Buffer Traps Effect on GaN-on-Si High-Electron-Mobility Transistor at Different Substrate Voltages

Substrate voltage (VSUB) effects on GaN-on-Si high electron mobility transistors (HEMTs) power application performance with superlattice transition layer structure was investigated. The 2DEG conductivity and buffer stack charge redistribution can be affected by neutral/ionized donor and acceptor traps. As the donor/acceptor traps are excessively ionized or de-ionized by applying VSUB, the depletion region between the unintentionally doped (UID)/Carbon-doped (C-doped) GaN layer may exhibit a behavior similar to the p–n junction. An applied negative VSUB increases the concentration of both the ionized donor and acceptor traps, which increases the breakdown voltage (BV) by alleviating the non-uniform distribution of the vertical electric field. On the other hand, an applied positive VSUB causes the energy band bending flattener to refill the ionized traps and slightly improves the dynamic Ron degradation. Moreover, the amount of electrons injected into the buffer stack layer from the front side (2DEG channel/Ohmic contact) and the back side (AlN nucleation layer/superlattice transition layer) are asymmetric. Therefore, different VSUB can affect the conductivity of 2DEG through the field effect, buffer trapping effect, and charge redistribution, which can change the electrical performance of the device.


Introduction
In recent years, GaN-on-Si power HEMTs have gained tremendous interest due to their superior electrical characteristics, such as high breakdown voltage (BV), low on-resistance (R on ), and fast switching speed [1][2][3]. Compared to GaN-on-SiC and GaN-on-sapphire, GaN-on-Si is most suitable for commercialization owing to the low cost and the scalability to large wafer sizes [3]. The epitaxial structure of a lateral GaN-on-Si buffer stack from bottom to top is usually composed of an AlN nucleation layer, called a superlattice, or graded AlGaN transition layer, a carbon-doped (C-doped) GaN layer, and an unintentionally doped (UID) GaN channel layer [4]. However, GaN buffer stacks usually have plenty of dislocations, lattice defects, and bulk traps due to the lattice and thermal expansion coefficient mismatch between GaN and silicon [5]. The dynamic R on of GaN-on-Si HEMT is related to the surface-and buffer-induced trapping. When the devices are switched from the off-to on-state, the hot electrons generated in the two-dimensional electron gas (2DEG) channel are injected into the surface and buffer layers. When the surface and buffer traps cannot immediately release the trapping electrons, the 2DEG channel is partially depleted to increase the dynamic R on [6].
The surface-induced trapping can be suppressed by the passivation and field plate. In contrast, the buffer-induced trapping caused by the compensated doping and crystalline defects is more challenging to address. During the epitaxial growth, the AlN nucleation layer is usually defective. It contains many donor-type impurities (e.g., Si, O), resulting in a lower effective energy barrier at the AlN/Si junction [7,8] for the enhanced charge injection to occur. Moreover, due to the energy barrier at the AlN-Si substrate interface, the electrons induced from the Si substrate may also be injected into the buffer with positive V top-to-sub (or negative V SUB ) by trap-assisted tunneling, thermionic emission, and Poole-Frenkel emission [7][8][9].
For the GaN-on-Si HEMT, carbon doped (C-doped) is a common technique to reduce buffer conductivity, reduce current collapse, and increase BV during the epitaxial growth process [10]. The carbon atoms can substitute Ga or N sites, occupy interstitial positions in the nitride, and form chemical complexes with intrinsic defects [11][12][13]. A dominant deep-acceptor trap at E V +0.9 eV (generally considered to correspond to the C N state) and a shallow donor trap at E C −0.11 eV (more likely related to the C Ga state) of the two main energy states are related to the carbon doping in the GaN buffer [10]. An appropriate ratio of acceptor trap and donor trap concentration may exist to achieve the optimum BV and buffer leakage current due to the vertical electric field modulation [14].
C-doped GaN layers could behave as p-type materials due to sufficient compensation doping and the Fermi level (E F ) in the lower half of the bandgap [15]. UID GaN layers are observed to show light n-type conductivity owing to the existence of nitrogen vacancies [4]. The neutral, ionized donor, and acceptor traps can affect the 2DEG conductivity and bufferstacks charge redistribution. The applied V SUB or V D at the UID/C-doped GaN interface may result in a behavior similar to the p-n junction [4]. The applied positive/negative V SUB causes different charging/discharging behaviors, which could be related to the nonuniform spatial distribution of dopant traps and the charge redistribution in the buffer stacks. The intrinsic trap time constants and the charge transport in the buffer stacks are both related to the response of donor/acceptor traps in the UID/C-doped GaN layers [15].
Recently, several studies discussed the impacts of V SUB on buffer-related performances, such as dynamic R on [4,16] BV [17,18] vertical leakage current [19,20], gate charge change [21], and output capacitance [22]. However, most works were limited to the individual properties mentioned above. There are few reports [17,18] to investigate the overall device characteristics such as BV, on-state drain current, and dynamic R on respective to V SUB for a GaN HEMT structure, thus, they are lacking unified models and discussions. Compared to these reports [17,18], which applied a V SUB of about ±60 V, we can apply a higher V SUB because the superlattice transition layer with a higher energy barrier can prevent electrons from injecting into the buffer stack layer at a high V SUB . In this study, we investigate the effects of a positive/negative V SUB on the above-mentioned device characteristics of the AlGaN/GaN HEMT having a superlattice transition layer. The impact of 2DEG conductivity and the charge trapping/de-trapping procedure are also investigated under the different V SUB and off-state V D .

Materials and Methods
The AlGaN/GaN heterostructures were grown on a 6-inch silicon substrate by metalorganic chemical vapor deposition (MOCVD), as shown schematically in Figure 1. The AlGaN/GaN HEMT structure consists of an AlN nucleation layer, a superlattice transition layer, a C-doped GaN layer, a GaN channel layer, and 22 nm Al 0.25 Ga 0.75 N barrier layer. The mesa isolation was performed by BCl 3 + Cl 2 mixed gas plasma etching. To perform the Ohmic contact formation at source and drain, Ti/Al/Ni/Au was deposited by e-beam evaporation and was annealed by rapid temperature annealing (RTA) at 840 • C for 30 s in N 2 ambient. The Ni/Au (50/300 nm) Schottky gate contact was deposited by the electron beam evaporation. Finally, 20 nm SiN was deposited as a passivation layer. The source-drain length, gate length, and gate width were maintained at 20, 3, and 25 µm, respectively. The buffer stacks can be considered insulated capacitors when the leakage current of each layer is smaller than the "displacement current"; i.e., I DISS = C TOT dV SUB /dt, C TOT is the series combination of the capacitances of UID GaN, C-doped GaN, and superlattice transition layer, respectively [15].
Micromachines 2022, 13, x FOR PEER REVIEW 3 of 10 the Ohmic contact formation at source and drain, Ti/Al/Ni/Au was deposited by e-beam evaporation and was annealed by rapid temperature annealing (RTA) at 840 °C for 30 s in N2 ambient. The Ni/Au (50/300 nm) Schottky gate contact was deposited by the electron beam evaporation. Finally, 20 nm SiN was deposited as a passivation layer. The sourcedrain length, gate length, and gate width were maintained at 20, 3, and 25 μm, respectively. The buffer stacks can be considered insulated capacitors when the leakage current of each layer is smaller than the "displacement current"; i.e., IDISS = CTOTdVSUB/dt, CTOT is the series combination of the capacitances of UID GaN, C-doped GaN, and superlattice transition layer, respectively [15].

Results and Discussion
As shown in Figure 2, the buffer trap charging/discharging mechanism of HEMT devices with a superlattice transition layer is demonstrated schematically. The substrate of GaN-on-Si power HEMT can be used as an independent contact terminal for applying VSUB [23]. The VSUB can modulate the 2DEG concentration through the back-gate effect. Figure 2(a) shows that the buffer acceptor and donor traps are partly ionized and neutralized at the initial state. The un-ionized neutral donor/acceptor traps may exist in the UID/C-doped GaN layer. The opposite doping polarities may lead to the formation of the p-n junction at the UID/C-doped GaN interface [4,9,15].

Results and Discussion
As shown in Figure 2, the buffer trap charging/discharging mechanism of HEMT devices with a superlattice transition layer is demonstrated schematically. The substrate of GaN-on-Si power HEMT can be used as an independent contact terminal for applying V SUB [23]. The V SUB can modulate the 2DEG concentration through the back-gate effect. Figure 2a shows that the buffer acceptor and donor traps are partly ionized and neutralized at the initial state. The un-ionized neutral donor/acceptor traps may exist in the UID/Cdoped GaN layer. The opposite doping polarities may lead to the formation of the p-n junction at the UID/C-doped GaN interface [4,9,15]. Figure 2b shows that the electrons are injected from the Ohmic contact or 2DEG channel into the buffer layer when applying a positive V SUB [24,25]. The Ohmic contact provides the reservoir of free electrons that can be injected into the buffer layer when applying a positive V SUB . Without applying V SUB , the acceptor trap pulls down the E F in the buffer/transition layer so that the E F is below the acceptor trap level (E TA ). Many acceptor traps are not ionized (charge-neutral) because the acceptor traps are deep. When the positive V SUB increases, the high vertical electric field induces electron trapping (ionization at the acceptor trap level) in the buffer layer of the acceptor trap [24,25]. Because a positive V SUB of 200 V is smaller than the traps-filled-limit voltage of donor traps (V TFL2 ), both with the positive V SUB and without the V SUB do not affect the concentration of the ionized donor trap. Micromachines 2022, 13, x FOR PEER REVIEW 4 of 10  Figure 2b shows that the electrons are injected from the Ohmic contact or 2DEG channel into the buffer layer when applying a positive VSUB [24,25]. The Ohmic contact provides the reservoir of free electrons that can be injected into the buffer layer when applying a positive VSUB. Without applying VSUB, the acceptor trap pulls down the EF in the buffer/transition layer so that the EF is below the acceptor trap level (ETA). Many acceptor traps are not ionized (charge-neutral) because the acceptor traps are deep. When the positive VSUB increases, the high vertical electric field induces electron trapping (ionization at the acceptor trap level) in the buffer layer of the acceptor trap [24,25]. Because a positive VSUB of 200 V is smaller than the traps-filled-limit voltage of donor traps (VTFL2), both with the positive VSUB and without the VSUB do not affect the concentration of the ionized donor trap. Figure 2c shows that the concentration of both ionized acceptor and donor traps increases with the negative VSUB. In addition to the inherent barrier of the AlN/Si junction, the superlattice transition layer makes the electrons more difficult to be injected into the buffer layer in the off-state under the high VD. The top and bottom asymmetric vertical leakage and the fundamental difference between the carrier injection/transport mechanisms induced a different buffer layer trapping. The acceptor traps in the carbon-doped GaN are mostly ionized at a VSUB of −100 V, resulting in a net negative charge [4]. This net negative charge lifts the buffer energy band, increasing the energy barrier of the transition layer to a higher value, and injecting fewer electrons to the buffer layer [16]. The transition layer with a superlattice buffer structure can block the electron injections from the AlN/Si junction to the buffer layer. At off-state and high VD, the electrons from the Si substrate  Figure 2c shows that the concentration of both ionized acceptor and donor traps increases with the negative V SUB . In addition to the inherent barrier of the AlN/Si junction, the superlattice transition layer makes the electrons more difficult to be injected into the buffer layer in the off-state under the high V D . The top and bottom asymmetric vertical leakage and the fundamental difference between the carrier injection/transport mechanisms induced a different buffer layer trapping. The acceptor traps in the carbondoped GaN are mostly ionized at a V SUB of −100 V, resulting in a net negative charge [4]. This net negative charge lifts the buffer energy band, increasing the energy barrier of the transition layer to a higher value, and injecting fewer electrons to the buffer layer [16]. The transition layer with a superlattice buffer structure can block the electron injections from the AlN/Si junction to the buffer layer. At off-state and high V D , the electrons from the Si substrate begin to overcome the energy barrier of the superlattice layer and inject into the buffer layer. When part of the buffer layer starts to conduct, the negative charge accumulates in the buffer layer, resulting in an increase in the buffer leakage current [15]. Figure 3 shows the energy band diagram when a different V SUB and low/high V D are applied. Figure 3a shows that without V SUB , the energy band has more bending as the V D increases. Figure 3b shows that at low V D , the applied positive V SUB causes the Si substrate to bend downward. The 2DEG channel tends to inject more electrons into the buffer stack, resulting in an entirely negative charge redistribution in the buffer layer and an increase in the concentration of the ionized acceptor trap. The positive V SUB can alleviate the energy band bending caused by applied high V D , and the electrons in the inversion layer formed at the Si substrate/AlN interface are less able to overcome the energy barrier of AlN and superlattice and inject into the buffer layer, avoiding the accumulation of negative charge in the buffer layer. Figure 3c shows that the negative V SUB can raise the energy band of the Si substrate even more and accumulate electrons in the inversion layer at the Si substrate/AlN interface at lower V D . After the electrons overcome the AlN energy barrier, the electrons are blocked by the superlattice layer and are not injected into the buffer layer. When the V D becomes larger, the bending of the energy band becomes more significant, which is equal to the applied drain voltage of |V D + V SUB | to the device. This causes the electrons accumulated in the Si substrate/AlN inversion layer to have higher energy to overcome the energy barrier of AlN and superlattice to inject into the buffer layer. This results in a significant amount of electrons injected into and accumulated in the C-doped GaN layer.
applied. Figure 3a shows that without VSUB, the energy band has more bending as the VD increases. Figure 3b shows that at low VD, the applied positive VSUB causes the Si substrate to bend downward. The 2DEG channel tends to inject more electrons into the buffer stack, resulting in an entirely negative charge redistribution in the buffer layer and an increase in the concentration of the ionized acceptor trap. The positive VSUB can alleviate the energy band bending caused by applied high VD, and the electrons in the inversion layer formed at the Si substrate/AlN interface are less able to overcome the energy barrier of AlN and superlattice and inject into the buffer layer, avoiding the accumulation of negative charge in the buffer layer. Figure 3c shows that the negative VSUB can raise the energy band of the Si substrate even more and accumulate electrons in the inversion layer at the Si substrate/AlN interface at lower VD. After the electrons overcome the AlN energy barrier, the electrons are blocked by the superlattice layer and are not injected into the buffer layer. When the VD becomes larger, the bending of the energy band becomes more significant, which is equal to the applied drain voltage of |VD + VSUB| to the device. This causes the electrons accumulated in the Si substrate/AlN inversion layer to have higher energy to overcome the energy barrier of AlN and superlattice to inject into the buffer layer. This results in a significant amount of electrons injected into and accumulated in the C-doped GaN layer.   Figure 4 compare the overview of DC characteristics under the positive/negative VSUB. Applying the positive VSUB tends to induce more electrons in the 2DEG channel. However, the maximum drain current (ID) is almost constant because the ionized acceptor traps caused by injection electrons from the 2DEG channel or ohmic contact could clamp the increasing 2DEG concentration [4]. Under the positive VSUB, the DC characteristics are similar to the device without VSUB. A slightly reduced off-state ID is observed with a positive VSUB due to the leakage current flow from the drain to the buffer layer. Part of the electrons are also captured by the acceptor trap of the GaN carbon-doped layer, resulting in a slightly reduced total leakage current. Figure 2b shows that due to the applied positive VSUB, the electrons from the Ohmic contact at the interface of source and   Figure 4 compare the overview of DC characteristics under the positive/negative V SUB . Applying the positive V SUB tends to induce more electrons in the 2DEG channel. However, the maximum drain current (I D ) is almost constant because the ionized acceptor traps caused by injection electrons from the 2DEG channel or ohmic contact could clamp the increasing 2DEG concentration [4]. Under the positive V SUB , the DC characteristics are similar to the device without V SUB . A slightly reduced off-state I D is observed with a positive V SUB due to the leakage current flow from the drain to the buffer layer. Part of the electrons are also captured by the acceptor trap of the GaN carbon-doped layer, resulting in a slightly reduced total leakage current. Figure 2b shows that due to the applied positive V SUB , the electrons from the Ohmic contact at the interface of source and drain with GaN or 2DEG are injected into the buffer layer, where the acceptor trap of the GaN carbon-doped layer is ionized for the lower V D (12 V) case. An applied negative V SUB ideally reduces the maximum I D due to the simple electric field effect or capacitive coupling [15]. The ionization of acceptor traps in the buffer layers depletes the 2DEG channel; both can degrade the 2DEG conductivity. The V th increases because the required gate voltage (V G ) for turning on the depleted channel is increased. As Figure 2c shows, off-state applied negative V SUB causes the ionized donor trap concentration in the UID GaN layer to increase. An increase in ionized donor trap concentration indicates that positive charges have accumulated in the buffer layer. Hence, the positive charges neutralize part of the leakage current flowing into the buffer layer. Therefore, the total off-state I D decreases, and the on/off current ratio increases. The subthreshold swing (S.S) of this HEMT device is related to the I D magnitude when V G is smaller than the threshold voltage (V th ). As the off-state leakage current reduces due to the negative V SUB , a reduction in S.S is observed. The transconductance (g m ) hike of this superlattice HEMT device with a negative V SUB applied is also observed. Firstly, the applied negative V SUB reduced the off current and increased the change in current per unit voltage, making the g m maximum larger. Secondly, when the negative V SUB is between −50 and −200 V, charge redistribution occurs in the GaN carbon doping layer. The higher maximum g m is related to the buffer stack charge redistribution due to the capacitance change of UID GaN, C-doped GaN, and superlattice transition layer [26]. drain with GaN or 2DEG are injected into the buffer layer, where the acceptor trap of the GaN carbon-doped layer is ionized for the lower VD (12 V) case. An applied negative VSUB ideally reduces the maximum ID due to the simple electric field effect or capacitive coupling [15]. The ionization of acceptor traps in the buffer layers depletes the 2DEG channel; both can degrade the 2DEG conductivity. The Vth increases because the required gate voltage (VG) for turning on the depleted channel is increased. As Figure 2c shows, off-state applied negative VSUB causes the ionized donor trap concentration in the UID GaN layer to increase. An increase in ionized donor trap concentration indicates that positive charges have accumulated in the buffer layer. Hence, the positive charges neutralize part of the leakage current flowing into the buffer layer. Therefore, the total off-state ID decreases, and the on/off current ratio increases. The subthreshold swing (S.S) of this HEMT device is related to the ID magnitude when VG is smaller than the threshold voltage (Vth). As the off-state leakage current reduces due to the negative VSUB, a reduction in S.S is observed. The transconductance (gm) hike of this superlattice HEMT device with a negative VSUB applied is also observed. Firstly, the applied negative VSUB reduced the off current and increased the change in current per unit voltage, making the gm maximum larger. Secondly, when the negative VSUB is between −50 and −200 V, charge redistribution occurs in the GaN carbon doping layer. The higher maximum gm is related to the buffer stack charge redistribution due to the capacitance change of UID GaN, C-doped GaN, and superlattice transition layer [26].    Figure 5 shows the BV characteristics measured using off-state leakage current with the positive/negative V SUB . Figure 5a,b show that the negative V SUB can increase the BV due to the increased ionized donor trap concentration. The BV was measured starting from V D = 0 V and gradually increased in voltage by 2 V/s. The electrons forming in the inversion layer between the Si substrate and AlN layer slowly accumulate energy to overcome the energy barriers to inject into the buffer stack. The electrons injected into the buffer layer have a sufficient reaction time to neutralize the ionized donor trap in the UID GaN with a positive charge. The ionized donor trap is more critical to BV than the ionized acceptor trap due to the alleviation of the non-uniform distribution of the high electric field caused by high V D [14]. Applying the negative V SUB (or positive V top-to-sub ) causes the donor/acceptor traps to ionize at the UID/carbon-doped GaN interface [27]. An increase in ionized donor trap concentration with a moderate acceptor trap can increase the BV and reduces the off-state leakage current [14]. Therefore, the concentration of ionized donor trap and BV increase when a negative V SUB is applied. On the other hand, Figure 2a,b show that the electrons from the 2DEG channel or the Ohmic contact are primarily injected into the buffer layer to increase the ionized acceptor traps without decreasing the concentration of the ionized donor traps [15]. The BV at the applied positive V SUB and V SUB = 0 are very similar due to the same concentration of ionized donor traps.

200
3.05 × 10 5 201 −11.78 9.16 81.7 701 2.3 × 10 −3 Figure 5 shows the BV characteristics measured using off-state leakage current with the positive/negative VSUB. Figures 5a,b show that the negative VSUB can increase the BV due to the increased ionized donor trap concentration. The BV was measured starting from VD = 0 V and gradually increased in voltage by 2 V/s. The electrons forming in the inversion layer between the Si substrate and AlN layer slowly accumulate energy to overcome the energy barriers to inject into the buffer stack. The electrons injected into the buffer layer have a sufficient reaction time to neutralize the ionized donor trap in the UID GaN with a positive charge. The ionized donor trap is more critical to BV than the ionized acceptor trap due to the alleviation of the non-uniform distribution of the high electric field caused by high VD [14]. Applying the negative VSUB (or positive Vtop-to-sub) causes the donor/acceptor traps to ionize at the UID/carbon-doped GaN interface [27]. An increase in ionized donor trap concentration with a moderate acceptor trap can increase the BV and reduces the off-state leakage current [14]. Therefore, the concentration of ionized donor trap and BV increase when a negative VSUB is applied. On the other hand, Figure 2(a) and (b) show that the electrons from the 2DEG channel or the Ohmic contact are primarily injected into the buffer layer to increase the ionized acceptor traps without decreasing the concentration of the ionized donor traps [15]. The BV at the applied positive VSUB and VSUB = 0 are very similar due to the same concentration of ionized donor traps. The dynamic Ron characteristics for the HEMT devices with different Vsub were also measured. After each dynamic Ron measurement, the device's initial state is fully recovered by shining microscopic light for 30 s, so that all trapped electrons were de-trapped [28]. Figure 6a shows that the device is switched to the on-state with a total switching time of 200 μs after 10 ms of high VD stress at off-state. As shown in Figure 6b, the positive VSUB slightly reduces the dynamic Ron, while the negative VSUB significantly increases the dynamic Ron. The dynamic Ron ratio for applying the positive VSUB is slightly lower than without VSUB (the off-state VD is from 100 to 300 V). As shown in Figure 3b, the positive VSUB The dynamic R on characteristics for the HEMT devices with different V sub were also measured. After each dynamic R on measurement, the device's initial state is fully recovered by shining microscopic light for 30 s, so that all trapped electrons were de-trapped [28]. Figure 6a shows that the device is switched to the on-state with a total switching time of 200 µs after 10 ms of high V D stress at off-state. As shown in Figure 6b, the positive V SUB slightly reduces the dynamic R on , while the negative V SUB significantly increases the dynamic R on . The dynamic R on ratio for applying the positive V SUB is slightly lower than without V SUB (the off-state V D is from 100 to 300 V). As shown in Figure 3b, the positive V SUB can relax the energy band bending of the Si substrate/AlN junction caused by high V D , reducing the electrons that can overcome the energy barrier injected into the buffer layer.
Although an ionized acceptor trap in a C-doped layer may screen the electric field of a positive V SUB , applying a positive V SUB slightly increases the 2DEG conductivity by the field effect. Due to the electron injections into the buffer layer from the 2DEG channel and Ohmic contact, the ionization of the acceptor trap concentration in the C-doped GaN layer increases by applying a positive V SUB . The donor trap concentration is almost similar when applying a positive V SUB and without V SUB . The literature has reported that the ionization donor trap in UID GaN can alleviate the dynamic R on degradation [9,14]. The ionization donor trap leads to a redistribution of the positive charge in the buffer stack. However, as shown in Figure 3c, the negative V SUB aggravates the Si substrate energy band bending, allowing the electrons to have higher energy to overcome the energy barriers between the AlN nucleation layer and superlattice transition layer injection into the buffer layer. The electrons injected into the buffer layer are not neutralized by the ionization donor trap (positive charge) in a short time. The 2DEG channel may be depleted, resulting in dynamic R on degradation. Therefore, under the negative V SUB , the dynamic R on degraded more seriously. can relax the energy band bending of the Si substrate/AlN junction caused by high VD, reducing the electrons that can overcome the energy barrier injected into the buffer layer. Although an ionized acceptor trap in a C-doped layer may screen the electric field of a positive VSUB, applying a positive VSUB slightly increases the 2DEG conductivity by the field effect. Due to the electron injections into the buffer layer from the 2DEG channel and Ohmic contact, the ionization of the acceptor trap concentration in the C-doped GaN layer increases by applying a positive VSUB. The donor trap concentration is almost similar when applying a positive VSUB and without VSUB. The literature has reported that the ionization donor trap in UID GaN can alleviate the dynamic Ron degradation [9,14]. The ionization donor trap leads to a redistribution of the positive charge in the buffer stack. However, as shown in Figure 3c, the negative VSUB aggravates the Si substrate energy band bending, allowing the electrons to have higher energy to overcome the energy barriers between the AlN nucleation layer and superlattice transition layer injection into the buffer layer. The electrons injected into the buffer layer are not neutralized by the ionization donor trap (positive charge) in a short time. The 2DEG channel may be depleted, resulting in dynamic Ron degradation. Therefore, under the negative VSUB, the dynamic Ron degraded more seriously.

Conclusions
The effect of VSUB on the power application characteristics of GaN-on-Si high electron mobility transistors (HEMTs) with a superlattice transition layer, including breakdown voltage (BV) and dynamic on-resistance (Ron), was investigated. The negative VSUB increases the BV due to the increasing ionized donor-trap concentration. However, the positive VSUB and without VSUB do not affect the ionization donor-trap concentration, so both BVs are similar. The applied positive VSUB can relax the energy band bending of the Si substrate/AlN junction caused by the high vertical electric field. Therefore, the reducing electrons inject the buffer stack layer. The dynamic Ron ratio can be slightly reduced by applying a positive VSUB of 200 V at high off-state VD. Although the applied negative VSUB increases the ionized donor trap concentration, it also raises the Si substrate energy band. More electrons overcome the energy barrier and are injected into the buffer layer, causing the buffer trap to capture more electrons, and depleting the two-dimensional electron gas (2DEG) channel, making the dynamic Ron degradation more serious. When applying the negative VSUB, S.S decreases, the on/off current ratio increases, and threshold voltage (Vth) shifts to positive due to the reduction in source-drain leakage current. When applying the

Conclusions
The effect of V SUB on the power application characteristics of GaN-on-Si high electron mobility transistors (HEMTs) with a superlattice transition layer, including breakdown voltage (BV) and dynamic on-resistance (R on ), was investigated. The negative V SUB increases the BV due to the increasing ionized donor-trap concentration. However, the positive V SUB and without V SUB do not affect the ionization donor-trap concentration, so both BVs are similar. The applied positive V SUB can relax the energy band bending of the Si substrate/AlN junction caused by the high vertical electric field. Therefore, the reducing electrons inject the buffer stack layer. The dynamic R on ratio can be slightly reduced by applying a positive V SUB of 200 V at high off-state V D . Although the applied negative V SUB increases the ionized donor trap concentration, it also raises the Si substrate energy band. More electrons overcome the energy barrier and are injected into the buffer layer, causing the buffer trap to capture more electrons, and depleting the two-dimensional electron gas (2DEG) channel, making the dynamic R on degradation more serious. When applying the negative V SUB, S.S decreases, the on/off current ratio increases, and threshold voltage (V th ) shifts to positive due to the reduction in source-drain leakage current. When applying the positive V SUB , the DC performances of the devices, such as on/off current ratio, S.S, and V th , remain almost the same as without an applied V SUB because both 2DEG conductivities remain nearly the same.