Total-Ionizing-Dose Effects at Ultrahigh Doses in AlGaN/GaN HEMTs

Total-ionizing-dose (TID) effects in AlGaN/GaN high-electron-mobility transistors (HEMTs) are evaluated by dc and low-frequency noise measurements. Devices with and without passivation layers are irradiated with 10-keV X-rays up to 100 Mrad(SiO2) under different bias conditions. Irradiated devices show significant electrical shifts in threshold voltage and transconductance. At doses < 10 Mrad(SiO2), the TID-induced effects are related to the passivation of preexisting acceptor-like defects via hole capture, which induces negative threshold voltage shifts and improvement of transconductance. At doses $>$ 10 Mrad(SiO2), dehydrogenation of defects and impurity complexes leads to the creation of acceptor-like defects, which degrade the transconductance, shift positively the threshold voltage, and increase the low-frequency noise. Effects are enhanced in unpassivated devices and when the gate is biased at high voltage.


I. INTRODUCTION
T HE demands on electronics working in harsh radiation environments have increased in the last several decades. Experimental fusion reactor facilities, nuclear waste depositories, and next-generation particle accelerators require chips able to withstand ultrahigh ionizing radiation doses on the order of >10 Mrad(SiO 2 ). For example, the trackers of the future High-Luminosity Large Hadron Collider (HL-LHC) at CERN, Switzerland, will be exposed to total ionizing doses (TIDs) up to 1 Grad(SiO 2 ) over ten years of operation [1].
On the other hand, power and RF systems often incorporate compound semiconductor devices, including GaN-based highelectron-mobility transistors (HEMTs). AlGaN/GaN HEMTs are widely used in high-power and radio frequency applications due to their enhanced carrier mobility in the twodimensional electron gas (2-DEG) and their high breakdown electric field [13]. The absence of gate dielectrics in AlGaN/GaN devices is a key factor for their high tolerance to ionizing radiation [14]. However, in the last ten years, several works have pointed out significant TID sensitivities of AlGaN/GaN HEMTs; these may prevent the proper functioning at the relatively low doses typical of space applications and are likely to be even larger in high-radiation applications [15], [16], [17], [18], [19], [20], [21], [22], [23].
X-ray and 1.8-MeV proton irradiations revealed that AlGaN/GaN HEMTs often exhibit threshold voltage shifts and reductions in transconductance [15], [16], [17]. These TID and DD effects are related primarily to trap activation and/or neutralization in the AlGaN and GaN layers with strong sensitivities to the bias applied during irradiation [15], [16]. The results of proton irradiations are often complicated to interpret, as they combine TID effects with displacement damage (DD) effects. X-ray irradiations in recent works focused on the exploration of pure TID mechanisms in AlGaN/GaN HEMTs. However, the cumulative dose in these studies is typically <1 Mrad(SiO 2 ), much lower than the TID often induced by proton irradiations, which is often >10 Mrad(SiO 2 ) [15], [16], [18]. Hence, the pure TID response of GaN HEMTs at ultrahigh doses is still unknown. Its exploration is useful for improving knowledge of basic degradation mechanisms in GaN-based HEMTs.
This work explores ultrahigh-dose effects in a developmentstage AlGaN/GaN HEMT technology through dc and lowfrequency noise measurements. The results evidence complex synergies between TID effects and electrical stress, each of which is strongly influenced by applied biases. The largest threshold-voltage shifts are observed for this technology when negative-gate bias is applied during irradiation, i.e., when devices are irradiated in the OFF-state. On the contrary, the largest transconductance degradation occurs under ON-bias irradiation. Unpassivated devices show larger sensitivity to TID irradiation and bias stress than passivated devices, reinforcing the strong role of hydrogen transport and reactions on the radiation response and reliability of GaN-based HEMTs.

A. Devices Under Test
The GaN HEMTs under test were fabricated in a development-stage AlGaN/GaN technology at the University of California at Santa Barbara (UCSB), USA [22]. The active region consists of 200 nm of unintentionally doped (UID) GaN that is grown by Ga-rich plasma-assisted molecular beam epitaxy on n-type free-standing GaN substrate having layer doped with Fe and C (Fig. 1). The transistor channel width is 150 µm and the channel length L g is 0.7 µm, with gate-todrain separation (L gd ) of 1 µm and gate-to-source separation (L gs ) of 0.5 µm. Transistors in this study are built in two configurations: 1) passivated by a thick SiN x layer (normal configuration) and 2) unpassivated (for comparison), as shown in Fig. 1(a) and (b).

B. Test Conditions
Irradiation tests were conducted using a 10-keV X-ray irradiator at a dose rate of 3.8 Mrad(SiO 2 )/h for a total exposure time of ∼29 h to reach 100 Mrad(SiO 2 ). All doses and rates are referred to equilibrium doses in SiO 2 for consistency in calibration and to facilitate comparison with other works [15], [16], [17], [18], [19]. Considering the relative atomic numbers, we infer from comparative studies of similar materials that the doses in GaN may be 2×-2.5× the quoted equilibrium SiO 2 dose [24], [25], [26].
The radiation exposure was stopped at several steps. At each step, devices were kept with all terminals grounded for about 60 s before transistors were electrically characterized. This minimizes annealing time and allows the device to stabilize before measurements. After completion, devices were annealed at room temperature (RT) for 27 h. Several bias configurations were applied to the HEMTs during irradiation and annealing, as shown in Table I. The dc static response and low-frequency noise of the transistors were measured at RT before exposure and at several irradiation steps. At least two devices were tested for each set of conditions; the representative results are shown below. The threshold voltage V th is defined as V gs−int − V ds /2; V gs,int is extracted in the linear region (V ds = 0.1 V) as the gate-voltage axis intercept of the linear extrapolation of the I d −V gs curve at the point of its maximum first derivative.

C. Evaluation of Electrical Stress Effects
Irradiations up to 100 Mrad(SiO 2 ) require about ∼29 h. To distinguish degradation induced by electrical stress from that induced by irradiation, the GaN-based HEMTs were stressed without X-ray under the bias conditions of Table I for the same amount of time required for devices to be irradiated up to 100 Mrad(SiO 2 ). Fig. 2 shows the electrical stress-induced degradation for (a) passivated and (b) unpassivated devices irradiated in "CUT-OFF"-bias condition, which is the case in which the largest parametric shifts are observed. The stressed HEMTs exhibit increases in leakage current and shifts of threshold voltage V th . The V th undergoes a negative and rapid shift in the first 1 h. After ∼29 h, the passivated HEMT biased in the "CUT-OFF" condition shows an increase of maximum drain current of 4%, while the passivated HEMT shows a decrease of maximum ON-current of 7%. For all biases, the unpassivated HEMTs show greater degradation than passivated devices.
Often electrical-stress-induced effects of HEMTs may be substantial and comparable to the TID-induced effects. For this reason, Sections III and IV report and compare the responses of biased irradiated and unirradiated devices at similar times.

A. TID Tolerance at Different Bias Conditions
The dc characteristics in the saturation regime (V ds = 5 V) of passivated GaN-based HEMTs are shown in Fig. 3 when devices are irradiated and annealed under the "CUT-OFF" and "ON"-bias conditions. During exposure, the drain leakage current at V g = 0 V increases significantly regardless of the applied bias, a signature of drain-to-gate leakage. The V th shifts monotonically to negative values as large as −270 mV in "CUT-OFF"-biased devices and to positive values up to 50 mV in "ON"-biased devices. The "ON"-biased devices exhibit degradation of the transconductance g m , which is evident as decreases in the slopes of the I d − V g curves for -3 V < V g < 0 V. Slight performance recovery is visible after 27 h of RT annealing, suggesting the formation of a significant density of stable defects. Fig. 4 shows the effects of applied bias on the degradation of the maximum drain current I on , defined here as  Degradation of maximum drain current I on as a function of dose in passivated GaN-based HEMTs. Dotted lines refer to the electrical stress-induced degradation, where devices were tested without X-ray exposure; continuous lines refer to test results with X-ray exposure. Irradiations were performed up to 100 Mrad(SiO 2 ) and then devices were annealed at RT for 27 h in different bias conditions. the drain current at V gs = 0 V and V ds = 5 V. Dotted lines refer to electrical-stress-induced degradation without X-ray exposure; continuous lines are obtained with X-ray exposure. GaN HEMTs irradiated in the "CUT-OFF" and "OFF" biases show increases (improvement) in I on by about 9% and 7% at 100 Mrad(SiO 2 ). RT annealing reduces this increase: I on = 4% after 27 h. The significant difference between the continuous and dotted lines demonstrates that the performance enhancement of "CUT-OFF" and "OFF"-biased devices is affected more strongly by TID-induced effects than by bias-induced effects. On the contrary, devices irradiated in the "ON"-bias condition exhibit a degradation, as evidenced by the 5% decrease of I on at 100 Mrad(SiO 2 ). The degradation for devices irradiated or electrically stressed in the "ON"-bias condition is similar, showing that bias-induced stress dominates the changes in device response.

1) Insensitivity to TID When Irradiated at 0-V Gate Bias:
The best TID tolerance is found for "GND"-biased and "ON"-biased devices, for which the gate bias is 0 V. This enhancement of the TID tolerance is most likely related to the limited charge yield at low electric fields [15], [16], [27]. The "ON"-biased devices show a slight V th increase and g m decrease, but this degradation is related to a bias-stressinduced effect, as shown by the overlapping of the continuous and dotted (red) curves. It is more likely that "ON"-biased devices suffer from hot-electron stress, which can induce the activation of acceptor-like traps through the dehydrogenation of O N -H complexes in the GaN buffer layer [15], [17], [19]. In the case of "ON"-biased devices, post-irradiation annealing continues to induce a slight increase of the V th , as "ON" bias is maintained for longer times.
2) Rebound of the V th Shifts at 10 Mrad(SiO 2 ) of Irradiated HEMTs: Irradiations in the "CUT-OFF" and "OFF" conditions induce an initial negative V th shift. At 10 Mrad(SiO 2 ), the V th is about -280 mV for the "CUT-OFF"-biased HEMTs and -263 mV for "OFF"-biased HEMTs. The negative V th shift and the increase of g m−MAX indicate the neutralization of acceptor-like defects [28]. After 10 Mrad(SiO 2 ), V th recovers somewhat, as shown by the positive trend versus dose. At 100 Mrad(SiO 2 ), V th is about -190 mV for both "CUT-OFF and "OFF"-biased HEMTs and it continues to recover during the RT annealing.
3) Rebound of g m−M AX at 10 Mrad(SiO 2 ) of Irradiated HEMTs: At doses <10 Mrad(SiO 2 ), the g m−MAX of irradiated HEMTs increases with cumulative dose at a higher rate compared to electrically stressed devices, consistent with accelerated acceptor neutralization [17]. At doses >10 Mrad(SiO 2 ), g m−MAX degrades, suggesting activation of defects in the active GaN layer [15], [17], [19]. Increases of acceptorlike defect densities in the 2-DEG may lead to TIDinduced transconductance loss and positive V th shifts, i.e., Degradation of (a) threshold voltage V th and (b) maximum transconductance g m−MAX as a function of dose in passivated GaN-based HEMTs. Dotted lines refer to electrical stress-induced degradation without X-ray exposure; continuous lines refer to test results with X-ray exposure. Irradiations were performed up to 100 Mrad(SiO 2 ); then, devices were annealed at RT for 27 h in different bias conditions. formation of N vacancies, Ga vacancies, and/or O N DX centers [15], [16], [18], [20], [29].

B. Low-Frequency Noise Responses
Additional insight into densities of defects contributing to charge trapping are obtained by low-frequency noise measurements [15], [16], [18], [30], [31], [32], [33]. The drainvoltage noise power spectral density S vd was evaluated in a frequency span between 1 Hz and 1 kHz at |V ds | = 0.1 V for several values of V gt = V gs -V th . The low-frequency noise of GaN-based HEMTs is caused primarily by fluctuations of the number of carriers induced by the capture and emission of carriers at individual defect sites in the GaN layer or at the AlGaN/GaN border, which is often affected by the atomic reconfiguration of single-defect sites in the GaN [27], [29]. Fig. 6 shows the low-frequency noise for (a) "CUT-OFF"-biased and (b) "ON"-biased HEMTs before irradiation, at 1 Mrad(SiO 2 ), at 100 Mrad(SiO 2 ), and after To investigate the density distributions in space and energy of the border traps, Fig. 7 plots the low-frequency noise magnitude at 10 Hz as a function of V gt = V gs -V th in GaN-based HEMTs irradiated in "CUT-OFF" and "ON"-bias conditions. When the slope |β| of the S vD − V gt curve is approximately equal to 2, the effective density of the border traps is uniform in space and energy [27], [28], [29], [33], [34]. "CUT-OFF"-biased devices show a significant increase of low-frequency noise levels at 100 Mrad(SiO 2 ) in agreement with Fig. 6(a). In both pristine and irradiated devices, the slope |β| of S vd -V g t is ∼2.1, indicating an approximately uniform spatial and energetic distribution of traps [27], [28], [29]. The slope |β| of S vd -V gt is ∼2.5 after the RT annealing, suggesting a less uniform density of generated traps in space and energy. On the other hand, "ON"-biased devices are characterized by constant noise with |β| equal to 2, indicating uniform and constant density of traps. RT annealing in "ON"-biased devices induces a slight increase in |β|, which is equal to 2.3 after 27 h of RT annealing.

IV. UNPASSIVATED DEVICES A. TID Sensitivity
This section analyzes the TID response of unpassivated devices that are otherwise similar to those in Figs. 3-7. The dc characteristics in the linear regime (V ds = 0.1 V) of unpassivated GaN-based HEMTs are shown in Fig. 8 when the devices are irradiated and annealed in the "CUT-OFF" and "ON"-bias conditions. The highest shift is visible in the "CUT-OFF" condition, where V th shifts to negative values by about ∼0.6 V and the leakage current increases by two orders of magnitude, from 2 × 10 −6 to 2 × 10 −4 A. Slight performance recovery is visible after 27 h of RT annealing, similar to passivated devices. The "ON"-biased devices are characterized by negative V th shifts, about -270 mV after 100 Mrad(SiO 2 ), which are smaller than those of "CUT-OFF"-biased devices.
The influence of irradiation bias on TID sensitivity is shown in Fig. 9, which summarizes the degradation of: (a) maximum drain current I on , (b) threshold voltage V th , and (c) maximum transconductance g m−MAX . The dotted lines refer to electrical stress-induced degradation without X-ray exposure; continuous lines are obtained with X-ray exposure. The I on variation of irradiated GaN HEMTs of Fig. 9(a) is mostly dominated by the negative shift of V th . The highest shift is visible in "CUT-OFF" and "OFF" biases, while the bias conditions with the smallest degradation are the "GND" and "ON" conditions. The clear separation of dotted and continuous lines indicates that the degradation induced during the exposure is mainly related to TID.
In contrast to the passivated HEMTs, the V th values of unpassivated devices irradiated in "GND" and "ON" conditions degrade with a similarly decreasing monotonic trend. On the other hand, similar to passivated devices, the "CUT-OFF" and "OFF"-biased transistors exhibit a rebound of the V th and g m values around 10 Mrad(SiO 2 ). At doses <10 Mrad(SiO 2 ), V th shifts to negative values and g m increases. At doses >10 Mrad(SiO 2 ), V th shifts toward more positive values and g m decreases. Fig. 10 shows V th as a function of the time for devices that were irradiated in the OFF-bias condition and then annealed at RT for up to 24 h in the same bias condition. In the passivated devices, V th recovers by 34 mV in the first 5 h and by 40 mV after 24 h. In unpassivated devices, V th recovers by 79 mV in the first 5 h and by 103 mV after 24 h. The plot shows that most annealing-induced shifts occur in the first 5 h with the highest shifts in unpassivated devices. The recovery then saturates, becoming approximately stable for annealing times over 15 h. The additional annealing that occurs in the unpassivated devices is most likely due to enhanced hydrogen diffusion and passivation reactions in these devices [21], [23], as discussed in Section V.
Low-frequency noise measurements for the "CUT-OFF"biased HEMTs show typical 1/ f low-frequency noise, as shown in Fig. 11(a). The noise magnitude was unchanged up to 1 Mrad(SiO 2 ) and increased by almost one order of magnitude after 100 Mrad(SiO 2 ), indicating activation of new border traps. Fig. 11(b) plots the low-frequency noise magnitude at 10 Hz as a function of V gt in the HEMTs irradiated in "CUT-OFF"-bias condition. The devices show significant increases of the low-frequency noise levels at 100 Mrad(SiO 2 ), corresponding to the V th increase and g m decrease visible in Fig. 11. In both pristine and irradiated devices, the slope |β| of Sid-V gt is ∼2.1 indicating an approximately uniform spatial and energetic distribution of traps [27], [28], [29] before and during the irradiation. During RT annealing, the value of |β| is 2.4, indicating a slightly non-uniform redistribution of the traps.

B. TID Mechanisms
These results suggest two main TID-related mechanisms as follows.
-First Mechanism: This is characterized by negative V th shifts, g m increases, and unchanged noise. In unpassivated devices, this response is visible under all bias conditions, with enhancement at high gate biases. It is likely that this mechanism results from the TID-assisted passivation of acceptor-like defects via hole capture [16], [32].
-Second Mechanism: This is characterized by positive V th shifts, g m degradation, and the increase of the low-frequency noise magnitude. It is visible only during the irradiation at high gate biases, i.e., "CUT-OFF" and "OFF" at doses >10 Mrad(SiO 2 ). It is more likely that this mechanism is related to the activation of acceptor-like defects via dehydrogenation of defects, as often observed in proton-irradiated GaN-based HEMTs at higher fluences [15], [16], [18], [29]. Fig. 12 plots the OFF-state leakage current I off flowing through the drain terminal when V gs = -7 V and V ds = 5 V in unpassivated HEMTs. In general, the value of I off increases by about one order of magnitude after devices are irradiated to 100 Mrad(SiO 2 ), with negligible contributions of electrical stress (dotted curves). Only GaN-based HEMTs in the "CUT-OFF"-bias condition (green curves) exhibit these large I off increases of about two orders of magnitude. This high I off degradation in "CUT-OFF"-biased HEMTs is induced by electrical stress, which is enhanced when the gate and drain are simultaneously biased at opposite voltages. The high electric field induced by high gate-to-drain voltage may lead to percolation-based transport through defect and/or impurity centers in the AlGaN layer [20], [29], [35], [36], [37]. Particularly at higher voltages, these stress conditions may also lead to impact ionization of carriers, leading to positive charge trapping in the passivation layer of these devices [37], [38].

C. Leakage Current
V. DISCUSSION   Fig. 13 shows the V th and gate leakage I g of passivated and unpassivated HEMTs irradiated and annealed under different bias conditions. The "CUT-OFF" and "OFF"-biased HEMTs have similar TID responses; only "CUT-OFF"-biased devices are shown for clarity. The trends of curves in Fig. 13(a)  among passivated and unpassivated devices are generally similar, except for passivated "ON"-biased devices, for which the degradation is dominated by electrical stress. TID-induced effects are enhanced when bias is applied to the gate, i.e., at V g = -7 V, corresponding to the OFF-condition. In "CUT-OFF" passivated and unpassivated devices, TID effects are caused by the two mechanisms described in Section IV. The first mechanism occurs at doses <10 Mrad(SiO 2 ) with negative V th shifts, while the second mechanism dominates at doses >10 Mrad(SiO 2 ) with positive V th shifts. "GND" and "ON"-biased devices exhibit only the first mechanism, inducing negative V th shifts.
Observed shifts in Fig. 13(a) are smaller for passivated devices than unpassivated devices. After 100 Mrad(SiO 2 ) in the "CUT-OFF"-bias condition, V th values for the passivated HEMTs shift by -0.34 V versus -0.68 V for unpassivated devices. The higher TID sensitivity of unpassivated HEMTs compared to passivated devices highlights the key role of contaminant absorption (e.g., oxygen, moisture) through the surface layers [21], [23]. SiN x passivation inhibits moisture absorption, limiting the formation of acceptor-like defects [21]. Hence, the composition and thickness of passivation layers are the important factors in determining the radiation tolerance and long-term reliability of GaN-based HEMTs. Fig. 13(b) shows the gate current I g normalized by its pre-irradiation value of I g0 for GaN-based HEMTs irradiated under different bias conditions. In general, the highest gateto-drain leakage currents are visible in unpassivated devices. After 100 Mrad(SiO 2 ), the value of I g /I g0 of unpassivated HEMTs increases by roughly one order of magnitude for  all cases except CUT-OFF devices. These show increases of two orders of magnitude, consistent with the stress-induced increase visible in I off in Fig. 12. In passivated devices, the increase of gate leakage is lower than one order of magnitude, regardless of bias applied during irradiation. Since the leakage depends only weakly on gate bias, the leakage in the passivated devices is dominated most likely by charge trapping in the SiN layer [35], [36], while unpassivated devices are most likely dominated by surface trap buildup at the top of the AlGaN [37], [38]. Both the traps in SiN and the surface traps are most likely charged positively after stress, based on the behavior of the device of Fig. 12, and consistent with the electric fields. The comparative responses between the two device types in Fig. 13(b) suggest that the surface trap density in the unpassivated devices exceeds the SiN-charged trap density [23].

VI. CONCLUSION
AlGaN/GaN HEMTs irradiated at ultrahigh doses show significant degradation due to TID and electrical stress, with magnitudes depending on irradiation bias. In general, ultrahigh doses induce negative V th shifts with magnitudes that are higher than the ones retrieved at lower doses of previous X-rays and gamma studies. The HEMTs irradiated with negative gate bias (OFF-state) exhibit the most negative threshold voltage shifts. At doses <10 Mrad(SiO 2 ), TID effects are related to the passivation of pre-existing acceptor-like defects via hole capture, which induces negative threshold voltage shifts and improvement of transconductance. At doses >10 Mrad(SiO 2 ), dehydrogenation of defect and impurity complexes leads to the creation of acceptor-like defects. These degrade the transconductance, shift the threshold voltage positively, and increase low-frequency noise. Slight performance recovery is visible after 27 h of RT annealing, most likely due to the formation of a significant density of stable defects.
Irradiation results on passivated and unpassivated HEMTs show that passivation layers may strongly affect oxygen and moisture absorption. The enhanced degradation of unpassivated devices in this study reinforces the key role that oxygen impurities and hydrogen play in the radiation response and long-term reliability of GaN-based HEMTs.