The GaN-based metal–insulator–semiconductor high electron mobility transistors (MIS-HEMTs) have been extensively studied for high power switching application, mainly due to greatly suppressed gate leakage and enlarged gate swing compared with the conventional Schottky-gate HEMTs.^[1,2] However, introducing an insulating layer would bring in additional defects and trap states which can lead to severe device instabilities.^[3,4] Even state-of-the-art devices suffer threshold voltage (V_th) drift introduced by positive or negative gate bias stress.^[5,6] Typically, the V_th would move positively under positive gate bias stress and negatively under negative gate stress, due to the trapping and detrapping of electrons at the insulator/AlGaN interface or in the insulating layer, respectively.^[7–9] Both processes are recoverable with different kinetics.^[9]

The conclusions for positive V_th instability were similar in Refs. [10–13]. A broad distribution of capture and emission time constant have been reported under positive gate stress and recovery conditions, respectively,^[14] for a distribution of energy and capture cross section of traps at or near the interface, and additional lateral trapping at the dielectric/III-N interface plane due to carrier hopping and thus a transport mechanism between the interface/border states.^[15,16] The instability of V_th under negative gate stress is a serious concern for depletion-mode devices when they need to be turned off and during off-state. However, there are different views of the mechanisms behind the negative bias-induced V_th instability. The recoverable negative drift of V_th due to electrons’ emission from interface states and border traps has been reported.^[7,17–19] There is non-recoverable negative drift of V_th in the harsh stress condition due to the formation of interface states as a result of broken H bonds at oxide/semiconductor interface.^[20] The mechanism of negative V_th drift was also introduced by the accumulation of holes under gate, and the holes were suggested to be generated by impact ionization or inter-band tunneling in the high electric field region.^[21] However, in the hole-barrier-free E-mode SiN/GaN MIS-FET, the holes, generated in high reverse-bias condition, could not accumulate at the interface and would flow through the gate dielectric, which accelerated the generation of new defects in the gate dielectric and resulted in the large positive V_th shifting.^[22,23] The positive V_th drift in fully recessed enhancement mode metal–insulator–semiconductor field-effect transistor (MOSFET) under negative gate stress has been reported to be due to the gate-injection and the following trapping in the dielectric layer and the inductively coupled plasma (ICP) recessed GaN channel layer.^[24]

In this paper, we study the instability of V_th in D-mode MIS-HEMT, with in situ SiN as gate dielectric under negative gate stress. It is observed that the drift of V_th is non-monotonic whether in the negative stress condition or after the stress. This is inconsistent with what has been reported. A physical model, based on the combination of the effect of interface states at SiN/AlGaN and the effect of the zener trap in the AlGaN barrier, is proposed for the V_th behavior in the D-mode MIS-HEMT. The results in this work provide a comprehensive insight into the instability of V_th in the D-mode MIS-HEMT.

The study was carried out on D-mode MIS-HEMT grown on a Si substrate. The epilayers, from bottom to top, consisted of a GaN buffer intentionally doped with C, an undoped GaN channel, an Al_0.22Ga_0.78N barrier, and 30-nm in situ SiN as gate dielectric and passivation layer. The cross section of the device is sketched in the inset of Fig. 1(a) (not scaled). The gate length and gate-source distance of the device under test are both 3 μm, and gate–drain distance is 20 μm. Figure 1(a) exhibits great effects on suppressing gate leakage current at both reverse and forward bias. The transfer and transconductance characteristics are shown in Fig. 1(b). The fresh device shows that Id_max is 414 mA/mm, V_th is −9.05 V, and G_m,max is 60 mS/mm.

Fig. 1.

	Figure Option ViewDownloadNew Window
	Fig. 1. (a) The Igs–Vgs characteristics of MIS-HEMTs with inset showing schematic cross section view of MIS-HEMT. (b) Transfer and transconductance characteristics at Vds = 10 V. (c) Schematic diagram of experimental procedure.

In order to avoid the influence of drain voltage on the experiment and focus on the region under gate, the drain voltage was set to be 0 V during the stress and the transfer measurements were carried out at V_d = 0.1 V in the intermediate monitoring process. Meanwhile, the gate voltage was swept from −10 V to −7 V in order to exclude the influence of the positive gate voltage. The V_th was defined as the gate voltage corresponding to the drain current of 1 μA/mm. The detail of experimental process is shown as Fig. 1(c).

As shown in Fig. 2(a), when the negative gate stress of V_g = −10 V is applied, the V_th shifts positively. It is unusual that the V_th continues shifting positively even after the stress has been removed. In addition, the maximum of transconductance (G_m,max) decreases during the negative stress and increases in the recovery process as shown in Fig. 2(b).

Fig. 2.

	Figure Option ViewDownloadNew Window
	Fig. 2. (a) Transfer and (b) transconductance curves of GaN-based MIS-HEMTs at different time.

To exclude the influence of the monitoring tests, the transfer measurements without any stress at certain time as mentioned above are conducted, and the results are shown in Fig. 3(a). Figure 3(b) shows that the V_th shifts positively at all times.

Fig. 3.

	Figure Option ViewDownloadNew Window
	Fig. 3. (a) Transfer characteristics measured in 2000 s without any stress, and (b) the variations of threshold voltage with time during the experiments in panel (a).

Figure 4 shows the difference between V_th under V_g = −10 V negative stress and that under V_g = 0 V (no stress). The amount of V_th drift with no stress is larger than that under stress (0 s<time < 500 s). However, when 1000 s<time ≤ 2000 s, the V_th shifts more positively after exerting stress than that under no stress.

Fig. 4.

	Figure Option ViewDownloadNew Window
	Fig. 4. Drift of Vth versus sampling time with and without negative biased stress.

To further investigate the mechanisms of V_th drift, variations of ΔV_th with stress time under the negative stresses of V_gstress = −15 V and −25 V excluding the influence of the monitoring tests are shown respectively in Figs. 5(a) and 5(b), where ΔV_th is the difference between the V_th drift under stress and the V_th drift (the degradation of G_m,max) under no stress. The V_th shifts negatively when V_gstress = −15 V. After the stress is removed, the V_th shifts first positively and then negatively. As shown in Fig. 5(c), the G_m,max decreases to 77.37% under −15 V stress and recovers to 96.46% as the recovery time reaches to 1000 s. For V_gstress = −25 V, the V_th shifts first positively and then negatively during the stress. As the stress is removed, the V_th shifts first positively and then negatively. As shown in Fig. 5(d), the G_m,max decreases to 71.79% under −25 V stress and recovers to 91.40% as the recovery time reaches to 1000 s.

Fig. 5.

	Figure Option ViewDownloadNew Window
	Fig. 5. Plots of ΔVth versus stress time and recovery time under stress voltage of (a) −15 V and (b) −,25 V. Plots of normalized Gm,max versus stress time and recovery time under stress voltage of (c) −15 V and (d) −25 V.

Before analyzing the drift of the V_th, the change of the G_m,max should be investigated first. In general, the traps in the SiN insulator and at SiN/AlGaN interface have effects on the electron mobility, acting as remote impurity scattering, thus G_m,max decreases.^[25] Experimental results and theoretical calculations indicate that the remote scattering rate decreases exponentially as the distance increases between the charges and the channel, which could be simplified as the thickness of AlGaN barrier layer.^[26,27] In this work, the thickness of AlGaN in MIS-HEMT is 20 nm. The remote impurity scattering could be neglected for the 20 nm AlGaN layer, which is shown in Fig. 6. As a result, the electrons trapping and detrapping at SiN/AlGaN interface and in the SiN insulator can induce the V_th drift but have little influence on G_m,max. However, the G_m,max decreases under the negative stress and increases in the recovery process, indicating that there is trapping behavior in the place closer to the channel, which is in the barrier here.

Fig. 6.

	Figure Option ViewDownloadNew Window
	Fig. 6. Plots of calculated mobility versus thickness of AlGaN barrier for different SiN/AlGaN interface charge deisities.

In the above analysis, the location where the trapping- and detrapping-behaviors occur is identified, and the mechanisms for the drift of V_th can be proposed in the following. Under the negative stress, a strong electric field exists in the barrier layer under the gate, especially at the edges of the gate on the source side and drain side. Electrons can be trapped in the AlGaN barrier, which can take place under high reverse electric field, when electrons tunnel from the valence band to the trap states in the AlGaN barrier in a process, which is sometimes referred to as zener trapping.^[20,28] The trapped electrons in the AlGaN barrier induce the V_th to positively drift and the G_m,max to decrease under the negative stress. This mechanism runs in parallel with the detrapping behavior of interface states at SiN/AlGaN,which induces the V_th to negatively drift and has little influence on transconductance. The effect of zener trap is strongest at the edge of gate, where the highest electric field exists. Trapped electrons lift the bands up as shown in Fig. 7(b). When the negative stress is removed, the V_th shifts negatively for electrons escaped from the trap in the barrier by thermal activation. On the contrary, as the stress is removed, the interface state at SiN/AlGaN falls again below the Fermi level, and is filled with electrons, which induces the V_th to positively drift. The evolution of V_th in the stress and recovery process under the two mechanisms are shown in Fig. 8.

Fig. 7.

	Figure Option ViewDownloadNew Window
	Fig. 7. (a) Two mechanisms under negative stress. Behaviors of (b) zener traps and (c) interface states in recovery process.

Fig. 8.

	Figure Option ViewDownloadNew Window
	Fig. 8. Plot of variation of Vth with time, induced by zener traps and interface states in negative stress and recovery process.

Considering the two mechanisms, figure 4 should be further analyzed. When no stress is applied, zener trapping dominates in the monitoring process. And it is hard for the thermal detrapping because bands lift up slightly with V_g = −10 V. As a result, the V_th shifts positively at all times under no stress. For V_gstress = −10 V, the detrapping of the interface states induces the V_th to less positively drift under negative stress (0 s< time < 500 s). And the retrapping of the interface states results in the more positive drift of V_th as the stress is just removed.

As shown in Fig. 5(a), for V_gstress = −15 V, the detrapping of interface states dominates the negative drift of V_th. As the −15 V stress is removed, the interface states, especially those with low energy level, can be retrapped rapidly, which results in a sharply positive drift of V_th in 2 s. The electrons are trapped in the AlGaN barrier due to the zener trapping under −15 V stress, which results in a relatively small positive drift, and then electrons in the AlGaN barrier can be detrapped through thermal process, resulting in the negative drift of V_th. On account of the relatively deep level, the electrons’ thermal detrapping dominates as the time goes on in the recovery process. Because the transconductance is mostly associated with zener traps in the AlGaN, the variation of G_m,max is relatively simple as shown in Fig. 5(c). The trapped electrons in the AlGaN enhance the scattering, indicating the decrease of G_m,max. And the thermal detrapping in the recovery process causes G_m,max to increase.

As shown in Fig. 5(b), when V_gstress = −25 V is applied, both the trap effect of interface states at SiN/AlGaN and the zener trap effect in the AlGaN barrier are enhanced. The zener trap is based on the tunneling process, whose time constant is on the order of picoseconds. As a result, zener trap effect dominates at the beginning of stress for the higher electric field under a stress of −25 V. As the time increases, the interface states dominate instead. The V_th shifts first positively and then negatively. When the stress is removed, the variation of V_th under stress of V_g = −25 V is similar to that under −15 V stress. It is noted that the negative drift of V_th is faster under −25 V stress. More electrons trapped in the AlGaN lift the energy bands higher, making thermal emission take place more easily. As the number of trapped electrons decreases, the energy band’s lifting up becomes low again and the thermal emission is weakened. As a result, the negative drift of V_th slows down.

In conclusion, the instabilities of V_th in D-mode MIS-HEMT, with in situ SiN as gate dielectric under different negative gate stresses are investigated. The effect of interface state at SiN/AlGaN and the effect of zener trap in AlGaN barrier layer are opposite to each other both under negative stress and in recovery process (no stress). The V_th negatively shifts under the stress due to the detrapping of the electrons at SiN/AlGaN interface, and positively shifts due to zener trapping in AlGaN. As the stress is removed, V_th positively shifts for the retrapping of interface states and negatively for the thermal detrapping in AlGaN. The complex non-monotonic evolution of V_th is induced by the two mechanisms together. However, it is the trap behavior in the AlGaN rather than the interface states that gives rise to the change of transconductance in the D-mode MIS-HEMT.