Hot-Electron Degradation of AlGaN/GaN High-Electron Mobility Transistors During RF Operation: Correlation With GaN Buffer Design

Comprehensive RF stress-test campaign has been performed over AlGaN/GaN high-electron mobility transistor employing different GaN buffer designs, including unintentional doping, carbon doping and iron doping. No signature of gate-edge degradation has been found, and good correlation emerges between the buffer composition, subthreshold leakage current, and permanent degradation of the RF performance. The degradation mechanism, more pronounced in devices with parasitic buffer conductivity, involves the generation of additional deep trap states, the worsening of the dynamic current collapse, and the subsequent degradation of RF output power.


INTRODUCTION
Long-term reliability of GaN-based HEMTs (High Electron Mobility Transistors) is a key aspect for the successful integration of the promising GaN-based electronics in high-power microwave applications.During RF operation, GaN HEMTs are subjected to the most stressful conditions, involving high current density, high electric fields and high channel temperature.Consequently, reliability evaluation should include RF accelerated testing as an essential part: RF operation can indeed induce failure modes and mechanisms that are not observed during DC or thermal storage tests [1].
Among the key technological aspects of GaN HEMTs, the compensation of the unintentional n-type conductivity of the GaN buffer layers (e.g., by means of iron or carbon doping) is of essential importance for the optimization of the 2DEG carrier confinement, the substrate insulation, and the control of the short-channel effects [2]- [4].With this letter, we present an analysis of the relevant interplay between compensation strategy, pinch-off properties, and degradation mechanisms in RF-tested AlGaN/GaN HEMTs.
Results can be summarized as follows: (i) degradation consists of an increase of dynamic current collapse, with consequent degradation of RF performances; no DC degradation is observed; (ii) in particular, the worsening of current collapse effects is caused by an increase of the concentration of two traps with EA = 0.79 eV and σc = 6 × 10 −13 cm 2 and EA = 0.84 eV, σc = 4 × 10 −14 cm 2 , detected in unintentionally doped (u.i.d.) and 10 17 cm −3 C-doped buffer devices; finally (iii) a good correlation has been found between the subthreshold leakage current in untreated devices and the amount of RF power degradation under RF operation, thus suggesting an enhancement of hot-electron effects during RF operation when the dynamic load line reaches deep pinchoff conditions and the devices with poor carrier confinement do not completely turn off.Since this mechanism is enhanced by short-channel effects and by poor carrier confinement, it strongly depends on GaN buffer compensation characteristics.

EXPERIMENTAL DETAILS
Devices under test belong to fourteen wafers differing only for GaN buffer design.HEMTs were fabricated using the same process steps and layout, with a 0.5μm Ni/Au gate; they adopted an AlGaN/GaN heterostructure with nominal 25% Al concentration and 23 nm AlGaN thickness, and different buffer compensation, including either no doping (type I), 10 17 cm −3 C-doping (type II), 10 17 cm −3 Fe-doping (type III), or 10 18 cm −3 Fe and 10 18 cm −3 C codoping (type IV).A PECVD silicon nitride passivation was employed during the process in order to reduce surface leakage and surface-related current collapse.Before starting the RF tests, devices were subjected to load-pull characterization in order to identify optimal matching conditions.Load matching point was chosen as a compromise between maximum output power and PAE.Fresh devices from each wafer were subjected to a 24 h CW RF test at 2.5 GHz with quiescent bias at (VDS = 30 V, ID = 30% IDSS) and driven into a 6 dB compression point.Base-plate temperature was set to 40°C.The characterization protocol performed prior to and after the stress tests includes static and pulsed ID-VD and ID-VG measurements and Drain Current Transient Spectroscopy (DCTS) [5].

RESULTS
Devices subjected to 24-hours RF test experienced a degradation on the RF output power (ΔPOUT) ranging from −0.05 dBm to −1.1 dBm.From static measurements performed prior to and after the stress, no worsening of the gate leakage current was detected, even in devices with ΔPOUT = −1 dBm (Figure 1a).This suggests that the gateedge degradation (e.g.caused by converse piezoelectric effects, time dependent breakdown, or electrochemical GaN oxidation [6]) is not the dominant degradation mechanism.Likewise, no correlation can be found between the degradation of RF performances and the variation of the DC parameters (on average the devices experienced an IDSS decrease of ∼4%, with a Pearson's correlation coefficient |r | of 0.18 with the degradation of the output power).Conversely, the main evidence of device degradation is a significant worsening of the current-collapse.Figure 1b shows the pulsed ID-VD characteristics of a representative device with u.i.d.buffer (type I) which experienced a ΔPOUT of −1 dBm.Though no significant degradation is found in the reference quiescent-bias point (VG,Q;VD,Q) = (0V;0V), a remarkable drain-current degradation (−25% at VDS = 4 V) is found in the hot quiescent bias-point (VG,Q;VD,Q)= (VTH+0.5V;25V).Figure 2 shows the correlation (with a Pearson's correlation coefficient |r | of 0.77) between the increase of current-collapse and the decrease of RF output power among all tested wafers.By investigating the causes of reported degradation, it has been noticed that the ΔPOUT is more pronounced in those devices that feature worse pinch-off properties and enhanced short-channel effects (Figure 3).Though featuring similar subthreshold-slope and gate leakage-current, devices grown on u.i.d.GaN buffer have higher drain-induced barrier lowering effects (DIBL) higher source-to-drain leakage current (ID,LEAK), and experience higher POUT degradation than devices grown on 10 18 cm −3 Fe and C co-doped buffer.An intermediate behavior is shown by devices employing the other types of buffer, leading to a remarkable correlation (|r| = 0.82) between the ΔPOUT and the initial drain-source leakage (Figure 4a).The source-floating gate-drain breakdown voltages of u.i.d. and 10 18 cm −3 Fe and C co-doped samples are 171V and 155V, respectively.No correlation is found between ΔPOUT and the average power dissipated during the RF stress (Figure 4b).These results provide an evidence on the correlation between hot-electron degradation mechanisms, already observed in GaN-based HEMTs [7]- [10], and parasitic unintentional conductivity of GaN buffer layers.When devices are driven towards pinch-off, they cross regions of operation characterized by significant current densities and relatively high VDS and electric field magnitude.If parasitic buffer conductivity is present, source-to-drain subthreshold leakage is maintained beyond pinch-off, possibly generating a high density of highly energetic carriers (hot-electrons) which can induce defects in the AlGaN and/or in the GaN layers, e.g., through the direct damage of weak lattice bonds or the dehydrogenation of Ga vacancies or N antisites complexes [9].More in details, double-pulsed gm vs VGS characteristics (Figure 5a and Figure 6a) show that the increase of current dispersion after the RF testing is manifested in the drop of the dynamic transconductance (up to 60% in u.i.d.samples) accompanied by no VTH degradation.As reported in [11], the degradation of the dynamic transconductance could be ascribed to the generation of additional trap states localized in the access regions.Thanks to drain current transient spectroscopy (Figure 5b and Figure 6b), it can be noticed that the worsening of current-collapse in u.i.d. and 10 17 cm −3 C-doped buffer devices is caused by an increase of the concentration of the traps with activation energy EA = 0.79 eV and capture cross-section σc = 6 × 10 −13 cm 2 (labelled E3) and EA = 0.84 eV, σc = 4 × 10 −14 cm 2 (labelled E4) [12], [13].On the other hand, though devices with 10 18 cm −3 Fe-and 10 18 cm −3 C co-doping feature higher initial current dispersion (mainly caused by the trap E2, with EA = 0.56 eV and σc = 5 × 10 −15 cm 2 [14]), they experience negligible decrease of the dynamic transconductance and negligible increase of trap density after RF operations.The Arrhenius plot with the signatures of the detected deep-levels is reported in Figure 7.

CONCLUSION
RF degradation of AlGaN/GaN HEMTs not affected by gate-degradation has been studied.The degradation of the RF output power is correlated with the initial subthreshold leakage current and, consequently, with buffer compensation.In devices with non-optimal buffer compensation, the degradation of the RF output power is caused by the increase of pre-existing defect density, and the worsening of the current collapse.Though featuring higher initial current-collapse, devices with 10 18 cm −3 Fe-and 10 18 cm −3 C co-doped buffer experience improved subthreshold behavior and more reliable RF operations.

Fig. 1 .
Fig. 1.(a) IG-VG characteristics and (b) pulsed ID vs VDS characteristics acquired prior to and after the stress of a sample with u.i.d.buffer experiencing a ΔPOUT of −1 dBm.No gate-leakage increase and no significant IDSS variation are found.The worsening of the current-collapse is the main evidence of device degradation induced by RF operations.

Fig. 2 .
Fig. 2. Good correlation is verified between ΔPOUT and the worsening of current-collapse after RF test.

Fig. 4 .
Fig. 4. (a) Good correlation is verified between initial subthreshold source to drain leakage current and RF output power degradation.ID,LEAK is defined at (VGS;VDS) = (VTH−1V;40V).(b) No correlation is found between power dissipated during the RF stress and the degradation of the RF output power.

Fig. 5 .
Fig. 5. (a) Pulsed gm vs VGS characteristics and (b) DCTS performed prior to and after the RF test on a sample with u.i.d.buffer (type I).Stress induces a remarkable gm droop in the hot quiescent bias-point (VTH+0.5V;25V),with the concentration increase of the pre-existing deep levels E3 and E4.

Fig. 6 .
Fig. 6.(a) Pulsed gm vs VGS characteristics and (b) DCTS performed prior to and after the RF test on a sample with 10 18 cm −3 Fe-& C-doped buffer (type IV).After the RF test, only 10% gm droop is found in the hot quiescent bias-point (VTH+0.5V;25V),with slight increase of pre-existing trap E1, and no relevant increase of trap E2.