor layer on a different substrate, such as GaN on SiC, or AlGaAs on GaAs. The growth systems used must be highly precise. The layers must be of nearly perfect quality and have atomically smooth interfaces. The most common growth methods are chemical vapor deposition (CVD), metal organic CVD (MOCVD), and molecular beam epita xy (MBE). Since epitaxi al growth involves a

PAGE 35

35 chemical reaction, where the melt processes were mainly physical, kinetics are very important. The kinetics of MOCVD growth are a subject or continued res earch at present, and a better kinetic understanding of reaction phenomena will ultimately lead to better material quality and higher performance [69]. MBE uses pure atomic sources, directed as ther mal beams, which react at the surface. The advantage of this method is that precise co ntrol over composition and doping can be achieved, however growth rates are typically slow, less than 1 m/h. Though slow, this system presents the ultimate performance in terms of control, clean liness, and characterizat ion capabilities. MBE systems operate under ultrahigh-vacuum (10-8 Pa), so there is minimal contamination, and characterization equipment can be integrated for in-situ SEM or XPS analysis of the films. High temperature baking steps are used to prepare the surface for MBE by removing native oxide or other adsorbed species. It is a relatively low temperature process (400-900 C), and combined with the low growth rate, many unique structur es can be created, such as the superlattice structure (periodic structure consisting of layers <10 nm thick). A schematic of an MBE reactor is shown in Figure 3-5 [69]. MOCVD has emerged as the choice method for growing GaN, and III-V compounds, due to its high degree of compositiona l control and uniformity, availabi lity of high purity sources, and large scale manufacturing comp atibility. This method is partic ularly advantageous for those materials that do not form stable hydrides or halides. Another advantage is that the organic precursor compounds are volatile at low temperatures, and liquid Ga is not required. In general, the precursors are the trimethyl-III (ex, trimethylgallium (TMGa), trymethalaluminum (TMAl)) and V-H3 (ex, arsine, AsH3, ammonia, NH3). Th e material is typica lly doped with zinc, cadmium, or silicon, so diethylzinc, diethylcadmium, and silane are used as the sources. Safety is

PAGE 36

36 a particular concern when dealing with these systems due to the dangers associated with these precursors (silane is highly flamma ble, arsine is toxic, etc). Th ere are three main components of an MOCVD system: gas delivery, reactor, and exha ust. Typically the organometallic sources are enclosed in bubblers, with hydrogen as the carr ier gas, and the group V source as a compressed gas. Extreme care must be taken in the delivery sy stem to avoid condensation or pyrolysis of the organometallic precursors. A general issue is the gas flow pattern and introduction into the chamber. To obtain atomically smooth interfaces, th e valves must be designed so that there will be no perturbations as flows start and stop. A schematic of an MOCVD reactor is shown in Figure 3-6 [68]. 3.3 Dopant Im purity dopant is used to cha nge the electrical properties of semiconductors. It is an important step in every device fabrication proce ss since it creates the p-n junctions that are the key active parts of the devices. The doping proc ess must be very tightly controlled and reproducible. It is criti cal to control the process both in terms of introducing the impurity concentration to design specifications and contro lling the area where the impurity is introduced. There are two methods of doping semiconductors: th ermal diffusion, which is typically used to create deep wells, and ion implan tation, which is used to create pr ecise shallow junctions or very high impurity concentrations [69]. 3.3.1 Diffusion Diffusion is a funda mental transport phenomenon that involves the re distribution of free material in response to a concentration gradie nt. The material flux is proportional to the concentration, and movement will be away fr om the high concentration region towards the region of lowest concentra tion. There are two basic atomic diffusion mechanisms in semiconductor crystals: movement through vaca ncy lattice sites, and movement through

PAGE 37

37 interstitial sites. The basic govern ing equation is known as Ficks first law of diffusion, shown in equation 3.1 below, with the diffusion coefficient, D, in cm2/s. The diffusion coefficient is temperature dependent and follo ws an Arrhenius law [69]. 2 2 x C D t C (3.1) The differential equation can be solved anal ytically for two importa nt cases: constant source and constant impurity material. In the fo rmer case, the boundary c onditions are: C(x,t=0) = 0, C(t,x=0) = Cs, C(x=inf, t = 0). The solution is shown below in equation 3.2, where erfc is the complimentary error function. In this case, the surface concentration is determined by the solubility of the dopant in th e crystal and the depth profile is determined by time. [69] Dt x erfcCtxCS2 (3.2) For the latter case, the boundary conditions are: C(x=inf, t = 0), and Q = const (total dopant quantity). The solution is a Gaussian function of the form shown in Equation 3.3. In this case, the material will move into the semiconductor, but the surface concentrati on will decrease. [69] Dt xe Dt Q txC42, (3.3) In most semiconductor thermal diffusion proc esses, both of these conditions are used. First, what is called a predeposition step is used to introduce the required quantity of material. This is considered a constant source diffusion proce ss. The source can be either a solid, liquid, or gas. After this step, a second step called a drive-in is used to quite literally drive the dopants into the material and redistribute to th e desired concentra tion profile. [69] Thermal diffusion gives very good reproducibility, and concentr ation profiles can easily be calculated using simulation software such as SUPREM. The problem with thermal diffusion is

PAGE 38

38 that any later diffusion or thermal oxidation step s will cause redistribution of the dopants that are already in the material, so the combined effect of multiple temperature cycles must be considered. It is therefore undesira ble for shallow junctions as they cannot be controlled as well. It is still used to form deep junctions and as an alternative to ion implantation when crystal damage is an issue. For compound semiconductors, ion implantation almost exclusively used. As mentioned in previous sections, high temperat ure processes for long time periods can cause preferential loss of one component, so these mate rials are more compatible with low temperature processing unless capping layers are used [69]. 3.3.2 Ion Implantation Ion implantation uses charged particles to introduce impurities into the material. The ion dose can be controlled by monitoring the ion cu rrent, making it very attractive for shallow or lightly doped junctions. The principle of operatio n is as follows: a heated filament breaks up source gas into charged ions, which are extrac ted from the source. The specific ion to be implanted is selected using a mass analyzer to fi lter by mass to charge ratio. The ions are then accelerated toward the sample at 10-400 kV, where they strike the samples and penetrate to a given depth, with a distribution determined by the ion dose, accelerating volta ge, and type of ion. A schematic of an ion implanto r is shown in Figure 3-7 [68]. There are two stopping mechanisms: electronic stopping, where the attractive power of electrons in the latti ce eventually slows the ion and st ops it, and nuclear stopping, where collisions with nuclei in the la ttice cause energy loss and ultimat ely stopping of the ion. It is possible for ions to travel down channels creat ed by the crystal planes, so most commercial systems tilt the sample at 7-10 degrees relative to the major plane. As the ion travels into the lattice, it will displace nuclei, causing lat tice damage and in the case of high doses, amorphization. When ion implantation is used, a post-implant anneal step is required to

PAGE 39

39 recrystallize the material and re move damage. In some cases, a sacrificial layer of oxide or nitride can be used to protect the surface from significant damage and alter the profile. The implanted profile typically follows a Gaussian distribution and can be calculated using SRIM software code [69]. 3.4 Film Deposition Film formation and patterning constitutes a ba sic device building block. There are actually four general types of th in films: thermal oxide, dielectrics, polysilicon, and metal. Since this work focuses primarily on compound semi conductors, thermal oxidation and polysilicon deposition will not be discussed in detail but will be briefly mentioned here. Thermal oxidation of silicon is well documented and is used to form both thin and thick native oxide layers. Typical growth conditions are either flowing dry oxygen slow growth rate, high quality or flowing wet oxygen (oxygen bubbled through wa ter) fast growth rate but lower quality at temperatures of 900-1200 C. The growth rate is dependent upon thickness, since for thicker films the oxygen must diffuse through the film to reach the silicon surface to react, and equations are available to determin e growth time to reach a given thickness [69]. Polycrystalline silicon is primarily used for the gate electrode in silicon MOSFETs because it has better reliability than alum inum. It is deposited in a CVD reactor by the pyrolysis of silane at 600 C to form silicon and hydrogen. The parameters that affect this process are primarily deposition temperature and the ad dition of any dopant gas [69]. 3.4.1 Chemical Vapor Deposition Chemical vapor deposition (CVD) is typically us ed to deposit dielectric films. There are three variations of the CVD process: atmospheri c pressure CVD, low pressure CVD, and plasma enhanced CVD (PECVD). The considerations in se lecting the process include temperature, rate,

PAGE 40

40 quality, morphology, and electrical properties. Atmospheric CVD is particularly advantageous in that it allows for a continuous feed system since no pumping is required, and it has high deposition rates. The reactant gas is simply in jected to a showerhead above the wafers on a heated chain track. The disadvantage to this system is the poor quality and the particle contamination associated with atmospheric pres sure processes. Low pressure CVD is very popular in the IC business because, while a ba tch system, it allows for high throughput and excellent uniformity. PECVD offers the advantag e of low temperature processing and the ability to integrate with a cluster tool, but uniformity can be poor [68]. Silicon nitride and silicon oxide are the dielectrics primarily used with compound semiconductors. Silicon nitride in particular is important as a passivation and capping material. It oxidizes very slowly, serves as a diffusion barrie r to water and sodium, and has a slower etch rate in buffered HF than silicon dioxide. Sili con dioxide can be depos ited by thermal oxidation on Si, or on any semiconductor by LPCVD or PECVD using silane, NO, and nitrogen. Atmospheric CVD can be used to deposit silicon films containing 412% phosphorous or boron (phosphosilicate or borosilicate glass PSG or BSG) Silicon nitride is very difficult to grow thermally, so LPCVD or PECVD are typically used to react silane and ammonia. It deposits as an amorphous dielectric, and can contain hydrogen. This is especi ally true in PECVD, where the nitride is rarely stoichiometric (Si3N4), and instead is denoted as SiNx, with up to 25% hydrogen. The quality of the film can be determined immediately by comparing the refractive index measured by ellipsometry to that of an ideal stoichiometric film [69]. 3.4.2 Metal Deposition While metal can be deposited using CVD from metalorganic sources, most metal films are formed by physical vapor deposition (PVD) Two common methods are evaporation and sputtering. The mechanism is simple the metal atoms are ejected from the source into a

PAGE 41

41 vacuum, where they land on the samp le of interest. In the evaporation method, a metal crucible is heated either resistively, inductively, or with an electron beam, and emits metal vapor. The vapor travels through the vacuum and la nds on all surfaces in the chambe r. Evaporation typically gives poor step coverage because the at oms are very directional as they approach the surface. This can actually be advantageous in a lift-off process (d escribed below). The deposition rate is monitored by a crystal monitor, which is a thin film osc illator. As particles accumulate, the resonant frequency of the oscillator changes in a very pr edictable manner, and therefore can measure film thickness with angstrom level precision. The de position rate will be proportional to the vapor pressure of the metal at the charge surface a nd the distance from the charge to the sample. A problem with evaporation is that it is difficult to deposit alloys. Alloyed charges cannot be used unless the vapor pressure of both metals is simila r at deposition temperature, otherwise there will be preferential deposition of the more volatile component. A rotating shutter can be used to alternate deposition, but this can be costly si nce multiple charges and heaters are required. Evaporation is used extensively in the co mpound semiconductor industry because of the capabilities to deposit a wide range of metals, and the capabilities to load multiple charges and form stacks of different metals [68]. Sputtering involved the physical bombardment of a target metal sample by ions (typically argon), resulting in the ejection of material into the chamber, which deposits on the surface of the sample. Because the process involves charged partic les, there is the possibility of damage to the semiconductor surface, but heating of both the sa mple and the metal and chamber is less of a concern. It is particularly easy to sputter alloys, and sputtered atoms are less directional, so step coverage is improved. Since the sputtering pr ocess is physical bombardment, it will not be

PAGE 42

42 selective, so sputtering of an alloy should result in a similar composition in the film and the original target. For these reasons, sputtering is us ed extensively in the silicon IC industry [68]. There are two methods of patterning metal fi lms additive and subtractive. In the subtractive process, the film is deposited first, th en patterned and etched. This is widely used for aluminum patterning, since a selective etchan t is readily available (PAN etch = phosphoric, acetic, nitric acid solution). The additive method, or lift-off method, is used for patterning metal stacks used in compound semiconductor devices. In this process, the pattern is formed in photoresist, then the metal is deposited over this film, then the film is removed, leaving behind the metal on the substrate. The yield is usually lower for this process, particularly for small features [69]. 3.5 Lithography Lithography is the most complicated, critic al, and expensive process in semiconductor fabrication. It is the process for forming pa tterns on the semiconductor substrates for film deposition, film removal, or doping. It is the prim ary driving force for growth in the industry, as it defines the minimum feature size. Improvements in lithographic techniques to make smaller features will improve chip performance more than any other technology described in this chapter. This step is repeated for almost ev ery layer in the device structure, and therefore represents about 35% of IC ma nufacturing cost. For these reas ons, it is the most significant economic factor in IC manufacturing [68]. 3.5.1 Photoresists Photoresists are polymers that are sensitive to radiation at a gi ven wavelength. When selectively irradiated through a ma sk, a chemical change will occur that makes them either more soluble or less soluble in what is known as a developer solution than the surrounding matrix that has not been irradiated. In this manner, a patt ern can be transferred from the mask to the

PAGE 43

43 semiconductor surface. This will selectively open or protect areas for doping, etching, and film deposition. There are two basic types of resists: positive tone and negative tone. In the positive resist case, the area that is e xposed through the mask is removed, therefore duplicating the mask pattern on the surface. In the negative resist case, the area that is not exposed is removed, leaving the area that was exposed. This creates the reve rse image from the orig inal mask. A pattern formed by positive and negative resists is shown in Figure 3-8. Both have applications in industry, some examples being as follows: posi tive resists are used to open areas for doping, dielectric window etching, and metal lift-off, and negative resists used to mask transmission lines in a metal film while the rest is etched off. Positive resists tend to offer better resolution and fewer environmental hazards, and therefore ar e used almost exclusively in industry [68]. Photoresists are applied by spin coating. The film thickness is a function of both the spin speed and the viscosity. After spin coating, the re sists are baked on either a hot plate or an oven to remove the solvent. Resists are identified by a 4-digit number. The first two digits are the series, and the last two digits identify the thickness. For exampl e, Shipley S-1808 is a 1800 series resist that is 0.8 m thick at 5000 RPM, and S-1818 is a 1800 series resist that is 1.8 m thick at 5000 RPM, while S-1045 is a 1000 series resist, and is 4.5 m thick at 5000 RPM. Optical resists typically cons ist of three components: a po lymer resin, a photoactive compound (PAC), and a solvent to control the mechanical properties. The typical positive resist uses novolac (N) as the resin and xylene and acetates as the solvent. Th e solvent is primarily used to adjust the viscosity for coating, then baked off before exposure. The most commonly used PAC is diazoquinone (DQ). Before exposure, th e molecule acts as an inhibitor, slowing the dissolution rate of the N resin, which is normally very soluble in aqueous solutions. Exposure to UV light causes the loss of N2 from the carbon ring. The active site is stabilized by a Wolff

PAGE 44

44 rearrangement. In the presence of water, the structure rearranges once more to form a carboxylic acid. The DQ molecule and the subsequent r eactions are shown in Figure 3-9. The carboxylic acid will readily dissolve in a base solution. The typical positive resist developing solutions are KOH-based (example: Clariant AZ-400K) or tetramethylammonium hydroxide (TMAH: (CH3)4NH4OH)-based (example: Microchem MF-321) solutions dilute d with water and surfactants added. The positive resists have bett er resolution because the unexposed areas are unaffected by the developer and the novolac resin is very resist ant to chemical attack. A disadvantage is that for some semiconductors, such as Si, there can be adhesion trouble. To get around this problem, the wafers are primed with hexamethyldisilazane (HMDS), which is an adhesion promoter [68]. Negative resists work by cross-linking unde r UV exposure, making the exposed areas less soluble. As such, there is a wider variety of negative resist resins, but one class is azidesensitized rubbers such as cyclized polyisoprene. A post exposure bake is usually required to complete the cross-linking reaction, and the deve loper is typically an organic solvent. The advantage of negative resists is the short exposure times required to promote cross-linking, and good adhesion to substrates. The disadvantage is that the resist swells during development, which can cause broadening and distortion of lines. Therefore negative resi sts are not suitable for high resolution applications [69]. For metal lift-off processes, an overhanging stru cture is desired to create a discontinuous film that can be removed. There are several op tions available for this. The one primarily used involves soaking the resist in either toluene or cholorobenzene. Thes e molecules will be absorbed by the top layer of resist and bond w ith some of the PAG sites, decreasing the sensitivity of the resist to light. The bottom part of the resist is unaffected. After exposure, the

PAGE 45

45 top layer will develop more slowly, creating an overhang. Other new technology involves the development of special lift-off resists (LOR), wh ich are used in a dual layer system. LOR is not photoactive, and is more soluble in the deve loper than the regular photoresist. During development, after the developer goes through the positive resist layer, it will dissolve the LOR isotropically, creating the overhanging structure [68]. A third class of resists that mu st be discussed are electron-be am resists. Nonoptical resists are typically long chain polymers. Irradiation of these polymer chai ns by either electron beam or X-ray radiation will cause either chain scission or cross-linking, creating positive and negative tone images, respectively. Most nonoptical lit hography is positive tone, and the most commonly used resist is polymethyl methacrylate (PMMA). Chain scission and cross-linking occur, but the rate of chain scission is much higher, therefore it creates a positive image. During development, the longer chains will be more soluble in the de veloper, and the chains that have been cut will develop faster. The developer therefore is just an organic solvent, typically a solution of methylisobutyl ketone (MIBK) and isopropanol (IPA). Like optical resists, the PMMA is dissolved in an organic solvent, such as anisole, for storage and coating. The solvent is baked off before exposure. PMMA itself has only fair sensitivity to the electron beam. For a lift-off process, a layer of a more sensitive polymer, such as a PMMA-MAA copolymer (MAA = methyl acryl acrylate) can be applied first and baked, th en coated with PMMA alone. This will create an overhanging structure that is suitable for lift-off, shown in Figure 3-10 [68]. 3.5.2 Optical Lithography Optical lithography is the true workhorse of the industry for pattern generation. A typical CMOS process can involve 15-20 lithography steps, almost all of which are achieved by optical lithography. In this method, the phot oresist is patterned using optical radiation, typically in the UV range using a mercury arc lamp. Some standa rd wavelengths are the g-line (465 nm) and I-

PAGE 46

46 line (365 nm). Resolution is limited by diffraction to about the wavelength of light used. Efforts are underway to reduce the wavele ngth by using xenon as a fill gas in the lamp or using excimer lasers. A simple optical lithography sy stem is shown in Figure 3-11 [68]. The basic feature characteristic to optical lithography is exposure through a photomask. The mask is a quartz plate with a chrome metal la yer on one side that is patterned. The mask is designed using computer software such as SPICE or L-Edit, then transferred to the mask using a high resolution lithography system such as e-beam lithography, or patterned using optical lithography using a transparency page from a hi gh resolution laser prin ter as the mask. The master mask can then be used to transfer th e pattern to the semiconductor substrates using photoresists as desc ribed above [68]. The basic components of an optical lithogr aphy system will now be described. As mentioned previously, the light source is a high-pr essure mercury arc lamp. There is a series of reflectors inside the lamp housing to collect light and direct it towards the wafer. There is then a series of apertures and homogenizers to ensure the beam is uniform over the wafer area, and collimate and shape the radiation. Finally, there is a filter to select the appropriate wavelength from the emission spectrum, and then there is a s hutter to control the exposure of the wafer [68]. There are multiple exposure methods, which will ultimately affect the resolution of the pattern and the lifetime of the mask The method used in research is referred to as hard contact. This means that the mask is pressed against the substrate. This close contact will achieve the highest resolution, but th ere will be considerable wear and da mage to the mask master. This is required to achieve submicron resolution. A seco nd method, referred to as proximity exposure, leaves a gap of anywhere form a few to tens of microns between the mask and the wafer. While this significantly reduces wear on the mask, it ha s poor resolution. Projec tion printers are used

PAGE 47

47 almost exclusively in industry. These systems us e a series of lenses to demagnify the mask pattern and focus it on the wafer. This met hod can achieve the high resolution of contact lithography with minimal mask wear. The first systems were simply 1:1 projectors. These systems have since been replaced with step and repeat systems. Rather than simply project the mask pattern onto the sample, a reticle containing just one cell or chip is imaged. A 10:1 or 5:1 reduction lens is used to demagnify the mask on to the wafer, and then mechanically stepped to the next cell. Since there is a demagnification factor associated with this process, ultra high resolution can be achieved. A state of the art step per system can pattern <0.5 um features over a 2.5 cm chip and accurately step the chips over a 40 cm wafer. A particular advantage of these systems is that they automatically adjust ali gnment and focus at each site, so wafer bowing can be compensated [68]. 3.5.3 Electron Beam Lithography To further reduce feature size, efforts have b een made to use short wavelength, non-optical sources, such as X-rays and electron beams. Electron beam lithography (EBL) systems are direct-write systems. These systems are essentially a SEM system with a built in pattern generator and stage translation components. They consist of an electron source (typically a W filament), extractor cage to accele rate the electrons toward the wafer, and series of apertures and condenser lenses to focus the beam. Deflection coils are places immediately before the pole piece to direct the beam over a given area. The e xposure is performed by a combination of beam deflection and stage translation, similar to a stepper system. The st age steps to one cell, called a stitch field, and the beam is deflected over that area to form the pattern. The stage then moves to the next field, and the beam is deflected, and so on. The stitch field is a square from 100-1000 m per side. Another consideration in the exposur e process is the proxim ity effect. Electrons

PAGE 48

48 penetrate the wafer and are scattered, and can re -emerge from the wafer. This is in fact one method of detection in an SEM. These backscattered electrons will affect the irradiation in neighboring areas of the pattern. It is possible to simulate thes e effects and correct the dose on regions that are close enough to ex perience this effect. A schematic of the electron beam system is shown in Figure 3-12 [68]. The major disadvantage of the EBL systems is that the throughput is extremely low since it is a serial direct-write process. Exposure of just one wafer can ta ke an hour or more because the stage has to step and expose every stitch field. For this reason, they are excellent for research level fabrication where small features are required (minimum feature size can be <50 um), but impractical for mass production. This represents one of the fundamental challenges for the industry over the next 10 years how to push th e limits of optical lithography to this resolution to maintain the throughput [68]. 3.6 Etching Etching is used to remove material through a mask patterned by lithography. There are two basic types of etched: wet and dry, both with th eir own advantages that will be discussed. The differences between wet and dry etching in term s of pattern transfer are shown in Figure 3-13 [69]. 3.6.1 Wet Etching Wet etching is a purely chemical attack of th e material. It is primarily used to etch dielectrics such as SiO2 and SiNx. There are three basic steps in the reaction mechanism diffusion of etchant to the surface, chemical reac tion, and diffusion away form the surface. That being said, there are two types of wet etch diffusion limited and reaction limited. The majority of wet etched are diffusion limited, so an acid spray can be used to continuously supply fresh etchant to the surface to enhance the etch rate and the uniformity. Sinc e both the reaction and

PAGE 49

49 diffusion are temperature dependent, precise control of the bath is required for reproducible etch rates [69]. Wet etching of SiO2 and SiNx is performed in a buffered hydr ofluoric acid solution, known as buffered oxide etch (BOE). Typical ratios are 6:1, 10:1, and 20:1 water to HF. The solution is buffered with NH4F to maintain a fresh supply of HF. The solution is selectiv e to silicon dioxide over silicon 100:1, and etch rate s are around 1000 /min. The reactions are shown below. SiO2 + 6HF H2 + SiF6 + H2O NH4F NH3 + HF Silicon nitride can also be etched in BOE, at a rate of about 1/3 that of oxide, depending on the quality of the nitride. PECVD nitride can be etched at nearly the same ra te as oxide if there is a significant amount of hydrogen incorporation. Silicon alone can be etched by a solution of HNO3 and HF, and selectively etched along the (100) plane over the (111) plane by KOH. Doping-selective etches are also av ailable, and a solution of HF/HNO3/CH3COOH will selectively etch heavily doped laye rs of Si. Defects etch preferen tially, so wet etching can be used to stain defects and make them visible in even an optical microsc ope to calculate defect density. Wet etches for GaAs are based on mixtures of peroxide and acid, such as H2SO4/H2O2/H2O. Similar mixtures using NH4OH or H3PO4 instead of H2SO4 can be used to selectively etch AlGaAs. [68] There is no known wet etchant for GaN except molten KOH or NaOH [70]. Wet etching is not widely used in industry any more, mainly due to the poor control, potential for contamination, and isotropy. The main problem is that the proce ss is isotropic, so it will undercut the resist layer as it etches vertically. This results in a loss of resolution. The main advantage is that it does not damage the surface a nd it can be selective, so it can be used for

PAGE 50

50 applications where overetch can be tolerated, but surface damage cannot, such as in opening dielectric windows for optical de vices like photode tectors [69]. 3.6.2 Dry Etching To achieve high resolution pattern transfer, a controlled, anisotropic etch is needed. This is where dry etching becomes relevant. There are three main types of dry etching: plasma etching, reactive ion etching (RIE), and high density plasma etching (HDP). Plasma is either a fully or partially ionized gas composed of equal amount s of positive and negatively charged ions and some neutral molecules. It is formed when a suffi ciently high electric field is applied to a gas to cause break down and ionization. In the basic plasma etch, the reaction proceeds as follows: First, the positive ions and free radicals are ge nerated in the plasma, then transported to the surface of the wafer, followed by adsorption on th e surface, reaction, desorption of volatile products, diffusion into the bulk, and removal through the vacuum system [69]. Plasma chemistry is very complex, and minor changes to the gas mixture can have significant effects on the etch process. Etching can have both chemical and physical forms. There is both a chemical reaction taking place to form volatile products that can be desorbed and removed, but it is also possible to have physical sputtering occu rring as well. In general, to obtain a good plasma etch, producing a reasonable etch rate and smooth surface, one must balance the chemical and physical etch processes. The earliest plasma etch systems, introduced in the 1970s, relied on high pressure, low power pl asmas. In this process, the etching depends heavily on the chemistry of the plasma, since it is primarily a chemical process. One of the first applications was photoresist stripping in an oxyg en plasma. Ion milling is the opposite of plasma etching in that it is a purely physical etch [68]. Reactive ion etching was developed to satisfy th e need for anisotropic and selective etch. A RIE system is a parallel plate reactor operated at a gas pressure of 100-1000 mTorr. This reaction

PAGE 51

51 is essentially chemically enhanced physical et ching, and thus physical damage can become a problem. Inductively coupled plasmas (ICP) and other HDP etch systems were developed to improve control over plasma parameters. The adva ntage of these systems is that the biasing is decoupled, so the wafer is powered independently of the plasma source, so there is less wafer damage. Also, these systems operate at a lower pressure (1-10 mTorr), but high plasma density, so they have high efficiency. These systems provide better critical dimensi on control, faster etch rates, and improved selectivity [68]. A few common plasma etch gases are: O2, CH4, H2, Cl2, BCl3, Ar, and SF6. A typical RIE/ICP system is equipped with this set of gases, and the gases can be combined. It is as much art as science to understand the pl asma chemistry and come up with a gas recipe that will etch a given material well. Typically a matrix of gas flows, pressures, and biases must be investigated to determine the optimum parameters. This set of gases can be used to etch silicon, compound semiconductors, dielectrics, and metals. [68-70] A few typical recipes wi ll be listed in Appendix C. 3.7 Packaging The packaging process involves dicing the wafer, picking the good devices, and bonding them on chip carriers that can be mounted on circu it boards. This step serves to both protect the chip and interface it with the outside world. A single chip package (SCP) supports a single electronic device, whether it be an integrated circuit (IC), or a simple FET. If multiple active devices are contained in one p ackage, it is referred to as a multi-chip module (MCM). The primary function of the package is to enable the device to perform its desired function reliably through the design lifetime of the system [21]. The first package level is the chip to carrier. Th e wafer carrier serves to decrease wire trace density from the IC level fine pitch of less than 100 um to the typical board pitch of greater than

PAGE 52

52 200 m. It also serves to encapsulate the chip to protect it from envir onmental and mechanical stress, and it acts as a heat sink for the chip, pr oviding some level of thermal management. There are three main first-level packaging technique s in the industry: wirebond, tape automated bond (TAB), and flip-chip bond. Wirebonding and flip-c hip bonding will be discussed in detail. TAB is used in the LCD industry and involves mounti ng and interconnecting IC on metallized flexible polymer tape. A schematic of the three packaging schemes is shown in Figure 3-14 [21]. 3.7.1 Wire Bonding Wire bonding is an interconnection technique th at involves attaching a fine wire between the I/O pads on the chip and the associated pins on the package. It is widely used in industry and in the early 2000s accounted for over 90% of all chip level interconnections. Typically a gold wire 25 m thick is bonded ultrasonical ly between the chip pad and the carrier pin. There are two options for forming the bond: ball bonding and wedge bonding [21]. Ball bonding is a thermosonic welding techni que. The primary adva ntage of ball bonding is that the bond formed is circular, so the wire can be pulled to the ne xt bond in any direction. The bond is formed under a controlled for ce (<100 g), modest temperatures (150-200 C), and ultrasonic excitation (60-120 kHz). Th e wire is then paid out from a spool to form a loop, where the next bond is made. This bond has to be a wedge bond or line fracture since no ball can be formed. After wire fracture, an electronic flam e off (EFO) fires to form the next ball bond. Process times can be <20 ms per bond cycle [21]. Wedge bonding is advantageous in fine pitc h situations, since the bond wire is only deformed 20-30% beyond the original diameter compared to 60-80% for ball bonds. It also demonstrates a higher yield, but there is a ma jor drawback: the first bond must be aligned perfectly along the axis between the first and second bonds. Otherwise, the first bond can pull off

PAGE 53

53 if it is moved in a direction off-axis. The bond is formed at a lower temperature (50-125 C), a controlled wedge bonding force, and ultrasonic ex citation. This method essentially applies a force to pinch the wire and the ultrasonic excitation forms the bond [21]. The advantages of wire bonding are the extrem ely high reliability, ease of automation, and high level of industry infrastructure support. Current automate d equipment can bond an entire chip in a fraction of a second. The disadvantages are that it is a serial process so it is an inherently slower bonding process relative to flip-chip, long bond lengths due to the loop of the wire can degrade chip performance, especially for RF packages, there is a larger chip footprint required again due to the wire loop and routing outside of the chip, and there is potential for wire movement and shorting during encapsulation [21]. 3.7.2 Flip-Chip Bonding Flip-chip bonding involves quite literally the face-down direct bonding of the chip pads to the carrier. This method uses sold er bump (sphere) interconnections to form both the mechanical and electrical bond, in contrast to wire bonding wher e an epoxy is used to mount the chip to the carrier and then the gold wires fo rm the electrical connection, with no contribution to mechanical strength. This bonding method is one of the most significant packaging developments to improve cost, reliability, and productivity. There are cons iderably more processing issues with this method, particularly compatibility between chip pad metallization and solder material and I/O routing issues. Solder bumps ar e deposited using optical lithography to define the bump pattern and electroplating or evaporation to deposit the metal. After patterning, the bumps are metal posts, and are then reflowed to homogenize the structure, since phase separation can occur during electroplating [21].

PAGE 54

54 The bond is formed by contacting the solder bumps on the chip to the bond pads on the carrier, and heating to the refl ow temperature. The solder bu mps melt and react with the bond pads, and then the structure is cooled. This is refe rred to as a controlled co llapse interconnection. With the chip facing down, alignment must be performed while the chips are separated. This is accomplished using either a camera with objecti ves pointing up and down inserted between the components, or with a series of mirrors in an optical microscope. After alignment, the chips are pressed together in a controlled manner, with temperature and force applied. Typical materials are the standard tin/lead (Sn/Pb) eutectic solder fo r industrial applications. Al so of interest due to the low melting point is indium (In) and for MEMS applications where no electrical connection is required, photoresist, such as SU-8, can be used [21]. 3.7.3 Encapsulation Encapsulation of chips significantly enhances re liability of the package, both for wire bonds and flip-chip bonds. It provides both chemical and mechanical protection for the IC. For flip-chip bonded packages, it si gnificantly reduces the strain on the solder bumps, greatly enhancing reliability of the structure. In the ca se of wire bonded packages it will seal the wires in place to prevent them from moving or breaking. Mo isture is a major contributing factor in packaging failures, so protection from moisture and other chemical attack is critical in electronic packaging [21]. An ideal encapsulant will protect the chip from moisture and chemicals. Mechanical considerations must be kept in mind. It s hould demonstrate good stre ss-strain behavior. The coefficient of thermal expansion (CTE) must be cl ose to that of the chip to ensure low stress between chip and underfill. The most common en capsulants fall into four categories: epoxy, cyanate ester, silicone, and ur ethane. Epoxy is by far the most common encapsulation material. The polymerization process is fast, clean, produc es no volatiles, and is very well studies and

PAGE 55

55 established. The other materials are high performa nce materials for specialized applications. The encapsulation process is typically a simple transfer molding pr ocess for wire-bonded packages. Encapsulation of flip-chip packag es is accomplished by application around the edge of the chip, and then a capillary force will pull the mate rial under the chip to fill the void [21]. 3.8 Board-Level Operations A printed wiring board (PWB), also called pr inted circuit board (PCB), is typically a composite of organic and inorganic materials. It contains internal and external wiring, power, and heat sinking. It carries all components of an entire system, so it can also be called a system board or motherboard. It is made of an insula ting structure with copper wires both on the surface and in internal layers. Electrica l components are soldered on the boa rd to form both the electrical and mechanical connection. FR-4 is the mo st commonly used composite board material, constructed of multiple plies of epoxy-resin impregnated onto a woven glass cloth. A cross section of a typical PCB is shown in Figure 3-15 [21]. 3.8.1 Printed Wiring Board Fabrication There are two methods of board fabrication: sequential build-up and pa rallel build-up. The sequential process follows the device chip fabrica tion process. A series of layers of dielectrics and metals are sequentially deposited and patt erned. The parallel build-up process involves fabricating each think film layer separately, then laminating all layers together. The processes used for PCB fabrication are primarily lithograp hy and etching. Most pro cessing is subtractive, with Cu metal deposited on the board from 10-50 m thick, followed by lithography with a negative resist and acid etching. An additive proc ess can be used to build thick films, however instead of a lift-off process, a combination of electroless and electr odeposition is used. The electroless plated copper can be used to select ively grow a copper film on the PCB material where there was no copper before. It will se lectively grow on windows opened up in the

PAGE 56

56 photoresist, but not on the photores ist itself. This is useful for growing a seed layer 1-5 m thick. Electrodeposition can be used to build up this laye r to be tens of microns thick. Vias can be formed either by etching or physic al drilling of the material. Pa ssive components can either be fabricated on the board (inductors from a spiral line, capacitors as two pl ate between a dielectric layer, and resistors as a line of high resistivity material) or solder ed on in the assembly step [21]. 3.8.2 Assembly and Testing PWB assembly is the process is attaching components on the board. Early assembly used through holes to the solder joints on the opposite side of the boar d from the components. By the early 2000s, nearly 80% of assembly was perfor med using surface mount assembly (SMA). With this technology, all components are mounted and so ldered on one surface. The main advantage of this is a significant reducti on in component area, so higher de nsity boards can be fabricated, even putting components on both sides of th e board. A schematic showing through hole and surface mount technologies is shown in Figure 3-16 [21]. The main processes for SMA are solder printi ng, component assembly, and reflow. Solder printing is accomplished using stencil printing. In this process, a squeegee is used to force a bead of solder through a stencil and onto the board. The solder paste usually consists of solder spheres (usually Sn/Pb 63/37), flux (to promote adhesion and remove oxides), and solvent (to adjust viscosity). Chip placement is automated with a camera for alignment, and a vacuum nozzle picks up components and drops them in their locations. For IC chips, a high degree of alignment is necessary due to the relatively small pitch. For pa ssive components, a chip shooter is used, which has high throughput and modest accuracy. This is sufficient since components will self-align to the center of the bond pads during reflow. The refl ow step is performed in a multiple zone oven on a moving conveyor. The typical zones include pre-heat, soak, re flow, and cool. There can be

PAGE 57

57 anywhere from 3 to 9 heating zones to finely control temperature and usually a couple cooling zones. The preheat step is to remove the solven t from the solder. The soak phase is to equalize the temperature of all components. The reflow phase is to melt the solder and form the electrical joint, and the cooling phase allows for the solder to solidify before exiting the oven. Control of the temperature profile is critical because this determines the yield, mechanical and electrical integrity, and reliability [21]. Through-hole soldering uses a wave-soldering machine. This type of system solders all components in one pass by passing the board with all components on it across a wave of molten solder where the bottom of the boa rd barely touches the solder. The solder will wet areas of exposed metal, but not areas cove red by the top layer of the board, called a solder mask [21]. There is a final test after assembly to ensure functionality. There is actually a test step before each package level. IC testing is performe d by the chip manufacturer. This is a simple sort that tests basic functionality. G ood chips are packaged and tested again after packaging. PCBs are tested for opens and shorts inside the board then again after assembly for solder joint continuity. At the system level, more extensive functional testing is performed to make sure the board carries out its intended operation [21]. 3.9 Reliability The reliability of a device is the probability that a system will operate within acceptable limits for a given period of time. Since the design lifetime is usually on the order of years to tens of years, accelerated testing must be performed to determine failu re rates. Device failure can be caused by thermomechanical, electrical, and chemi cal mechanisms. An ideal failure distribution follows a bathtub curve, shown in Figure 3-17. This implies that at the beginning of the life cycle, there will be a high failure rate, referred to as infant mort ality, in which devices fail due to defects introduced in manufacturing. After the ini tial failure period, there will be a long stable

PAGE 58

58 operation period with few failures. Finally, there will be another period of high failure rate as the devices reach the end of their de sign life cycle. This curve is re ferred to as the failure density function, which is the time derivative of the cumula tive failure function, wh ich is the fraction of a group of devices that has fa iled ay any given time [21]. The main chemical failure mechanism is co rrosion. A physical failure mechanism may be due to creep or other physical aging of polymer encapsulants. Electrical failure can be caused by electromigration. Most failures on the system level are thermomechanical failures. They are caused by packaging stress and thermal expansion co efficient mismatch. This can be manifested in the form of chip cracking, solder joint fail ure, and delamination. Accelerated testing can be performed to investigate all of these failure m echanisms. Some accelerated tests include: thermal cycling or thermal shock (cycle from -20 C to 120 C, for example), thermal soaking (baking for long time), mechanical vibration or applied force (she ar applied to solder joints), high voltage or power cycling, and high humidity (environm ental chamber at say 90% humidity and 90 C) [21]. A set of standards has been developed by severa l different groups, each with different acceptance criteria. The most stringent reliability standard s are the US military standards. A more thorough discussion of electrical reliability will follow in Chapter 5.

PAGE 59

59 Figure 3-1. A Czochralski growth system

PAGE 60

60 Figure 3-2. A float-z one refining system Figure 3-3. A Bridgman furnace

PAGE 61

61 Figure 3-4. Standard flat orientations for Silicon Figure 3-5. A MBE reactor

PAGE 62

62 Figure 3-6. A MOCVD reactor Figure 3-7. An ion implantor

PAGE 63

63 Figure 3-8. Positive and negative photoresist patterning

PAGE 64

64 Figure 3-9. Photolysis of DQ upon UV exposure Figure 3-10. SEM image of a dual layer PMMA -MAA/PMMA resist structure for lift-off

PAGE 65

65 Figure 3-11. An optical lithography system Figure 3-12. An electron beam lithography system

PAGE 66

66 Figure 3-13. Comparison of pattern transfer via wet and dry etching Figure 3-14. Schematic of packaging schemes

PAGE 67

67 Figure 3-15. Cross section of a PCB Figure 3-16. Demonstration of throu gh-hole and surface mount technology

PAGE 68

68 Figure 3-17. Failure density function

PAGE 69

69 CHAPTER 4 HIGH ELECTRON MOBILITY TRANSISTORS 4.1 Overview Current work has focused on fabrication of AlGaN/GaN HEMTs on various substrates. As discussed previously, the SiC substrate is of v ital importance when heat transfer and high power operation is critical, such as in an application integrating Si-based el ectronics with GaN. A particularly novel substrate has been demonstrated by Picogiga. It uses a bulk poly-SiC wafer, with a Si layer bonded on the surface, and the GaN grown on the Si. This could potentially integrate the low cost Si appro ach as a growth template and th e high thermal conductivity of SiC. Devices have been fabricated on both substrates, and performance results will be compared in this chapter. 4.2 Silicon Carbide Substrates SiC substrates are ideal for GaN device fabrication because they offer a thermal conductivity that is more than an order of magnitude higher th an sapphire (nearly 400 W/cm-K vs 20 W/cm-K), and a much better lattice match to GaN (3.5% vs 13%). The major drawback is cost. SiC wafers are by far the most expensive (h undreds of dollars per wafer compared to tens of dollars per sapphire wafer), making the devi ces cost-prohibitive except in high performance markets where extremely high thermal conductivity is a must. Devices fabricated in SiC typically do produce the best performance, and to that end, all devices presented in this section will be compared to SiC devices as a reference [71, 74]. The first section of this chapter will focus on HEMTs fabricated on SiC substrates. As mentioned previously, laser drilling presents an interesting opportuni ty for novel processing schemes and in particular 3-D integration. Th erefore to study the eff ect of laser drilling on devices, HEMTs were fabricated in a SiC reference sample and a laser drilled SiC sample. For

PAGE 70

70 more details about the laser drilling process, please refer to Chapter 8 of this work. HEMT devices were fabricated on a reference sample as well as a laser drilled sample, both from the same wafer, which had an AlGaN/GaN he terostructure grown by MOCVD on a 4H-SiC substrate by SVT Associates. The laser drilling was performed before HEMT fabrication on an as-received sample. The drilling test, procedures, and characterization is described in Chapter 8. The first step in processing the HEMT is a dr y mesa etch for device isolation. This is performed in an ICP etch for 50 s at 40W RF power and 150W ICP power, using Cl2/Ar (10 /5 sccm) as the etching gas and a chamber pressure of 2 mTorr (etch rate under these conditions ~ 1000 /min). STR-1045 photoresist was used to pr otect the mesas. A standard ohmic metal scheme (Ti/Al/Pt/Au, 200/600/400/800 ) was used in a lift-off process using S-1808 photoresist to define the ohmic contacts, wh ich were then annealed at 900 C in N2 for 1 minute using rapid thermal annealing (RTA). The ohmic contacts were studied to determine the preliminary effect of laser drilling. The results of this characterization are presented in Chapter 8. The final metal pattern, which supplies the pads for testing and makes the connections to the source, drain, and gate, was patterned and deposited again in a lift-off process. This metal scheme was Ti/Pt/Au (200/300/3000 ). The gates were written using e-beam lithography. A special bilayer resist system (7% PMMA-MAA/4% PMMA) was used. The different polymers have different sensitivities to the elec tron beam, so upon development, there will be an overhanging structure to aid in the lift-off of the Schottky metal (P t/Au 300/1500 ). E-beam lithography allows for selective-area gate deposition, as well as submicron gates. These were written to be 250 nm gate length. A device cross-section is shown in Figure 4-1 [71, 74]. Device testing demonstrated nearly identical performance in DC and RF modes on both an undrilled device and on a device fabr icated immediately next to a through-hole drilled at 355 nm.

PAGE 71

71 An optical microscope picture is shown in Figure 4-2, with VDS-IDS, VGS-IDS, VGS-Gm, f vs H21, and f vs U plots shown in Figures 4-3 to 4-5. A detailed description of electronic testing and figures of merit is includ ed in Appendix B [71, 74]. 4.3 Silicon On Polycrystalline Silicon Carbide Substrates SopSiC substrates (Silicon on poly-SiC) are ve ry promising due to the combination of the low-cost approach of silicon and the high therma l conductivity of SiC. While poly-SiC does have a lower thermal conductivity (about 250 W/cm-K) than single crystal SiC, it is still higher than sapphire, GaN, or Si (both 120 W/cm-K). An intriguing approach involves the use of the Smart CutTM technology already demonstrated in large volume industry (SOI substrates) to produce SopSiC substrates [72-73]. The AlGaN/GaN HEMT wafer is then grown on the silicon on polySiC (SopSiC) composite substrate. The thermal conductivity of the composite SopSiC substrate is comparable to that of polycrystalline SiC and superior to Si. For power amplifier applications, ability to extract heat and the allowed thermal budget of operation is critical. High thermal conductivity substrates may make it possible to design a power amplifier without an auxiliary cooling system and lead to further cost savings (materials/operation) along with weight/volume reduction. Thermal simulations of nitrid e HEMTs operating at power densities of 5 W/mm grown on such substrates indicate ju nction temperatures fairly similar to those of devices on polycrystalline SiC substrates [72-73]. Lastly, good rf quality factors for passive elements such as capacitors, inductors, a nd transmission lines may be more easily integrated when using high resistivity substrates [73]. The starting substrates were 4 in. diamet er high resistivity (111) Si and conducting polycrystalline SiC. The Si was initially oxidized and implanted with a high dose of H+ ions, cleaned, and bonded to the SiC wafer. Removal of the Si by Smart Cut splitting was followed by reclaiming of the top Si wafer and surface pr eparation of the SopSiC wafer. The SopSiC

PAGE 72

72 wafer consists of the SiC substrate, followed by 0.1 m of thermal SiO2, and 0.2 m of (111) Si. The Molecular Beam Epitaxy (MBE) growth in the rf N2 plasma-assisted mode of the HEMT structure began with deposition of an ~0.5 m thick GaN buffer, 1.4 m additional buffer of Al0.25Ga0.75 N, and was followed by the active regions of a 100 thick GaN channel, then a GaN/AlN 10 period superlattice w ith total thickness 32 nm and average Al content of 32 at.%, a Al0.25Ga0.75N barrier (50 ) and undoped 20 GaN cap la yer. The sheet resistance of the HEMT was 456 / sq, with a sheet carrier density of 7 1012 cm2 and electron mobility at 300 K of 1400 cm2 /V-s. A cross-sectional transmission electron microscopy (TEM) micrograph of the entire structure is shown in Figure 4-6. The stru cture shows single crystal nature and clean, sharp interfaces, while there is the usual density (~5 09 cm2) of threading dislocations originating from the lattice mismatch at the hetero-int erface. The ~500 m thick polycrystalline SiC provides electrical isolation, h eat dissipation and mechanical st rength. This is followed by ~200 nm of SiO2 which establishes the qu ality of the wafer bonding and finally ~470 nm of high resistivity (111) Si which provi des a suitable surface for growth of hexagonal GaN and also provides electrical isolation. Device fabrication followed the st eps discussed in the previous section for SiC HEMTs, with the exception be ing that the gate length was written as 0.5 m in the e-beam lithography system [73]. Figure 4-7 shows a typical IDS-VDS characteristics from a HEMT. At top, the device 2x100 m2 gates with 2 m gate-drain spacing. All of the devices showed excellent gate-drain breakdown voltages in the range 40-350V for gate -drain spacing of 2-125 m, where the reverse breakdown voltage was defined as the voltage at which the reverse current density was 1A-cm2. These correspond to breakdown fi elds in the range 3.2-20 4 V/cm and are comparable to HEMTs on other substrates with the same sheet carrier density. The maximum drain-source

PAGE 73

73 current of 400 mA was obtained in an 800 m gate width HEMT, while the highest current density of 375 mA/mm was obtained in the 200 m gate width devices [74]. We were able to extract the variation in kn ee voltage as a function of gate-drain spacing from the IDS-VDS characteristics, as shown in Figure 4-8. A minimum value of 2.12 V was obtained for a 2 m spacing device. This furt her confirmed the good electron mobility in the two-dimensional electron channel of the HEMT. With the increase of gate drain spacing, the drain resistance and knee voltage proportionally in creased. Figure 4-9 s hows typical transfer characteristics from a 200 m gate width devi ce with 2 m gate-drain spacing. The maximum extrinsic transconductance was ~ 110 mS/mm. The maximum intrinsic transconductance was of 143 mS/mm, which was estimated with 111minRsggmex where gmex is the extrinsic transconductance, gmin is the intrinsic transcondu ctance and the Rs is the gate-source resistance. The gate source resistance was extracted from the s parameters with an equivalent circuit. The drain current of ~350 mA/mm is in line with that expected fo r this sheet carrier density. One frequently reported problem fo r AlGaN/GaN HEMTs is that the rf power obtained is still much lower than expected from the dc characteristics [76-81]. This problem is manifested by a collapse in drain current or frequency disp ersions in transconductanc e and output resistance, leading to severely reduced output power and power-added efficiency. Several mechanisms have been identified, including the presence of surface states between the gate and drain which deplete the channel in this region with time c onstant long enough to disrupt modulation of the channel charge during large signal operation or of trap states in the buffer layer [75-77,80]. Several studies have shown that the use of diel ectric passivation layers can be effective in reducing the effects of surface states. We ha ve employed gate lag measurements on surface passivated HEMTs as a metric for establishing the presence of buffer traps [79]. In this method,

PAGE 74

74 the drain current (IDS) response to a pulsed gate-source voltage (VG) is measured. Figure 4-10 shows the normalized IDS as a function of drain-source voltage (VDS) for both dc and pulsed measurements of a SiNX passivated HEMT. In the data, VG was pulsed from V to 0 at 0.1MHz and 10% duty cycle. The differences between dc and pulsed drain currents are consistent with the presence of buffer traps. This suggests that further optimization of the buffer growth conditions are necessary. 4.4 Silicon Substrates GaN on silicon HEMTs are entering commercia lization, with Nitronex Corporation offering devices in the 2-4 GHz range. These devi ces are particularly attractive due to the low cost of the Si substrates (dollars per wafer). Fa brication can prove difficu lt though, as the lattice mismatch can be difficult to overcome. This factor is offset by the fact that 4-inch Si fabrication process can be easily integrated, wheras SiC and Sapphire are still in the 3-inch wafer level. Si has a moderate thermal conductivity, making it competitive with SiC in all but the highest power markets. GaN on Si is available from Picogiga as well, and some interesting future work could involve a direct comparison of GaN on Si HEMTs and GaN on SopSiC HEMTs.

PAGE 75

75 Figure 4-1. Cross-Section of an AlGaN/GaN HEMT on SiC Figure 4-2. Image of reference FET (left) and FET adjacent to hole drilled at 355 nm (center, right)

PAGE 76

76 Figure 4-3. I-V Characteristics of reference HEMT (top) and drilled HEMT (bottom)

PAGE 77

77 Figure 4-4. I-V characteristic s of reference HEMT (top) and drilled HEMT (bottom)

PAGE 78

78 Figure 4-5. RF characteristics of refere nce HEMT (top) and drilled HEMT (bottom)

PAGE 79

79 5 W/mm 10W/mm 15W/mm reference Figure 4-6. Simulation results of junction te mperature as a function of substrate. Figure 4-7. TEM image of SopSiC structure

PAGE 80

80 012345 0 50 100 150 200 250 300 350 400 Vg = 0 Vstep = -1 V Figure 4-8. DC I-V curve for SopSiC HEMT Vds = 4 V Figure 4-9. DC I-V curve for SopSiC HEMT

PAGE 81

81 -6-5-4-3-2-10 0 20 40 60 80 100 120 Vds = 7 V DC 1 kHz 100 kHz Figure 4-10. Pulse curve for SopSiC HEMT

PAGE 82

82 CHAPTER 5 RF RELIABILITY 5.1 Overview Recently AlGaN/GaN HEMTs have made rapi d progress in their characteristics for operation as high power microwav e devices due to their excelle nt properties of high breakdown field of 3 106 V/cm and high sheet carrier density. With this progress, many groups are studying AlGaN/GaN structures inte nsively for their insertion into applications, such as; for Lband applications including wire less base station of mobile communication systems, for C-band and Ku-band applications on satellite communication systems or fixed wireless access systems. Since AlGaN/GaN HEMTs are able to operate at higher voltage w ith higher power density than its competitors, the reliability of device at the circumstance of high voltage, high temperature becomes a very critical issue. The degrada tion mechanisms may be quite different from the conventional III-V GaAs and InP based HEMTs due to the devices operating at a much higher voltage and current region. An example is the phe nomena of sudden degradation. In this case, the drain current of the nitride HEMT suddenl y increases in the pinched-off condition and reaches failure. To date, there are only limited re ports on the long-term reliability of the GaNHEMTs. However, the commercially available mu ltiple-device reliability test systems are not suitable for the nitride based HEMTs due to high voltage and high pow er requirements. There have been sporadic reports, which explain degradation mechanisms with their own unique models [81-86]. Unfortunately, these repo rts draw conclusions without an adequate amount of lifetime-test data. Multiple channel re liability DC and RF stress test systems are needed to produce enough data for meaningful de gradation analysis. We have seen that many issues have an affect on HEMT performance and reli ability, including the material quality, strain state, surface cleaning process, and the choice of etch and contact metal scheme.

PAGE 83

83 In device fabrication, we have developed new contact schemes fo r n-and p-type Ohmic contacts to GaN HEMTs, LEDs and MOS-HEMT gas sensors based on borides. Figure 5-1 shows changes in IDS as a function of time at 350C for HE MTs with different combinations of Ohmic and Schottky contacts [87]. Devices wi th conventional Ti/Al/Pt/Au Ohmics show significant decrease in drain cu rrent and RF performance over tim e at 350 C due to reactions between the GaN and the contact metals. By cont rast, the choice of boride-based contacts in addition to use of Pt/Au gate metal show far le ss change in both DC and RF performance [8788]. Our newly developed surface cleaning processes combined with improved passivation materials have led to GaN gated MOS-diodes and enhancement mode MOSFETs and superior mitigation of current collapse in HEMTs. Current studies aim to further improve the passivation between source/drain contacts and the gate when the AlGaN is at the surface. This can be accomplished by improved epitaxial growth t echniques, device design and passivation. Traditionally, lifetime prediction for device operation has usually relied on accelerated testing at elevated temperature and then extrapol ation back to room temperature operation. This technique frequently fails for scaled, high curr ent density devices found in modern technologies, as shown in Figure 5-2 [89]. Devi ce failure is driven by electric field or current mechanisms or low activation energy processes that are masked by other mechanisms at high temperature. Device degradation can be driven by failure in either active stru ctures or passivation layers [8991]. 5.2 Research Plan Stress Testing The research plan for stress testing is as follows The first test to be performed is RF stress measurements under CW power of 3dB at a Vds of between 30 and 60 V. It has been reported the power degradation has been related to the tr aps in the buffer layer-within 10 hours of stress, the power degradation could reach 0.5 dB. Second, we will study the stability of gate leakage

PAGE 84

84 current, Imax, gm, and threshold voltage (Vth) high-temperature (>150 C) DC 3-terminal aging test in the pinch-off state will be conducted and focused on gate leakage current equivalent to high power operation. Third, we will study the effects of fabrication techniques on current degradation mesa isolation vs. ion implantation isolation, different Ohmic and Schottky metalliation, Schottky vs. metal oxide semiconduc tor (MOS) based gate contact, and different passivation approaches will be included in the study. Fourth, we will study the effect of DC operation conditions on the degradation different drain voltages, drain currents, and junction temperatures will be used in the study. Finally we will study accelerated life tests performed at elevated temperatures. Effectively our protocol will be a combinati on of Si industry-standard approaches and areas specific to compound semiconductors (suc h as surface passivation and a higher defect concentration). We will run sufficient samples to get decent statistics and determine the lifetimelimiting factors at different temperatures. The us ual Arrhenius approach to obtaining operating lifetime from an extrapolation of high temperatur e data clearly is a problem since mechanisms that occur at high temperature may not really be a factor at lower temperatures. We will use a full range of device and materials characteriza tion methods to better understand the factors determining the reliability of GaN HEMT tec hnology. Failed devices can be analyzed to determine if macroscopic defects (dislocations) ha ve played a role in th e device break down. We largely have these capabilities in house so we can provide quick turnaround on device fabrication. 5.3 Stress Test System A block diagram of a single frequency 32-channel test system is shown in Figure 5-3. The system we are going to build consists of 4 RF driver boards with an output to 2-way splitters,

PAGE 85

85 creating 8 outputs at approximately 5 W each. A voltage-controlled osci llator is used to synthesize the RF signal, then a series of amplifie rs and attenuators is used to tune the output to the required power level, shown schematically in Figure 5-4. The splitter boards have isolators, directional couplers, and diode de tectors to detect both forward a nd reflected power as an output voltage, detailed in Figure 5-5. The signal then goes into the device boards. These have the input and output matching networks fabricated on the circuit board or inte grated on the GaN chip, w ith a thick copper layer on the back side for heat sinking and as a co mmon ground plane. A hole is bored through the boards to the copper for the device cavity. The de vice is mounted directly on the copper using Diemat, and wire bonded, as shown in Figure 5-6. This pre-matched circuit board is designed as disposable board. On the back side of the board, thermoelectric heaters are clamped on with a thermostat for temperature sensing. This allows for temperature control. The heaters will be controlled by PI D controllers with a computer interface. The device boards also have DC control and monitoring capabilities. The RF output from the boards goes to a detector diode board. The diode detectors work by rectifying the RF signal into a DC voltage output. This output must be calibrated since the response will be different at different fr equencies. This can be done by usi ng a synthesized sweeper to input a known power to the diode, and simply measuring the output with a digital multimeter. A curve can then be fit to this data and an equation and co nstants extracted, as shown in Figure 5-7. We have performed the thermal simulations of th e HEMT for different power levels and device configuration. The HEMT junction temperatur e can be easily higher than the physical device by more than 100 C [92-93]. It is very important to de termine the junction temperature during the

PAGE 86

86 reliability test. We propose to acquire a therma l image system to determine the actual junction temperature during the stress measurements. This equation can be input to LabView to perform the direction during acquisition. All control and acquisition is processe d in LabView. A full summary of the data that is processed is as follows. DC source/drain bias and gate bias RF generator output pow er, and temperature are controlled. DC source/drain bias and gate bias, RF forward a nd reverse input power, RF output power, and temperature are monitored and plot ted as a function of time in LabView while acquiring data. The data acquisition summary is shown in Figure 5-8. The modular approach we have taken allows for considerable flexibility and expandability to the system. Additional test cap acity can be added on down the road, and the frequency of the system can be changed by simply changing the driver components. Most importantly though, the independent device te st boards can be easily changed out to accommodate any type of device, so both packaged devices and bare die can be tested by simply redesigning the test board. This also then does not limit the system to purely nitride device testing. The system can be used to test any de vice so long as the board is designed with the proper matching components.

PAGE 87

87 02468101214161820222426 -70 -60 -50 -40 -30 -20 -10 0 10 At 350oC Regular Boride Source,Drain Pt/Au Ni/Au Pt/Boride/Au Ni/Boride/Au% IDSsat Time (Days) Figure 5-1. Changes of IDS as a function of time at 350C for HEMTs with different combinations of Ohmic and Schottky contacts. Figure 5-2. RF power output as a function of time

PAGE 88

88 Figure 5-3. A single freque ncy 32 channel RF and DC reliability test stand Figure 5-4. The RF source

PAGE 89

89 Figure 5-5. The splitter board Figure 5-6. A pre-matched circuit board

PAGE 90

90 Figure 5-7. Calibration and curve fitting of a diode detector. Figure 5-8. Data acquisition.

PAGE 91

91 CHAPTER 6 GAS SENSING 6.1 Overview There is significant interest in developing Ga N diodes for gas sensor applications. The gate region can be functionalized so that current changes can be detected for a variety of gases, liquids, and biomolecules. Hydroge n sensors are particularly inte resting with the emerging fuel cell vehicle market. The use of simple Schottky diode structures using thin Pt contacts allows for detection of hydrogen at concen trations of hundreds of ppm at wide temperature range temperatures (room temperature-500C) [105-114]. Th ere are also applicati ons for detection of combustion gases for fuel leak detection in spacecraft, automobiles and aircraft, fire detectors, exhaust diagnosis and emissions fr om industrial processes [105-110]. Simple two-terminal semiconductor Schottky diode s with Pt or Pd gates have been shown to be particularly effective hydrogen sensors. A cr oss section schematic and an optical image of a Schottky diode sensor is shown in Figure 6-1. Typically the sensing mechanism is ascribed to the dissociation of the molecular hydrogen on the Pt gate contact followed by diffusion of the atomic species to the oxide/semiconductor interf ace where it changes the piezo-induced channel charge and effective barrier height on Schottky diode structures. This effect has been used in Si, SiC, ZnO and GaNbased Schottky diode com bustion gas sensors [94-109]. Three-terminal transistor structures have been investigated to a much less extent but may have advantages because of the presence of the capability for cu rrent gain. AlGaN/GaN HEMTs are an attractive option as the hydrogen sensor because they can op erate over a broad range of temperatures and form the basis of next-generation microwav e communication systems so an integrated sensor/wireless chip is possible [115-116].

PAGE 92

92 6.2 Hydrogen Sensors 6.2.1 Gallium Nitride High Electron Mobility Transistor Sensors AlGaN/GaN Schottky diodes with Pt sensing metal demonstrate improved sensitivity under reverse bias and elevated temperature. A detection limit of 10 ppm was achieved under reverse bias with a current increase of 14%. Th e detection limit was further decreased to 1 ppm by operating the sensor at 150 C, still demonstrating a current ch ange of nearly 50%. This is a considerable improvement over the forward bi as detection limit of 100 ppm, with a current change of 4% [117]. The device layer structures were grown on C-plane Al2O3 substrates by Metal Organic Chemical Vapor Deposition (MOCVD). The layer structure included an initial 2 m thick undoped GaN buffer followed by a 35nm thick unintentionally doped Al0.28Ga0.72N layer. Mesa isolation was achieved by using an inductiv ely coupled plasma system with Ar/Cl2 based discharges. The Ohmic contacts were form ed by lift-off of Ti (200)/Al (1000)/TiB2 (200)/Ti (200)/Au (800). The metals were de posited by Ar plasma-assisted RF sputtering. The contacts were annealed at 850 C for 45 sec under a flowing N2 ambient in a Heatpulse 610T system. A 100 thick Pt Sc hottky contact was deposited by e-beam evaporation for the schottky metal. The final step was deposition of e-beam evaporated Ti/Au (300/1200) interconnection contacts. The individual devices were diced and wirebonded to carriers. These were then placed in an environmental test cham ber and connected to an electrical feedthrough for testing. Mass flow controllers were used to control the gas flow through the chamber, and the devices were exposed to either 100% pure N2, or H2 concentrations of 500 ppm down to 1 ppm in N2 and temperatures from 25 to 500 C [117].

PAGE 93

93 Devices were tested under both forward and re verse bias conditions at room temperature (25C) in a nitrogen atmosphere. There was an increase in current under both forward and reverse bias conditions upon exposure to hydrogen, as shown in Figure 6-2. This is consistent with previously discussed mechanisms in which the hydrogen molecules dissociate into hydrogen atoms through the catalytic action of the Pt gate contact, and diffuse to the Pt/AlGaN interface [96-97, 102, 106, 109, 118-123]. The hydr ogen atoms form a dipole layer, lower the Schottky barrier height, and increase net positi ve charges on the AlGaN surface as well as negative charges in the 2DEG ch annel. The calculated barrier height decrease for 500 ppm and 100 ppm hydrogen is 5 meV and 1 meV, respectively. The ideality factors were calculated to be 1.25 and 1.23 in 500 and 100 ppm hydrogen, respectiv ely, compared to 1.26 in 100% nitrogen [117]. However, a plot of hydrogen sensitivity (def ined as the drain current change over the initial drain current) versus bias voltage shows different charact eristics for forward and reverse bias polarity conditions at 500 ppm of H2, as shown in Figure 6-3. For the forward bias condition, there is a maximum sensitivity obtain ed around 1 V and further increase of bias voltage reduces the sensitivity. The sensitivity fo r the reverse bias condition is quite different and it increases proportionally to the bias volta ge. We propose the follo wing mechanism for the change in sensitivity under forward and reverse bias conditions: (1) The initial increase in the sensitivity is due to the Schottky barrier height reduction. (2) Further increase in forward bias allows electrons to flow across the Schottky barrier. These excess electrons bind with H+, form atomic hydrogen, and gradually destroy the dipole layer at the interface, therefore losing the hydrogen detection sensitivity. (3) For the revers e bias condition, electrons given away by the hydrogen atom may be swept acro ss the depletion region. At higher reverse bias voltage, a

PAGE 94

94 higher driving force is applied to the electrons to move across th e depletion region. Thus the dipole layer is amplified at the Pt/AlGaN interface for higher reve rse bias voltage. Due to this dipole layer amplification, the detection sensitivity is enhanced at higher reverse bias voltage. The change in current upon exposure to h ydrogen is directly related to the hydrogen concentration, as demonstrated in Figure 6-4, wh ere current was monitored as a function of time, with the hydrogen concentration being increased from 1 ppm to 500 ppm, alternating between hydrogen and nitrogen exposure [117]. It is also evident in Figure 65 showing the detection sensitivity as a function of hydrogen concentrations, that the diodes are much more sensitive under re verse bias conditions. A detection limit of 100 ppm is achieved under forwar d bias, but the reverse bias detection limit is an order of magnitude lower, 10 ppm. The change in current at 10 ppm is 14% and over 200% at 500 ppm under reverse bias conditions. This de monstrates again that the same sensor under reverse bias has a much lower detection limit, as forward bias operation re sults in changes of 2575% over the 100-500 ppm range. This is consiste nt with published reports indicating improved sensitivity under reverse bias [124]. The reliabil ity of the hydrogen sensor may be quite different under the two bias voltage polaritie s, since different degradation mechanisms in GaN devices are accelerated by either the presence of high voltage depletion regions (reverse bias) or current injection (forward bias in this experiment) [89]. Temperature effects were studied next. A hydr ogen concentration of 25 ppm was chosen to study the temperature effects sinc e it demonstrated a large res ponse at room temperature. The diode current was measured in both 25 ppm H2 and 100% N2 atmospheres at temperatures from 25 C to 500 C. The percentage change in current as a function of temperature is shown in Figure 6-6. At temperatures above 550 C, the device was irreversibly damaged. It is interesting

PAGE 95

95 to note the peak in the response at around 200 C. This is due to competing effects on the surface. It is expected that with increasing temperature, the cracking of the hydrogen on the surface will become more efficient. This will result in more molecular hydrogen, which will also diffuse to the metal-semiconductor interface faster at increased temperature. This effect is countered by basic kinetics. There will also be a decreased surface con centration of hydrogen, since there will be more transport to and fr om the surface at elev ated temperature. With this in mind, studies were performed to see if even lower detection limits were possible under elevated temperatur e conditions. A temperature of 150 C was chosen since it demonstrated a large response without pushing th e temperature so high that the device was damaged. As shown in Figure 6-7, a repeatable detection limit of 1 ppm was observed. Given the relatively large response even at 1 ppm, there is speculation that even lo wer detection limits can be achieved, but 1 ppm was the limit of the testing system. 6.2.2 Wireless Hydrogen Sensor Network We have demonstrated a wireless hydrogen se nsing system using commercially available wireless components and GaN Sc hottky diodes as the sensin g devices. Our sensors have achieved ppm level detection, with the added adva ntages of a very rapid response time within a couple of seconds, and rapid recovery. The sensors have shown current stability for more than 8 months in an outdoor environment. Our wire less network sensing system enables wireless monitoring of independent sensor nodes and transm its wireless signals. This is especially useful in manufacturing plants and hydrogen-fuelled au tomobile dealerships, where a number of sensors, possibly with each dete cting different chemicals, woul d be required. We have also developed an energy-efficient transmission prot ocol to reduce the power consumption of the remote sensor nodes. This enables very long li fetime operation using batteries. Experimental

PAGE 96

96 results showed that a 150 meter transmission di stance can be achieved with 10 mW total power consumption. The entire sensor package can be built for less than $50, making it extremely competitive in today's market [115-116]. The sensor devices are based on the technology described in the previous section, and demonstrate comparable characteristics. An inst rumentation amplifier is used for the detection circuit to sense the change of current in the device. The cu rrent variation, embodied as a change in the output voltage of the dete ction circuit, is fed into the microcontroller. The microcontroller calculated the corresponding current change and c ontrolled the ZigBee transceiver to transmit the data to the wireless network se rver. The block diagram of the sensor module and the wireless network server are shown in Figure 6-8 [115-116]. The Zigbee compliant wireless network supports the unique needs of low-cost, low-power sensor networks, and operates within the ISM 2.4 GHz frequenc y band. The transceiver module is completely turned down for most of the time, and is turned on to transmit data in extremely short intervals. The timing of the system is s hown in Figure 6-9. When the sensor module is turned on, it is programmed to power up for the first 30 seconds. Following the initialization process, the detection circuit is periodically powered down for 5 seconds and powered up again for another 1 second, achieving a 16.67% duty cycle. The ZigBee transc eiver is enabled for 5.5ms to transmit the data only at the end of ev ery cycle. This gives a RF duty cycle of only 0.09% [115-116]. A web server was developed using MATLAB to share the collected sensor data via the Internet. The interface of the server program, s hown in Figure 6-10, illustrates three emulated sensors with different baseline currents. If any of the sensors current increases to a level that indicates a potential hydrogen leakage, the alarm would be triggered. A client program was also

PAGE 97

97 developed to receive the sensor data remotely. As shown in Figure 6-10, the remote client was able to get a real time log of the system for the past 10 minutes via the client program. In addition, a full data log obtained via accessing the se rver via a ftp client as the server program incorporates a full data logging f unctionality. When an alarm was tr iggered, the client was able to deactivate the alarm remotely by clicking a bu tton on the interface. The server program for the wireless sensor network could also report a hydrogen leakage emergency through phone line. When the current of any sensors exceeded a certain level, indica ting a potential hydrogen leakage, the server would automatically call the phone-dial program, reporting the emergency to the responsible personnel [115-116]. The sensor module was fully integrated on an FR4 PC board and packaged with battery as shown in Figure 6-11 (a). The dimension of the sensor module package was: 4.5 2.9 2 inch3. The maximum line of sight range between the sensor module and the base station was 150 meters. The base station of the wireless sensor network server was also integrated in a single module (3.0 2.7 1.1 inch3) and ready to be connected to laptop by a USB cable, as shown in Figure 6-11(b) and (c). The base station draws its power from the laptops USB interface, thus do not require any battery or wall transformer [115-116]. Field tests have been conducted both at Univer sity of Florida and at Greenway Ford in Orlando, FL. The setup at Greenway Ford was aimed to test the stability of the sensor hardware and the server software under ac tual operational environment. Tw o sensor nodes were installed and the test was started on the 30th of August 2006. Six sensor modul es and the server have been functioning to date [115-116]. The outdoor tests at University of Florida have been conducted for several times, to test the sensors response to different concentrations of hydrogen at different distances. The tested

PAGE 98

98 hydrogen concentrations include: 1%, 4%, and 100%, and the distance from the outlet of hydrogen to the sensor ranges from 1 foot to 6 feet. Hydrogen leakage was successfully detected for all these cases, triggering th e program to send alarm to cell phone. Figure 6-12 shows the test result with four running sensor modules and 4% hydrogen at 3 feet away from the hydrogen outlet. The sensors were tested sequentially so the effect s of each sensor can be isolated. The test results also display the reliability of the wireless network as it is able to collect the data from each individual sensor. Initial results of field testing indicated that re liability of the sensors could be a concern, as the single diode sensors showed a periodic rise and fall in the current level, which can be attributed to a temperature effect. This is s hown in Figure 6-13. There was also a long term current degradation, which was attributed to ohm ic contact degradation wi thin the device due to the continuous bias. As a result, TiB2-based ohmic contacts have been employed, which have been proven to improve stability for long term op eration [125]. To avoid the thermal effects, the sensing device was redesigned to employ a differential detection scheme. This involved a reference diode, which is encapsulated, and an active diode, which is open to the ambient. Detection is achieved by monitoring the differe nce in current between the two devices as opposed to measuring an absolute current level. Test results demonstrating the operation of the device are shown in Figure 6-14. It is demonstrated in this figure that at room temperature and elevated temperature, the reference diode doe s not show a current change upon exposure to hydrogen, while the active diode does. The combin ation of differential diode sensors and boridebased ohmic contacts has significantly improved st ability and reliability. Recent field data is shown in Figure 6-15, and the data can be m onitored in real-time at the following website: 2http://ren.che.ufl.edu/ap p/realtimeSensing.htm [115-116].

PAGE 99

99 Figure 6-1. A Schottky diode hydrog en sensor shown in cross-section (top) and optical image (bottom)

PAGE 100

100 Figure 6-2. Forward and revers e bias plot of diode curren t in varying atmospheres Figure 6-3. Percentage change in curr ent as a function of bias at 500 ppm H2

PAGE 101

101 Figure 6-4. Time dependence of Schottky diode sensor when switching from pure nitrogen atmosphere to hydrogen concentrations from 1-500 ppm Figure 6-5. Percentage change in current as a function of hydrogen concentration under both forward and reverse bias.

PAGE 102

102 Figure 6-6. Percentage change in current as a function of temp erature at a hydrogen concentration of 500 ppm. Figure 6-7. Time dependence of Schottky diode sensor when switching from pure nitrogen to 1 ppm hydrogen under reverse bias at 150 C

PAGE 103

103 AC Wall TransformerRe g ulator Re g ulator Battery Backup Control Sensor Reference Zi g Bee T ransceiver Internet Wireless Network ServerInternet Server Cell Phone MicrocontrollerBattery/Ener g y Harvest Figure 6-8. Block diagram of sensor module and wireless network server Warm up: 30 seconds 1s 5s Monitor T x data Monitor T x dataMonitor T x data Monitor Tx dataPower Up Power Down1s 5s 1s 5s 1s 5s Figure 6-9. System timing of at wireless sensor node Figure 6-10. Interface of on line hydrogen le vel monitoring

PAGE 104

104 (a) (b) (c) Figure 6-11. Photo of sensor system (a) Sensor with sensor device; (b) Sensor and base station; (c) Computer interface with base station Figure 6-12. Field test re sults of four sensors Figure 6-13. Field test results for single diode sensors over a period of one week

PAGE 105

105 Figure 6-14. Differential diode test results using 1% H2 at room temperature and elevated temperature Figure 6-15. Field test results over a month period for differen tial diode sensors with boridebased ohmic contacts

PAGE 106

106 CHAPTER 7 THERMAL SIMULATIONS OF TH REE-DIMENSIONAL PACKAGE 7.1 Overview GaN-based devices are optimal for high power, high temperature, and high speed applications. Silicon dominates the low power low frequency IC market with CMOS technology. A new hybrid approach involves vertically integrating GaN and Si devices to obtain the best performance from both devices. A desi gn investigated in this work involves the integration of a Si modulator ch ip and a GaN power amplifier (PA) in a vertical stack, with a heat sink on top integrated with the antenna. Ther mal management is critical in such a design since Si device are much more sensitive to te mperature and cannot operate reliably above 150180 C. The integration of these two types of devi ces must, therefore, be optimized for heat transfer to efficiently draw the heat from the GaN device before it affects the performance of Si devices. A first-generation schematic of the MCM 3D integration of GaN and Si components on high resistivity Si is shown in Figure 7-1. The Si modulator chip and GaN PA chip are flip-chip bonded, providing room for expansion to include more circuits and functi ons. The ideal vertical stack design is shown in Figur e 7-2. The Si modulator is bon ded to the common ground plane and GaN PA chip using polydimethylsolixane (PDM S). The PDMS layer also has a secondary function as a thermal insulator for the Si modul ator, due to the low th ermal conductivity (~0.1 W/m-K). In the real package, metal vias would be added to make the electrical connections between chips and to the common ground plane. 7.2 Finite-Element Modeling To study heat transfer through the package, finite-element modeling was used to look at various parts of the package. This method breaks down the actual geometry into a mesh, which is

PAGE 107

107 an interconnected system of nodes, with averaging occurring in the space between the nodes. The physical differential equation governing the variable of interest is applie d at each node, creating an enormous system of equations. The equati ons are solved simultaneously and the results plotted. The particular advantage of this met hod is that it can solve differential equations in geometries that defy analytical solutions becau se the mesh is essentially a breakdown of the geometry into a number of connected linear regions To that end, by enabling a finer mesh, one is able to achieve more accurate results, but cons iderably more computing power is required. By performing a finite-element analysis us ing FlexPDE software, a package design to optimize heat transfer could be realized. A brief discussion of heat transfer principles will begin this section to allow for a better understanding of the simulations performed. From there, multiple generations of simulations will be discussed and results presented. 7.2.1 Heat Transfer There are three modes of heat transfer: conduction, convec tion, and radiation. Energy transfer through conduction is a ccomplished through direct molecu lar collisions and through free electrons. The ability of any give n material to conduct heat is referred to as thermal conductivity, k, and is analogous to electrical conductivity. Conductive heat transfer at steady state is described by Fouriers first law of heat conduc tion, also known as the Fourier rate equation, shown below. It is analogous to the mol ecular momentum transfer equation [126]. Tk A q The thermal conductivity is independent of di rection, and is primarily a function of temperature and is a fundamental physical prope rty of a conducting medium. Depending on the material, it can be considered consta nt over a modest temperature range.

PAGE 108

108 Convective heat transfer involves energy exch ange between a surface and adjacent fluid. Natural convection occurs when there are no extern al forces driving the fluid, just the natural circulation due to density differences in the fluid from temperature varia tion. Forced convection occurs when there is an external driving force su ch as a pump or fan. Convective heat transfer is described by Newtons law of cooling, al so known as Newtons rate equation [126]. Th A q In this equation, h is the convective heat tran sfer coefficient. It is not a constant and generally is a function of system geometry, fluid, flow properties, the di fference in temperature, and the type of convection present. Radiative heat transfer occurs between tw o surfaces and does not require a medium to propagate. This is the only method of heat tr ansfer in a vacuum. The energy emission by a perfect radiator (black body) is given by the Stefan-Boltzmann law of thermal radiation [126]. 4T A q This equation must then be corrected fo r deviations from black body behavior. It is common for all three methods of heat transfer to be taking place simultaneously. For this work, the effects of radiative heat transfer are minimal compared to conduction and convection. Consider as an example a situation where a hot gas is contacting a 3-material composite wall, with a cold gas on the other side. The heat transfer can be described as a series sum of convective heat transfer between the hot gas and the first surface, conduction through the first material of thermal conductivity k1, conduction through the second material of thermal conductivity k2, conduction thr ough the third material of thermal conductivity k3, and convection again at the wall. This is shown schematically in Figur e 7-3. Considering these walls

PAGE 109

109 ot be infinite in the y and z directions, we can solve the Fourier equation in the x-direction by integrating with the boundary conditions T(x=0) =T1 and T(x=L)=T2 to obtain the result below [126]. T L kA q This resembles Newtons rate equation. Since th e system is at steady state and no heat is generated or destroyed, the heat transfer through each layer must be equal. Therefore the total heat transfer can be written by equating each term: c c hhTTAhTT L Ak TT L Ak TT L Ak TTAhq 4 43 3 3 32 2 2 21 1 1 1 This can be rearranged to determine the total change in temperature: AhAk L Ak L Ak L Ah qTTc h ch1 13 3 2 2 1 1 From this equation, one can see that heat tr ansfer can be described in a manner analogous to Ohms law for electrical circuits. This give s rise to the term thermal resistance, where R=1/hA, L/kA. This example is shown as the equi valent of a series circuit. One can imagine a geometry including materials in parallel, which would produce an equation similar to a parallel circuit. The equation can be ge neralized in an analagous manner to Ohms Law with an equivalent thermal resistance instead of electrical resistance [126]: iT EqT EqTRR R T q, By performing an energy balance on a given control volume of dimension x, y, and z, it is possible to obtain a transient differential equa tion for general heat transfer. This equation cab be written as follows in the limit as x, y, and z approach zero [126]:

PAGE 110

110 vv Dt DT cqTkv2 In this equation, q refers to the total energy generation, refers to the viscous work rate per unit volume, and potential and ki netic energy terms are simplified in the right hand side of the equation. From here, the equation can be furt her simplified for the pur poses of the simulation by taking the case where there is no fluid motion and all heat transfer is by conduction. This equation, which is the one used in the therma l simulations, can be written as follows [126]: qTk Dt DT cp2 In this case, T is the temperature, t is time, is density, CP is the heat capacity, k is thermal conductivity (W/cm-K), and PD is the dissipated power (generated in the device). The latter can be determined from the product of bias voltage and drain current thr ough the HEMT, divided by the HEMT layer volume [126]. To solve the differential equation, boundary conditions must be specified. The initial conditions refer to the values at the start of the time period of interest. For this equation, temperature is the only variable, and therefore the initial condition can be considered to be room temperature. The boundary conditions refer to the valu es of the variable of interest at the system boundaries. There are three types of boundary conditi ons for this type of problem: isothermal boundary, which is a constant temp erature (T=const), insulated boundary, where there is no heat flow across the boundary (dT/dx = 0), and boundaries with a function descri bing the temperature. This type of boundary is typically described by a change in the me thod of heat transfer, such as conduction transferring heat to a surface and conv ection removing heat at the surface. This can be written as follows, assuming a convective heat transfer coefficient and temperature far from the surface as Tinf [126]:

PAGE 111

111 x T kTThsurfsurf inf Perhaps the most difficult part of the simula tion is setting up the boundary conditions to be realistic. For this work, convective boundary cond itions were typically used because it is the most realistic case (the package will operate in air, not in an insu lated or isothermal environment, thus there will be convective heat transfer at the boundary). The major problem with convective boundaries is the convective heat transfer coefficient, h. Given that h is usually a function of temperature and geometry, it can be very difficult to determine a reasonable value to use in the problem, and empirical measurements are typically used to calculate a value. There are a number of correlations available for calculation of h, wh ich are typically empirically fitted functions of various dimensionless gr oups, defined below [126]: PrRe Pr Pr2 32 Pe GrRa TLg Gr k c k hL Nup One must pay careful attention to the conditions where each case is valid. Perhaps the best correlation for natural convection is the Churchill and Chu correlation. It is used for flow over vertical plates and shows good results over 13 orde rs of magnitude of the Rayleigh number (Ra). The Ra number is defined as the product of the Grashof number (Gr) a nd Prandtl number (Pr) [126].

PAGE 112

112 2 278 169 61Pr 492.0 1 387.0 825.0 Ra Nu For forced convection, there are many more correlations, some describing laminar flow and some describing turbulent flow, and the many different regimes of each. For laminar flow, the Sieder-Tate correlation provides very good re sults for flow through pipes, and the DittusBoelter generally gives good results for tu rbulent flow under common conditions [126]. 4.08.0 14.0 31PrRe023.0 86.1 Nu L D Pe Nuw b Note that most correlations are calculated fo r fluid flow through pipes or for flat surfaces. The heat sink design for this simu lation provides an intere sting geometry, as the stacked fins of a heat sink cannot be considered i ndependent infinite plates or enclosed pipes for fluid flow purposes. Furthermore, heat sinks typically ha ve a fan blowing over them to improve heat transfer. This creates a forced convection case, where a free-standing he at sink would present a natural convection case. To this end, rather than using the true dimensions of the heat sink, one can use the hydraulic diameter as a substitute for length or diamet er in the calculations above. It is defined as a ratio of surface area to perimeter as follows [127]: P A Dh4 This length quantity is useful in situations where the equations are solved using perfectly circular geometries, but the actual geometry is elongated. One could view the area between the fins of the heat sink as elongated channels. An alternative would be to determine the value

PAGE 113

113 experimentally by machining a custom heat si nk, putting a thermocouple on the top surface, a thermocouple and heat source on the bottom surface, and therefore knowing Q, A, and T, one can back out h from the convectiv e heat transfer equation [127]. 7.2.2 First Generation Simulation: Ga N Power Amplifier on Heat Sink As an initial test of the model, a simplified system was studied. This system consisted of a GaN HEMT on the heat sink. This structure re sembles the schematic for the first generation device. Figure 7-4 shows a schematic of the device centered on the integrated heat sink/antenna. The assumed initial values were: 3 W/mm dissipated power in the GaN power amplifier, a 5 m thick GaN layer, 500 m thick SiC layer, and an effective conv ective heat transfer coefficient, h, of 30 W/m2C. The heat sink was assumed to be ~20 mm2 area, with 11 fins, 10 mm fin height, and 0.94 mm fin width, which are the dimensi ons required for a 2.4 GHz antenna. The finned heat sink acts as a large thermal mass for heat re moval. From these starting values, each layer was optimized by systematically varying the thickness to find the minimum steady state temperature. The most realistic boundary condition is a convectiv e boundary on all sides [92-93]. The first variable studied wa s thermal conductivity. This is ma nifested in the design in the substrate selection. Figure 7-5 shows the e xpected maximum temperatures reached by the integrated power amplifier and modulator at fixed power and layer thickness for different substrates. Clearly the use of sapphire substrates is not an effective solu tion because of their poor thermal conductivity. The resultant high device te mperature would be a major detriment to the reliability of the GaN power amplifier in particular since it is very difficult to prevent reaction of the gate metal at 400 C during ex tended operation. The surprising result from Figure 7-5 is that use of a bulk GaN substrate, while still inferior to SiC, would still maintain the die temperature

PAGE 114

114 at less than 200 C at 3 W/mm. Of course, the op timal substrate is single crystal SiC, with the highest thermal conductivity [92-93, 128-131]. The thickness of the substrate also plays an important role in the thermal characteristics of the MCM. Figure 7-6 shows that, for a SiC substr ate, if the substrate is thinned down from the usual 600 m to 100 m, there is an improvement of ~50 C in the operating temperature of the chip at a fixed PA power of 3 W/mm. This is due to lowering of the thermal resistance by thinning down the chip; however, there is also a slight rise in opera ting temperature below 100 m. Once the substrate becomes too thin, there is an increase in the lateral thermal resistance, which will drive up the junction temperature. These are important considerations when designing the interconnect vi as. The dry etch rates of SiC are usually below 1 m/min and, thus, the SiC substrate is often thinned down to reduce the etch time for via hole formation; therefore, 100 m would be an optimal substrate thickness [92-93]. Figure 7-7 shows the effect of PA power de nsity on the temperature rise of the MCM, assuming a standard thickness SiC substrate with no vias or wires in the structure for simplicity. At high power levels even the use of a SiC substrate is unable to keep the MCM from reaching 400 C at 10 W/mm. The GaN PA can theoretica lly operate at this te mperature if it uses thermally stable metallization, but the Si m odulator would be inopera ble. The PA operating temperature can be reduced to below 200 C even at 10 W/mm by optimizing the substrate thickness as discussed above and shown in Figure 7-7. The design of the device also plays an important role, and using a fingered de sign, consisting of 4 gates with 250 m gate width, provides an improvement of ~50 C over a single gate with 1 mm gate width [17-19, 92-93, 131132].

PAGE 115

115 Free convection is the only heat dissipati on mechanism through the outer vertical-surface of the MCM. The amount of heat dissipation th rough free convection is five orders smaller as compared to the conduction through the integrated heat sink antenna. The thermal resistance at the convective boundary is ~85 K/W, compared to ~2x10-4 K/W for conduction through the package. Therefore, the temperature rise at th e center of the chip on the top surface was highest (around ~ 110 C), and there was a ~40 C difference between the center and edge of the structure, based on a power amplifier density of 3 W/mm. To investigate the temperature rise while the device is turning on, the transient e quation was used. The temperature profile reached steady state in the range of ~30 sec [92-93]. Other MCM design considerations that were also considered, but were not found to have a significant effect on the operating temperature, were the GaN epi-layer thickness (~5 C difference for 10 m versus 3 m) and the wire thickness and lengt h (~2-8 C difference for wire lengths from 2-100 mm and thicknesses of 0.2-10 m) [92-93]. 7.2.3 Generation II: Combined Amplifier and Silicon Chip After initial study and optimizati on of heat transfer through the GaN PA on the heat sink, it is logical to go back and consider the more complex problem of modeling the entire 3-D package. The optimized GaN PA parameters were used as starting points for this work: 3 W/mm dissipated power in the GaN power amplifier, a 2 m thick GaN layer, 100 m thick SiC layer, and an effective convective heat transfer coefficient, of 30 W/m2C. The heat sink was, again, assumed to be ~20 mm2 area, with 11 fins, 10 mm fin he ight, and 0.94 mm fin width. The additional layers added in a vert ical stack to complete the 3-D structure were the PDMS layers for bonding (4 m each), an Au common ground plane (5 m), and the Si modulator chip (500 m) [92-93].

PAGE 116

116 Figure 7-8 shows the expected maximum temper atures reached by the power amplifier for various Si layer thickness and fi xed power density of 3 W/mm. The effect is small, changing by only approximately 10 C over an order of magnitude thic kness range. Therefore, standard thickness Si substrates can be used for the m odulator device without incurring any substantial losses in thermal management. It would be beneficial however, cost permitting, to use thin Si membranes for the modulator, using PDMS as a chip carrier. This could prove essential technology if this extra 10 C drop will make the difference between the function and failure of the Si devices, since these are the most temp erature-sensitive component in the MCM [92-93]. The typical temperature distribut ion in both GaN and Si layers is shown in Figure 7-9. The Si temperature distribution is considerably broader, due to lateral spreading from the high thermal resistance PDMS layers. This effect is highlighted in Figure 7-10, which shows a typical cross sectional temperature distribut ion in various layers in the de vice. Note from the scale that this shows only a small area, 2.5 mm on either side of the gate. The temperature in all layers reaches the same value within 2 mm of the ga te, and is sharply peaked in the GaN layer, becoming broader down through the Au and Si layers with the PDMS layers spreading the heat. The effects of both PDMS and Au thickness were al so studied, and the e ffect over a range of 15 m was negligible (<3 C) [92-93]. After optimization of all layer thicknesses in the structure, the device power level was studied. Previous work had shown that optimization of the GaN device design could drive down the high power operating temperatur e considerably from a reference case. Now, due to the high thermal resistance from the PDMS layers, the ju nction temperature of the HEMT rises again to 350C at high power levels, even with all layers optimized, as shown in Figure 7-11. Again, the GaN PA can operate at this temp erature, but the Si modulator w ould be inoperable, even with

PAGE 117

117 thermal insulation. It is, therefore, logical to attempt to drive down the operating temperature by removing the high thermal resistance layers. Since the distribution is sharply peaked at the gate, the removal of just the PDMS layer closest to the gate would driv e down the temperature considerably, as shown in Figure 7-12. Comp lete removal of PDMS would decrease the temperature another 10 C, but would create some difficulty in bonding the Si modulator to the rest of the package [92-93]. The purpose of the top PDMS layer is for thermal insulation of the Si modulator and to serve as a dielectric between the GaN PA a nd the antenna ground plane. Since the operating temperature can be decreased by over 200 C by removing this layer, alternatives were considered. By using a more conventional SiNx passivation, the dielec tric qualities can be preserved, and the improvement in heat transfer is still considerable for thin films, as shown in Figure 7-13 [92-93]. The design of the device also plays an important role. Again a multiple finger design, can provide an improvement of ~40 C over a single gate with 1 mm gate width with the proper gate pitch, as shown in Figure 7-14. The effect of the boundary conditions was inve stigated, as shown in Table 7-1. It is assumed that a reasonable forced convecti on heat transfer coefficient was 30 W/m2C, whereas a free convection heat transfer coefficient would be 10 W/m2C. The fixed value boundary would correspond to the package being mounted on a chip carrier, which would function as another heat sink. The most realistic cas e would be a fixed temperature near 25 C on the bottom and free convection at the top. Note that this improves the heat transfer considerably [92-93].

PAGE 118

118 Table 7-1. Effect of boundary c onditions on package temperatures Top Forced Free Forced Forced Free Bottom Forced Free Free Fixed Fixed GaN 124 227 141 57 57 Si 96 198 111 25 25 Antenna 92 196 109 25 25 Figure 7-1. First-generati on prototype of integrated heat si nk/antenna incorporating GaN PA and Si modulator SiC GaN Thermal Insulator Silicon I/O Routing Figure 7-2. Concept of the 3-D multi-layer stru cture combining heat-sink antenna, GaN RF and silicon signal pro cessing electronics

PAGE 119

119 Figure 7-3. Example of heat tr ansfer through a composite wall Figure 7-4. Mesh used for finite element modeling.

PAGE 120

120 Figure 7-5. Effect of substrat e thermal conductivity on the junc tion temperature of the HEMT Figure 7-6. Effect of silicon carbide substrate thickness on the junction temperature of the HEMT

PAGE 121

121 Figure 7-7. Effect of powe r dissipation on the junction temperature of the HEMT Figure 7-8. Effect of Si thickness on the junction temperature of the HEMT

PAGE 122

122 Figure 7-9. Typical temperature distribution under base operating conditions in the GaN layer (top) and Si layer (bottom)

PAGE 123

123 Figure 7-10. Typical temperature pr ofile in a cross section of the structure zoomed in to the area near the gate Figure 7-11. Effect of HEMT power dissipation on the maximum temperature in both active layers

PAGE 124

124 Figure 7-12. Effect of the removal of top PDMS layer and both PDMS layers on the maximum temperature in both active laye rs at varying power levels Figure 7-13. Effect of SiNx passivation thickness on the junc tion temperature of the HEMT

PAGE 125

125 Figure 7-14. Effect of gate pitch on the juncti on temperature of the HEMT in a multiple finger configuration

PAGE 126

126 CHAPTER 8 UV LASER PROCESSING 8.1 Overview Excimer laser processing has emerged over the past 30 years as a powerful micromachining technique. They are particularly flexible tools due to their short wavelength, high efficiency, and high power. A sampling of the many processes that use excimer lasers are as follows: photoablation, etching, micron-scal e lithography, doping, texurizing surfaces, planarizing and cleaning surfaces, and medical applications. The most notable development in the medical field is laser keratectomy with a 193 nm laser, known today most commonly as LASIK. It is also used in the electronics industry in several capaci ties, most notably being for via hole formation in dielectric films, but it is also used to selectively remove metal to rework circuits or masks where there are electrical shor ts. Pulsed laser deposition is one of the most studied applications, and offers several advantages over conventional CVD or sputtering, the most notable being stoichiometric ablation of an alloy. Pulsed long wavelength lasers such as the CO2 and YAG lasers, were first developed commercially. They are used primarily for bulk micr omachining, particularly engraving, of metal or ceramic parts. These lasers are quite limite d though in that they are incapable of producing high-resolution features, reflection from the surface becomes more efficient at larger wavelengths, thus requiring more input power, and longer wavele ngth photons have less energy, so photochemical reactions cannot be initiated effici ently. For this reason the primary ablation is heating and subsequent sublimation, which can cr eate damage in the material and debris on the surface [133]. UV laser processing is advantageous in that UV photons are capable of excitation of gas, liquid, or solid molecules, which is a first step in the photoablation process. UV photons are

PAGE 127

127 strongly absorbed by most materials, so with d ecreasing wavelength, the ra nge of materials that can be processed increases considerably to where nearly everything except quartz can be machined at 193 nm. UV light can also penetr ate a laser induced plasma plume above the ablation site, allowing the etch to continue. Longer wavelength photons cannot penetrate this plume therefore the etch can be self-terminating as it reaches a point where the plume cannot dissipate before the next pulse occurs. UV lasers are also capable of producing high peak power, which is important for rapid processing and initi ation of ablation processes, and high average power, which is important for a rapid, repeatable process (i.e. if each pulse has the same average energy, one can count the number of pulses for a precise etch de pth control) [134]. Since there is no effective wet etchant fo r SiC, vias are currently formed using conventional dry etching techniques, such as Re active Ion Etching (RIE) or Inductively Coupled Plasma (ICP) etching. The problem with these pr ocesses is that the typical etch rates in F2or Cl2based plasmas range from 0.2-1.3 m/min. The typical substrate thickness is on the order of hundreds of micrometers and, even for thinned substrates of 50 m, etching can take up to four hours under ion energy condition where mask erosion is minimal. Another disadvantage of the dry etching process is that the mask material must be robust, typi cally metals such as Ni, Al, or Cr. The deposition, patterning, and removal of the mask add complexity to the process, making it difficult for mass fabrication. There are also sidewall roughening and micromasking issues with dry etching, making the proce ss very inconsistent [135]. Laser drilling appears to show promise for fabrication of via holes in SiC substrates. This is a maskless process, so there is no additiona l processing required beyond the actual drilling. The drilling is computer controlle d, and the laser is focused only on the area where the hole is desired, so there are no loading e ffects observed, allowing for very reproducible results. With the

PAGE 128

128 computer controlling, there is no physical mask to be made, which adds considerable flexibility and tuneability to the process. The etch rates are consid erably higher (2-14 m/sec) than dry etching, but this is a serial process. With the hi gh etch rates, however, a single hole can be drilled within ten seconds with no additional processing required, making this process still faster than dry etching. The biggest issue that must be a ddressed is surface debris and the reliability of HEMT devices fabricated on laser drilled substrates [135]. Debris formation is in fact the motivation behind such an extensive study of drilling systems before purchase. The most ideal situ ation would be purely gaseous ablation products, which can easily be removed by a nozzle near the ablation point attached to a vacuum pump. In a real situation, the ablated material is red eposited on the surface of the sample and nearby mechanical components. Much like in a plasma et ch, there will be volatile products, which will be easily removed, and nonvolatile products, which will remain on the surface. XPS can be used to identify the components remaining on the surf ace, and once known, met hods of protecting the surface can be determined. As an example, an XPS study of polyimide ablation shows a carbon rich surface, implying that oxygen and nitrog en formed volatile products, most likely NOx and COx species, while the carbon was left as soot which can be difficult to remove [136]. One interesting method of protecting the surface is by preventing the formation of such debris. A proposed method for this is to perform the ablation in either a vacuum or a chamber with a light gas atmosphere such as He, or flowing He over the samp le. The reason for this is that the light atoms or vacuum will decrease the probab ility that the debris atoms will form clusters and increases the likelihood that they will stay in the gas phase and be swept away [136]. Typically a cleaning step must be performed after ablation. Th e exact nature of the clean will depend on the material and the nature of the debris formed. One possible method involves

PAGE 129

129 using the laser again at a low energy to ablate th e debris. This usually just results in moving the debris around on the surface to a safer location. For organic contaminants, an oxygen plasma can be used to ash the debris. Other chemical methods are also used, such as a rinse with acetone, an acid etch, or either combined with ultrasonic agitation. Finally, one of the more promising general use methods would be to coat the material with a sacrificial layer, such as photoresist, prior to ablation. This operates under the same principle as coa ting a sample with photoresist prior to dicing. The resist can be removed after ablation to leave the clean surface [136]. 8.2 Solid State Laser Processing Initial efforts to study via hole formation used industrial standard marking systems, which use solid state lasers. These syst ems are typically excellent for laser scribing of identification numbers, cutting, or welding of metal parts. It was ultimately determined that while these lasers are excellent for metal processing, the wave lengths do not couple well with semiconductors, ceramics, or polymers. This is due primarily to the increased refl ection at these longer wavelengths relative to UV lasers (discussed in a later section). This implies that more laser power will be required to ablate material, which in turn will cause more sample heating. Metals, having a high thermal conductivity, can dissi pate the heat more efficiently [133]. 8.2.1 Background: Diode Pumped Solid State Lasers Diode-pumped solid state (DPSS) lasers operate by using a laser diode to pump a crystal medium for gain. The crystal is typically ne odymium-doped either yttrium aluminum garnet (Nd:Y3Al5O12), better known as Nd:YAG, or yttrium orthovanadate (Nd:YVO4). The neodymium replaces yttrium in the crystal lattice and emits at 1064 nm. While there are two types of crystals used, both are referred to as YAG lasers. Such lasers are capable of extremely high power by operating in Q-switched mode, which places an optical switch in the laser cavity

PAGE 130

130 that is timed to open when the maximum population inversion is achieved. This maximizes the efficiency of the laser at its emitting wavelength [137]. With a 1064 nm base IR laser system, a series of nonlinear optics crys tals can be used to generate higher order harmonics to alter the wavelength of the system. For example, lithium triborate (LiB3O5, LBO) can be used to generate the second harmonic frequency in the visible range of the spectrum at 532 nm. This is the method commonly used to create green laser pointers. A typical third harmonic generator is potassium dihydrogen phosphate (KH2PO4, KDP). This is used to generate the frequency triple d harmonic at 355 nm, creating now a UV laser. A frequency quadrupled laser can also be form ed, emitting at 266 nm. While these capabilities comes a cost. While a UV DPSS laser may be desira ble, the price paid is power output. Typically the second harmonic generation is only 20% effi cient relative to the YAG emission, thus the power level of a frequency doubled system is only 20% of a 1064 nm YAG system for the same laser. High harmonics will be less efficient, so one can imagine that for a UV DPSS system, one must start at a high output power laser to obta in any reasonable power output level [137]. The source is an optical beam, so the pattern shape is always circular, with the minimum feature size determined by the beam size, which is typically no smaller than 10 m and can be up to 30 or 50 m. Patterning is then achieved by trepanning, which is the process of translating the beam in the shape of the desired pattern to re move material. DPSS systems do offer advantages though in that they require no maintenance of the laser itself, so the only periodic maintenance cost will be upkeep on the optics. Since the DPSS laser is pumped by a so lid-state source rather than a gas discharge, the pulse frequency can be very high (over 100 kHz). The low power output is compensated by this, so high etch rate s can be achieved, and these systems are used for high throughput/low maintenance applications [138].

PAGE 131

131 8.2.2 Experiments: Frequency Effects The early work on this project built on the pos tdoctoral work of Dr. Suku Kim at UF, who explored via hole formation in SiC using laser pr ocessing [135]. The primary goal of this work has been proof-of-concept for the via hole proc ess and to evaluate systems available on the market with the goal of purchasing one that will se rve the purposes of research at UF. To that end, samples were sent to several companies fo r simple processing on their system to evaluate compatibility and limits. The primary methods us ed to quantify performance were feature size and the degree of damage and surface debris, bo th determined using a combination of optical microscopy and SEM. The first studies evaluated the performan ce of systems at Lenox Laser and US Laser Corporation. These were selected as they were highly reputable marking systems, however they were typically used for metal processing. They both used a 1064 nm YAG laser, and US Laser provided the capabilities to study both frequency doubled and frequency tripled modes. This allowed us to effectively study the effects of decreasing the wa velength. Performance improved significantly when the UV region was reached, as predicted by the literature and mentioned in the discussion both above and below. In this set of experiments, a HEMT structure was used, and devices were fabricated on the samp le after laser drilling [71, 74]. An AlGaN/GaN HEMT structure was grown by MOCVD on a SiC substrate (4H-polytype, N-doped, ~1017cm-3, thickness was ~400 m). Laser drilling was empl oyed in both directions, i.e. from front-to-back and vice versa. Vi a holes with nominal diameters of 50 m were obtained by laser drilling with a Nd:YVO4 laser ( = 1064 nm), frequency doubled Nd:YVO4 laser (532nm) or frequency tripled Nd:YVO4 laser (355nm). The pulse frequency was up to 60 kHz at an average power of 11 W, leading to an average pulse energy of 183 J. Mesas were

PAGE 132

132 formed in the epi layer by ICP etching in a Cl2/Ar plasma (10 sccm/5 sccm) for 1 minute at 150 W ICP power with a chamber pressure of 2 mt orr. Ti/Al/Pt/Au contacts 100 m square were deposited by e-beam evaporation and patterned by lift-off. These were annealed at 900C for 1 minute under flowing N2. The anisotropy and surface morphology of the resulting holes were examined by SEM [71, 74]. Typical average ablation rates of up to 50 m/sec were achieved, depending on the energy density of the laser pulse. The sidewa ll roughness of the laser drilled holes was comparable or better than those achieved with pl asma etching. The absence in the laser drilled features of vertical striations, resulting from transfer of mask sidewall roughness during conventional plasma etching, is clearly an ad vantage of the former method. Since the laser drilling works by ablation of the SiC, there is also less of a problem with micromasking leading to so-called grass on the SiC surface as there is with plasma etching. However, the 1064 nm laser light source limits the feature size up to 4050 m. For the features less than 40 m, it is difficult to control the dimension and shape. We found that the use of a 355 nm laser not only can reduce the dimensions of the drilled featur es, but also can improve the morphology of the drilled surfaces. As an example, Figure 8-1 shows SEM cross-sections of holes drilled at 1064 nm (left) and 532 nm (center). The latter produces an improved sidewall surface roughness, which is further improved at 355 nm (right). The SE Ms in Figure 8-2 of holes drilled from the front of the wafer at 1064 nm (above) and 532 nm (below) show improved roundness and cleaner holes at 532 nm with similar amounts of surface redeposition in the two cases [71, 74]. Figure 8-3 shows a SEM image of holes drille d in the HEMT/SiC substrates at 532 nm from the front (above, left) and b ack (below, left) of the wafer a nd also at 355 nm from the front (above, right) and back (below, right); minor top-side surface damage is visible on holes drilled

PAGE 133

133 from the front, while some epitaxial layer cracking is visible on holes drilled from the back-side. This is an indication of high thermally-induced st resses in the latter case and is clearly not acceptable for device fabrication [71, 74]. Figure 8-4 shows SEM at an angle of holes dr illed from the front at 1064 nm (top), from the back at 355 nm (center), and from the fr ont at 355 nm (bottom). The shorter wavelengths produce significantly cleaner and rounder holes, along with minimal surface damage. The left picture in Figure 8-5 shows the undrilled alloyed Ti/Al/P t/Au-based metallization in transmission line measurement (TLM) patterns. The dimension of the square metal pads is 100 x 100 m2. The center picture shows the same pattern on a substrate drille d with the 1064 nm laser. In this case, there is significant debris caused by the drilling, contaminating the areas surrounding the vias. The picture on the right shows a hole drilled with a 355 nm laser in the TLM region, without damaging the TLM pattern. Ta ble 8-1 shows the electrical data from the TLM patterns before and after the laser drilling at two different wavelengths. The sheet and contact resistance from a 355 nm laser drilled samp le are very similar to the undrilled reference sample [71, 74]. 8.3 Excimer Laser Processing Since the UV solid state laser performed much better than the IR solid state laser, it makes sense to try an even shorter wavelength using an excimer laser. A brief background section on excimer laser processing will be presented next, followed by results from a comparison of excimer and solid state laser drilled holes. 8.3.1 Background: Excimer Lasers An excimer laser is a gas-phase system that uses electronically excited molecules to emit high intensity pulses of UV light. The term was originally meant as a shorthand for excited dimmer, which implies two identical atoms. Mo st excimer gases are actually heteronuclear

PAGE 134

134 diatomics, thus the proper term, though not widely used, would be exciplex. A table of excimer molecules and their wavelength is listed in Table 8-2. Typically excimer lasers use a mixture of <1% halogen (F2, Cl2) gas, 2-5% rare gas (Xe, Kr, Ar), and the remaining 94-95% a buffer gas (Ne, He). The bulk of this discussion will refer to an ArF excimer gas at 193 nm. The primary reason for this is because this is the wavelength of the system that will be purchased, and therefore it is the most relevant wavelength. Note though that since an excimer laser is a gaseous source, the source gas and thus wavelength of the system can be changed by simply purging and filling the chamber with a different mixture. The optics must also be changed to be compatible with the new wavlength as well, but the entire assembly can be purchased for ~$5K [138]. The principle of operation is very complex. The buffer gas absorbs most of the initial excitation energy, then an electr on is transferred to the halogen, creating a negative ion that bond rapidly with the rare-gas positive i on to produce an ion pair in an excited state. As the molecule falls back to a dissociated ground state, it emits at a characteristi c wavelength (bound-free transition) [136]: Kr+ + F+ Kr Kr+F+ Kr Kr + F + Kr + h (248 nm) The gain in this system is inherently high (up to 2x104 in a single pass) since the density of excited states exceeds the unbound gr ound state population in the laser cavity. Excimer systems therefore do not require the high reflectivity cavity mirrors to provide optical feedback and can operate with a rear reflector and an uncoated front window. Excimer lasers are pumped by a fast electrical discharge. In this type of situati on, the excimer action must be achieved before breakdown and arcing occur, so pul ses are short (10-50 ns) [136]. All systems will degrade with time as the halogen gas slowly reacts and absorbing impurities form in the gas mixture. For this reason, a periodic maintenance issue with excimer

PAGE 135

135 lasers is gas changing. The performance can be restored by purging th e laser cavity and adding fresh gas. Gas lifetime is usually stated in term s of the number of pulses before the output power drops to half of the fresh gas output pow er. This depends on the wavelength (shorter wavelengths have shorter lifetimes) and laser design, but a typical range is 107-1010 pulses. For a system that is regularly used, this amounts to a gas change every ~2-3 months [138]. The beam size depends on the geometry of th e discharge electrodes in the chamber, but typically ranges from 5-10 mm x 10-30 mm with a gaussian po wer profile along the short axis and a fairly level power output al ong the long axis. Because it is an optical beam, this can be focused and scanned serially, or exposed through a mask for parallel patterning. This is a particular advantage of excimer lasers over soli d-state lasers, which are a point source. Excimer lasers are, in fact, advantageous over all other UV light sources because they operate by direct transitions between excited and ground states as opposed to DPSS lasers, which operate on principle of generating harmonics to triple or quadruple the frequency and are inherently inefficient. They are also advantageous over UV lamps as they emit directed power rather then requiring collectors and reflectors. It is for this reason that excimer lasers are considered promising for deep-UV lithog raphy applications [136]. Given the size of the beam, it is typically desirable to use some pattern generation process for image formation. There are several options for doing this. The first method is trepanning, which was described in the DPSS processing secti on. This is typically not used in excimer processing because the option is available to use masks. The masks are usually molybdenum plates with patterns formed in them reflecting the geometry of the image. For finer features, one can even use chrome on quartz lithography masks as well. As one can see, it is possible to pattern very complex features in a semi-parallel fashion with an excimer laser. The pattern can

PAGE 136

136 then be stepped across a sample (i.e. a 4 x 4 ar ray of squares can be stepped twice in each direction to create an 8 x 8 arra y). The third, and most intriguing, method is to coordinate beam scanning with stage translations to control th e vertical dimension as a function of x and y position. This can create stepped, sloped, or rounded walls [138]. The ultimate resolution of the beam will be determined by geometry, depth, optics, and material properties. Patterning features to micron scale resolution is quite possible with appropriate quality fo cusing optics, and 10 m is possible with a standard system. There will be an aspect ratio limitation as well though, with limitations from about 10:1 to 20:1, depending on the material. Hard materials such as SiC will require a large fluence, and there will be a demagnification factor associated with the projec tion process to achieve said fluence. This will thus limit the feature size and aspect ratio achieva ble. For polymers and thin films, it is much easier to achieve high resolution patterns because the material couples with the laser very well [138]. Upon irradiation with a UV lase r beam, four processes can occur: photolysis, electron ejection/ionization, electr on-hole pair generation, and heating. Photolysis refers to molecular bond breaking, i.e. the following reaction: AB + hv A + B The bond that is broken in the process can re present a bond between diatomic molecules, resulting in individual atoms A and B; a bond within a molecule, creating fragments; a bond between a molecule and a surface, resulting in desorption; or a bond within a molecular solid, which is broken. This is the mechanism for lase r ablation. This process occurs when the photon energy from the laser beam excites a molecular bond, as typically the case with UV wavelength

PAGE 137

137 lasers, and is the most relevant reaction for the purposes of this work. A schematic of the photoablation process is s hown in Figure 8-6 [136]. If the photon energy (6.7 eV for ArF) exceed s the work function (3 -5 eV) of a metal surface, a photoelectron will be ejected. This is relevant particularly with metal surfaces, and the ejected electron can actually se rve to enhance a chemical etch by assisting in desorption of reaction products. Electron-hole pair generation occurs in semiconductors when the photon energy exceeds the bandgap of the material. Th e effects can be the same as photoelectron emission in metals, but it can also have potentially implications for reliability testing as the additional electron-hole pairs on a biased device w ill cause a higher current level and more stress on the device [136]. Sample heating is also a relevant process, a nd occurs when photon energy is converted into excited electronic states, which decay into ground state atomic vibrations, thus raising the temperature. Each pulse is able to rais e the temperature very rapidly (up to 1010 K/s) because a large amount of energy is imparted into a small volume in a short time. This is complementary to the ablation process as it forces the sublima tion or desorption of nonvol atile ablation products. At UV wavelengths, most of the photon energy goes into exciting molecular bonds rather than pure heating. The temperature can get very high on a per pulse basis (AC temperature rise), but the repetition rate is relatively low that the sample always reaches equilibrium temperature between pulses (DC temperature rise). For high pu lse rates or long wavelengths, such as in a DPSS laser, the AC temperature may never reach th e ambient value, leading to a rise in the DC temperature of the sample. This can cause damage to the material [136]. A distinction must be made between ablation and etching. In the context of laser processing, ablation refers to material removal purely by the interaction of the laser beam with

PAGE 138

138 the material, while etching refers to the laser ex citation of a chemical etchant for enhancement of the etch. To this end, pure laser ablation has b een studied to date in this work. Laser etching could possibly be studied to selectively etch ma terial, such as enhanced etching of GaN using KOH [136]. Ablation can be broken down into 3 areas: high ly absorbing materials, weakly absorbing materials, and organics. Highly ab sorbing materials are capable of being intensely heated by the laser pulse. Ablation then occurs primarily by sublim ation. Metals fall into this category, and it is for this reason that metals t ypically machine poorly in UV systems. There is usually an area around the periphery of the patter n where there is beam-induced da mage in the form of an area that was melted during ablation. Another concer n is thermal conductivity. Metals prove difficult to machine because, though they are highly abso rbing, they tend to conduct the energy away very rapidly, thus requiring high fluence lase rs to continuously provi de sufficient energy on target. This is epitomized in an example of a Cu film. Bulk Cu is very difficult to machine with a UV laser and requires several J/cm2 for ablation to occur, however ablation will occur at 0.1 J/cm2 in a 1000 Cu film on an insulator such as SiO2. This is because the low thermal conductivity of the dielectric confines the laser energy to the Cu layer [136, 138]. In weak absorbers, such as semiconductors several processes may occur. First, basic photoablation can occur if the ener gy is larger than the bandgap, which will always be the case for an ArF laser. Second, multiphoton absorption can lead to electron ge neration and ionization, which will cause an avalanche breakdown process and generation of a plasma condition, resulting in material ejection. This is typically the case for insulators, which have a bandgap larger than the photon energy. Third, defects wi ll play a major role by acting as localized absorption centers, which can cause heating, and subsequent breakdown and ejection. Ablation

PAGE 139

139 of organics is typically photochemical, as organic bonds absorb str ongly in the UV region. Polymers are usually a low thermal conductivity mate rial, so there is heating occurring, but the primary ablation mechanism of photochemical [136]. 8.3.2 Experiments: Laser Type After settling on the UV laser as the ideal candidate for our system, we sought out a company with expertise in this area. It was he re that we began working with Jeff P. Sercel Associates (JPSA Laser). They were able to machine samples using a DPSS 355 nm system and a 193 nm ArF system. As an initial test, samples of the SopSiC HEMT material from Picogiga (see Ch. 4) were sent to them. While there is a wavelength variation be tween the two systems, the primary comparison is between the DPSS an d excimer laser performance. The wavelength effects are not as pronounced within the UV regi on as they are between visible and UV. Devices were not fabricated on these samples, so el ectrical measurements cannot be compared. Samples were processed from the top surf ace down to the bottom surface. For the DPSS laser, the wavelength was 355 nm, with a pulse rate of 100 kHz and laser power of 1W. The excimer laser was an ArF laser at 193 nm, with a power of 5W and pulse rate of 1 kHz. Optical images are shown in Figure 8-7, and SEM images in Figure 8-8, with DPSS on the left and ArF excimer on the right. It is clear even from th e optical images that there are burn marks on the surface of the DPSS laser sample. In addition, self-terminating effects discussed previously were witnessed with the DPSS laser due to the high pu lse rate and high aspect ratio. This placed a severe limit on the feature size for thick samples, with an aspect ratio of ab out 5:1. Given that the sample can be up to 500 m thick, a minimum feature of over 100 m is not acceptable. The excimer laser was able to form features down to 30 m in diameter in the same sample. In

PAGE 140

140 addition, there is less surface debris on the surf ace of the sample processed with the excimer laser, where it is just a small radius of material. To further examine the drilling processes, th e samples were cross-sectioned with a dicing saw. The SEM images are shown in Figure 8-9, with the DPSS hole on the left and excimer on the right, and low magnification on the top and high magnification on the bottom. The excimer hole shows some taper towards the bottom of the hole. This is characteristic of excimer laser processing, and is actually beneficial for the pur pose of the via hole proc ess because the tapered hole will facilitate metallization. The degree of ta per can be controlled with the laser parameters, and if perfectly straight holes are necessary, either a higher power or ove rablation can be used. The disadvantage of higher power is that some c ontrol over the etch rate is lost. The excimer holes show a smoother surface with less b eam-induced damage. There are obvious molten regions on the DPSS hole, visi ble at high magnification. As another test, an array of holes in InP a nd GaAs was formed. The optical image is shown below in Figure 8-10. An 8x8 array of holes were drilled through a 150 m thick wafer with an entrance hole diameter of 110 m and exit hole diameter of 100 m, with a pitch of 200 m. While these were processed serial ly, it would be possible to create a mask that could be stepped, such as the one described above (a 4 x 4 array st epped twice in both the x and y direction). This test showed very good feature control and reproducibility. Ultimately it was decided that the JPSA IX -260 machining system was perfect for the needs of the research group and the university. It offers the resolution and flexibility of materials that are necessary. It will be configured initially for 193 nm, but can be altered to perform at any excimer gas wavelength. In addition, high-resolu tion optics are available to achieve micron-level resolution. It has been fitted with an air-bearing stage that provides 0.1 m resolution and 1 m

PAGE 141

141 repeatability over the entire 6-inch range of m ovement. A system has been ordered and will be installed in spring 2008.

PAGE 142

142 Table 8-1. Electrical data demonstrating the e ffect of laser drilling on etching and contact properties Process Drill Direction Rs (/sq) Rc (-cm2) Mesa Depth () Reference No drill 368 23 2.97 x 10-5 950 1064 nm drilled Back to front 644 5.74 x10-6 350 1064 nm drilled Front to back 421 1.53 x10-5 0 355 nm drilled Back to front 361 44 6.33 x 10-5 945 355 nm drilled Front to back 362 22 2.83 x10-5 945 Table 8-2. Excimer laser gases and wavelengths Wavelength (nm) Emitting Molecule Relative Pulse Energy 157 F2 0.05 193 ArF 0.6 222 KrCl 0.1 248 KrF 1.0 308 XeCl 0.8 351 XeF 0.5 Figure 8-1. SEM cross sections of holes drilled at 1064 nm (left) and 532 nm (center)

PAGE 143

143 Figure 8-2. SEM images of hole drilled from front at 1064 nm (above) and 532 nm (below)

PAGE 144

144 Figure 8-3. SEM image of holes dr illed at 532 nm from front (above left) and back (below, left) and 355 nm from front (above, righ t) and back (below, right)

PAGE 145

145 Figure 8-4. SEM at an angle of holes drilled from the front at 1064 nm (top), from the back at 355 nm (center), and from front at 355 nm (bottom)

PAGE 146

146 QuickTime and a decompressor are needed to see this picture. Figure 8-5. Photographs of TLM patterns with Ti/Al/Pt/Au based metallization on an undrilled sample (left), drilled with 1064 m laser (center), drilled with 355 nm laser (right). Figure 8-6. The UV photoablation process.

PAGE 147

147 Figure 8-7. Optical microscope images of front and back of via holes formed in poly-SiC substrates with a DPSS laser (left) and excimer laser (right) from the top semiconductor surface (top) and bottom SiC surface (bottom) Figure 8-8. SEM images of via holes formed in poly-SiC substrates with a DPSS laser (left) and excimer laser (right) from the top semiconductor surface.

PAGE 148

148 Figure 8-9. SEM images of cross-sections of vi a holes formed in poly-Si C substrates using DPSS laser (left) and excimer laser (right) at low magnification (top) and high magnification (bottom) Figure 8-10. Optical image of a rray of holes formed in GaAs

PAGE 149

149 CHAPTER 9 ELECTROLESS METAL DEPOSITION 9.1 Overview While the electroless metal deposition project was initially studied for metallization of via holes in SiC, the most interesting device applicat ion that has emerged has been its use for hard mask deposition for dry etching. To this end, the following section will provide an overview of the fundamentals of Ni deposition, followed by an overview of a photovoltaic device that has been fabricated in SiC using electroless Ni as an etch mask. 9.1.1 Nickel Plating Electroless (or catalytic) metal deposition is a subset of the general category of nonelectrolytic metal coating processes. All me tal depositions from solution follow the same general mechanism, in which the metal in solution is reduced by a source of electrons to the solid state on the surface to be coated, shown in the following reaction: M+x (aq) + xeM (s) In a typical electrodeposition process, this reaction is quite straightforward, with the electrons being supplied by an exte rnal power supply. In this case oxidation occurs at the anode, producing the metal ions from a solid source in to solution, and at th e cathode the ions are deposited onto the surface of interest by the reduction react ion given above. An electroless plating method follows this same principle, but the reactions occur simultaneously in solution with no external source required. This process is, therefore, cons iderably more complex and not completely understood as of yet [139]. Nickel is a metal of interest both for via hole filling and as an etch mask for SiC during dry etching. Since the etch rates are so slow, the mask material mu st be quite robust, since it will be sputtered away during the etching process. Th is calls for a very thick mask, approximately 10

PAGE 150

150 m. This cannot be deposited using standard microelectronic fabrication methods such as evaporation or sputtering, but it can be acco mplished using electroless deposition. [139] The procedure for electroless deposition is st raightforward. First, a nonmetallic substrate is cleaned and etched. The substrate is then activated and sensitized in a tin and palladium process, which is the catalytic activator for the electroless bath. The Pd surface initiates the deposition reaction. At this point, the process becomes autocatalytic, and small islands form around the adsorbed Sn/Pd sites, spreading la terally until they meet, forming a continuous conductive surface. Electroless nickel has actually been proposed as an alternative to copper, offering the advantage of better stability and process development [139]. The electroless nickel process was developed industrially in the 1950s. It is widely used in industry as a coating process due to the wear and corrosion resistance of Ni. The standard reducing agent for Ni is sodium hypophosphite (NaH2PO2). If the Ni salt and sodium hypophosphite are allowed to react, the Ni is precipi tated as a sludge, which is of little use. An organic acid is typically added as a complexing agent and buffer. It was first shown by Scholder and Heckel in 1931 that the product formed in this reaction is actually a Ni-P alloy, with the phosphorous inclusion coming from the decomp osition and incorporation of the hypophosphite ion in solution. One mechanism for the reaction is shown in the following reactions: H2PO2 + H2O HPO3 2+ 2H+ + H(acid solution) H2PO2 + 2OHHPO3 2+ H2O + H(alkaline solutions) Ni2+ + 2HNi + H2 This mechanism explains the observed simulta neous reduction of nickel and evolution of hydrogen gas. Many other mechanisms have been proposed, including si multaneous catalytic and electrochemical processes. This is a testament to the complicated nature of the process [139].

PAGE 151

151 Acid solutions present many benefits, includ ing higher deposition rates, greater stability, and improved physical characteristics of the de position. The general reaction is as follows: Ni2+ + NaH2PO2 + buffers, complexors, stabilizers (Ni + P) + H2 + NaHPO3 This reaction takes place on a catalytic surface, either activated using Sn/Pd or an acidactivated seed metal surface. Iron has been discusse d as a seed metal, but it is a deep level trap for most semiconductors, making it of little use. Also discussed have been Pt and Ni. The most common salts used for the Ni source are the chlori de and sulfate, in conc entrations of 4-10 g/L. Sodium hypophosphite is virtually the only reducing agent for Ni. Th e reduction potential for Ni is .2 V, so this reaction requires a strong re ducing agent. The buffers and complexors are typically single ingredients, and almost all are or ganic acids. These stabi lize the pH of the bath and mildly complex the Ni. Again, moderators and wetting agents can be added if necessary. Temperature is the most important factor that influences deposition rate, with essentially no deposition occurring below 50 C, and most baths operating between 80 and 100 C [139]. Alpha-Voltaics An alpha-voltaic device is very similat to a solar cell, except it operates by producing power from alpha-particle radiati on rather than solar (visible) ra diation. This is particularly useful for portable power supply applications, in which a radiation source is coupled to the devices, thus producing a completely self-contai ned long-lifetime power supply. Alpha-voltaics are particularly attractive over be ta-voltaics due to the less strict shelding requirements for alpha particles. The device consists of an alpha ra diation source coupled with a semiconductor p-n junction, which will collect the electron-hole pairs created by the particle as it penetrates the material. The limiting factor in the development of these devices has been the radiation damage to the semiconductor. While the radiation sour ce is stable over a long time period (Americium-

PAGE 152

152 241 is typically used, which has a half-life of 432.7 years), the semiconductor is damaged within a matter of hours, thus reducing the power output. To this end, SiC is particularly attractive due to its inherent resistance to radiation damage. A schematic of such a device is shown in Figure 9-1 [140]. 9.2 Electroless Plating Procedures Initial testing used a basic solution formulated in the lab. It operates on the principle of the reactions described above. Most industrial solutions are based on this on a fundamental level, with many proprietary chemicals added to improve stability and to tailor the solution to specific applications (i.e. high rate, selective deposition, etc). While this is useful in determining the fundamentals of the process, it is impractical for real processing because the solution will be inherently unstable. Therefore when investigating a plating process for a real application, such as the MIPS device described above, commercial solu tions were used. Results and observations from both are described. 9.2.1 Solutions Made In-House Electroless Ni deposition on Si has been demo nstrated using the following bath conditions: NiSO4 (29 g/L), NaH2PO2 (17 g/L), sodium succina te (15 g/L), and succinic acid (3.2 g/L). A 500 mL solution was made of these components, as well as 200 mL solutions of the following two pretreatment solutions: SnCl2-HCl (0.1 g/L-0.1 mL/L) and PdCl2-HCl (0.1 g/L-0.1 mL/L). Initially, various preparation methods were used to determine the best method of initiating the reaction. If the aqueous Pd pretreatment is used, then, of course, Pd will adsorb on all surfaces contacting the solutions, causing Ni to plate everywhere. Since we are searching for a patternable plating process, this is clearly undesirable. Ev aporated Au, Ni, and Pt (1000 ) were also investigated, and it was found that only Ni functioned as a catalytic su rface to in itiate the reaction. Using sputtered Ni as a seed metal, fo llowed by activation in the pretreatment solutions

PAGE 153

153 for 2 minutes each, with rinsing between, then plating at ~60 C and a pH of 5.8, a deposition rate of approximately 2 m/h has been demonstrated [141-143]. Initial efforts resulted in de lamination of the film. This was first thought to be caused by stress due to the high plating rate. The dela mination phenomenon was still observed even when no plating was taking place. Annealing could help with adhesion, but there was still film peeling around the edges. The plating bath is slightly acidic, so when plating for hours, it will slowly attack the native oxide on Si. This will undercut the metal, resulting in delamination. Surface preparation is, therefore, essentia l. The oxide must be removed from the substrates by etching in BOE in the case of Si. To ensure there is no oxide on the surface between cleaning and deposition, a cleaning step was used in the sputtering system. This step biases the sample chuck, causing the ions to essentially sputter the chuck in stead of the metal. This will remove all oxide and result in the best adhesion. The next issu e is oxide formation on the metal surface. It has been observed by testing the sel ectivity to bare Si vs SiO2 that even activated SiO2 will not plate. The same trend appears to follow for most metal oxides. Before plating, the metal surface must be cleaned to remove the oxide. Several methods ha ve been investigated to protect the Ni seed metal surface. The first method is simply by etching the NiO in NH4OH, which is selective to NiO over Ni. The procedures and times for this pro cess have not been investigated in detail. The method currently used involves sputtering a thin la yer of Ti on top of Ni. This will protect the surface until plating. The Ti and TiO2 that form can both be etched in BOE, and then rinsed and put straight in the plating bath without being dried, which would expose it to potential oxidation [144]. The next step is to make the process pattern able. If a seed metal is activated and then selectively removed in a lift off process, the pretreatment will follow on those areas, leaving

PAGE 154

154 activated seed metal areas, which will plate at a rate varying with temperature, and the nonactivated substrate, which will not plate. Wo rk is in progress to optimize this process. Preliminary results show that an increase in pla ting rate and decrease in plating selectivity is observed by increasing the temperature to ~90 C; therefore, the most optimal temperature appears to be ~70-80 C. Studies of the pretreatment proce ss show that different seed layers require different pretreatment pro cesses. It appears that the SnCl2 process may actually have a negative impact on the Ni process with seed metal, with improved uniformity, rate, and initiation, all at lower te mperatures being observed with simply a 30 second PdCl2-HCl pretreatment. The pretreatment step is critical, si nce if the sample is pretreated too long, there will be bridging of the features. The plating rate is highly temperature dependent, with observed rates of ~2 m/h at 60 C and ~7 m/h at 75 C. This process has been used to form a pattern consisting of a uniform layer of Ni with holes pa tterned that could be etched, as shown in Figure 9-2. The films were deposited at varying temper atures and, therefore, have varying thickness [144]. The plating bath as formulated is highly unstable. The deposition rate is temperature dependent and pH dependent, which are conditions for potential runaway reactions. Note that in the reactions presented above in acid solution, 2 H+ ions are evolved and only one Hion. Therefore the pH of the solution will decrease as the reaction pr oceeds. The reaction is buffered by the organic acid, but if the buffering capacity is exceeded, the pH can rapidly decrease, accompanied by a temperature increase, culminati ng in a runaway reaction where the all of the nickel precipitates as a sludge, accompanied by si gnificant evolution of hydrogen. This entire process can occur in less than one minute. The reaction can be stopped by making the solution neutral again through the addition of NaOH [144].

PAGE 155

155 The next step is to repeat these results on Si C. The transition proved quite difficult. Initial efforts to duplicate the process were failures in that the metal film always delaminated. Ultimately, it was determined that surface conditi ons are absolutely critica l to the process. The SiC samples from one company resulted in delami nation after less than on e minute in the plating bath, while samples from a different company, wh ich will have a different surface condition, will plate easily. It was determined that the homemade solution is too unstable for a reproducible process. Commercial plating solutions are readily available and testing has produced stable and reproducible results requiring cons iderably less solution and effort. 9.2.2 Commercial Plating Solutions The first commercial plating solution invest igated is known by the trade name Nickel-B from Transene. This solution was chosen because it resembled the solution produced in the lab in that it requires an activator (therefore is selective) and grows Ni films at a very high rate (desirable to reduce process time). A water bath in a double beaker system was used to maintain a constant temperature of 64-66 C, and a 31:1 solution of activator to Nickel-B was used. A growth rate of 0.2 m/min was observed. This film actually gr ew too fast and resulted in a very strained film, which peeled off after about 1 m was grown. Based on the observations from the Nickel-B ex periments, it was determined that many of the problems were due to the fast growth rate. A different solution, known as Nickelex, also from Transene, was recommended due to the slower gr owth rate. The primary disadvantage for this solution is that it is not selective and a seed la yer is required, thus some work is necessary to make a pattern in the film. The seed layer wa s e-beam evaporated Ti/Ni (200/800 ). The Ti layer is meant to improve adhesion. Since the film requires a clean seed layer, the samples must be acid etched to remove any native oxide and create a clean Ni surface for plating. The etch

PAGE 156

156 uses a 3:1 HNO3:BOE solution for 10s. This will selec tively remove NiO. The plating then proceeds at 65-70 C using a double beaker water bath to control temperature, and the plating rate is 50 nm/min. A major drawback to using such long processing times is that bubbles form on the surface as H2 gas is evolved as a reaction product. Th is results in locali zed pitting of the surface. The bubble effect can be reduced by directing solution over the surface using a pipette every 3-5 minutes, which is the same time scale as bubble formation. More advanced recirculators could be purchase d for large-scale implementation. This process can be made patternable using the same method as electrop lating operations deposit a seed layer over the entire surface, then pa ttern the areas to be plated with photoresist. The areas opened with photoresist will then be plated up and the areas covered will not. This method is preferred to performing lift-off and then sele ctively plating because the form er method results in a more uniform electric field on the surface and thus mo re uniform plating. The latter method provides no effective charge dissipation a nd thus each nickel pattern will have a different surface potential and thus plating rate. Other problems encountered was that, much like previous observations, the solution will deposit on any rough su rface, hence a film will grow on the rough back side of Si test wafers and on any scratched surfaces in th e plating beaker. All components must therefore be cleaned thoroughly in an aggressive aqua regia etch before plating [145]. The seed layer can be removed by plasma etching. Conveniently the plasma etch conditions used for SiC etching contain a sufficient physical compone nt that the seed Ni can be easily etched away, while the plated Ni is sufficien tly thick that it can withstand the etch to the desired depth. The plasma etch was performed in an ICP etch system using a SF6/O2 (25/2.5 sccm) plasma. The ICP power was 500 W and the RIE power was 200 W. The SiC etch rate was 200 nm/min [145].

PAGE 157

157 For the MIPS project, this allowed for formation of 8 m mesas in SiC (etching down to the n-type layer). The Ni can then be stripped very rapidly and selectively using an aqua regia etch (1:3 HNO3:HCl). Further processing involves the deposition of the n-type ohmic contact, which is Ni, patterened by lift-off and annealed at 700 C. The p-type ohmic contact is also deposited by evaporation, and consis ts of Al/Ti/Au, annealed at 200-300 C. A passivation layer of 1000A SiNx is deposited by PECVD. This represents the entire device structure, shown in cross section in Figure 9-4. An optical image of a 10 finger device is sh own in Figure 9-5 [145].

PAGE 158

158 Figure 9-1. An -voltaic battery

PAGE 159

159 Figure 9-2. Optical microscope pi ctures of Ni plated on Si (4 m on top, 11 m on the bottom), with exposed holes

PAGE 160

160 Figure 9-3. Photo (top) and optical microscope image (bottom) of Ni hard mask for SiC trench etching.

PAGE 161

161 Figure 9-4. Cross-sectio n of SiC MIPS device

PAGE 162

162 Figure 9-5. Photo (top) and optic al microscope image (bottom) of MIPS device fabricated using an electroless Ni etch mask.

PAGE 163

163 CHAPTER 10 FLIP-CHIP BONDING 10.1 Overview Flip-chip bonding is one of the biggest advances in packaging technology. It has been used primarily as a method of packaging high performa nce electronic components to achieve shorter bond lengths, smaller chip size, and a higher I/O density. A new trend that is emerging is the use of flip-chip bonding for device integration, par ticularly for MCM electro nic devices and MEMS devices. This process involves individually fabricating and optimizing components (i.e. an amplifier and a sensor) and then integrating them using flip-chip bonding. This can be viewed as a system-on-package (SOP) approach, rather than system-on-chip (SOC) approach, which involves directly fabricating all components on on e chip. The disadvantage of the SOC approach is that process integration is particularly di fficult, such as integration of Si and compound semiconductors on one chip, and typically performance of one or both components must be sacrificed in order to achieve the desired integration. Standard flip-chip processing has been discussed in Chapter 3. This chapter focuses on the usage of flip-chip technology to integrate electronics component s (in this case a VCSEL) and MEMS (a DBR structure) to create a hybrid device. The goal of this work is to create a tuneable filter. The DBR structure is a GaAs/AlGaAs superl attice and the emitter is a Si-based reflector to demonstrate the concept. The ultimate goal would be to fabricate a tuneable filter where the DBR structure is electrostatically actuated using me tal arms. By raising and lowering the DBR, one could change the resonant frequency of the device, thus creating a tuneable filter. This approach is a very hot research topic and many such in tegration schemes have been proposed. Another such idea involves bonding a phot odetector directly to an amplifier chip [146-148].

PAGE 164

164 The device structure is shown in Figure 10-1. It consists of an Au reflector on a Si substrate, surrounded by poly-Si actuator arms bonded to a DBR structure, which is a GaAs/AlGaAs superlattice. The bond pads are 50 m squares. The DBR structure is formed by an RIE etch, and there is a 5x6 ar ray of devices on a die. After bonding, the substrate is removed from the DBR chip by etching a sacrificial AlAs layer in HF. The actuators are then released by another HF etch. Different HF c oncentrations are used to selec tively etch each sacrificial layer [146-148]. 10.2 Bonding Studies Initial tests on this project involved findi ng a compatible materials system for bonding. While each component can be fabricated separate ly, a compatible bonding material that could be deposited in-house needed to be studied. Seve ral candidates were propos ed, including In, Au, and SU-8 polymer. The first efforts focu sed on metal bonds. In each case, bonding was performed at a matrix of conditions of temper ature and pressure. For the metal bonds, large area die were used for bonding, then cross-sectioned a nd observed in the SEM. This was used to look at the amount of voids that formed and to look at the grain size. This is particularly important because the In film was rough as-d eposited, so there is a considerable amount of voids formed at the interface. In another test, one side was totally removed, equivalent to the post-processing that would be experienced by the device chip, a nd surface scans were performed to study the roughness at the in terface [146-148]. In the case of the polymer SU-8 bonding, diffe rent tests were performed. The polymer was chosen because the range of temperatures and the reliability was more compatible for the desired process, and polymer was used because only a m echanical bond was needed. In this case, the materials were bonded at a temperature and pressure matrix and the die were then pried apart.

PAGE 165

165 Though systems are commercially available to sh ear die apart and measure the stress, none are available at UF. To that end, the quality of the bond was more a qualitativ e assessment as to at which temperature the bond was hardest to shea r off. After the die we re split, they were observed under the microscope to look for the deformation of the pad and any delamination. Based on these images, an optimized bonding progr am was built and devices were successfully bonded. All bonding tests were performed at UF, while the design and pre/post processing and testing were performed at AFRL [146-148]. 10.2.1 Metal Bonding Again, the first systems studied were In-A u and Au-Au pads. Since the actual device die were fabricated at a foundry at some time and cost it is desirable to preserve this material as much as possible. Therefore, test die were fabricated from Si with either just the bond pads patterned or just a metal film deposited, and the DBR die were fabricated as they would be for the real device. After bonding, the package underwent the rel ease processing, and the mesas were observed for cracking or complete breakage [146-148]. The first material set tested was Au-Au. This is the simplest system to fabricate. Experiments were performed to test the bonding over a pressure range of 1-1.5 kg and a temperature range of 300-400 C. It was ultimately determined that the optimal pressure was 1.3 kg and temperature was 375 C. While this was moderately su ccessful in testing with dummy die, testing of real samples yiel ded either cracking or complete destruction of th e die. This is because the high temperatures and relatively high pressures from the bonding process were not compatible with the MEMS process [146-148]. The next material set tested was In-Au. This yielded considerably bett er results in testing dummy die. A temperature range of 162-248 C and pressure range of 200-400g were tested.

PAGE 166

166 Cross-section SEM was used at first to study th e quality of the bond form ed. A typical result is shown in Figure 10-2. From the image, one can see that there is a significant quantity of voids in the In layer, and a large grain size (>2m). The bond program was optimized as a stepped pressure profile. This will allow for the In to soften at a lower pressure (200 g) as the temperature is increased, then a hi gher pressure (400 g) is used to perform the controlled collapse above the melting point (200 C), followed by cooling. The low bonding pressures can be achieve using In at the bonding metal because it is quite sticky, and it has a low melting point, making it an ideal candidate. In is usually deposited using lift-off of a thermally evaporated resist. Films are usually 5-10 m thick, so a very thick resist is required for the process. Ultimately these films did not stand up mechanically to the release steps either. While the device remained intact due to the low temperature and pr essure, the HF release step is not compatible with In [146-148]. 10.2.2 Polymer Bonding Since no electrical contact is required, polymers can be i nvestigated. SU-8 has been demonstrated as having a low temperature and pr essure bonding requirement, and is resistant to HF. As a test study, bond pads were deposited on Si, the pieces were bonded at a matrix of temperatures and pressures, 105-165 C and 800-1200 g. After manual separation, both die were observed in an optical microscope, as shown in Figure 10-3. One can see that at the lowest temperature, the bond quality was poor as each set of pads adhere d to the substrate, but were easily separated where they were bonded together. At the highest temperature, on the other hand, the pads bonded to each other, but deformed and be gan to shift. At the middle temperature, there is a solid bond formed, but minimal deformation. It was therefore determined that the optimal bonding profile is 135 C, 1 kg pressure, for 10 minutes. This particular temperature was chosen

PAGE 167

167 because it offered the lowest temperature to stil l make a solid bond that exceeded the strength of the bond form the pad to the substrate. Real devices were then bonded using this method and successfully released. An optical microscope im age of the device die is shown on Figure 10-4, and an image of a bonded die after rel ease is shown in Figure 10-5 [146-148]. 10.3 Device Demonstration After device fabrication, th e die were bonded to carriers and wire bonded for testing. Reflectance was measured using a wh ite light source directed at th e reflector structure. A tuning voltage range of 0-10V was invest igated and a reference sample c onsisting of Au evaporated on Si was used as a reference reflectance sta ndard. The results are shown in Figure 10-6. A wavelength tuning range of 930-990 nm was ach ieved. Though the design range matches the actual wavelength range, the design actuation vo ltage vs wavelength curve does not correlate well compared to the real measurement. The reason for this is mainly due to stress factors that were not accounted for. The DBR was not plan ar, and actually formed a concave cavity. This was verified using interferometry, with the da ta shown in Figure 10-7. When a simulation was performed using a nonplanar cavity, the results correlated well with real device results, shown in Figure 10-8. All design, simulati on, and testing was performed at AFRL, while the bonding was performed at UF [146-148].

PAGE 168

168 Figure 10-1. A tuneable filter device Figure 10-2. SEM cross-section of In film bonded to Au film

PAGE 169

169 Figure 10-3. Optical microscope of test die after separation at 165 C (top), 135 C (middle), and 105 C (bottom) Figure 10-4. Optical microscope images of actual MUMPS die (left) and DBR die (right)

PAGE 170

170 Figure 10-5. Optical image of a bonded device Figure 10-6. Device testing results using a white light source compared to design simulations using a planar cavity

PAGE 171

171 Figure 10-7. Interferometry data showing the curvature of the DBR Figure 10-8. Device testing results using a white light source compared to calculations based on a curved cavity

PAGE 172

172 CHAPTER 11 FUTURE WORK 11.1 Processing Studies In pursuit of a functional prototype of the 3-D package concept described above, the goal for HEMT fabrication is to de sign and fabricate high power devi ces compatible with the laser drilling process. Work must be done to investigate fully the effect of laser drilling by examining HEMT performance before and after laser drilling. On the packaging level, work will also include investigation of novel oxide dielectrics for device passiva tion and packaging. To further push the state-of-the-art of AlGaN/GaN HEMT technology, e-beam lithography based T-gate or mushroom gate process needed to be used to achieve very small gate contact area, which improves the RF characteristics a nd, at the same time, have th e large trace on top, minimizing gate resistance and further enhancing the RF performance. 11.2 Reliability The topic of reliability is very critical for the realization of any practical uses of the nitride HEMT technology. Typical rf stre ss test systems can be quite e xpensive ($250K for 8 channels, up to $1M for >50 channels), so we are in the pr ocess of building one reliability test set at UF using a design adapted from Sandia National Labs. After completion of this system, there is a tremendous amount of interesting dc and rf reliability tests that can be done as there is little work performed to date. Primarily, there should be an investigation into the failure mechanisms of AlGaN/GaN HEMTs, by stressing devices under a combination of DC, RF, and thermal stress. FIB and TEM analysis can be performed to look at the semiconductor layers. Typically, for the conventional III-V devices, failure rate can be linearly related to th ermal stress, therefore accelerated testing is performed at elevated temperat ure. This is not the case for rf nitride based

PAGE 173

173 HEMTs, therefore the ultimate goal of such a test system would be to derive an accelerated test program.[ref] Beyond this goal, the system will be of great use for any novel device structure. At present, the temperature controller boxes are assembled and the final components for the RF driver boxes. DC power supplies for source -drain bias must be purchased, along with the required PCI boards for monitoring power outpu t and supplying gate bias control. The programming must be performed in LabView to automate the system to control DC bias, RF power, and temperature, while monitori ng DC current and RF power output. 11.3 Gas Sensors A significant amount of work has been performed in our group on gas sensors. At present, the most marketable sensor is the wireless hydrog en sensor. The sensitivity range and response is competitive with other sensor technologies. One advantage of using GaN HEMT sensors is that the current change will be directly related to the gas concentration. Therefore work should be done to correlate the relationship and integrat e into the programming in the data acquisition program. This would then allow the user to se t a given threshold valu e in terms of a real parameter, rather than just hard coding a curr ent change threshold that will not have much meaning to an end-user. 11.4 Simulations The work to study heat transfer through the 3D package has been wrapped up, but finiteelement simulations will be important for all new devices at the packaging level. In particular, this type of simulation could be performed on hy drogen gas in a room to look at the gas flow pattern and better choose sensor installation sites, or even re design the sensor package to incorporate a concentrator to better detect small leaks. It will also be beneficial to revisit thermal simulations after reliability studies have been performed. The reliability study will give a good

PAGE 174

174 idea of the device response to thermal stress, which can be fed into a thermal simulation to intelligently design a package to search for problem areas and remove heat as necessary. 11.5 Laser Drilling Laser drilling has been studied thus far to ev aluate systems for purch ase and installation at UF. The installation of a JPSA system will occur January 2008. This system will be a great asset to the research group. Laser proce ssing has been used primarily in i ndustry, and to that end, there has been little published, and little studied in depth. First of all, this piec e of equipment can serve as a versatile rapid patterning system. For rapid pr ototyping applications, there will be no need to make a mask or go to a special job shop for micromachining applications. A wide range of materials, from polymers to ceramics, semiconducto rs, and metals can be patterned to various depths using this system. This opens up the po ssibilities for collaboration with new groups in Mechanical Engineering, for example, who woul d have great interest in creating controlled microstructures on a rigid surface. A second area of major importance is the analysis Again, since this has been an industrial application, little research has been performed to study the laser machining process. While the ablation mechanism can be studied through simula tions and high speed photography, little true materials analysis has been performed to study la ser-induced damage in pa rticular. The extensive analysis equipment available at the Major Analy tical Instrumentation Center at UF will allow future group members to perform a full battery of tests on materials after laser ablation. In particular, XPS and TEM will be useful, with TEM used to look at the thickness of a damaged layer after laser processing and XPS used to probe the surface, to look at changes in the bond structure after ablation. AES will also be a usef ul tool to study redeposition to determine the nature of redeposited species and come up with methods of protecting th e material from debris contamination.

PAGE 175

175 11.6 Electroless Plated Metal While this phase of the project has been completed, this process will prove very useful for device applications. Dr. Brent Gila in Materials Sc ience Department has used this process with a commercial plating solution to deposit films se veral microns thick with good uniformity and reproducibility. The device structur e is particularly interesting in that the device uses SiC semiconductor, and Ni forms a good ohmic contac t to this material. Therefore, the device structure uses the nickel film as an etch mask and then can be left on to form self-aligned ohmic contacts. Further work can be done to integrat e this concept to device processing for niche applications. 11.7 Flip-Chip Bonding Work has been completed on most flip-chip bonding projects at present. This is still an active area of research, and is particularly impor tant for 3-D integration. There are two primary areas in which work can be done. First is the device level packaging process for RF HEMTs. Wire bondings introduce parasitic impedances due to the long tr ansmission lines and curved nature of the bond wires. These have proven very difficult to model for the design of matching circuits. Flip-chip bonded devices should have a shorter transmissi on line distance in a straight line path. This should improve RF performance of packaged devices. The installation of the SolderJet system will allow for reliable in-hous e bumping of devices, so a direct comparison between devices processed at the same time from the same wafer should be possible. The next area of interest is the physical design of the pack age to optimize performance and reliability. Flip-chip bonding can be used for phot odetectors to allow for back-side illumination, which will enhance the efficiency A particularly new area of study is component integration to create hybrid packages, similar to the 3-D project described in this work. The integration of CMOS, compound semiconductor devices, and MEMS is possible throu gh flip-chip bonding,

PAGE 176

176 such as the project studied in Ch apter 10. Further work on this type of packaging level project is possible to create novel device structures.

PAGE 177

177 CHAPTER 12 CONCLUSION W ide bandgap semiconductor devices present interesting challenges to those of us involved in the field. The group III-nitrides are a partic ularly new class of compound semiconductors and show superior properties fo r high speed, high temper ature, and high power applications. They will enable us to expand to areas we have never been with semiconductors before. GaN-based electronic devices are begi nning to see commercial production in 2006, but there is still work to be done to push the state-of-the-art particularly in terms of packaging and in taking advantage of emerging processing methods such as e-beam lithography. 3-D packaging is a novel concept that will be necessary, particularly in mobile technology. As real estate on circuit boards is decreased and the need for shorter interconnects becomes necessary, a 3-D approach can be taken to mini mize the chip footprint and interconnect distance and increase reliability. An especially novel conc ept involves integrating GaN-based high power amplifier devices with established Si-based CMOS technology. Considerable work must be done to solve the processing challenges presented to realize this stru cture, but the base technology has been well established in other fi elds. Clearly it is an exciting time to be involved in widebandgap semiconductor processing a time when Si and GaAs are being pushed to their theoretical limits, and nitrides are ju st beginning to show their potential.

PAGE 178

178 APPENDIX A EQUIPMENT OPERATION A.1 Purpose This chapter is meant to serve as a database of operating procedures for the equipment in Dr. Rens labs. A.2 Spin-Coating 1. Turn on spin coater with power switch. 2. Place the appropriate size chuck on the spinner 3. Press the foot pedal to activate spinner, ch eck spin speed on digital readout (in kRPM). Adjust using know if necessary. The time r eadout is on the left and is set to 30s. DO NOT CHANGE THE SPIN TIME. 4. Turn on vacuum pump, place sample on the chuc k, perform another test spin to remove any dust particles on the sample 5. Apply photoresist using disposable pipette. Only a small amount is needed, a drop about the size of the sample will be more th an sufficient (a drop the size of a quarter will coat an entire 4-inch wafer) 6. Press the floor pedal to activate spinning. 7. After the spinner stops, remove the sample with tweezers. Use the ones by the spinner so you do not get photoresist on your personal sample tweezers. 8. Shut down: turn off vacuum pump (some times the switch gets dust in it and quits working. If this happens, take the pump to Jim to have it cleaned) and turn off the spinner A.3 Mask Aligner (MJB-3) 1. Turn on N2, air compressor at source 2. Enter clean room and turn on corresponding ball valves 3. Turn on power to mask aligner and lamp power source 4. When lamp power source displays rdy, pr ess start to fire the lamp. cold should appear if the lamp started alright, otherwise it will go back to rdy. 5. Allow lamp to warm up (power level will di splay when it is ready). Prepare photoresist while waiting. 6. Turn on the vacuum pump, load mask on mask holder so that the chrome side will face the wafer, and press vacuum mask button to activate the vacuum system on the mask holder. 7. Load the mask holder into the aligner into the slot on the front. 8. Load the wafer by pulling out the slide, placing the wafer in the correct orientation, and sliding back into the system. 9. Put the wafer in contact mode to verify he ight. The wafer should not move in contact (both handles forward) but should be free in separation mode (back handle forward, front handle back) 10. Perform alignment in separation mode

PAGE 179

179 11. Place in contact, set correct exposure tim e, and press exposure to perform the exposure 12. Remove the wafer from contact (bot h handles forward) and develop. 13. Remove the mask by reversing step 6 14. Shut down system: turn off power to aligner, lamp, and vacuum pump, turn off air and N2 ball valves and sources A.4 Raith E-Beam Lithography 1. Load sample on the appropriate chuck. Use electrostatic chuck for full 4-inch wafers. Make sure at least one clamp is touching the sample for grounding. Make sure the sample is oriented correctly relative to the electron beam, NOT the ccd camera. The SEM image is rotated 90 clockwis e relative to the CCD camera. 2. Place the chuck in the system, log in to the software, sample handling window, click load to begin the loading process. This is automated. 3. When loading is completed, th e system will ask fo r a series of values. Typical values are accelerating voltage = 10 kV, aperture size = 30 mm 4. Set up alignment window with proper coordi nates (determine from GDS file if necessary) and set the activ e area in the GDS file. 5. Move to sample so that the edge is a pproximately under the pole piece of the SEM using CCD camera. This will put the starting po int near the sample, but not at risk of accidental exposure. 6. Turn on the SEM (beam on). Adjust focus roughly. 7. Use the SEM to find the first alignment mark Place the mark roughly at the center of the screen using the crosshairs. Read this point. 8. Use either the SEM or stage translations to find the second and third alignment marks. Read these points to the alignment window Click adjust bloc k to calculate the coordinate transform. I prefer to set up th e coordinates roughly first so you do not get lost on the sample. 9. Find a particle on the surface of th e resist. It should be small (<5 m) or have very sharp points (radius <500 nm). Use this to fine tune focus and adjust stigmation and aperture alignment. Remember that in the SEM, anything focused at high magnification will remain focused in low magnification. 10. Use the crosshairs at high magnification to m ove a clear reference point to the center of the screen. Perform write field alignment at this point. 11. Use the stage translation window to move to the Faraday cup on holder. Turn on the beam and in the beam current window click measure. This will read the beam current from the picoammeter. 12. Since the alignment marks have been found, use the lightning bolt icons to move back to each point and fine tune the alignment at a higher magnification. 10-15KX works well for 2-5 m marks. A good way to do this is to use the annotation to measure out half of the mark width in each direction fr om center using the crosshairs, then fit the mark to the annotation. 13. Set up the exposure, making sure the dose is set and calculating the other values, check the layer, and click on the adjust working ar ea button to ensure that the software will

PAGE 180

180 move the sample to the working area rather than just starting the exposure from where the chuck is located. 14. Begin exposure. The system should handle everything from here but periodically check on it. The time is about 1.5 times the calculated time. 15. Unload the sample by clicking on the unload sample button. The procedure is automated and will take about 10 minutes. 16. Remove the samples from the chuck and place the chuck back in the system. A.5 Plasma-Therm PECVD 1. Turn on system N2 cylinder 2. Start vacuum pump (master and blower switches) 3. Switch on power supply for the system (next to pump) 4. Turn on chiller: open water lines (supply and return, close cro ssover valve), press start/stop button 5. Check the system to verify utilities star tup air, water, nitrogen, +15V, 15V, +24V boxes are all green 6. Turn on interlocks at the bottom right of th e system in the following order primary mech pump, lock pump, turn switch to enable (machine power switch remains off) 7. Turn off hold button on screen and turn on standby button 8. Pump the system by going to the Service menu, then to Mainetnance > Pump > System. Use the Windows pull down me nu to access the system overview diagram if necessary, and check the pressure of the chamber. It should be 0 mTorr 9. Load the try into the chamber for hea ting. Go to the Service menu, then Mainetnance > Wafer Handling. Click on th e Load button. After the process is complete, click Exit. 10. Set the temperature of the chamber. Go to the Utilities menu, then Set Standby Temps. There is some offset in the software, so a setting of 283 C will correlate to 255 C on the temperature controller (lower le ft cabinet on the front panel of the system). Heating will take 2-3 hours. 11. Turn on the gases required for the reac tion. Go to the Service menu, then Mainetnance > Evacuate Gas Lines. C lick on the box corresponding to each gas line, wait for evacuation, then turn on the gas. A stable flow s hould appear. Begin with SiH4, and NEVER evacuate more than one line at a time. The CF4LK line is attached to the N2O cylinder. 12. Set up the system for the reaction. Go to the Service menu, then Manual Mode. Input the process parameters from the recipes as necessary. 13. Perform a test run Set the time to 2 min. Click on the Gas button to begin gas flow to the chamber, and wait for the flow to stabilize (boxes turn gr een). Click Pressure and wait for the chamber pressure and temperat ures to stabilize. Click RF to activate the RF power. 14. Click the Purge button to purge the cham ber with nitrogen after each run. Perform the purge at least 3 times for 1 minute each. 15. Unload the tray and vent the loadlock (Ut ilities menu, go to Loadlock > Vent)

PAGE 181

181 16. Load a sample and repeat the process. Ad just the process time as per the thickness requirements. It is best to run a test sample and check th e thickness using ellipsometry before running a real sample. 17. To shut down the system, begin by turning o ff the gas lines. Turn off the gas at the cylinder and then use the Evacuate Gas Line s procedure to completely evacuate the lines. 18. Close the gate valves on the system ( Utilities > Close Gates > System) 19. Set the temperatures back to room temperature 20. Turn off the Standby button 21. Close the interlocks and silence the alarm 22. Turn off the pump, power supply, and chiller 23. Turn off the system nitrogen. The last me ssage displayed should be System Nitrogen Is Low. A.6 Evaporator (CHA) 1. Turn on the roughing pump in the chase 2. Close the high vacuum gate valve with th e switch and turn off the ion gauge 2. 3. Vent the chamber with the Vent switch. Venting is complete when the chamber door can easily be opened. Turn off the vent switch when complete. 4. Load the sample and check the metal levels by opening the shutter and rotating through the crucibles. 5. Pump the chamber by pressing the Rough switch. 6. When the pressure is <25 m Hg, turn off the roughing switch and turn on the high vacuum switch to open the gate valve. 7. Turn off the roughing pump, wait 2-3 hours for the system to achieve high vacuum (<1x10-6 mTorr) 8. Turn on the chiller, wait ~ 10 minutes for it to warm up. 9. Turn on the power supply, wait ~10 minutes for it to warm up 10. While waiting, set the deposition parameters as per the log book 11. Turn on the high voltage (key and toggle sw itch) and the beam sweeper (toggle switch) 12. Use X and Y controls to adjust the beam position to the center of the crucible. The shutter will open after 2 minutes. 13. Monitor the time during deposition. Do not run the system for more than 5 minutes at a time. Always cool down for 5 minutes between depositions. 14. When deposition is complete, wait 10 minutes and then turn off the power supply in the chase. Wait another 10 minutes and turn off the chiller. 15. Vent the chamber, remove the sample, and pump the chamber again. A.7 Old Evaporator 1. Turn on roughing pump and turbo pump for warm-up 2. Close the high vacuum gate valve 3. Vent the chamber by turning on both nitrogen valves. 4. When gas is escaping the bottom of the bell chamber, venting is complete. 5. Use the hoist to raise the chamber up. 6. Load the sample, lower the hoist

PAGE 182

182 7. Close the nitrogen valves, open the va lve to the roughing pump and chamber 8. Wait for the chamber to pump down. Switch to the turbo pump by closing the roughing pump valve opening the valve when the pressure = 100 m Hg. Turn off roughing pump 9. Wait for the chamber to pump. Switch to th e cryo pump by closing the turbo pump valve and chamber valve and opening the gate valve when the pressure = 10 m Hg. Turn off turbo pump. 10. Wait 4-6 hours for the system to reach high vacuum (pressure < 1x10-6 mTorr) 11. Turn on chiller, wait 10 minutes for warm up 12. Turn on power supply, wait 10 minutes for warm up. Turn on the deposition controller and set the deposition parameters as per the log book while waiting. 13. Turn on the high voltage (key and push bu tton) and the beam sweep (push button). 14. Verify that the beam position is near the center of the crucible. The beam should not need adjusting. 15. Monitor time during deposition. Do not depos it for more than 5 minutes to avoid damage to the system. Always wait 5 minutes between depositions. 16. When deposition is complete, wait 10 minutes and then turn off the power supply. Wait another 10 minutes and turn off the chiller. 17. Vent the chamber, remove the sample, and pump the chamber again. A.8 Thermal Evaporator 1. Turn on the control panel, roughing pump and turbo pump 2. Open the ball valves for nitrogen an d house chilled water supply and return 3. Vent the chamber using the toggle valve at the system interface 4. Use the hoist toggle switch to raise the chamber top 5. Remove the glass chamber wall and clean with acetone if necessary 6. Load the boat with metal charges and repl ace the glass wall. The relationship between metal mass loaded and final film thickness is linear, and a calibration curve should be established with all new metals. 7. Lower the chamber top and begin roughing the chamber by turning the 3-way valve to the rough position 8. When the chamber pressure reaches 100 m Hg, change the 3-way valve to the foreline position, wait a minute to pump out the lin e, and then open the turbo pump valve 9. Turn on the high vacuum gauge. Wait for the chamber to pump down (<1x10-5 mTorr) 10. To deposit, turn on the current supply. Manually ramp up the current. Dwell for 1 minute at a low setting relative to the de position current for initial heating. 11. Slowly ramp the current to an intermediate setting and watch the metal charge for melting. Dwell until the charge is completely melted. 12. Ramp the current to the deposition level a nd open the shutter. Evaporate the entire charge. 13. Slowly ramp down the current supply, turn it off and wait 5 minutes for cooling. 14. Turn off the turbo pump valve, turn the 3-way valve to off, and vent the chamber. 15. Clean the chamber wall and pump the chamber.

PAGE 183

183 APPENDIX B ELECTRICAL MEASUREMENTS B.1 Purpose This section is meant to provide information on the standard electrical measurements that are performed to quantify HEMT performance. Background and system setup diagrams will be given when applicable. A thorough knowledge of electri cal testing is a useful tool. Just as much diagnostic information, if not more, can be gather ed from a round of electrical testing as from a full battery of analytical equipment. B.2 Ohmic Contacts B.2.1 Overview Transmission Line (TLM) testing is used to quantify the ohmic contacts. A typical transmission line consists of a series of ohmic contacts with varying spacing on one mesa. By measuring the current-voltage curve between each contact, one can calculate the resistance from Ohms Law (V=IR) as the simple slope of the line. The resistance is a linear function of the spacing between contacts, so from here one can fit a curve if these va lues are plotted. The resulting slope will be the sheet resistance (Rs), which is reported in ohms/sq. This is the resistance purely of the semiconductor material. In an ideal situation, th e intercept of this line will be zero, but since the contacts themselves ha ve some resistance, the slope will represent the transfer resistance (RT), which is reported in ohm/mm. Thes e factors are combined into a specific contact resistance (Rc), which is defined as Rs/RT, and is reported in ohm-cm2. A good ohmic contact will have Rc ~1x10-6 ohm-cm or smaller. B.2.2 Procedure An optical image of a TLM pattern is shown in Figure B-1 to demonstrate the terminals. Use the parameter analyzer and probe station for this test, as shown in Figure B-2. Set up the

PAGE 184

184 parameter analyzer for a diode (Vf-If) measuremen t. A typical test range is from V to 1 V. Probe the TLM pads, starting with the largest sp acing. Determine the slope of each line (the intercept should be zero). This can be done by eith er curve fitting in Excel after testing, as shown in Figure B-3, or directly measuring the tang ent on the parameter analyzer. The resistance is equal to 1/slope from Ohms Law (V=IR, therefor e R=V/I). Plot the resistance as a function of the gap between pads, as shown in Figure B-4. Pe rform a linear fit to determine the slope and the intercept of this line. The sheet resistance (Rs) is equal to 100*slope. The transfer resistance (RT) is equal to intercept*0.1/2. The 0.1 comes from the pad length, in mm, and the 2 comes from the number of pads. B.3 Direct Current B.3.1 Overview DC testing is performed to check for basic function of HEMTs. This primarily means checking for current saturation and gate modulation. To this end, two curves are typically taken source-drain voltage vs s ource-drain current at vary ing gate voltages and ga te voltage vs sourcedrain current at a constant drain voltage. The former measurement is used to check for saturation, and the latter used to check the gate characteristics. Ideally, the current will linearly increase at low voltages and then level off at saturation. In sapphire or Si HEMTs, the curren t will actually decrease linearly with increasing drain voltage past the saturation voltage. This is due to the sel f-heating of the device, and therefore is not seen in SiC HEMTs. By extrapolating the linear region forward and the saturation level back, the knee voltage can be found as the inters ection of these lines. Knee voltage is a function of gate to drain distance, and thus increases linearly with this ga p. The current level is typically normalized and reported as mA/mm or A/mm, where mm refers to the total gate width. A state-of-the-art GaN HEMT approaches the theoretical limit of 1.2 A/mm. This test is also used to check for gate

PAGE 185

185 modulation. In a depletion mode device, the gate voltage is modul ated from 0 in steps of -1 V until pinch-off is reached. The current saturation level should decrease with increasing reverse gate bias. Typically pinch-off occurs between -2 and -7 V, and if nothing is reached by -10 V, there is most likely a problem with the gate of the device. The second curve taken is the gate voltage vs drain current curve. Typically the gate voltage is taken from beyond pinch-off (-10 V, fo r example) to 0.5 V. One should see the drain current leveling off beyond Vg=0, so this is why it is taken to slightly beyond 0. The drain current should linearly decrease with increasing re verse gate bias as pinc h-off is approached, and be nearly flat at a very low current level (nA) beyond pinch-off. A line ca n be extrapolated back to the x-axis to determine the threshold voltage The transconductance (gm, reported normalized as mS/mm or S/mm) can also be determined from this curve. It is defined as dI/dV, so it amounts effectively to the first derivative of this curve. It can be calculated manually by calculating dI and dV between two consecutive points for th e entire curve. A good maximum gm is above 200 mS/mm, which implies a very sharp transition from on to off states. B.3.2 Procedure An optical image of a HEMT is shown in Figur e B-5 to demonstrate the terminals. Use the parameter analyzer and probe station for this te st, as shown in Figure B-6. Set up the parameter analyzer first for FET Vd-Id measurement. Th e drain bias range depends on the source-drain spacing, but 0-10 V is typically a good starting point. For a depl etion mode device, the gate should go from 0 to pinch-off in V steps. If th e pinch-off voltage is not known, -5 or V is a good starting point. A sample curve is shown in Figure B-7. After performing this measurement, cha nge the measurement mode to FET Vg-Id measurement. The drain bias will be constant in this mode, and should be set to just beyond

PAGE 186

186 saturation (5 or 6 V in the samp le curve) and the gate voltage should be set beyond pinch-off (10 V is typically good). A sample curve is shown in Figure B-8. B.4 Pulse (Gate Lag) B.4.1 Overview A pulsed gate test is used to look at tr aps in the semiconductor material. In this measurement, the gate voltage is pulsed from be yond pinch-off to on at a given rate (typically 1 kHz-1 MHz range) and at a given drain bias. The tr aps in the material, particularly surface states and buffer traps, will cause the system to respon d differently to a pulsed gate as opposed to DC operation. This measurement is used to generate the same curves as DC measurements so that a direct comparison can be made. A high density of surface states is manifested as a low drain current under pulsed conditions as compared to DC (under 80%). A high density of buffer traps is manifested as a shift in threshold voltage und er pulsed conditions (as mu ch as 2 V or more). B.4.2 Procedure A schematic of the test setup is shown in Fi gure B-9. A DC power supply is used to apply source-drain bias. A resist or of known value is then used to back out the current from Ohms Law. There will be a voltage drop across the re sistor, measured by the voltage at the power supply compared to the voltage measures on screen at the oscilloscope. Therefore, from V=IR, I = V/R. It is important to choose a resistor that will provide the appropriate sensitivity so that the voltage drop can be measured, but is not so large that signif icant digits are lost. Typical values are 10 and 48 The gate voltage is supplied from the pulse generator, pulsing from pinch-off (typically V) to the desired gate vo ltage with a 10% duty cycl e. This type of test setup means that every point must be calculate d as there is no automated measurement on this system.

PAGE 187

187 When measuring the Vd-Id curve, typically onl y the Vg = 0 curve is used. The pulse will therefore be from -10 V to 0 V. To take the point at Vd = 1 V, the oscilloscope center line is set at 1 V (note that this is before the resistor so this will be the true drain voltage), the pulse is turned on, and the DC power supply is ramped up so that the bottom of the pulse step is at the center line. The reading on the DC power supply is recorded for curre nt calculation. This process is repeated for each point that one wishes to meas ure. Typically the drain voltage step is 1 V. A typical curve relative to the DC curve is shown in Figure B-10. To measure the Vg-Id curve, the oscilloscope cen ter is set at the same voltage that the DC Vg-Id curve was taken at. The pulse magnitude is changed to the various gate voltages that one intends to measure. For example, starting at a pu lse from V to 0 V, the next step may be say V to .5 V, and so on. Typical step size is 0.5 or .25 V. At each step, the DC power supply voltage is adjusted so th at the bottom of the pulse fall s at the center line. The same current calculation applies. A typica l curve is shown in Figure B-11. B.5 Leakage/Isolation B.5.1 Overview The source-gate and gate-drain Schottky diodes are used to check for gate leakage. The characteristics should match an ideal diode. The is olation can be tested between ohmic contacts on two adjacent mesas. If mesa etching was su ccessful in confining the 2DEG, the isolation current will be very low (nA). If the cu rrent between two adjacent mesas is high (A-mA), further etching is required. Leakage testing is es sentially diode characterization. The source-gate or drain-gate diode is test ed, and a low current level a nd good turn-on are ideal. B.5.2 Procedure Isolation testing is set up like TLM testing, and the two are typically performed at the same time. The difference is where a TLM pattern is measured between ohmic pads on the same mesa,

PAGE 188

188 isolation is measured between adjacent mesas. If the mesa has not been etched deep enough, current will flow between the two mesas (mA level). If the mesa has been sufficiently etched to confine the electrons, there will be no current fl ow (nA level). A typical voltage range for this measurement is V to 40 V. A good isolation curve is shown in Figure B-12. When designing a mask, it is desirable to put a set of pads on separate mesas cl ose to each other (10-20 um) for isolation measurements. A schematic for a diode measurement is shown in Figure B-13. This measurement uses the parameter analyzer in diode (Vf-If) mode. Forw ard and reverse bias measurements should be made separately to avoid damaging the device. Forward bias should never exceed 5 V, and reverse should never exceed V. It is important to set the co mpliance level very low (<10 mA forward, <1 mA reverse) to avoid damage. Th e ohmic pad (SMU3 source or drain) is ground. A typical curve for forward bias is shown in Figure B-14. B.6 Breakdown/High Power B.6.1 Overview High power testing is typically used to ch eck the breakdown voltage of HEMTs. Extreme care must be taken to protect the tester and the device as much as possible. Testing is usually performed in Fluoroinert for high breakdown devices. The breakdown of air occurs around 400 V, so this solution, which is a Teflon-based liquid, will allow for te sting beyond this level. Typically the device is operated at or near pinc h-off to prevent damage and to obtain the true breakdown voltage of the material. The current level will be either flat or slightly increasing, then demonstrate a sharp increase at the breakdo wn voltage. In an ideal careful test, one can approach this voltage slowly and just reach the breakdown point and stop before the device arcs and is destroyed.

PAGE 189

189 B.6.2 Procedure Breakdown measurements below 400 V can be performed with the curv e tracer, as shown in top of Figure B-15. On this system, the co mpliance setting is a power level, so it is a combination of current and voltage Therefore the compliance can be reached at high current and low voltage or high voltage and low current. Wh ile the compliance on this system will protect the system, it can still create hazar dous voltages or currents to th e user. Set the sweep for a DC Vd-Id measurement. It is best then to begin at the lowest settings a nd gradually work up. The sweep is set to go from high to low power, a nd is represented at a percentage of maximum power. Therefore the setting shoul d be gradually increased, with single measurements, and then when the desired point is reached, a sweep back can be used to obtain the full I-V curve. The curve that is generated is shown in Figure B-16. If the breakdown voltage is too high to meas ure with the curve tracer, the Glassman High Voltage power supplies can be used. A DC power supply must be connected to the system to control the voltage. An applied voltage of 010 V will correspond to 0-max power (1 kV, 6 kV, 20 kV) on the high voltage supply. The high voltage is connected to the drain terminal through a large (1 M) resistor to limit the current for safety. Th e source terminal is connected to a digital multimeter to measure the source-d rain current. The gate voltage is controlled through a second DC power supply to maintain the gate voltage ne ar pinch-off. The breakdown is determined by monitoring the current level and l ooking for a sudden sharp rise. A schematic of the test system is shown in the bottom of Figure B-15. B.7 Small Signal RF (S-parameters) B.7.1 Overview This test is used to look at the frequency re sponse of a HEMT. It inputs a small RF signal over a large frequency range (typically 10 MHz to 40 GHz or more) and measures what are

PAGE 190

190 referred to as the s-parameters, which have both a real and imaginary component on a Smith chart. From these, several different gain modes can be calculated. In particular, the unitary gain cutoff frequency (H21) and the maximum gain (U) are calculated and plo tted against frequency. At the point where each of these respective curves reaches zero gain, Ft and Fmax are recorded. These values will be a function of gate and dr ain DC voltage as well. Typically they are maximized when Vd is at saturation and Vg is at the point where Gm is maximized (typically Id = 25% Isat) The device s-parameters can be simulated using an equivalent circuit model to look at the parasitic resistances and capacitances in the device structure. By inputting the measured data and then using the simulation to attempt to match this data, one can extract the values for the parasitics in the system. B.7.2 Procedure First, the RF cables and Ground-Signal-Gr ound (GSG) probes must be connected. Care must be taken as RF components ar e very sensitive to dus t. The DC bias must be set-up to be supplied either through a Bias Tee or through the network analyzer. A schematic of the test setup is shown in Figure B-17. After starting up, the network analyzer must be calibrated and the frequency range must be set. A typical range is 50 MHz-40.05 GHz. Calibration is achieved using the SOLT method (short, op en, load, through). The s-parameters are taken with the probes in each configuration (open = probes lifted, or in individual pads, short = ground and signal shorted, load = 50 load between signal and ground, a nd through= probe 1 and probe 2 are connected via transmission lines). A calibration standard with each pattern is available. It is best to change the viewing options to a smith chart to verify calibration. An open probe will be at the

PAGE 191

191 far left of the smith chart (infinite resistance), short will be on the far right (0 resistance), and load will be in the center (50 resistance). After calibration, the device is placed on the pr obe station, oriented with the gate towards port 1 and the drain towards port 2. After lowering the probes, a DC I-V curve is measured to ensure that the device is active. The RF m easurement is performed using MMICAD software. After initializing the software to connect to th e network analyzer, perfor m a 2-port measurement. The software program (HEMT.ckt) will calculate U and H21 and plot a smith chart and the U, H21, and Gain as a function of frequency. It will also model the equivalent circuit with parameter estimates. A sample s-parameter plot is shown in Figure B-18, with data in red and blue, and model in green and brown. A sample fre quency plot is shown in Figure B-19, with the locations of FT and Fmax. If either parameter is beyond the range of the instrument, as in this case, a line is extrapolated from the last point to the axis with a slope of dB/decade. B.8 Large Signal RF (Load-Pull) B.8.1 Overview This test is used to look at the power output of a HEMT. In a load-pull test, RF power is input at a fixed frequency and th e output power is measured. Th e curve produced is input power vs output power, going from low to high power. Fr om this curve, one can calculate the actual gain of the device (Pout/Pin) and the power added efficiency (PAE). Note that this measurement uses the dBm unit scale, which is a logarithmi c absolute power unit, standing for a decibel referenced to 1 mW power (1 dBm = 1 mW, 10 dBm = 10 mW, -10 dBm = 0.01 mW). A typical curve is shown in Figure. The gain is usually linea r over a large range, then begins to decrease as the power output saturates. This point is referred to as the compression point. Each dB below the constant gain level is referred to as a dB of compression. Typically devices are operated at the 13 dB compression range.

PAGE 192

192 Typically this measurement requires an RF s ource that has the capability to set a single frequency with variable power, pow er meters at least for input and output (some systems also use reflected power to measure the tr ue power input to the system), tuners for impedance matching, and a DC bias system. A considerable amount of tim e must be put into this measurement, as each component must be measured and calibrated out so that only the characteristics of the device are calculated. B.8.2 Procedure A schematic of the load-pull system is shown in Figure B-20. All components must be calibrated first in order to make the measurement. This process can take several days. First, the network analyzer must be started and calibrated as described in the previous section. The tuners are then connected to the network analyzer for calibration. S-parameters are taken at specified impedance points on the tuner. This procedure ty pically takes 8-10 hours pe r tuner as the tuner must move to each position. The Maury load-pull software manages all of the data acquired. The power meters must be set to the correct frequency, and it is best that they are zeroed as well. Sparameters are taken for the directional coupler, power meters, bias tees, and probe tips. All of this data is assembled by the Ma ury software for calibration. When a device is placed on the probe station, a DC curve is taken first. The tuners must then be moved to perform impedance matching. When the input and output impedance is matched, the DC bias is then applied, and then RF power is applied. A sweep is performed to measure input and output power as a function of the input power from the RF generator. Input and output power and gain are then plotted as a function of input power for the device. A typical curve is shown in Figure B-21. While most of this is performed by the software, it is important to note that the dB scale is logarithmic, so while gain is defined as the difference between output and input power, this cannot be calculated in dB. The power must be converted to mW,

PAGE 193

193 subtracted, and then converted back to dB. The correlation is PdBm = 10log(PmW), or reversed, PmW = 10PdBm/10.

PAGE 194

194 Figure B-1. Optical image of TLM pattern. Figure B-2. Test setup for TLM measurement

PAGE 195

195 Figure B-3. Typical I-V curves for TLM measurement Figure B-4. Calculati on of Rs, Rt, and Rc from TLM measurement

PAGE 196

196 Figure B-5. Optical image of GaN HEMT Figure B-6. Test setup for DC measurement

PAGE 197

197 Figure B-7. Typical DC Vds-Ids curve Figure B-8. Typical DC Vg-Ids curve

PAGE 198

198 Figure B-9. Test setup fo r gate lag measurement Figure B-10. Typical pulsed Vds-Ids curve compared to DC

PAGE 199

199 Figure B-11. Typical pulsed Vg-Ids curve compared to DC Figure B-12. Typical isolation data

PAGE 200

200 Figure B-13. Schematic of test se tup for gate leakage measurement Figure B-14. Typical leakage data

PAGE 201

201 Figure B-15. Test setup for breakdown meas urement <400 V (top) and >400 V (bottom)

PAGE 202

202 Figure B-16. Typical breakdown data Figure B-17. Test setup for small signal RF measurement

PAGE 203

203 Figure B-18. Typical s-parameter data Figure B-19. Typical H21 and U calculation for Ft and Fmax

PAGE 204

204 Figure B-20. Test setup fo r load-pull measurement Figure B-21. Typical load-pull data

PAGE 205

205 APPENDIX C PROCESSING RECIPES C.1 Purpose This appendix is meant to serve as a database of processing recipes. There may be more than one recipe that works for any given process, so this is by no means a definitive set. It does, however, represent a list of reci pes that have consistently worked for me over the years. C.2 Photolithography C.2.1 Mesa Etching (S-1045) 1. Spin 1045, 3000 RPM, 30s 2. soft bake for 5 minutes at 110 C (hot plate) 3. expose for 30 sec, develop 20s in MF-322 4. hard bake 15 minutes, 110 C (hot plate) C.2.2 Lift-Off (S-1818) 1. spin 1818, 4000 rpm, 30s 2. soft bake 110 C, 5 min 3. toluene soak, 30s, N2 dry 4. expose 25s, develop 30s in MF-321 C.2.3 Lift-Off (S-1808) 1. Spin 1808, 4000 RPM, 30s 2. soft bake for 1 minute at 80 C (hot plate) 3. soak in toluene for 1.5 minutes 4. soft bake for 10 minutes at 90 C (oven) 5. expose for 15 sec, develop for 1 minute in MF-321 6. Descum in O2 plasma for 1 min 7. Remove oxide using 5% HCl or BOE for 1 min C.2.4 Lift-Off (LOR/1808) 1. spin LOR, 5000 RPM, 30s 2. soft bake 150 C, 2 min 3. spin 1808, 5000 RPM, 30s 4. soft bake 110 C, 5 min 5. expose 30s, develop 15s in MF-321 6. Descum in O2 plasma for 1 min 7. Remove oxide using 5% HCl or BOE for 1 min

PAGE 206

206 C.2.5 Image Reversal (5214) 1. spin 5214, 4000 RPM, 30s 2. soft bake, 100 C, 1.5 min (oven) 3. expose 10s with mask 4. bake, 110 C, 2 min (hot plate) 5. flood expose, 30s 6. develop 30s in AZ400K:H2O (1:4) 7. hard bake, 110 C, 30 min (oven) C.2.6 Air Bridge 1. spin PMGI, 3000 RPM, 30s 2. soft bake, 180 C, 20 min 3. spin 1045, 3000 RPM, 30s 4. soft bake, 110 C, 5 min 5. expose 25s, develop 45s in MF-322 6. hard bake, 110 C, 15 min 7. Etch PMGI in O2 plasma 8. Remove photoresist in acetone 9. Bake at 180 C for 1 hour to round the edges of the PMGI 10. Deposit seed Au layer in evaporator (1000 A) 11. Pattern air bridge level using 1045 resist (see previous section for processing) 12. Electroplate Au (~5 m) in opened areas 13. Remove photoresist with acetone 14. Remove seed layer by plasma etching C.3 E-Beam Lithography C.3.1 Etching (PMMA) 1. Spin PMMA to desired thickness (2% = ~1000, 4% = ~2000), 4000 RPM, 30s 2. Remove photoresist form one corner for conductivity in the SEM using acetone 3. Soft bake, 195 C, 10 min 4. Expose at 10 kV, 100 C/cm2 5. Develop 30s in MIBK:IPA (1:3), Rinse with IPA C.3.2 Lift-off (PMMA/PMMA-MAA) 1. Spin PMMA-MAA copolymer to desi red thickness (7% = ~2000), 4000 RPM, 30s 2. Remove photoresist from one corner fo r conductivity in the SEM using acetone 3. Soft bake, 195 C, 10 min 4. Spin PMMA to desired thickness (2% = ~1000, 4% = ~2000), 4000 RPM, 30s 5. Remove photoresist from one corner fo r conductivity in the SEM using acetone 6. Soft bake, 195 C, 10 min 7. Expose at 10 kV, 100 C/cm2

PAGE 207

207 8. Develop 30s in MIBK:IPA (1:3), Rinse with IPA C.3.3 T-gate (PMMA/PMMA-MAA/PMMA) 1. Spin 4% PMMA, 4000 RPM, 30s 2. Remove photoresist from one corner fo r conductivity in the SEM using acetone 3. Soft bake, 195 C, 10 min 4. Spin 7% PMMA-MAA copolymer, 4000 RPM, 30s 5. Remove photoresist from one corner fo r conductivity in the SEM using acetone 6. Soft bake, 195 C, 10 min 7. Spin 2% PMMA, 4000 RPM, 30s 8. Remove photoresist from one corner fo r conductivity in the SEM using acetone 9. Soft bake, 195 C, 10 min 10. Expose at 30 kV, 70 C/cm2 11. Expose at 30 kV, 130 C/cm2 12. Soak in chlorobenzene, 6s 13. Develop 60s in Methanol:IPA (1:1) 14. Develop 30s in MIBK:IPA (1:3), Rinse with IPA C.4 Plasma Etching C.4.1 Inductively Coupled Plasma Etch With Cl2 (AlGaN/GaN and Au) Temp = 25 C Pressure = 2 mTorr ICP Power (RF2) = 150 W RF Power (RF1) = 40 W Gas flows: Cl2 = 10 sccm Ar = 5 sccm Etch Rate = ~800 A/min for GaN (strongly dependent on material quality) ~1500 A/min for Au C.4.2 Reactive Ion Etch With CF4 (SiO2, SiNx) Temp = 25 C Pressure = 175 mTorr RF Power = 50 W Gas flows: CF4 = 50 sccm C.4.3 Reactive Ion Etch With O2 (polymers) Temp = 25 C Pressure = 315 mTorr RF Power = 50 W Gas flows: O2 = 30 sccm

PAGE 208

208 C.5 Plasma Enhanced Chemical Vapor Deposition C.5.1 Silicon Nitride (SiNx) Temp = 255 C Pressure = 900 mTorr RF Power = 30 W DC Bias = 0 Ref Power = 0 Gas flows: SiH4 = 100 N2 = 400 NH3 = 27 Dep Rate = ~115 A/min Refractive index = ~1.7-1.9 C.5.2 Silicon Dioxide (SiO2) Temp = 255 C Pressure = 900 mTorr RF Power = 30 W DC Bias = 0 Ref Power = 0 Gas flows: SiH4 = 200 N2O = 200 Dep Rate = ~300 A/min Refractive index = ~1.38-1.46 C.6 Wet Etching C.6.1 Buffered Oxide Etch (SiNx, SiO2) SiNx etch rate = ~1000 A/min SiO2 etch rate = ~6000 A/min Aqua Regia (Ni) Formulation = HNO3:HCl:H3PO4 (3:1:1) Ni etch rate = ~1 m/min C.7 Thermal Evaporation C.7.1 Indium Current = 200 A Deposition Rate = 125 A/s 1 In charge = ~1 m film

PAGE 209

209 C.8 Flip-Chip Bonding C.8.1 Indium-Gold Pressure = 200g for contact, 400g during heating Temperature = 174 C Time = 10 min C.8.2 Gold-Gold Pressure = 1.3 kg Temperature = 375 C Time = 10 min C.8.3 Polymer (SU-8-SU-8) Pressure = 1 kg Temperature = 135 C Time = 10 min

PAGE 210

210 LIST OF REFERENCES 1. A. P. Zha ng, L. B. Rowland, E. B. Kaminsky, J. B. Tucker, J. W. Kretchmer, A. F. Allen, J. Cook, and B. J. Edward, Electron. Lett. 39, 245(2003). 2. W. Saito, Y. Takada, M. Kuraguchi, K. Tsuda, I. Omura, T. Ogura and H. Ohashi, IEEE Trans Electron Dev. 50, 2528 (2003 ). 3. A.P. Zhang, L.B. Rowland, E.B. Kaminsky, V. Tilak, J.C. Grande, J. Teetsov, A. Vertiatchikh and L.F. Eastman, J. Electron. Mater.32 388(2003). 4. W. Lu, V. Kumar, E. L. Piner and I. Ad esida, IEEE Trans. Electron Dev. 50, 1069 (2003 ). 5. P. Valizadeh and D. Pavlidis, IEEE Trans. Electron. Devices, 52, 1933 (2005). 6. S. J. Pearton, J. C. Zolper, R. J. Shul, F. Ren, J. Appl. Phys. 86, 1 (1999) and references therein 7. M. Hikita, M. Yanagihara, K. Nakazawa, H. Ueno, Y. Hirose, T. Ueda, Y. Uemoto, T. Tanaka, D. Ueda and T.Egawa, IEEE Tr ans. Electron Devices, 52, 1963 (2005) 8. S. Nakazawa, T. Ueda, K. Inoue, T. Tanaka, H. Ishikawa and T. Egawa, IEEE Trans. Electron Devices, 52, 2124 (2005 ). 9. T. Palacios, S. Rajan, A. Chakraborty, S. He ikman, S. Keller, S.P. DenBaars and U.K. Mishra, IEEE Trans. Electron Dev. 52, 2117 (2005) 10. U. K. Mishra, P. Parikh, and Y. F. Wu, Proc. IEEE 90, 1022 (2002). 11. 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, J. R. Shealy, IEEE Trans. Electron Dev., 48 479(2001). 12. S. Keller, Y.-F. Wu, G. Parish, N. Ziang, J. J. Xu, B. P. Keller, S. P. DenBaars, U. K. Mishra, IEEE Trans. Elec tron. Dev., 48 552 (2001). 13. V. Adivarahan, M.Gaevski, W.H. Sun, H.Fatima, A. Koudymov, S. Saygi, G. Simin, J. Yang, M.A. Khan, A. Tarakji, M.S. Shur and R. Gaska, IEEE Electron Device Letters, 24, 514 (2003). 14. A. Tarakji, H.Fatima, X.Hu, J.P.Zhang, G. Simin, M.A. Khan, M.S. Shur and R. Gaska, IEEE Electron Dev. Lett. 24, 369 (2003). 15. S. Iwakami, M. Yanagihara, O. Machida, E. Chino,N. Kaneko, H. Goto and K. Ohtsuka, Jpn. J. Appl. Phys. 43, L831 (2004). 16. R. Mehandru, S. Kim, J. Kim, F. Ren, J. Lothian, S.J. Pearton, S.S. Park and Y.J. Park, SolidState Electron. 47, 1037 (2003).

PAGE 211

211 17. A. Zeitouny, M. Eizenberg, S. J. Pearton, and F. Ren, J. Appl. Phys. 88, 2048 (2000). 18. B. Luo, F. Ren, R.C. Fitch, J.K. Gillespie, T. Jenkins, J. Sewell, D. Via, A. Crespo, A.G. Baca R.D. Briggs, D. Gotthold, R. Birkhahn, B. Peres and S.J. Pearton, Appl. Phys. Lett. 82, 3910 (2003). 19. R. Khanna, S.J. Pearton, F. Ren, I. Kravchenko, C.J. Kao and G.C. Chi, Appl. Surf. Sci.252255, 1846 (2005). 20. J.H. Lau, Low Cost Flip Chip Te chnologies (McGraw-Hill, New York, 2000) 21. R. Tummala, Fundamentals of Microsyste ms Packaging (McGraw-Hill, New York, 2001) 22. M. Schimpf, U.S. Laser Corp, Telephone Interviews, Jan-Aug 2005 23. J.P. Sercel, JPSA Laser Co rp, Telephone Interview, July 2005 24. J.C.J.Verhoeven, J.K.M.Jansen, R.M.M.Mattheij and W.R.Smith, Mathematical and Computer Modelling 37 419( 2003). 25. S. Thomas III and J. J. Brown, in Handbook of Advanced Plasma Processing Techniques, ed. R. J. Shul and S. J. Pearton (Springer, Berlin 2000). 26. P. H. Yih, A. J. Steckl, J. Electrochem. Soc. 142 2853 (1995). 27. P. Leerungnawarat, D. C. Hays, H. Cho, S. J. Pearton, R. M. Strong, C.-M. Zetterling, and M. stling, J. Vac. Sci. Technol. B 17 2050 (1999). 28. S. Tanaka, K. Rajanna, T. Abe, and M. Esashi, J. Vac. Sci. Technol. B 19 2173 (2001). 29. B. Li, L. Cao and J. H. Zhao, Appl. Phys. Lett. 73 653 (1998). 30. J. R. Flemish, K. Xie and J. H. Zhao, Appl. Phys. Lett. 64 2315 (1994). 31. F. A. Khan and I. Adesida, Appl. Phys. Lett. 75 2268 (1999). 32. H. Cho, P. Leerungnawarat, D. C. Hays, and S. J. Pearton, S. N. G. Chu, R. M. Stong, C.-M. Zetterling and M. stling and F. Ren, Appl. Phys. Lett. 76 739 (2000). 33. P. Chabert, N. Proust, J. Perrin and R. W. Boswell, Appl. Phys. Lett. 76 2310 (2000). 34. 2P. Chabert, J. Vac. Sci. Technol. B 19, 1339 (2001). 35. J. J. Wang, E. S. Lambers, S. J. Pearton, M. Ostling, C.-M. Zetterling, J. M. Grow, F. Ren, and R. J. Shul, J. Vac. Sci. Technol. A 16 2204 (1998). 36. F. A. Khan, L. Zhou, V. Kuma r, J. Electrochem. Soc.149 G420 (2002). 37. S. J. Pearton, F. Ren, A. P. Zhang and K. P. Lee, Mat. Sci. Eng. R30 55 (2000)

PAGE 212

212 38. H. Morkoc, S. Strite, G.B. Gao, M.E. Lin, B. Sverdlov, M. Nurns, J. Appl. Phys. 76, 1363 (1994) 39. T. P. Chow, R. Tyagi, IEEE Tr ans. Electron. Dev. 41, 1481 (1994) 40. Z.Z. Bandic, P.M. Bridger, E.C. Piquette, T.C. McGill, R.P. Vaudo, V.M. Phanse, J.M. Redwing, Appl. Phys. Lett. 74, 1266 (1998) 41. R. Gaska, J.W. Yang, A. Osinsky, Q. Chen, M.A. Khan, A.O. Orlov, G.L. Snider, M.S. Shur, Appl. Phys. Lett. 72, 1266 (1998) 42. Y. Uzawa, Z. Wang, A. Kawakami, B. Komiyama, Appl. Phys. Lett. 66, 1992 (1995) 43. F. Ren, J.C. Zolper, Wide Energy Bandgap Electronic Devices (World Scientific Press, Singapore, 2003) 44. H. Marchand, J.P. Ibbetson, P.T. Fini, P. Kozodoy, S. Keller, S. DenBaars, J.S. Speck, U.K. Mishra, Mater. Res. Soc. Int. J. Nitride Semi. Res. 3, 3 (1998) 45. R. Radar, G. Jacob, J. Hallais, R. Frucgart, J. Cryst. Growth 31, 197 (1975) 46. J. Karpinski, S. Porowski, S. Mitokowska, J. Cryst. Growth 56, 77 (1982) 47. S. Nakamura, T. Mukai, M. Senoh, Appl. Phys. Lett. 64, 1687 (1994) 48. H. Amano, M. Kito, K. Hiramatsu, I. Akashi, Jpn. J. appl. Phys. 28, L139 (1989) 49. S. Nakamura, M. Senoh, S.I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y.S. Ugimoto, Jpn. J. Appl. Phys. 35, L217 (1996) 50. S. Nakamura, M. Senoh, S.I. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, Y.S. Ugimoto, Appl. Phys. Lett. 68, 2105 (1996) 51. S. Li, Semiconductor Physical Electronics (Springer, 2006) 52. M.A. Khan, A.R. Bharrari, J.N. Kuznia, D.T. Olson, Appl. Phys. Lett. 63, 1214 (1993) 53. M.A. Khan, J.N. Kuznia, D.T. Olson, W. Schatt, J. Burm, M.S. Shur, Appl. Phys. Lett 65, 1121 (1994) 54. L. Esaki and R. Tsu, IBM Research, RC 2418 (1969) 55. R. Dingle, H.L. Stormer, A.C. Gossard, W. Weigmann. Appl. Plys. Lett. 33, 665 (1978) 56. T. Mimura, S. Hiyamizu, T. Fujii, K. Nanbu. Jap. J. Appl. Phys. 19, L225-227 (1980) 57. D. Delagebeaudeuf, P. Delescluse, P. Etienne M. Laviron, J. Chaplart, N.T. Linh. Electron Lett. 16, 667-668 (1980)

PAGE 213

213 58. M.A. Khan, J.M. Van Hove, J.N. Kuznia, D.T. Olson. Appl. Phys. Lett. 58, 2408-2410 (1991) 59. M.A. Khan, D.T. Olson, J.N. Kuznia. Appl. Phys. Lett 65, 1121-1123 (1994) 60. Q. Chen, J.W. Yang, R. Gaska, M.A. Khan, M.S. Shur, G.J. Sullivan, A.L. Sailor, J.A. Higgings, A.T. Ping, I. Adesida. IE EE Electron Device Lett. 19, 44-46 (1998) 61. Y.-F. Wu, B.P. Keller, P. Fini, J. Pusl, M. Le, N.X. Nguyen, C. Nguyen, D. Widman, S. Keller, S.P. Denbaars, U.K. Mish ra. Electron. Lett. 33, 1742-1743 (1997) 62. Wakejima, Matsunaga, Okamoto, Ota, Ando, Nakayama, Miyamoto. IEEE MTT-S International Microwave Sy mposium Digest, 2006, 1360-1363 63. Nitronex Corporation, Available Pr oducts. 64. Choi, Pophristic, Peres, Cha, Spencer, Ea stman. Semiconductor Science and Technology 22 (2007), 517-521 65. Higashiwaki, Mimura, Matsui. Phys. Stat. Sol. (A) 204, No. 6, 2042-2048 (2007) 66. Palacios, Dora, Chakraborty, Sanabria, Kelle r, DenBaars, Mishra. Phys. Stat. Sol (A) 203, No. 7, 1845-1850 (2006) 67. Y.-F. Wu, A. Saxler, M. Moore, R.P. Smith, S. Sheppard, P.M. Chavarkar, T. Wisleder, U.K. Mishra, P. Parikh, IEEE Electron Device Letters, v 25, n 3, March 2004, p 117-19 68. Campbell. The Science and Engineering of Mi croelectronic Fabrication. Oxford University Press, 2001 69. Sze. Semiconductor Devices: Physics and Technology. Wiley & Sons, 2002 70. Pearton, Zolper, Shul, Ren, Appl. Phys. Rev. 86, #1, 1 July 1999 71. T.J. Anderson, F. Ren, L. Covert, J. Lin, S.J. Pearton, T. Dalrymple, C. Bozada, R. Fitch, N. Moser, R. Bedford, M. Schimpf. Journal of Electronic Materials. Vol. 35, No. 4, 2006, 675-679 72. A.Minko,V.Hoel,E.Morvan,B.Grimbert, A. Soltani, E.Delos, D.Ducateau, C.Gaquiere, D.Theron, J.C.Jaeger, H.Lahreche, L.Wedz ikowski,R.Langer and P.Bove, IEEE Electron Dev.Lett.25,453 (2004). 73. T. J.Anderson, F. Ren, L. Voss, M. Hlad, B. P. Gila, L. Covert, J. Lin, S. J. Pearton, P. Bove, H. Lahreche, and J. Thuret, J. Vac. Sci. Technol. B 24, 2302 (2006). 74. Travis Anderson, F. Ren, S.J. Pearton, M. Mastro, R. Holm, R. Henry, C. Eddy, J.Y. Lee, K.Y. Lee, J. Kim. J. Vac. Sci. Technol. B. 24 (5) Sept/Oct 2006

PAGE 214

214 75. 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. R. Shealy, IEEE Trans. Electron Dev., 48, 479(2001). 76. S. C. Binari, K. Ikossi, J. A. Roussos, W. Kr uppa, D. Park, H. B. Dietrich, D. D. Koleske, A. E. Wickenden and R. L. Henr y, IEEE Trans. Electron Dev., 48, 465(2001). 77. R. Vetury, N. Q. Zhang, S. Keller, and U. K. Mishra, IEEE Trans. Electron. Dev. ED48, 560(2001). 78. B. M. Green, K. K. Chu, E. M. Chumbes, J. A. Smart, J. R. Shealy, and L. F. Eastman, IEEE Electron. Dev. Lett., 21, 268(2000). 79. B. Luo, J. W. Johnson, J. Kim, R. M. Mehand ru, F. Ren, B. P. Gila, A. H. Onstine, C. R. Abernathy, S. J. Pearton, A. G. Baca, R. D. Briggs, R. J. Shul, C. Monier and J. Han, Appl. Phys. Lett. 80,1661 (2002). 80. A. Vertiatchikh, L. F. Eastman, W. J. Scha ff and I. Prunty, Electronics Lett. 38,. 388 (2002). 81. Binari, S.C., Klein, P.B., and Kazior, T.E., Proc. IEEE, 2002, 90, pp. 104858. 82. T. Kikkawa, Japanese J. Appl. Phys., vol. 44, pp. 4896-4901, 2005. 83. D.C.Dumka, C. Lee, H.Q. Tserng and P. Saunier, 2004 Electronics Letters, vol. 40, no. 24. 84. T. Kikkawa, K. Imanishi, M. Kanamura, K. Joshin, 2006 CS-MANTECH Technical Digest., pp.171-174, 2006. 85. S. Singhal, T. Li, A. Chaudhari, A.W. Hans on, R. Therrien, J.W. Johnson, W. Nagy, J. Marquart, P. Rajagopal, J.C. Roberts, E.L. Piner, I.C. Kizilyalli, K.J. Linthicum, Microelectronics Reliability 46 (2006) pp.1247. 86. H. Kim, V. Tilak, B. M. Green, H.Y. Cha, J. A. Smart, J. R. Shealy, and L. F. Eastman, 2001 IEEE 39th Annual Internationa l Reliability Physics Symposium, pp214-218. 87. R. Khanna, L. Stafford, S.J.Pearton, T.J. Anderson, F. Ren, I.I. Kravchko, A.Dabiran, A.Osinsky, J.Y.Lee, K.Y.Lee and J.Kim, Electron. Mater. 36, 379(2007). 88. R. Khanna, S.J. Pearton, F. Ren, I.I. Kravchenko, Applied Surface Science 253, pp. 2340 2344 (2006). 89. A. Koudymov, M.S. Shur, G. Simin, R. Gaska. Phys. Stat. Sol. (RRL) 1, No. 3, 116-118 (2007) 90. W. Kuang, R. Trew, G. Bilbro, Y. Liu. Mater. Res. Soc. Symp. Proc. Vol. 955 (2007) 91. A. Koudymov, M.S. Shur, G. Simin, C. Lee, K. Chu, P.C. Chao, J. Jimenez, T. Balistrerei. International Semiconductor Device Resear ch Symposium Proceesings (2007) 1-2

PAGE 215

215 92. T. J. Anderson, F. Ren, L. Covert, J. Lin, S. J. Pearton. Journal of Vacuum Science and Technology B, 24 (1), Jan/Feb 2006. 93. T. J. Anderson, F. Ren, L. Covert, J. Lin, S. J. Pearton. Journal of the Electrochemical Society, 153 (10), G906-G910 (2006) 94. P. Tobias, A. Baranzahi, A. L. Spetz, O. Kordina, E. Janzen and I. Lundstrom, IEEE Electron. Dev. Lett. 18 287(1997). 95. A. Vasiliev, W. Moritz, V. Fillipov, L. Bartholomus, A. Terentjev and T. Gabusjan, Sens. Actuators B49 133(1998). 96. J. Kim, F. Ren, B. Gila, C. R. Aberna thy, and S. J. Pearton, Appl. Phys. Lett. 82 739 (2003). 97. J. Kim, B. Gila, C. R. Abernathy, G. Y. Chung, F. Ren, and S. J. Pearton, Solid State Electronics, 47 1487 (2003). 98. A. Lloyd Spetz, P. Tobias, L. Unus, H. Svenningstorp, L.-G. Ekedahl and I. Lundstrm, Sens. Actuators B70 67(2000). 99. E. J. Connolly, G. M. O'Halloran, H. T. M. Pham, P. M. Sarro and P. J. French, Sens. Actuators A99 25(2002). 100. G. W. Hunter, P. G. Neudeck, R. S. Okojie, G. M. Beheim, V. Thomas, L. Chen, D. Lukco, C. C. Liu, B. Ward and D. Makel, Proc. ECS Vol 01-02 212(2002) 101. R. Neuberger, G. Muller, O. Ambacher, M. Stutzmann, Phys Status Solidi A 185 85 (2001). 102. J. Schalwig, G. Muller, O. Ambacher M. Stutzmann, Phys Status Solidi A 185 39 (2001). 103. G. Steinhoff, M. Hermann, W. J. Schaff, L. F. Eastmann, M. Stutzma nn, M. Eickhoff, Appl. Phys. Lett. 83 177 (2003). 104. M. Eickhoff, R. Neuberger, G. Steinhoff, O. Ambacher, G. Muller, M. Stutzmann. Phys. Status Solidi B 228 519 (2001). 105. J. Schalwig, G. Muller, M. Eickhoff, O. Ambacher, M. Stutzmann. Sens Actuat B 81 425 (2002). 106. M. Eickhoff, J. Schalwig, G. Steinhoff, O.Weidmann, L.Gorgens, R. Neuberger, M. Hermann, B.Baur, G.Muller, O. Ambacher, and M. Stutzmann, Phys. Stat. Solidi C 6 1908 (2003) 107. H. Svenningstorp, P. Tobias, I. Lundstrm, P. Salomonsson, P. Mrtensson, L.-G. Ekedahl and A. L. Spetz, B57 159(1999). 108. L. Chen, G. W. Hunter, and P. G. Neudeck, J. Vac. Sci. Technol A 15, 1228 (1997); J. Vac. Sci. Technol. A16 2890(1998).

PAGE 216

216 109. A. Baranzahi, A. Lloyd Spetz and I. Lundstrm, Appl. Phys. Lett. 67 3203 (1995). 110. B.S. Kang, R. Mehandru, S. Kim, F. Ren, R. Fitch, J. Gillespie, N. Moser, G. Jessen, T. Jenkins, R. Dettmer, D. Via, A. Crespo, B.P. G ila, C.R. Abernathy and S.J. Pearton, Appl. Phys. Lett. 84 4635 (2004). 111. B.S. Kang, Y.W. Heo, L.C. Tien, D.P. Nort on, F.Ren, B.P. Gila and S.J. Pearton, Appl. Phys. A 80 1029 (2005). 112. A. El. Kouche, J. Lin, M.E. Law, S. Kim, B.S. Kim, F. Ren and S.J. Pearton, Sensors and Actuators B: Chemical 105 329 (2005). 113. S.J. Pearton, B.S. Kang, S. Kim, F. Ren, B. P. Gila, C.R. Abernathy, J. Lin and S.N.G. Chu, J. Phys: Condensed Matter 16 R961 (2004). 114. S. Kim, B. Kang, F. Ren, K. Ip, Y. Heo, D. Norton and S.J. Pearton, Appl. Phys. Lett. 84, 1698 (2004). 115. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z.N. Low, J. Lin, S.J. Pearton, F. Ren. NHA Conference Proceedings 2007 116. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z. N. Low, J. Lin, S. J. Pearton, F. Ren, A. Dabiran, A. Osinsky, J. Painter. Hydrogen Safety Report, July 2007 117. T. Anderson, H.T. Wang, B.S. Kang, C. Li, Z. N. Low, J. Lin, S. J. Pearton, F. Ren, A. Dabiran, A. Osinsky, Journal of the Electrochemical Society (submitted October 2007) 118. Jun-Rui Huang, Wei-Chou Hsu, Yeong-Jia Chen, Tzong-Bin Wang, Kun-Wei Lin, HueyIng Chen and Wen-Chau Liu, Sensors Actuators B117, 151(2006). 119. B. S. Kang, S. Kim, F. Ren, B. P. Gila, C. R. Abernathy and S. J. Pearton, Sensors Actuators B 104, 232(2005). 120. K. Matsuo, N. Negoro, J. Kotani, T. Hashizume and H. Hasegawa, Appl. Surf. Sci.244, 273 (2005). 121. J. Song, W. Lu, J. S. Flynn and G. R. Brandes, Solid-State Electr onics, 49, 1330(2005). 122. L. Voss, B.P. Gila, S.J. Pearton, Hung-Ta Wang, F. Ren. J. Vac. Sci. Technol. B 23, 6 (2005) 123. H.-T. Wang, B.S. Kang, F. Ren, R.C. Fitch, J.K. Gillespie, N. Moser, G. Jessen, T. Jenkins, R. Dettmer, D. Via, A. Crespo, B.P. Gila, C.R. Abernathy, S.J. Pearton. Appl. Phys. Lett. 87, 172105 (2005) 124. C.-W. Hung, H.-L. Lin, H.-I. Chen, Y.-Y. Tsai, P.-H. Lai, S.-I Fu, W.-C. Liu. Electron Device Letters, 27, 12 (2006)

PAGE 217

217 125. H.-T. Wang, T.J. Anderson, B.S. Kang, F. Re n, Changzhi Li, Zhen-Ning Low, Jenshan Lin, B.P. Gila, S.J. Pearton, A. Dabiran, A. Osinsky. Applied Physics Letters, 90, 252109 (2007) 126. Welty, Wicks, Wilson, Rorrer. Fundamental s of Momentum, Heat, and Mass Transfer. Wiley & Sons, 2001 127. Bird, Stewart, Lightfoot. Tran sport Phenomena. Wiley & Sons, 2002 128. J.Zou, D.Kotchetkov, A.A.Balandin, D.I. Florescu and F.H. Pollack, J.Appl. Phys. 92 2534(2002). 129. D.I.Floescu, V.M.Asnin, F.H.Pollack, R.J.Molnar and C.E.C.Wood,J .Appl.Phys. 88 3295 (2000). 130. D.I.Florescu, V.M. Asnin, F.H. Pollack, A.M. Jones, J.C. Ramer, M.J. Schurman and I.Ferguson, Appl.Phys.Lett. 77 1464(2000). 131. D. Selvanathan, F.M.Mohammed, A.Tesfayesus and I.Adesida, J. Vac. Sci. Technol. B22, 2409 (2004). 132. M. W. Cole, D. W. Eckart, W. Y. Han, R. L. Pfeffer, T. M onahan, F. Ren, C. Yuan, R. A. Stall, S. J. Pearton, Y. Li and Y. Lu, J. Appl. Phys. 80, 278 (1996). 133. US Laser Corporation, 134. JPSA Laser Corporation, 135. S.Kim, B.S. Kang, F. Ren, J.D Entremont, W. Blumenfeld, T. Cordock and S.J.Pearton, J. Electron. Mater. 33, 477 (2004). 136. James Brannon. Excimer Laser Ablation a nd Etching. (ed. Woody Weed). American Vacuum Society, 1993 137. Wikipedia, Diode Pumped Solid St ate Laser. 138. D. Pulfer, JPSA Laser Corp. Telephone Interview. May 2006-May 2007 139. L.J. Durney, Grahams Electroplat ing Engineering Handbook (Springer, 1984) 140. S. Bailey. Alpha-Voltaic Power Source Designs Investigated. 141. N.J. Pinto, P. Carrin, J.X. Qui ones, Mater.Sci. Eng. A 366, 1 (2004) 142. G.A. Shafeev, L. Bellard, J.-M. Themlin, C. Fauquet-Ben Ammar, A. Cros, W. Marine, Surf. Coat. Tech. 80, 224 (1996)

PAGE 218

218 143. K. Aliberti, W. Ruff, H. Shen, P. Newman, M. Giza, W. Sarney, M. Stead, J. Damman, R. Mehandru, F. Ren, Proc. SPIE Laser Radar Technology And Applications IX 5412, 99 (2004) 144. Anderson, T. Fabrication. Characterizat ion, and Packaging Technology for High Speed Compound Semiconductor Devices. Qualifying Exam. University of Florida. 2005 145. B. Gila. University of Florida. Interview. December 19, 2007 146. E. Ochoa, T. Nelson, R. Bedford, J. Ehret, L. Starman, M. Harvey, T. Anderson, F. Ren. IEEE Photonics Technology Letters. Vol. 19, No. 6 (2007) 147. E. Ochoa, L. Starman, R. Bedford, T. Nelson, J. Ehret, M. Harvey, T. Anderson, F. Ren. J. Microlith., Microfab., Microsys t. 6 (3), 1 (Jul-Sep 2007) 148. Ochoa, Edward. Hybrid Micro-Electro-Mech anical Tuneable Filter. Diss. Air Force Institute of Technology, 2007.

PAGE 219

BIOGRAPHICAL SKETCH Travis Jam es Anderson, a native of Miami, Florida, began his higher education at the Georgia Institute of Technology in Atlanta, GA, where he had the opportunity to work as a research assistant for a prominent faculty me mber. After garnering much experience, he graduated from Tech in May 2004 with a bachelors degree in chemical cngineering. That fall, Travis enrolled in the Ph.D. program in the Chem ical Engineering Department at the University of Florida to further explore hi s research interests. In spring 2005, he joined Dr. Fan Rens research group and began the substantive research used to develop this di ssertation. Since joining the group Travis research has focused largely on the maturation of compound semiconductor device technology, studying novel se nsors and transistors, systems integration, and reliability. Having attained a high level of proficiency in th is research area, Travis graduated in May 2008, with a Doctor of Philosophy in chemical engineering.