Final Report

Hydrogen Effects on GaAs MicrowaveSemiconductors

Prepared for:

California Institute of TechnologyJet Propulsion Laboratory

Pasadena, California

Report Number: SMC97-0701Purchase Order Number: 000732172

July 1997

Shason Microwave Corporation

1120 NASA Road One, Suite 106Houston, TX 77058

Phone: 281.333.1950 Fax: 281.333.1954

TABLE OF CONTENTS

SECTION PAGE

Introduction 1

I. History 1

II. Failure Effects 5

a. MESFETs 7

b. PHEMTs 10

c. Indium Phosphide (InP) HEMT 18

III. Hydrogen Sources 19

IV. Possible Solutions 21

a. Thermal Treatment of Packages 22

b. Package Materials with Low H2 Absorption 23

c. Gate Barrier Metals 24

d. Non-hermetic Packages 25

e. Use of Hydrogen Getters 26

V. Recommendations of High Reliability Applications 28

a. Existing Designs 29

b. Designs in Development 30

Appendix 1. 34

LIST OF ILLUSTRATIONS

Illustration Page

Figure 1. Schematic Diagram of MESFET 7

Figure 2. Comparison of MESFET Idss in Hydrogen and Nitrogen 9

Figure 3. Median Life of GaAs MESFETs in 10% Hydrogen 10

Figure 4. Schematic Diagram of a typical PHEMT 11

Figure 5. Change in transconductance and Idss versus time in 4% hydrogen 12

Figure 6. Median life versus ambient temperature for GaAs PHEMTs 14

Appendix 1. A partial list of GaAs Device failure data 35

LIST OF TABLES

Table Page

Table 1. Comparison of median life for PHEMTs and MESFETs from the same 17 Process line, using calculated and measured data

Table 2. A listing of parameter shifts seen for exposure to hydrogen 19

Table 3. Hydrogen contributions from typical microwave module components 21

HYDROGEN EFFECTS ON GaAs MICROWAVE

SEMICONDUCTORS

INTRODUCTION

Hydrogen gas present in hermetic packages containing gallium arsenide (GaAs) field-

effect-transistors (FETs) and microwave monolithic integrated circuits (MMICs) has a

deleterious effect on their performance and lifetime. This effect was first reported in

approximately 1989, and other documents have been published since that first report. In recent

years, work has been performed to characterize the problem and efforts have been directed

toward solving it.

Initial work in GaAs development and reliability investigations failed to identify the

hydrogen problem, as most work was performed in nitrogen or other inert environments. A

study of the problem has been performed and this report details the history and effects of the

hydrogen failure mechanism, the attempts to solve the problem and recommendations for using

GaAs devices in high reliability applications.

I. History

Small, lightweight, high performance microwave amplifiers and other circuit functions

have been made possible by the use if GaAs or other compound semiconductor devices.

Compact size and high-efficiency performance of the circuits make them ideal for use in

applications having limited space and power availability.

Reliability investigations, beginning in the early 1980’s, studied failure modes and

mechanisms and began developing reliability enhancements needed to meet stringent reliability

requirements of military systems. Failures from these tests reported activation energies ranging

from 0.4 to 2.5 electron volts (eV), and median lives ranging from 2E5 hours (22.8 years) to

1.6E9 hours (1.8E5 years) at 150°C [1]. It is generally accepted that current MESFET

technology, from a stable process, will yield activation energies of 1.5 to 2.0eV with lifetimes of

greater that 1E6 hours at 150°C [26, 27, 28, 29]. A table giving partial listing of this early

reliability work is included as Appendix 1.

Most of the early reliability work, however, was performed in laboratory conditions and

in inert environments, to insure studies resulted in evaluation of devices and not their

environment. In 1989, degradation of GaAs FETs and MMICs caused by hydrogen gas trapped

inside the devices’ hermetic packages was reported. It was shown that hydrogen gas in

quantities as low as 0.5% of ambient atmosphere can cause significant degradation at elevated

temperatures (125°C), in a relatively short period of time (168 hours). It was proposed in this

work that the mechanism was due to catalytic conversion of molecular hydrogen to atomic

hydrogen by platinum in the gate. It was thought the atomic hydrogen diffused into the

semiconductor and compensated the silicon dopant (donors), thus causing a reduction in current

and gain of the device. A secondary mechanism observed was reduction in the Schottky gate

barrier height, thought to be due to modification of the gate-semiconductor interfacial layer [2].

Investigations to determine sources of hydrogen in hermetic packages, and efforts to

eliminate the sources were reported in 1991. The primary source was found to be outgassing

from ferrous metals used in package fabrication, and electroplating was also found to be a

contributor. Various studies to characterize the hydrogen content of the package material found

that hydrogen levels could be reduced, but not eliminated, and that one possible solution was to

use hydrogen free alloy or to passivate package surfaces [3].

Studies in 1993 concluded that hydrogen desorption from ferrous metals was a function

of metal thickness, with the greatest percentage being observed in thicker samples (0.01, 0.04

and0.06 inch test coupons). It was also found that moisture would increase in hermetic packages

as oxides were reduced by desorbed hydrogen. Amounts of desorbed hydrogen in this study was

dependent on processes used in package preparation (sealing technique, anneal, etc.), with one

package vendor using an anneal cycle that reduced desorbed hydrogen to undetectable levels.

Plated samples gave consistent results [4]. Work in this area has been continued by Lockheed-

Martin.

Reports of hydrogen work increased in 1994 with six papers on device results and one,

hydrogen related, workshop held. The increased 1994 activity signaled acceptance that hydrogen

was an industry wide problem.

The Hydrogen Effects on GaAs Devices workshop, sponsored by IEEE, 1994

International Reliability Physics Symposium (IRPS), was attended by 21 organizations, and was

conducted as an open forum, with three industry representatives presenting test data on the

problem. Results indicated that all field-effect devices tested, exhibited a problem and that

degradation could occur with fractions of a percent of hydrogen [5].

Three papers on metal-semiconductor-field-effect-transistors (MESFETs) and three on

pseudomorphic HEMTs (PHEMT) were published in 1994. Tests of 0.25 micron (uM) titanium-

platinum-gold (Ti/Pt/Au) gate MESFETs exhibited a sudden reduction of transconductance and

an increase in operating for 1.0% and 0.1% hydrogen. This data indicated an activation energy

of 0.4eV, and approximately linear dependence on hydrogen concentration [6]. Another

investigation was on Ti/Pt/Au and titanium-palladium-gold (Ti/Pd/Au) gate MESFETs of

different sizes. This report indicated increases in channel current and pinch-off, with

degradation times dependent on FET size and construction, hydrogen concentration and whether

the gate had platinum or palladium. Platinum was found to produce larger changes, at earlier

times, when compared to palladium [7]. MESFET tests at two temperatures in 10% hydrogen

yielded an activation energy of 1.0eV, with degradation being a sudden decrease in current and

transconductance [8].

Hydrogen tests on Indium-Phosphide HEMTs and GaAs PHEMTs with Ti/Pt/Au gates

showed that InP HEMTs degraded more rapidly and that channel current increased as opposed to

decreasing in the GaAs PHEMT. Source resistance for both were found to remain unchanged

and the amount of degradation was found to depend on platinum thickness, no degradation was

observed when platinum was reduced to 50 angstroms. Also, samples constructed with Ti/Al

gates did not exhibit degradation [9]. Limiter amplifiers, constructed with 0.25uM, Ti/Pt/Au

gate GaAs PHEMTs were tested under five different hydrogen percentages and seven

temperatures. Work from these tests yielded a model, which accounts for activation energy and

pressure acceleration terms when calculating lifetimes. Observed failure was a decrease in

operating current [10]. Tests on 0.25 uM, Ti/Pt/Au gate GaAs PHEMTs under four hydrogen

concentrations and three temperatures showed a dependence on temperature and hydrogen

concentration. Changes observed were a sudden decrease in channel current, decrease

transconductance, an increase in low-field channel resistance and a 30millivolt increase in

Schottky barrier height. Activation energy was measured at 0.34eV at low concentration and

1.73eV in nitrogen. Package bake (350°C, 7 hours), prior to seal was found to double median

life for the samples tested [11].

Investigations continued in 1995, as manufacturers addressed the extent of the problem

and began to look for solutions to the problem. GaAs MESFETs exposed to hydrogen and

deuterium degraded in each case. Data from this experiment indicates that hydrogen diffusion

occurs at the platinum sidewalls, and not at the gold surface of Ti/Pt/Au gates [12]. Experiments

conducted on low noise PHEMT amplifiers in two package types (aluminum-silicon (A40 and

Kovar) indicated that post-plating package bakes of 250°C for 168 hours reduces hydrogen

sufficiently to meet a twenty year mission life in spacecraft environments [13]. There was test

data reported on 0.15uM gate PHEMTs exposed to various concentrations, with and without

bias, and at two temperatures. Failure time was dependent on concentrations for amounts above

25%, but appeared less dependent on concentrations below that amount. As in other work,

failure was indicated by a sudden drop in channel current [14]. An industry survey established

the problem as industry wide, and that potential solutions included thermal treatment of

packages, use of hydrogen getters, or the use of new barrier metals in gates (replacing platinum

or palladium) [15].

In 1996, results of tests on two hundred 0.25uM Ti/Pt/Au gate GaAs PHEMTs was

reported. This experiment was performed to verify the model reported in [10], to determine if

hydrogen degradation was bias dependent, and to determine which temperature (ambient or

junction) established failure rates [16]. A report describing failure mechanisms, observed effects

on device parameters, and an outline of possible solutions to the problem was published [17].

Initial work on development of a physical model of the failure mechanism, accurately

duplicating experimentally observed results was reported [18]. Experiments were performed

using hydrogen plasma in an attempt to passivate the surface of GaAs PHEMTs, thus stabilizing

breakdown voltage and improving power performance of these devices. Large decreases in

current and mobility were observed much like tests of devices subjected to gas flows [19].

After a slow start, significant effort has been expended on this problem. As the above

history indicates, it was five years between discovery of the problem and reported device work

from others. This indicated a reluctance to accept the problem as real, and a delay in

accumulating data from testing. It is now realized the problem exists industry wide and work is

continuing to determine the best solution to the problem. Much more must be completed before

the problem is no longer a factor.

II. Failure Effects

The hydrogen problem, as described earlier, has been observed in MESFETs, PHEMTs

and InP HEMTs, and the effects have been different for each of these technologies. However,

one general statement can be made about the problem; devices subjected to hydrogen

atmospheres will change operating performance, with the time to change being a function of

temperature and hydrogen concentration. Reports of degradation have been limited to amplifier

functions. No observations of the problem have been reported for heterojunction bipolar

transistors (HBT), PIN diodes, switch and phase shifter functions. Table 2, at the end of this

section, is a summary of observed changes.

When FETs or MMICs are tested in hydrogen with RF applied, gain degradation is

observed concurrently with the change in operating current a transconductance. Since gain

degrades concurrently with reduced Idss and gm, ivestigators have performed degradation studies

utilizing cost effective storage tests; failure criteria used has been either a 10% or 20% change in

DC parameters. It should be noted that changes are not catastrophic, as seen in typical wear-out,

therefore some system applications may function adequately even with the observed degradation.

The exact failure mechanism is not known, but has been under investigation for several

years. One popular theory assumes that molecular hydrogen is converted to atomic hydrogen

through a catalytic reaction with platinum or palladium in the gate structure and diffuses into the

Schottky metal and channel, resulting in: 1) compensation of donors by monatomic hydrogen, 2)

a shift in barrier height and 3) a neutralization of impurities near the Schottky interface causing a

barrier height shift.

Tests have shown barrier height shifts when GaAs Schottky diodes (gates) are subjected

to hydrogen [11, 20, 21, 22, 23]. It has been reported that atomic hydrogen generated by plasma

results in changes in the barrier height of Ti Schottky diodes when compared to control samples

and that the hydrogen-impurity complexes near the interface may be the cause of this shift [22,

25]. Others report reduced barrier height and a neutralization of shallow dopants [11, 22].

Thermal recovery of these changes indicated an activation energy of 0.6eV. Others have shown

[24, 25] that silicon-hydrogen complexes do exist in Si doped GaAs when subjected to hydrogen

gas, supporting the compensation theory. Some have observed increases in current, which does

not support the compensation of donors, indicating there is, in some cases, another mechanism at

work. P.C. Chao, et. al. [9] proposed that hydrogen affects on pinning levels at the gate

metal/semiconductor interface may result in a decreased built-in potential and, therefore,

increased current in InP HEMTs subjected to hydrogen. Additional study is needed to

understand this phenomenon.

II.a. MESFETs

The degradation seen in MESFETs, shown in the schematic diagram of Figure 1, is

generally seen as a sudden and rapid change in transconductance (gm) and saturated drain current

(Idss), followed by partial recovery as time continues. The two top plots of Figure 2 demonstrate

the change in Idss when MESFETs are subjected to 10% hydrogen at 140°C and 180°C. Figure

5 shows changes in Idss and gm when devices are subjected to alternating exposures of 4%

hydrogen and 100% nitrogen, at 270°C. Other parameters change, although not always as severe

as these two parameters.

a)

b)

Figure 1. Schematic diagram of a typical MESFET. a) View showing the activechannel, gate, source, drain and nitride overlay. b) Enlarged view of the gate-

semiconductor interface

W.O. Camp, et. al. [2] observed MMIC gain reduction (from 0.0 to 5.0dB per 30dB of

gain) when subjected to 125C for 168 hours in sealed Kovar packages. Residual gas analysis

(RGA) showed a correlation of gain reduction to packages containing hydrogen. As little as

0.5% hydrogen content led to degradation. A series of tests in 100% hydrogen, at 150C for 4

hours, determined that hydrogen and platinum in the gates was responsible for the change, and

that reduced current and transconductance accompanied the gain changes. However, devices

from some manufacturers did not show change, leading the author, and manufacturer, to believe

those devices were immune to the problem. Longer exposure times would have shown change in

all devices with Ti/Pt/Au or Ti/Pd/Au gates. This data was not conducted to establish median

life and activation energy, but it did identify the problem and was the foundation for all other

work conducted in this field. It was the conclusion from the authors that donor compensations

were the cause of failure.

Source

Nitride

DrainGate

Silicon Doped GaAs(Active Channel)

Semi-insulating GaAs Substrate

Gold

Platinum

Titanium

GaAs

Delaney, et. al. [6], reported that 0.25uM, Ti/Pt/Au gate MESFETs were tested in 1.0%

hydrogen at 175°C and 200°C and 0.1% at 200°C. Transconductance degraded sharply and

current at fixed bias increased. Data analysis yielded a 0.4eV activation energy with a 100°C

median life (10% gm decrease) of approximately 1200 hours; time to 10% current change was

approximately 1E4 hours. The author noted degradation was approximately linear with

hydrogen concentration.

W. Roesch [7] indicated a decrease in gm and an increase in current, along with a large

percentage increase in pinchoff voltage. One significant difference in this data is the long time

to “onset” of changes. Also, the increase in current in this and [6] is not supported by the donor

compensation theory. This author reported no change in 1000 hours at 185°C and a need to raise

temperature to 250°C for reasonable test times. In addition, this author noted several items not

previously reported; 1) devices with Pt degraded faster than those with Pd, 2) large FETs

degraded faster than small ones and 3) changes caused by hydrogen were opposite those from

normal wear out, causing a possible enhancement to lifetimes in the presence of hydrogen. The

data observed in [6] and [7] are an indication a mechanism other than donor compensation is

probable in these processes.

The magnitude of change in 10% hydrogen, as reported by Decker [8], reaches 25%

before beginning recovery. Concurrently, abrupt increases in breakdown voltage and decrease in

pinchoff is observed, which is consistent with donor compensation. Some recovery is seen as

devices are maintained in the hydrogen environment. Time to 10% Idss degradation was 32

hours at 180°C and 395 hours at 140°C. Figure 1 shows the device parameters when exposed to

hydrogen and nitrogen environments at these two temperatures.

Figure 2. Comparison of MESFET Idss in Hydrogen and Nitrogen Environments

Reproduced from Ref. [8]

This data resulted in an activation energy of approximately 1.0eV and a 100°C median life of

800 hours. Figure 3 is a plot of median life for hydrogen and nitrogen testing for MESFETs

from the same process.

Figure 3. Median Life of GaAs MESFETs in 10% Hydrogen and 100% Nitrogen

Reproduced from Ref.[8]

Losses in gm and Idss were observed when commercially available MESFETs were

tested at 250°C in 5% hydrogen [12]. Experiments were performed with different gate metals

(Ti/Pt/Au, Ti/Pt, Ti) and the devices with Pt exposed had a much higher incorporation of

hydrogen (21 times) in the gate metal film for the Ti/Pt/Au gate after on hour at 250°C in 5%

hydrogen. This implies hydrogen diffuses at the Pt edge, and could lead to a gate process which

would not be effected by hydrogen.

II.b. PHEMTs

In general, PHEMTs, shown schematically in Figure 4, degrade in the same manner as

MESFETs, with at least one report [5] that PHEMT degradation occurs faster. A discussion of

hydrogen exposure work for this technology follows.

Figure 4. Schematic diagram of a typical PHEMT. Reproduced from [27]

NOTE: Drawing is not to scale.

P.C. Chao, et. al. [9] reported that PHEMTs with Ti/Pt/Au gates exposed to 4% hydrogen

at 270C exhibited changes similar to those seen in MESFETs. Transconductance and Idss

decreased, pinchoff increased, and there was no change in source resistance. Figure 5 shows a

plot of Idss and gm during alternate exposures to hydrogen and nitrogen. Note that the current

recovers when baked in nitrogen, however it does not recover to the original value. The author

reported similar recovery after four days in nitrogen at room temperature. Increased pinchoff

and no change in source resistance indicates the change is dominated by changes at the

metal/semiconductor interface. To further understand the degradation effects, device current and

Semi-insulating GaAs Substrate

GaAs Buffer

Undoped InGaAs

AlGaAs Spacer

Silicon Planer Doped Layer

n AlGaAs

nGaAs

Source Gate Drain

transconductance versus gate bias was measured and compared before and after a 10 minute

exposure. Comparison of the data shows the curves are shifted by the amount of pinchoff

change, which would indicate the change is due primarily to increased built-in potential and not

to donor compensation. Devices fabricated with Ti/Al gates were subjected to the same

conditions and performance did not shift, indicating the change due to hydrogen environments is

associated with Pt in the gate. Gates with varying thickness of Ti and Pt were exposed for 30

minutes. It was found that current changes are strongly sensitive to Pt thickness and relatively

insensitive to Ti thickness; the current was significantly less sensitive when Pt was reduced to 50

angstroms.

Figure 5. Change in transconductance (gm) and Idss versus time in 4% hydrogen and

100% nitrogen, at 270°°C. Reproduced from Ref. [9].

Data analysis from exposures to multiple temperatures and hydrogen concentrations

produced mathematical models, which account for the temperature and H2 interdependence on

lifetime [10]. Hydrogen concentrations from 0.0005 to 0.5% and temperatures ranging from 100

to 250°C were sued and 20% Idss degradation was used as the failure criteria. Adams, et. al.

[10] developed the following expression:

t = A Pnexp(Ea/kTj)

where:

t = median life (hours)

A = a proportionality constant

P = hydrogen partial pressure

Ea = activation energy

K = Boltzmann’s constant

Tj = junction temperature (°K)

Adams, et. al. [10] proposed a second model which replaces Ea in (1) with:

Ea = Ea – B ln P

Where B ln P is an interaction term, which in interpreted as the Freundlich potential, an

expression useful for describing absorption rates of gases at surfaces. This is the first work

reported on a model which allows the calculation of lifetimes with known values of hydrogen.

Hu, et. al. [11], reported on more testing of power and low noise PHEMTs with four

partial pressures and three temperatures which verified that time to failure was hydrogen partial

pressure dependent. PHEMTs sealed in packages made of Ni/Au plated Tungsten, with Ni/Au

plated Kovar lids were used for this test. Ten packages were vacuum baked at 350°C for seven

hours prior to assembly and five were left untreated; this allowed test to determine the effect of

temperature treatments on degradation. Observed failures in both groups were large, sudden

decrease in Idss (as much as 40%), with partial recovery as test time continued. Median time to

failure of the untreated packages was 800 hours at 125°C, and approximately 1600 hours for the

treated group. Although improvement was seen for the baked samples, the conditions used were

not sufficient to solve the problem. Additional testing in the four concentrations yielded data

which indicated the failures were dependent on both temperature and partial pressure. Failure

analysis indicated degradation was loclaized under the gate due to modification of the effective

gate voltage. Some compensation of donors were thought to occur in the later stages of a 9000

hour life test.

(1)

Saito, et. al. [13] reported on test performed to determine the maximum amount of

allowable hydrogen for spacecraft use of PHEMTs, assuming a twenty year mission life. In

addition, package bake experiments were performed to determine bake conditions necessary to

insure hydrogen levels in the package remained below the maximum allowable levels. Low

noise PHEMTs with 0.25uM Ti/Pt/Au gates were tested in 1% and 3% hydrogen at 125°C and

200°C. Median life at 125°C and 3% hydrogen was found to be 2200 hours, with an activation

energy of 0.52eV; degradation observed was a decrease in Idss and gain. From their work, the

authors concluded maximum allowable hydrogen to be 0.04% at 125°C and 0.46% at 70°C to

meet a 20 year mission life.

Package bake experiments determined a 250°C bake for 168 hours would reduce

hydrogen content from plated Kovar housings to 0.04% and plated Al/Si (A40) 0.004%. The

authors concluded most of the hydrogen evolved from plated nickel in the housings, and could

not be sufficiently baked to yield reliable assemblies. Some of these findings contradict some of

the results reported in [3], [4] and [11].

K. Decker reported [16] on testing of 200 GaAs PHEMT transistors to answer two

questions not answered in previous reports. The specific questions the tests were designed to

answer were:

• Does ambient or channel temperature drive this mechanism?

• Does electrical bias effect FET hydrogen degradation?

Tests of unbiased samples in nitrogen, 0.01, 0.1 and 1.0% hydrogen at four temperatures. Tests

were conducted for 2650 hours and the observed degradation in hydrogen was reduced Idss and

transconductance; in contrast, no degradation was observed in nitrogen. Figure 6 is the

Arrhenius plot for the hydrogen data and reference nitrogen life test data for a PHEMT amplifier

manufactured with the same process.

Figure 6. Median Life versus ambient temperature for GaAs PHEMTs.

Reproduced from Ref. [16].

Note that in the hydrogen test the unbiased samples (baked) and biased samples fit the same plot

only if they are plotted using ambient temperature for both. This indicates that the temperature

driving the failure mechanism is ambient temperature instead of junction temperature.

Activation energy was found to be 0.73eV. Data analysis found the data to fit the model

developed by Adams, et. al. [10] and described in (1):

t = A Pnexp(Ea/kT)

where:

t = the time to 10% Idss degradation

T = the ambient temperature in Kelvin

P = the hydrogen partial pressure in torr (%H2 x 7.6)

K = Boltzmann’s constant (8.615E-5)

(1)

Using the measured activation energy of 0.73eV and solving for n and A, then

Ea = 0.73eV

A = 5.46E-6

N= -0.7935

An extremely important finding from this work was that the ambient temperature (Ta), not

junction temperature (Tj), drives the hydrogen degradation mechanism. Examination of (1), the

time to failure expression, shows temperature to be a critical variable effecting time to failure; A

and n are assumed to be determined by the process and P is a function of the assembly. Time to

fail is exponential in 1/T, resulting in decreased life as temperature increases. Wear-out failures

are a function of junction temperature and, prior to the results of [16], it was assumed junction

temperature would apply to hydrogen degradation.

Bias applied to transistors causes heat to be generated, resulting in increased junction

temperatures and decreased lifetimes for wear-out failures. Common bias conditions for

PHEMT power devices can result in junction temperatures being 30 to 40C above ambient

temperatures. The 300uM gate-width devices tested in [16] were biased at 2.0V and 0.05A with

a resulting junction temperature rise of 24C.

Using data from Figure6 for PHEMTs tested in 1.0% hydrogen, and assuming a junction

temperature rise of 40C, a comparison of time to 10% Idss degradation has been made for Ta =

60C (Tj = 100C).

Ta = 60C; median life = 1.0E5 hours (11.4 years)

Tj = 100C; median life = 9.0E3 hours (1.03 years)

This is an increase of 11 times when based on ambient temperature, and demonstrates the

importance of ambient temperature determining lifetime of devices subjected to hydrogen.

Establishing that bias does not effect degradation also confirms that tests performed

without bias are valid and allows much simpler and cost effective tests to continue the study of

this problem.

In related work at University of California, Santa Barbara, S.S. Shi, et. al. [19], reported

on efforts being made to stabilize the surface of PHEMT power devices in order to make

breakdown voltages more reproducible and enhance power performance. Earlier work had

shown surface treatment with hydrogen ions to stabilize the surface through removal of excess

arsenic in the material. Excess As can form AsGa antisite defects which may be the cause for

non-reproducible breakdowns and degradation such as power slump. During surface treatment,

device parameters were found to degrade, and degradation was in prportion to the hydrogen ion

dose. When treated with the optimum dose for stabilization, maximum drain current dropped

10.8%, transconductance decreased 3.2%, and gain-to-date leakage current at 15 volts reverse

bias decreased 77%. Further exposures on van der Pauw structures resulted in decreased

mobility and donor concentration, with the most severe decreases at higher hydrogen ion doses.

This data shows a donor neutralization in support of [2]. Further, the authors observed recovery

of donor concentration and mobility after a 5 minute, 40°0C anneal, which explains recovery of

device parameters after exposure to higher temperatures.

To compare MESFET and PHEMT hydrogen sensitivities of devices manufactured on the same

process line, MESFET data from [8] has been compared to calculations for PHEMTs tested in

[16]. Reported hydrogen exposures of PHEMTs did not include enough data to compare

lifetimes of PHEMTs and MESFETs from the same process line and under the same exposure

conditions. However, models developed in [10] and verified in [16] allow calculation of

expected failure times for PHEMTs at various hydrogen partial pressures. Calculations for

median time to failure (MTTF) for PHEMTs in 10% hydrogen have been performed for devices

in [16], utilizing the expressions and parameters measured in that work. The calculated MTFF

for PHEMTs is then compared to data obtained from Figure 3 for MESFETs, from the same

manufacturer, measured in 10% hydrogen. The failure criteria for both was selected as a 10%

degradation of Idss. Results are shown in Table 1.

Table 1. Comparison of median life for PHEMTs and MESFETs from the same process

line, using calculated and measured data.

This comparison supports, as reported in [5], that PHEMTs are more sensitive to hydrogen than

MESFETs manufactured on the same process line.

In summary, every reported PHEMT test resulted in decreases in transconductance and

Idss. However, it is still not known what mechanism is responsible for the decreases. Changes

in barrier height were observed, and some believe the Idss shift to be due to donor compensation,

while others believe the shift is due to changes in material near the gate region. As can be seen

in the above summaries of work performed to date, and the fact that a consensus on the cause is

lacking, more work is needed to fully understand the failure mechanisms.

II.c. Indium Phosphide (InP) HEMT

Only one report relative to hydrogen effects on InP HEMT is available for review [9].

InP HEMT devices provide higher gain, mobility and cutoff frequencies, and have the high

breakdown characteristics common to the HEMT family of devices. Due to these characteristics,

they are favored in millimeter wave applications. During investigation of susceptibility to

hydrogen, observed failures were not the same as generally observed in other device

technologies. P.C. Chao, et. al. [9], reported that exposure to 4% hydrogen at 270°C resulted in

decreased gm, but Idss increased instead of decreasing. As in PHEMTs, some recovery toward

pre-exposure values was observed and room temperature exposure to nitrogen for four days

resulted in recovery. InP HEMTs and GaAs PHEMTs from [9], above, were processed on the

same device line and the result is a direct comparison of these two technologies. Retest was

performed on nine wafer lots, processed over a 18 month period. Eight of the lots exhibited the

same changes, while Idss degraded on devices from the ninth lot. Samples from the eight lots

had a mix of devices with and without nitride, thus ruling out nitride effects as a failure cause.

DEVICEPHEMTMESFET

150oC, 10% H2 Median Life96 hours, calculated from [16]200 hours, from [8] data

The increase in current was not consistent with compensated donors, but the cause of observed

changes was not understood and must be studied further. Test temperatures of 270°C may have

introduced extraneous effects. However, unpublished preliminary data, at lower temperatures,

confirms the observed trend of increasing current for InP HEMTs under hydrogen exposure.

In summary, all MESFETm PHEMT and Inp HEMT devices changed characteristics

when tested in hydrogen environments. Not all devices changed in the same manner, the cause

of which is not understood. It is possible that surface conditions (traps, defects, etc), stress from

gate metal or differences in process history cause these effects to be different. The problem is

still under investigation, and much work remains before the effect is fully understood. In Table

2, a summary of changes observed to date is given.

Table 2. A listing of some parameter shifts seen for exposure to hydrogen. When available,

the median life and activation energies are given.

III. Hydrogen Sources

Several sources of hydrogen in hermetic packages have been identified, with desorption

from ferrous metal package materials being the primary source [3, 4]. Hydrogen can be

“trapped” in the metal at structural imperfections; grain boundaries, precipitate interfaces,

dislocation cores, etc. These hydrogen trap sites will increase the metal hydrogen solubility by

orders of magnitude, and this trapped hydrogen can then be desorbed from the metal during

Device Idss gm Gain H2 Temp. Ea Ref.

Type (%) (oC) (eV)MESFET decrease decrease decrease 100 150 -- [2]MESFET increase decrease -- 0.1, 1.0 175, 200 0.4 [6]MESFET increase decrease -- multi 250 -- [7]MESFET decrease decrease decrease 10 140, 180 1 [8]MESFET decrease decrease -- 5 250 -- [12]PHEMT decrease decrease -- 4 270 -- [9]PHEMT decrease decrease -- multi 100 - 250 -- [10]PHEMT decrease decrease -- multi 60, 175, 200 0.3 - 1.0 [11]PHEMT decrease -- decrease 1, 3 125, 200 0.53 [13]PHEMT decrease decrease -- 0.01, 0.1, 146, 170, 0.73 [16]

1 186, 210PHEMT decrease decrease -- plasma 25 -- [19]

InP HEMT increase decrease -- 4 270 -- [9]

heating, such as that seen during burn-in. Some sources of the hydrogen [4] are reaction of H2

or H2O to free interstitial H at the metal surface; melt and casting processes; hydrogen

annealing, brazing or stress relief; air annealing in humid atmospheres; acid cleaning or

corrosion; and electroplating or electrocleaning. Gold and nickel plating used on packages is

permeable to hydrogen diffusion and may also be a source of hydrogen [4, 13]. During burn-ins

and steady state lifetests, untreated hermetic packages will desorb trapped hydrogen and levels of

1 – 2% will accumulate in the package [17]. As seen in previous sections, degradation will occur

at much lower levels.

Microwave absorber, sometimes used in package cavities to suppress oscillations and

interactions between microwave functions within the package, is a source for hydrogen [13].

The absorber is constructed of powdered iron fillings suspended in a carrier such as silicone

rubber or other plastic, and outgassing of hydrogen from this source may be dependent on

thermal treatments received by the absorber prior to package sealing. Y. Saito. Et. al. [13],

reported that outgassing from microwave absorber accounted for 16.3% of the total hydrogen

from sealed housing samples baked for 336 hours at 150°C; the samples had not been heat

treated prior to sealing. The author stated that hydrogen from this material would be much less

at 125°C due to a 1.6eV activation energy, measured during unpublished work at TRW.

Other metals used in microwave modules are also known to outgas hydrogen, but studies

of amounts and rates of outgassing are incomplete. One early indication is that cold rolled steel

is much than Kovar; Invar is also known to outgas significant hydrogen. It is generally found

that each metal tested has its own characteristics and that each candidate for use must be

characterized before use.

It is possible other materials utilized in module assemblies may be sources of hydrogen.

Epoxy is suspected by some to be a source; tests have shown that epoxies used in assemblies will

outgas hydrogen, but not to the same extent as Kovar and other packaging materials [13].

Subassemblies and circuit functions in complex microwave modules can be made up of many

components. Items such as circulators have metal housings, ferrite pucks made of iron powder,

circuit substrates and metal films and may be sources for hydrogen. Capacitors, resistors,

interconnects and substrate materials within the package should be characterized for possible

hydrogen outgassing. One method for determining hydrogen outgassing utilizes baking of test

coupons in sealed ampules, followed by analysis of the ampule contents with residual gas

analysis (RGA) after baking is complete. Another technique involves total melting of test

coupons and performing gas chromotography on the gasses formed during melting. This

technique must be used with caution, however, as hydrogen containing compounds can be

broken down and give erroneous indications of hydrogen.

Many potential sources of hydrogen exist in complex microwave modules and

evaluations assessing the risk or outgassing, utilizing some of the techniques described above,

should be performed before a design is considered complete. Table 3 is a listing of module

components and their role in contributing to the hydrogen problem.

Table 3. Hydrogen contributions from typical microwave module components.

IV. Possible Solutions

Several potential solutions have been suggested to eliminate the hydrogen degradation

problem. Suggestions include pre-seal package thermal treatment [4, 13], the use of package

materials with low hydrogen absorption [4], a change of barrier metals in gates [6, 15], the use of

non-hermetic packages [6, 15] and the use of hydrogen getter materials in the package [6, 15].

Each of these possibilities will be discussed in the following paragraphs.

IV. a. Thermal Treatment of Packages

Investigations have shown that thermal treatment of packages prior to sealing can

significantly reduce absorbed hydrogen [4, 13]. Monatomic hydrogen is the only form of

Yes

Isolator/CirculatorsInterconnects

Microwave absorber

Hydrogen contributor?YesYesNoYes

ProbableUnknown

Ferrous package materialsElectroplating

Ceramic substratesEpoxy

Component

hydrogen capable of diffusing through metals as an interstitial solute, but hydrogen can be

“trapped” in the metal at structural imperfections and incoherent boundaries such as grain

boundaries, dislocations, vacancies, micropores, precipitate interfaces, inclusions and particle

boundaries [4]. Absorbed hydrogen in excess of the lattice solubility will segregate to trap sites

within the metal and increase total hydrogen solubility by orders of magnitude. This hydrogen

trap site can be desorbed from the metal by thermal treatment. Each trap site has an associated

activation energy for out diffusion, and results in strong and weak trap sites. Studies using

thermal treatments of 150°C for 200 hours showed that thicker samples of Kovar desorbed

significantly larger amounts of hydrogen than thin samples. As an example, 0.06, 0.04 and 0.10

inch thick samples desorbed 4.0%, 2.0% and <0.02% hydrogen, respectively. The exact cause of

the differences are not known, but it is expected that it is due to differences in grain structure and

manufacturing processes. Samples showed increasing inclusion density with decreasing

thickness; it is assumed the finer grain structure of thicker samples will have strong trap sites and

the course structure of thin samples contain more weak traps. The weaker traps would be more

inclined to release trapped hydrogen during manufacturing processes (annealing, etc.) and thus

have less hydrogen to desorb during subsequent bakes. Microwave module housings are

generally made from sheet stock much thicker than 0.06 inches, and will have more available

hydrogen in strong trap sites and therefore may require thermal treatments to desorb the

hydrogen. Thermal treatments have shown that hydrogen can be baked out to acceptable levels

[4, 13]. One manufacturer has used this technique to produce modules acceptable for use in

spacecraft applications [13]. Proper bake procedures must be established to prevent bondability

problems with the post-bake housings. Bake temperatures of 350°C will successfully evolve the

hydrogen, but bondability begins to degrade. At elevated temperatures, Ni can diffuse through

gold and Ni oxides can form on the surface and degrade bondability. Y. Saito, et. al. [13],

reported that optimization of bake conditions for Au/Ni plated Kovar housings were established

to be 250°C for 168 hours in one. Post plating bake reduced evloved hydrogen from 0.6% to

0.0033% and bonding tests showed bond strength to be >7grams. A bake of 250°C for 168 hours

is severe and may damage the package finish. Studies of gold-over-nickel plating have shown

that nickel can diffuse through the gold and form a nickel oxide of the gold surface [4], causing

bondability problems. Schuessler, et. al. [4], reported most hydrogen could be desorbed from

Kovar samples with a 200 hour bake at 150°C, which is much less severe than the conditions

reported in [13]. More investigations are required before a treatment condition can be adopted.

IV.b. Package Materials with Low H2 Absorption

Materials with low hydrogen absorption have been studied for use as packaging

materials, with at least one material type used to solve the problem. An aluminum-silicon

(Al/Si) based alloy (A40) is a potential material for this use, and has been used by one

manufacturer for high reliability applications [13]. As in the Kovar case, post plate baking

reduced hydrogen levels to 0.002% after a 250°C, 168 hour bake. Aluminum does not absorb

hydrogen, and various alloys have been investigated for use in lightweight module applications

[31]. Alloys are necessary to gain the required rigidity needed in thin-wall applications such as

those needed in space applications.

A possible choice for packages with low hydrogen content is to manufacture them from

Kovar which has had hydrogen desorbed from it. Schuessler, et. al. [4], reported that one

package manufacturer was able to supply heat treated Kovar which did not desorb detectable

levels of hydrogen in subsequent bake experiments. The author stated the treatment was

performed at high temperatures and must be performed prior to plating.

When Kovar or nickel-plated Kovar are subjected to heat treatment in air, an oxide layer

forms and blocks hydrogen desorption [3]. Bare Kovar was treated for 100 hours at 320C and

nickel plated Kovar was treated for 24 hours at 320C. Subsequent desorption bakes

demonstrated that hydrogen levels had been reduced from 0.7 to 0.9% hydrogen to non-

detectable levels. Passivation of this type would be required prior to plating. An approach such

as this could be used if acceptable plating and module seal techniques could be developed for

oxidized metal.

IV.c. Gate Barrier Metals

The hydrogen degradation problem is caused when available hydrogen reacts with

platinum or palladium in the device gate structure. This catalytic reaction produces mono atomic

hydrogen which then diffuses through the gate metals and reacts with silicon dopants within the

material, thus causing degradation. A possible solution to this problem is to replace the gate

barrier metal (Pt, Pd) with a barrier metal which does not react with hydrogen. Amorphous thin

films of the Ta-Si-N type have shown good barrier properties and are thermally stable [30].

Tests have been performed on these materials with no observed hydrogen degradation [15].

However, additional tests are needed to determine if this material, which is applied via a

sputtering process, is adaptable to high production processes. Some areas of concern are sputter

damage and gate size repeatability.

Another approach would be to replace Pt and Pd with barrier materials such as tungsten

(W), molybdenum (Mo) or other suitable barrier to gold in the gate structure. Unpublished data

on materials such as these have shown excellent results, but again these materials must be

studied for adaptability to production processes. These materials are applied via evaporation, but

the temperatures at which they melt are much higher than that for Pt, Pd, Ti and Au. It remains

to be seen if normal gate lithography processes can be repeatable and adaptable to a high rate of

production. Unfortunately, work of this nature is considered proprietary by investigators, and

data is not yet available on them.

Work conducted at Texas Instruments (TI) has produced alternate gate metals with

dramatically improved performance in hydrogen environments. Activation energies have

increased from 1.0eV to 1.4eV, and time to failure in 10% hydrgogen at 150°C has increased

from 200 hours (1.2 weeks) for Ti/Pt/Au gates to 2.0E5 hours (22.8 years) for alternate gate

metals. This is three orders of magnitude improvement in MTTF, and gives a MTTF of 1.0E7

hours (1141 years) at 100°C, which is certainly sufficient for most applications. Qualification

and producability test have begun, and a form of this gate structure is planned for production

January, 1998 [31]. Metals used in this alternate metal approach are proprietary and cannot be

given.

IV. d. Hon-hermetic Packages

In typical hermetically sealed modules, desorbed hydrogen is trapped inside module

housings and cannot readily escape. Us of non-hermetic packages would allow the hydrogen a

path to escape to surrounding environment and thus would not be available to degrade devices.

Several considerations must be taken into account if this approach is to be used. However, the

use of a non- hermetic module should be considered very carefully. The use of non-hermetic

packages would allow ambient atmospheres to be present in the package. The presence of

moisture and ionic contamination may result in performance degradation or failure [32]; all

components within the assembly would need to be impervious to such moisture effects before

this approach would be viable. To date, there isn’t appreciable data available on susceptability to

moisture of GaAs and InP based devices. Studies pertaining to long term performance in this

condition would be required before this approach could be deemed acceptable. Other

contaminants (particles, chemicals, etc) would possibly be available for package ingress.

Strenuous pre-use environmental controls to prevent possible contamination would be required if

this approach were to be considered.

IV. e. Use of Hydrogen Getters

Hydrogen getters installed within the hermetic package could be a solution to the

degradation problem. Before the getter could be used, some considerations must be addressed:

(1) Getter capacity – The getter must have the capacity to react with and retain the total

quantity of hydrogen expected to be available through outgassing during operating

life of the module assembly. Information on total hydrogen to be desorbed and the

getter capacity with respect to that quantity would be required.

(2) Getter pumping/reaction rate – Desorption of hydrogen will occur over time at a rate

determined by package materials and thermal history. Investigations on this

desorption rate would be required and compared to the gettering rate of any material

used.

(3) Getter useable life – Data pertaining to the life of the getter, taking into account the

thermal cycles, reaction rates and total reaction would be needed. Most high

reliability applications of GaAs-based modules require long lifetimes and will subject

the assemblies to many thermal cycles. The getter material would be required to

remain effective during the mission life.

(4) Physical stability – Reactions with hydrogen may generate new complexes within the

getter or module cavity. This reaction must be able to occur without creating

particles or other contaminants within the sealed assemblies.

Hydrogen getters are commercially available from Allied Signal Aerospace (HMC

Getter, patent pending) [36]. This product shows gettering capability sufficient to be used in

microwave packages containing GaAs devices. The capabiltiies are listed as being able to

remove hydrogen over a temperature range of –55 to 150°C, and some other general data from

the getter data sheet are:

• Hydrogen reaction is irreversible

• Zero vapor pressure

• Maintains hydrogen level to less than 1 part per million (ppm)

• Maintains dew point to less than –100F

• Is bondable

• Mold to size is available

Test data shows, when the getter is used with Emerson & Cuming Eccosorb MFS-124 RF

absorber, with a volume ratio of Getter/Absorber = 1:2, that hydrogen outgassing from the

absorber is dramatically reduced. Control samples of the absorber produced 18,080ppm (1.8%)

hydrogen when baked at 125°C for 3792 hours, while the 1:2 mixture produced only 0.19ppm

(0.000019%) hydrogen during the same bake. This is a reduction of five orders of magnitude

and appears to be more than sufficient to solve hydrogen problems in microwave modules. The

getter system is available in several forms.

A getter of the type described above could be the solution to hydrogen degradation in

hermetic packages. Texas Instruments has utilized getters similar to the one described above

[33]. Engineers at TI have developed a getter design methodology which utilizes data on module

component desorption rates, getter pumping rates, total hydrogen available, mission thermal

profile and maximum allowable hydrogen within the module to design getters for application in

microwave modules using GaAs PHEMTs. The design is performed and then the getter is sized

to have capacity in excess of the design. Most of the work in this area is considered confidential

and cannot be given here.

Another possible getter could be designed containing titanium and platinum or palladium

[34]. The getter could be fabricated with alternating contiguous layers to form a sandwich-like

structure, using sputtering or evaporation. In a hermetic package, the getter would be secured to

the package or might be a coating applied to the lid used to seal the package. The amount of

titanium used would be designed to absorb all the hydrogen expected to be desorbed from the

package.

Titanium will absorb up to 67 atomic percent hydrogen [35], making it an excellent

choice for capturing and holding hydrogen. Either a palladium or platinum coating of over the

titanium serves two purposes; 1) it prevents oxidation or titanium, which would block hydrogen

flow into the titanium, and 2) the palladium would catalytically convert ambient hydrogen,

allowing it to diffuse into and be absorbed by the titanium.

V. Recommendations for High Reliability Applications

Hydrogen degradation of GaAs and InP based microwave semiconductor devices has

been identified as a serious problem, especially for high reliability applications where this non-

catastrophic degradation may reduce system performance to below acceptable levels. Due to

recent developments, however, acceptable solutions to the problem make the devices a viable

technology for use in space and other high reliability applications.

Solutions to the hydrogen problem will involve two basic scenarios and may involve

either short term or long term fixes for the problem. In the first case, existing designs utilizing

GaAs MMICs or discrete devices will have been committed to applications and a risk assessment

pertaining to potential hydrogen problems will be required. In the second case, designs under

development will require a hydrogen susceptability assessment and, if required, a solution put in

place to eliminate the risk of hydrogen degradation.

Short term fixes to the hydrogen problem could include making sealed packages non-

hermetic or designing and installing a hydrogen getter in the package. Getter use would require

packages capable of unsealing and resealing after installation of the getter material. If either of

these approaches are to be used the precautions noted in section IV should be followed.

Longer term fixes could include use of hydrogen getters, use of low hydrogen absorption

material for packages, using thermally treated packages and designing with devices which have

new gate barrier metals. Each of these have shown the potential for solving hydrogen

degradation problems. However, all except getters require more time to implement and would

best be used with new designs.

For each of the cases to be discussed in the following paragraphs, the user’s system

designers will be required to define and supply allowable failure rates for the planned mission

life, minimum acceptable performance for each functional block containing the devices and an

estimate of the mission temperature profile. These parameters, in conjunction with data for

device hydrogen degradation rates will be used to determine any actions necessary to meet or

exceed the system mission life.

IV. a. Existing designs

Existing designs will require an immediate risk assessment to determine if a solution to

the hydrogen degradation problem is required. The need for implementing a solution should be

based on the comparison of device degradation data, for each microwave function, to minimum

performance requirements established by the system engineers.

Acceptable data to be used in the risk assessment can be generated by exposing devices

to hydrogen, be obtained from prior test of devices, be deduced from prior hydrogen degradation

studies of the process, or by examination of changes in steady state lifetests during qualification.

In most cases, steady state lifetests are performed during qualification and the pass-fail

criteria will be based on the end of test performance (gain, etc. are required to meet

specifications). Typical test steady state lifetests are conducted for 1000 hours at 125°C with

DC bias. If hydrogen degradation is a problem, many of the designs will show gain and current

decreases under these conditions, but will still meet gain specifications after the test.

Examination of the before and after data would reveal the shift. To insure that data shifts are due

to hydrogen, unsealing samples and subjecting them to bakes in a nitrogen atmosphere will be

necessary. Partial recovery will indicate a hydrogen degradation problem and a solution to the

problem will be required if the data indicates the test time does not accurately simulate the

required mission life. If degradation is not evident, or if recovery is not achieved with a post-test

bake, the degradation will not be due to hydrogen.

IV. b. Designs in Development

Designs being developed for use in future applications should be characterized for

hydrogen degradation and, if needed, a solution which will eliminate the problem should be

included in the module design. The chosen solution will be vendor dependent, since not all

vendors will have the capability to implement all of the potential solutions. However, data

should be acquired which will allow a demonstration that the mission life can be met.

When the proposed solution is to replace the gate materials with one not affected by

hydrogen, it is recommended that qualification data should be available which shows the devices

are capable of passing, as a minimum:

• multiple temperature lifetests

• thermal cycle

• thermal shock

• susceptability to hydrogen

• susceptabiltity to moisture

The data should include hydrogen exposure test data that includes time to failure, pressure

dependence and the activation energy for hydrogen degradation.

When heat-treated packages or low moisture rate package materials are propsed,

analytical data showing that the amount of hydrogen to be expected during the mission life will

not degrade device performance to below acceptable levels.

Use of getters will require prediction equations which clearly show the getter is capable

of maintaining hydrogen levels to a value below that which would cause device degradation, and

that the getter will not degrade during the intended mission.

Implementing solutions to hydrogen degradation problems in designs utilizing GaAs of

InP based microwave devices will make these devices an excellent choice for use in applications

requiring lightweight and high efficiency performance.

REFERENCES

[1] EIA/JEDEC JC-14.7, Committee on GaAs Reliability, “JC-14.7 Failure Mechanism/Acceleration Factor List”

[2] W.O Camp, Jr., R. Lasater, V. Genova, R. Hume, “Hydrogen effects on reliability ofGaAs MMICs”, 11th Annual GaAs IC Symposium. Technical Digest, 1989, p.203-6

[3] P.W. Schuessler, D. Feliciano-Welpe, “The effects of hydrogen on device reliability andinsights on preventing these effects”, Hybrid Circuit Technology, vol. 8, no. 1, January 1991,p. 19-26

[4] Phillip Schuessler, Stephan Gonya, “Hydrogen Desorption from Base and ProcessedPackaging Alloy"” NIST Conference, April 1993.

[5] S. Kayali, “The Effects of Hydrogen on GaAs Device Reliability”, Notes from IRPSWorkshop on GaAs, April 11, 1994

[6] M.J. Delaney, T.J. Wiltsey, Min-Wen Chiang, K.K. Yu, “Reliability of 0.25uM GaAsMESFET MMIC Process: Results of Accelerated Lifetests and Hydrogen Exposure” GaAsReliability Workshop Digest, 1994

[7] William J. Roesch, “Accelerated Effects of Hydrogen on GaAs MESFETs”, GaAsReliabiltiy Workshop Digest, 1994

[8] K. Decker, “GaAs MMIC Hydrogen Degradation Study”, GaAs Reliability WorkshopDigest, 1994

[9] P.C. Chao, M.Y. Kao. K. Nordheden, A.W. Swanson, “HEMT Degradation in HydrogenGas”, IEEE Electron Device Letters, Vol. 15, No. 5, May 1994

[10] S.B. Adams, J.A. MacDonald, W.W. Hu, A.A. Immorlica, A.R. Reisinger, F.W. Smith,“Reliability of GaAs PHEMT MMICs in Hydrogen Ambient”, GaAs Reliability Workshop Digest,1994

[11] W.W. Hu, T.H Parks, P.C. Chao, A.W. Swanson, “Reliability of GaAs PHEMT UnderHydrogen Containing Atmosphere”, GaAs IC Symposium, 16th Annual Technical Digest, 1994,p. 247-250

[12] D.C. Eng, R.J. Culbertson, K.P. MacWilliams, “The effects of hydrogen and deuteriumincorporation on the electrical performance of a GaAs MESFET”, GaAs IC Symposium, 17 th

Annual Technical Digest 1995, p. 140-3

[13] Y. Saito, R. Griese, J. Kessler, R. Kono, J. Fang, “Hydrogen Degradation of GaAsMMICs and Hydrogen Evolution in the Hermetic Package”, Microwave and Millimeter-waveMonolithic Circuits Symposium Digest, 1995, p. 119-122

[14] G. Kelley, M. Cobb, D. Weir, M. Welch, M. Weig, “effects of Temperature andConcentration on Hydrogen Degradation of Psuedomorphic HEMTs”, GaAs ReliabilityWorkshop Digest, 1995

[15] Sammy Kayali, “Hydrogen Effects on GaAs, Status and Progress”, GaAs ReliabilityWorkshop Digest, 1995

[16] K. Decker, “GaAs PHEMT Hydrogen Sensitivity Study”, GaAs Reliability WorkshopDigest, 1996

[17] Sammy Kayali, “Hydrogen Effects of GaAs Device Reliabilty”, Gallium ArsenideManufacturing Technology Conference, 1996

[18] David P. Rancour, Sammy A. Kayali, “modeling of Hydrogen Effects in GaAs FETs”,GaAs Reliability Workshop Digest, 1996

[19] Song S. Shi, Ying-Ian Chang, Evelyn L. Hu and Julia J. Brown, “Surface Passivation ofGaAs-Based PHEMT by Hydrogen Ion Irradiation”, Material Research Society SymposiumProceedings, Volume 421, 1996, p. 401-406

[20] S.X. Jin, H.P. Wang, M.H. Yuan, H.Z. Song, H. Wang, W.L. Mao, G.G. Qin, Ze-YingRen, Bing-Chen Li, Xiong-Wei Hu, Guo-Sheng Sun, “Controlling of Schottky barrier heights forAu/n-GaAs and Ti/n-GaAs with hydrogen introduced after metal disposition by bias annealing”,Applied Physics Letters, vol. 62, no. 21, May 1993, p. 2719-21

[21] S.X Jin, L.P. Wang, M.H. Yuan, J.J. Chen, Y.Q. Jia, G.G.Qin, “Effects of hydrogen onthe Schottky barrier of Ti/n-GaAs diodes”, Journal of Applied Physics, vol. 71, no. 1, January1992, p. 536-8

[22] Y.G Wang, S. Ashok, “A study of metal/GaAs interface modification by hydrogenplasma”, Journal of Applied Physics, vol. 75, no. 5, March 1994, p. 2447-54

[23] D.E. Aspnes, A. Heller, “Barrier height and leakage reduction in n-GaAs-platinum groupmetal Schottky barriers upon exposure to hydrogen”, Journal of Vacuum Science Technology, B1(3), July-Sept., 1983, p. 602-607

[24] J. Chevallier, W.C. Dautremont-Smith, C.W. Tu, S.J. Pearton, “Donor Neutralization inGaAs (Si) by Atomic Hydrogen”, Applied Physics Letter, vol. 47, no. 2, July 15, 1985, p. 108-110

[25] A. Jalil, J. Chvallier, J.C. Pesant, R. Mostefaoui, “Infrared spectroscopic evidence ofsilicon related hydrogen complexes in hydrogenated n-type GaAs doped with silicon”, AppliedPhysocs Letters, 50(8), 23 February 1987, p. 439-441

[26] C. Canali, F. Castaldo, F. Fantini, D. Ogliari, L. Umena, E. Zanoni, “Gate Metallization‘Sinking’ into the Active Channel in Ti/W/Au Metallized Power MESFETs”, Electron DeviceLetters, vol. EDL-7, no. 3, March 1986, p. 185-187

[27] W.W Hu, P.C. Chao, P. Ho, R.J. Finke, A.W. Swanson, “Reliability of State-of-the-ArtGaAs Psuedomorphic Low-Noise HEMTs”, GaAs IC Symposium, 13th Annual Technical Digest,1991, p. 191-194

[28] William J. Roesch, “Thermo-Reliability Relationships of GaAs Ics”, GaAs IC Symposium,10th Annual Technical digest, 1988, p. 61-64

[29] Peter Ersland, Jean-Pierre Lanteri, “GaAs FET Switch Reliability”, GaAs IC Symposium,10th Annual Technical Digest, 1988, p. 57-60

[30] J.S. Chen, E. Kolawa, R.P. Riuz and M.A. Nicolet, “Stable Pt/Ge/Au Ohmic Contact to n-GaAs with a Ta-Si-N Barrier”, Material Research Society Proceedings, vol. 300, 1993, p. 255-260

[31] Private conversations with Ken Decker, Texas Instruments, Inc.; unpublished data

[32] S. Kayali, G. Ponchak, R. Shaw, “GaAs MMIC Reliability Assurance Guidelines forSpace Applications

[33] Private conversations with John Bedinger, Texas Instruments, Inc.; unpublished data

[34] Private conversations with S. Kayali, Jet Propulsion Laboratory

[35] Max Hansen, “Constitution of Binary Alloys”, McGraw-Hill, 1958

[36] Allied Signal Aerospace, “HMC GETTER DATA SHEET”, Technical Contacts are H.Mike Smith, 816.997.2603 and Jim Schicker, 816.997.2494

Appendix 1

A partial listing of results from early GaAs device reliability tests

FAILURE MECHANISM ACCELERATION FACTOR ACCELERATING MEDIAN LIFE REPORTED BY: SOURCE DATEDESCRIPTION (Activation Energy, Current CONDITIONS AT FAILURE (Author, Etc.) (Publication, Etc.)

Density Exponent, Etc.) SITE: (Temp., Voltage, Etc.) Hours/Temp.

FETs

Sinking Gate Activation Energy=2.5eV 245, 260, 275, 290, 310oC >1.6E9/150oC Roesch, et. al. Man-Tech 1988 Sinking Gate Activation Energy=1.3eV DC bias @ Vds=8V, 1/2 Idss, M/A COM Internal Unreported 1987 -

Tj=225, 245, 260m 290oC Ersland, et. al. 1988

Burn-out (Breakdown) Unknown (early failure) <150, <190, <225oC Russel, et. al. IRPS 1986 Au-Ga Intermetallic 24,000 hour life test Tch = 140, 150oC Postal, et. al. IRPS 1983 Schottky Degradation Tch = 58, 190, 225oC Riley GRW 1987

W-Ni gate contamination 4,000 hour test Tch = 83oC Postal, et. al. IRPS 1983

INTERDIFFUSION

Interconnect Activation Energy=2.4eV 250, 275, 300oC >8E7/150oC Roesch, et. al TQS Man-Tech 1988 Airbridge Activation Energy=0.43eV 200, 225, 250oC >2E5/150oC Roesch, et. al TQS Man-Tech 1988 Nichrome Thin Film Activation Energy=1.03eV 125, 175, 200oC 6E5/150oC Roesch, et. al TQS Man-Tech 1988

ELECTROMIGRATION Ohmic N-Factor = 3.5 (203oC) 0.455, 0.91 mA/cm2 Roesch, et. al. TQS Man-Tech 1988

Ohmic Metal Activation Energy=1.5eV <180, <240, <270oC Riley, et. al. GRW 1987 Interconnect N-Factor = 1.5 (300oC) 0.455, 0.91, 1.365 mA/cm2 >8E7/150oC Roesch, et. al. TQS Man-Tech 1988

Airbridge N-Factor = 4 to 5 (250oC) 1, 2, 4 mA/cm2 >2E5/150oC Roesch, et. al. TQS Man-Tech 1988 Nichrome Thin Film N-Factor = 3.0 (200oC) 2.5, 5, 4.5 mA/cm2 6E5/150oC Roesch, et. al. TQS Man-Tech 1988 Gate Metal Voiding Activation Energy=1.5eV Tch = 150, 190, 225oC Russell, et. al. IRPS 1986

Gate Metal Voiding Activation Energy=1.65eV Tch = 180, 240, 270oC Riley, et. al. GRW 1987 Interconnect N-Factor = 1.5, Ea = 0.7eV 150, 175, 200oC >1.2E6/150oC Thompson Internal 1990

03.9, 1.0, 1.6 mA/cm2 J = 4E5 AT&T Bell Labs

INTEGRATED CIRCUITS

Microwave Amplifier Activation Energy=1.75eV 225 & 240oC 8.4E6/150oC Rubalcava, et. al TQS Data Sheet 1989 Digital Counter Activation Energy=1.65eV 260 & 275oC 3.5E6/150oC Ingle, et. al. TQS Data Sheet 1989

MMIC Switch (sinking gate) Activation Energy=1.34eV HTRB @ 225, 250, 260oC Ersland & Lanterni GaAs IC 1988RF biased @ 200oC M/A Com Symposium

Preamp Activation Energy=1.3eV Tch = 185, 210, 235oC 2.5E5/150oC Spector & Dodson GaAs IC 1987AT&T Bell Labs Symposium

MMIC Amplifier Activation Energy=0.64eV Tch = 225, 255, 275oC 2E5/125oC Christianson, NRL IRPS 1992

JC-14.7 Failure Mechanism/Acceleration Factor List

A partial list of GaAs Device failure data - Reproduced fromEIA/JEDEC JC-14.7, Committee on GaAs Reliability, Minutes

SESSION III

HYDROGEN EFFECTS

October 12, 1997

Anaheim, California

Sponsored by JEDEC Committee on GaAs

In cooperation with the Electron Society of the Institute of

Electrical and Electronics Engineers, Inc.

Hydrogen Evolution of Packaging Materials

Test Plan

A. Identify the baseline hydrogen content of populated modules with no absorber or hydrogen getter.Evaluate (6) samples. The modules will be processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

B. Identify the baseline hydrogen content of populated modules fabricated with absorber. Evaluate (3)samples with Emerson & Cuming Eccosorb CRS-124-RF absorber. The modules will be processedin the following manner:• standard seam seal process (16 hour, 150C vacuum bake)• condition at 125C for 168 hours• measure hydrogen content inside the module (RGA)

C. Identify the hydrogen content of populated modules fabricated with absorber. Evaluate (3) sampleswith Emerson & Cuming Eccosorb CRS-124-RF absorber attached to a cover. The absorber wasapplied to the cover with Dow Corning Sylgard 577. The cover and absorber were vacuum baked at165°C and minus one atmosphere for 72 hours, prior to sealing. The modules will be processed in thefollowing manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

D. Indentify the hydrogen content of populated modules fabricated with getter. Evaluate (2) sampleswith Allied Signal getter compound, lot 118 attached to a cover. The compound had a tacky surfaceand was attached directly to the cover. The getter was 0.023 inches thick with a surface area of 1.20square inches. The modules will be processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

E. Indentify the hydrogen content of populated modules fabricated with absorber and getter. Evaluate(2) samples with Emerson & Cuming Eccosorb CRS-124 RF absorber attached to a cover and AlliedSignal compound, lot 118, loose in the module. The absorber was applied to the cover with DowCorning Sylgard 577. The cover and absorber were vacuum baked at 165°C and minus oneatmosphere for 72 hours, prior to sealing. The getter was previously used in another experiment(populated module no absorber). The getter was 0.023 inches thick with a surface area of 1.08 squareinches. The modules will be processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

F. Indentify the hydrogen content of modules fabricated with absorber and getter. Evaluate (1) samplewith Emerson & Cuming Eccosorb CRS-124 RF absorber attached to the cover and Allied Signalgetter compound, lot 118, loose in the module. The absorber was applied to the cover with DowCorning Sylgard 577. The getter was previously used in another experiment (populated module noabsorber). The getter was 0.023 inches thick with a surface area of 1.08 square inches. The moduleswill processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

G. Identify the hydrogen content of populated modules fabricated with absorber and getter. Evaluate (2)samples with Emerson & Cuming Eccosorb CRS-124 RF absorber attached to a cover and AlliedSignal getter compound, lot 198HC, loose in the module. The absorber was applied to the cover withDow Corning Sylgard 577. The cover and absorber were vacuum baked at 165°C and minus oneatmosphere for 72 hours, prior to sealing. The getter was 0.017 inches thick with a surface area of1.08 square inches. The modules will be processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

H. Indentify the hydrogen content of populated modules fabricated with absorber and getter. Evaluate(1) sample with Emerson & Cuming Eccosorb CRS-124 RF absorber attached to a cover and AlliedSignal getter compound, lot 198HC, loose in the module. The absorber was applied to the cover withDow Corning Sylgard 577. The getter was 0.017 inches thick with a surface area of 1.08 squareinches. The modules will be processed in the following manner:• standard seam seal process (16 hour, 150°C vacuum bake)• condition at 125°C for 168 hours• measure hydrogen content inside the module (RGA)

Summary of Gas Analysis

Sample SampleIdentification Conditioning Hydrogen Content (ppm)

A populated module 360no absorber 343post seal temperature conditioning 494168 hours at 125C 511

407121range = 121 – 511average = 373

B populated module 46100 (4.61%)CRS-124 absorber 46500 (4.65%)post seal temperature conditioning 47200 (4.72%)168 hours at 125C range = 4.61 – 4.72%

average = 4.6%

C populated module 526vacuum baked CRS-124 absorber, 61872 hours at 165C, -1 atmosphere 548post seal temperature conditioning range = 526 – 618168 hours at 125C average = 564

D populated module none detectedAllied Signal H2 getter, lot 118 none detected(0.023” thick)post seal temperature conditioning168 hours at 125C

E populated module none detectedvacuum baked CR124 absorber, none detected72 hours at 165C, -1 atmosphereAllied getter (0.024” thick) loose, lot 118post seal temperature conditioning168 hours at 125C

F populated module none detectedunbaked CRS-124 absorber,Allied getter (0.016” thick) loose, lot 118post seal temperature conditioning168 hours at 125C

G populated module none detectedvacuum baked CRS-124 absorber, none detected72 hours at 165C, -1 atmosphereAllied getter (0.024” thick) loose, lot 198post seal temperature conditioning168 hours at 125C

H populated module none detectedunbaked CRS-124 absorber,Allied getter (0.016” thick) loose, lot 198post seal temperature conditioning168 hours at 125C