The simplest definition of reliability is quality over time. Since time is involved in reliability, it is often measured by a rate. Just as quality is usually measured in terms of rejects (or un-quality), reliability is measured in terms of failures (or un-reliability).
Traditionally, the measurement of electronic failures has been straightforward. If one assumes all failure rates are constant, as they might be in a large system or machine, then a MEAN time between failures (MTBF) would be expected. In contrast, most integrated circuits, including GaAs devices, follow the lognormal distribution, which rarely approximates a constant rate.
Historically, failure rates were measured in percent failed per thousand hours of operation. The modern unit of failure commonly used today is failure-in-time (FIT). A FIT is also a unit of failure (or a Failure IT) that is equivalent to one failure per billion device hours. For comparison, one FIT is equivalent to 0.0001% per thousand hours, and 1% per thousand hours is equivalent to 10,000 FIT. However, a single rate is not sufficient to describe the reliability of semiconductors since their failure rates change over their lifetimes.
Generally, semiconductors have a very low wear-out failure rate early in life, and then have increasing failure rates as they wear out. At a point when about half of the devices fail in a group of circuits, the failure rate begins decreasing again. A very small part of an IC's population may fail early in life. These early failures have been associated with manufacturing or assembly defects. The early failures are sometimes called "infant" failures. As semiconductor reliability improves and more samples are stressed, the early failures become easier to detect and eliminate.
Failure mechanisms in GaAs device technologies can be significantly different than those observed for traditional Si devices. First of all, the metallization used is primarily composed of Gold, which is more conductive than aluminum used in conventional silicon device processing, and is also less susceptible to electromigration (electromigration is a diffusion process, diffusivity scales with melting point and Gold’s melting point is much higher than Aluminum). Gold can also be less susceptible to corrosion than Aluminum. Lastly, Gold eliminates the potential for Au/Al intermetallic problems during assembly since Gold bond wires are typically available.
Secondly, the active device used in mature GaAs ICs is the MESFET. Unlike a Si MOSFET, the gate is formed by a Schottky metal contact to the channel, instead of using a gate oxide. This eliminates the primary failure mechanisms found in MOS devices. Because of this Schottky configuration, the MESFET is relatively immune to surface effects and ionic contamination which plague silicon devices. In addition, GaAs devices are not susceptible to radiation degradation caused by the sensitivity of gate oxides in Si CMOS devices. Newer GaAs active devices, pHEMTs and HBTs, also have advantages over MOS devices and similar immunity to typical silicon surface problems.
The last major component of the process is the bulk wafer material itself. GaAs is actually a semi-insulator except in areas where it is implanted with silicon or in epitaxial layers. Because of its higher bulk resistivity, roughly 1000 times more resistive than silicon, GaAs is much less sensitive to the isolation and latch-up problems associated with silicon and silicon CMOS. There are other GaAs properties that lend themselves to better reliability, like lower electric fields at peak electron velocity, but they are minor compared to the major groups that have been discussed.
Common failure mechanisms in GaAs based devices
Interdiffusion mechanisms
The primary failure mechanism for MESFET pHEMT ICs and HBTs are "sinking gates." Sinking gates are caused by gate metal interdiffusion into the channel. This interdiffusion causes parametric shifts in several device parameters because the effective channel thickness is reduced. The largest change is decreased channel current so that parameter is typically used as the failure criterion. A 20% change in channel current is a common definition of a MESFET failure. In addition to channel current changes in an FET with sinking gates, channel resistance increases and the magnitude of the voltage required to pinch-off an FET is reduced (this usually means pinch offs are more positive). Sinking gates have never been catastrophic and they are self-limiting in a sense, because as the channel current decreases so does the power in the FET and thus the temperature is lowered causing the gates to sink more slowly. Eventually, one could expect the channel to be severed completely by the gate and become open, but this condition is rarely reached. The sinking gate mechanism has been observed at various temperatures and biases, but degradation is accelerated by temperature without bias or RF drive.
Gate degradation can be observed using cross-sections formed with a Focused Ion Beam (FIB). The movement of the metal gate at the GaAs surface is dramatic after high temperature aging. Some metal voiding is also present in the degraded gate, because of the mass of material that has moved into the GaAs. Operation at the maximum rated temperature (150°C) would be expected to exceed 2,000 years before a 1dB change could be observed. This expected longevity of sinking gates is acceptable in terms of commercial reliability goals, and is not considered as a threat to device lifetimes under normal operating conditions. Although gate sinking can be induced by high temperature acceleration, it has not yet been observed under nominal use conditions.
The failure mechanism for first layer interconnect begins with an interdiffusion mechanism. The interconnect is composed of a layered structure of titanium, platinum, and gold. When these metals interdiffuse, the resistance of the interconnect increases. Auger studies indicate that the metals intermix, and the whole stack becomes homogeneous. On a percentage basis, the resistance change can be as high as 250%. But on an absolute basis, a 50% change is roughly as much as the process window is wide, or 40 milli-ohms per square cm.
Implanted resistors have been studied to evaluate Ohmic contact failure mechanisms, but Ohmic degradation mechanisms have been elusive. Implanted resistor degradation has been found to be caused by changes in the contact resistance. Failure analysis on degraded FETs has shown that Ohmic metal does diffuse into the GaAs, but the physical diffusion seems to have a minimal effect electrically on the FET performance, especially compared to sinking gates. In general, Ohmic annealing is beneficial to circuit performance.
Electrostatic discharge failures
Electrostatic Discharge (ESD) has been the leading cause of failure in the field, and ESD failures scale inversely with device size. Therefore, efforts to reduce ESD sensitivity by design and handling countermeasures will become increasingly more important as device sizes continue to decrease.
The problem is difficult to model and analyze since the various sources of electrostatic energy—such as the human body, testing equipment, and accumulated free charge—all have different electrical characteristics. In addition, the abrupt and intense nature of a typical ESD event forces the devices that absorb the discharged energy to operate under high injection conditions, where the analysis is quite complex. Providing adequate protection against ESD also requires effective thermal distribution within the discharge area in order to avoid either dielectric damage, semiconductor melting, or metal spiking. To address these issues appropriately, each pad must be protected by a device capable of sustaining the discharged energy with no internal damage, preferably without compromising process complexity, total chip size, and electrical performance.
Electromigration
While metal/semiconductor interdiffusion is the most common wear out mechanism, and it occurs in GaAs contacts, interconnects, and resistors, electromigration is another common failure mechanism, and it also occurs in interconnects and resistors. If life testing is conducted under bias, electromigration can eventually occur, which causes catastrophic open circuits.
The failure mechanism for plated gold interconnect and air bridges in MMICs is electromigration (See illustration of air bridges in an InP RF device in figure 3). Under high current density stress, mass transport occurs because of the "electron wind" in the metallization. Voids form along the plated gold, and eventually the interconnect fuses open, the nitride passivation will crack, and molten gold will flow out of the failure site. Less than a 25% change in plated gold resistance has ever been observed before the catastrophic failure, and usually the pre-fusing degradation is negligible.
Figure 3. InP Heterostructure Barrier Varactor (HBV) that utilizes air bridges to minimize losses due to extrinsic resistance.
Other failure modes
The failure mechanisms for capacitors is Time Dependent Dielectric Breakdown (TDDB). Many tests have been conducted to evaluate dielectric performance. After the capacitor dielectric was changed to plasma deposited nitride in 1985, lifetest failures were non-existent. As the capacitor dielectric thickness was reduced from 2000 angstroms to 500 angstroms, voltage acceleration testing became more and more effective. Several improvement efforts have reduced the defect density in capacitors and they are now approaching theoretical maximums for voltage breakdown and longevity under voltage stress.
Other possible (less common) failure mechanisms are surface charge effects, leakage effects, ohmic contact degradation, burn-out, channel compensation, Schottky contact degradation, carrier diffusion, substrate via cracking, sidegating, gate electromigration, passivation cracking, interconnect-airbridge contact degradation, hydrogen-gate interdiffusion, capacitor dielectric breakdown, interlevel dielectric breakdown, and Ohmic contact electromigration.
Specific concerns in Space applications:
Hot electron effects
Another important degradation and failure mechanism that was not discussed in the previous section is the hot electron degradation effect, which is particularly important in the use of III-V devices under cryogenic conditions. The traditional temperature acceleration in life testing is more commonly used to predict reliability of GaAs devices, by simply evaluating experimentally the activation energy and then substituting actual use conditions into the Arrhenius equation. The problem with temperature-accelerated stress experiments is that activation energies for GaAs tend to be quite high, and extrapolation to use conditions can give values for predicted meant time to failure that are too optimistic. One of the reasons is that thermally induced degradation is based on interdiffusion of the different materials involved, and this depends on temperature and not so much on bias conditions. On the other hand, a high-bias stress which is related to the hot electron instabilities in the channel, may present a worse case scenario when devices are operated at room or low temperatures in a real environment for a long time period. Such phenomena have been attributed to the formation of deep level defects generated during hot electron and impact ionization conditions. These are caused by the presence of large electric fields in the device channels and barrier layers. Hot electron effects are quite likely to develop in RF applications (mainly in high electron mobility transistors HEMTs) because in order to be operated at microwave and millimeter wave frequencies the peak channel electric fields are very large even for low drain biases. Hot electron degradation has been shown to cause threshold voltage shifts, breakdown walk out, transconductance and cutoff frequency degradation, and the so called “power slump”.
Other reliability concerns specific to space applications
Some of the unique space requirements for these devices demand reliability for five years of operation at temperatures near 80 Kelvin. As some of these GaAs amplifiers will be routinely expected to provide in excess of 100mW of power, large thermal gradients are likely to add an additional concern in device reliability. Further concerns include the inherent lack of hermeticity. Due to the lack of availability of a non-absorbing window material many of these devices do not operate under hermetically sealed conditions. They are assembled in a block, which has an opening for collecting the signal. Environmental degradation of GaAs and GaAs metallization is then also a relevant study to be included, and it is important to simulate pre-launch storage conditions.
Table 2. Space reliability of Compound semiconductor Transistors (most based on GaAs).
Specific device
|
SEU
|
TID
|
Displacement damage
|
Device parameter monitored
|
Operating
thermal range and/or electrical stress
|
Ea from life testing (or expected mean time to failure)
|
ESD class
|
Other information
|
C-HIGFET Complementary heterostructure insulated gate FET- 1k SRAM
|
LET > 90 MeV.cm2/mg several ions
Ref [1]
|
|
|
Bit error cross-sections under dynamic conditions
|
5 K to 400 K
|
|
|
|
AlGaN/GaN HEMTs (high electron mobility transistors)
|
|
|
40 MeV protons, fluence up to 5 x 1010cm-2 – max frequency of oscillation changed by 5 to 25% [11]
|
Drain-source current vs drain-source voltage, gate current, reverse breakdown, maximum frequency of oscillation.
|
|
|
|
GaN based device, in optoelectronic applications they are 1order of magnitude harder than GaAs.
|
InAlAs/InGaAs?GaAs HEMTs
|
|
|
|
Transconductance degradation (used 10% to get Arrhenius plots in [20]).
|
|
Ea=1.8eV, MTTF=2x107h at Tch=125C and Jd=200mA/mm [20]
|
|
|
AlGaAs/InGaAs pseudomorphic high electron mobility transistors (PHEMTs) used for high speed switches
|
|
|
|
Gate current, pinch-off voltage, channel on-resistance, drain saturation current
|
As high as 250ºC (will last > 9,000 hours at this temperature)
|
Ea is assumed, MTTF estimated to be 108 hours at 50ºC [2]
|
|
Change in RF insertion loss was < 0.1 dB or < 2% degradation in power loss- Primary Failure mechanism is gate sinking
|
HFETs (heterojunction field effect transistors)
|
|
|
Transistors were still functional after 5 x 1014 - 24 GeV protons – Vth shifter from 75 to 120 mV, 10-5% decrease in transconductance [9]
|
Various: transconductance, threshold voltage, drain saturation current (gate body leakage), drain current.
|
They “work” from 4 K [7] up to 400ºC (673 K) [8], however, there is an exponential increase in gate body leakage and in drain current.
|
On-going life testing at JPL indicates less than 5% degradation in reverse saturation current after 1000 hours at 240ºC [10]
|
<1 for isolated devices [10]
|
Devices work over a wide temperature range, however, device characteristics do change with temperature, requiring characterization at the desired use T.
|
MESFETs, MODFET, N and P HIGFETs
|
|
Co60 at dose rate 106 rads/hr. After 80 Mrads, transconductance degraded only 2.3%, drain saturation current increased only 5.5% [6]
|
|
Various: transconductance, threshold voltage, saturation current
|
|
|
|
Threshold voltage shift ranged from 5 to 14 mV depending on device type.
|
HBT (heterostructure bipolar transistor) AlGaAs/GaAs
|
|
|
|
RF output power and current gain, collector current.
|
Up to 218
C junction temperature [16].
|
Ea = 1.75eV life tests extrapolate MTTF = 1.5 x 108 h at junction T = 125ºC and Jc=3kA/cm2 [15]
Separate tests performed at 26 kA/cm2 found Ea=0.42eV with MTTF of 2020 h at junction T=218ºC (corresponding to base plate T=110ºC [16]
|
|
Most mature technology in the HBT family. Strong dependence of degradation on electric field (not just temperature).
|
InGaP/GaAs HBTs
|
|
|
|
Current gain (beta)
|
|
Ea= 2eV, MTTF 106 h at junction T=200C and J=60kA/cm2 [17]
Separate study found Ea=0.7eV for J=50kA/cm2 [18]
|
|
Reliability is strongly dependent on fabrication processes, different studies find widely differing predicted MTTF
|
InP RF devices:
InP based devices are of interest in space applications mainly for communications, and include digital and mixed signal. InP devices have also demonstrated superior radiation tolerance to similar GaAs devices. Some of the advantages of this material for devices related to aerospace technologies include ultra high speeds (over 100 GHz), low phase noise, good thermal conductivity, and very good radiation hardness.
Table 3. Space reliability in InP devices (or using InP substrates):
Specific device
|
SEU
|
TID
|
Displacement damage
|
Device parameter monitored
|
Operating
thermal range and/or electrical stress
|
Ea from life testing (or expected mean time to failure)
|
Other information
|
InP based HEMTs (high electron mobility transistors)
AlInAs/InGaAs/InP
|
|
|
|
10% degradation in transconductance for Arrhenius plots [12]
|
Up to 2V drain voltage (Vd)[12]
|
For Vd= 1 V, Ea =1.8eV, projected life time 3x107 hours at 125C channel temperature. For Vd=2V, Ea decreased to 0.8eV [12]
|
High frequency performance is increased at Vd=2V operation with potential realiability problem – presently an order of magnitude lower than GaAs counterparts.
Hot electron degradation has also been reported to cause permanent negative threshold voltage shifts [19].
|
InP HBT (heterojunction bipolar transistors) InP/InGaAs
|
|
|
90Sr/90Y 100mCi beta-radiation source upt to 2.7x1016 e/cm2 (620 Mrad (InGAAs))- only 9% decrease in collector current and 7.5% decrease in beta [13]
|
Transistor curves (Ic at various Ib vs Vce), DC current gain (beta), beta vs collector current, and diode saturation current
|
|
|
Polyimide passivation appears to increase radiation resistance compared to earlier studies [13].
|
Resonant tunneling diodes (RTDs) – InGaAs and AlAs on InP substrates
|
|
No change in I-V characteristics after 1 Mrad(InP) using Co60 gamma source [14]
|
Slight change in IV characteristics after 3.5x1011/cm2 55 MeV protons and after 5 x1010/cm2 high energy neutrons [14]
|
RTD I-V characteristics
|
|
|
Slight change in IV curve after neutron irradiation recovered completely at room temperature after a few hours.
|