GaAs and GaN MESFETs and HFETs Processing Technology and Reliability Relationships, A Review

Although accelerated life testing of low noise and power GaAs MESFETs under d.c. bias and RF operation has been conducted, some failure mechanisms remain to be of concern. We will address these concerns and will develop failure models to include AlGaAs/GaAs HFETs. The new set of reliability physics models then will form the starting point for development of physics based failure models for GaN HFETs devices. Processes in effect in GaN, but not in GaAs, owing to higher fields and much larger field, temperature, and strain coupling will be included. In this vein, we discuss the state of the art of testing and failure mode analyses of GaAs devices and comment on the relevance to the proposed work. The same philosophy will then be extended to GaN based HFETs. A dominant failure mode of power GaAs MESFETs is catastrophic burn-out which is difficult to interpret. The 'long-term' catastrophic failure is the final result of parametric degradation while the 'instantaneous' burn-out is caused by sudden events, typically electrical overstress. Other failure mechanisms inducing parametric degradation are surface degradation, backgating, gate electromigration, and degradation of Schottky and ohmic contacts. The status for GaN is much less clear since gate leakage and the combined effects of high field, heat, and strain are all inter related and combined may cause the 'instantaneous' burn-out. Only after these mechanisms are minimized, can the 'long-term' parametric degradation become evident. 1.1.a. Failure Modes and Mechanisms of GaAs MESFETs Surface states lower the maximum field in the gate drain region, due to the captured negative charge which decreases the impact ionization, and increases the gate-drain breakdown voltage, VB. Deep levels associated with surface states result in 'gate-lag’ and transconductance dispersion which are strongly correlated since the characteristic capture and release times of surface states are longer than that of the applied signal. The magnitude of gate-lag and gm(f) dispersion is proportional to the surface state density. Gate-drain burn-out is caused by avalanche breakdown which depends on surface characteristics and device layout and technology. A recessed gate design would improve it. Source-drain burn-out is of thermal origin at the drain contact, a remedy for which is the n drain edge geometry resulting in uniform current flow and current spreading. The burn out might be dominated by the substrate/buffer region reaching extreme temperatures leading to the sudden increase in drain current. These phenomena are expected to occur in GaN HFETs as well, and the detailed analyses is necessary in order to develop the appropriate degradation model. 1.1. b Failure Modes and Mechanisms of AlGaAs/GaAs HFETs The AlGaAs/GaAs HFET degradation mechanisms, beyond those for GaAs MESFETs, include deep levels in the barrier and changes in the 2DEG concentration. The I-V collapse in the dark, and persistent photo-conductivity are more related to the material quality than to the long-term device stability. The decrease in 2DEG density might be due to carrier de--confinement, enhanced by field-aided impurity diffusion at the heterointerface (would also occur in GaN HFETs). The defects, present or created by high field (temperature, strain) followed by hot electron capture, would reduce the available carriers. These anomalies also cause high levels of LF noise. Similar effects must undoubtedly take place in the GaN system. Extensive analyses coupled with test heterostructures will be undertaken to uncover the nature of these anomalies for the failure model development. Electromigration plays an important role in GaAs HFETs since GaAs, being a binary compound, may have a wide variety of surface conditions (various native oxides and their clusters, surface states etc.). Further, electromigration is influenced by conductor-line material parameters and inhomogeneities, as well as structural features of the conductor layout, etc. Avoidance of sharp corners or transitions from a wide conductor to a narrow one helps mitigate the problem. Therefore, the electromigration effect must be studied with the same microscopic geometries as those used for both GaAs and GaN devices and knowledge gained will be applied to the physics based reliability models. 2. GaN FET physics of failure/degradation The degradation mechanisms germane to GaN, in addition to those present in GaAs, are primarily related to surface traps, metal semiconductor and inter-metal diffusion, compound formation, interface and bulk defect states. However, local high fields (>> GaAs) coupled with strain and temperature as well as the increased hot phonon generation will alter the key GaN HFET degradation mechanisms. Clearly, the GaAs model can be a starting point to be followed with its expansion to incorporate the GaN specific mechanisms. A variety of "trap" related device effects have been observed which include transconductance frequency dispersion, current collapse, light sensitivity, gateand drain-lag transients, and restricted microwave power output. The preliminary activity directed toward characterizing these effects parallels similar developments in the GaAs-based technology. Electron capture-emission by surface and bulk traps affects the 2DEG density resulting in current collapse, and transconductance dispersion. Because the associated characteristic time is ~ 1ns<τ<1 s, the trapping limits device performance even at relatively low frequencies. In addition, the thermally activated traps contribute significantly to LF noise. Understanding the origin of the traps in GaN-based transistors, their physical and energy location, and the physical mechanisms involved in the trapping is critical for not just the optimization of device performance, but for reliability modeling and reliability optimization for the GaN HFETs. We report (Figure 1) on the study of the nature of bulk (GaN and AlGaN), surface states, and interface states along with their effects on gate leakage followed by the effect of passivation, including the in situ deposited Si3N4. Degradation caused by surface states and preexisting bulk and interface states are reversible. However, when new defects begin to be created and their density cascades due to the combined effects of high field, heat and strain, the resulting device degradation becomes catastrophic. Figure 1. Figures showing the location of two traps as well as the drain current collapse with wavelength. 2.1 Bulk buffer traps Bulk traps in early GaN MESFETs and HFETs have been investigated. The fitted photoionization thresholds located the two dominant defects at ~ 1.80 and 2.85 eV below the conduction band edge, and after the 0.55 eV and 0.2 eV Franck-Condon correction, respectively. The DX centers plausibly associated with O have been observed in AlGaN may also be present. The enhancement of the optically induced drain-current recovery for photon energies at or above the band gap, Eg, of GaN has been measured and in contrast, no such increase for photon energies above Eg of AlGaN has been observed, hence, placing the traps to be within GaN buffer layer. However, it is obvious that further investigations are necessary. For instance, Trap 2 appears to be correlated with MOCVD growth pressure since trap 2 density increases at lower pressures. Obviously, this area requires further insightful investigations in order to determine the nature of the traps present. 2.2 Bulk Barrier traps Bulk barrier defects results in the trapping of carriers injected from the gate and/or hot carriers injected from the channel and leads to current reduction by reducing 2DEG density. The lateral and vertical fields also enhance the charge emission from the barrier traps. The field effects, therefore, are particularly important and must be taken into account. Localized trapping centers within the bandgap in the vicinity of the gate where the gate potential defines the energy position of the trap level with respect to the Fermi level have been investigated. The trap levels depend on the Al mole fraction. Activation energies and concentration of the traps must be determined with respect to Al mole fraction and incorporated into the failure model.