Improving Breakdown Voltage and Threshold Voltage Stability by Clamping Channel Potential for Short-Channel Power p-GaN HEMTs

This paper proposes a novel p-GaN HEMT (P-HEMT) by clamping channel potential to improve breakdown voltage (BV) and threshold voltage (VTH) stability. The clamping channel potential for P-HEMT is achieved by a partially-recessed p-GaN layer (PR p-GaN layer). At high drain bias, the two-dimensional electron gas (2DEG) channel under the PR p-GaN layer is depleted to withstand the drain bias. Therefore, the channel potential at the drain-side of the p-GaN layer is clamped to improve BV and VTH stability. Compared with the conventional p-GaN HEMT (C-HEMT), simulation results show that the BV is improved by 120%, and the VTH stability induced by high drain bias is increased by 490% for the same on-resistance. In addition, the influence of the PR p-GaN layers’ length, thickness, doping density on BV and VTH stability is analyzed. The proposed device can be a good reference to improve breakdown voltage and threshold voltage stability for short-channel power p-GaN HEMTs.

For power applications, low on-resistance R on and high breakdown voltage (BV) for GaN HEMTs are very desirable [7]. In order to realize low on-resistance, a short length scheme is always chosen as channel resistance under the gate is the main part of the total resistance for AlGaN/GaN HEMTs [10]. However, the short channel GaN HEMTs often suffer from the adverse drain-induced barrier lowering (DIBL) effect [11], namely, degradation of forward-blocking characteristics and negative threshold voltage (V TH ) shift at high drain bias [12,13]. In order to suppress the DIBL effect-induced BV degradation, Pinchbeck et al. proposed a GaN HEMT with extended gate length to achieve reduced short channel effect and improved BV [14]. In addition, Lu et al. proposed a dual gate AlGaN HEMT to achieve high BV, low on-resistance, and high threshold voltage characteristics [15]. However, those methods are not suitable for short-channel p-GaN HEMTs to suppress the BV degradation and V TH instability, which owns a p-GaN layer to achieve enhancementmode function.
In this work, we proposed a novel p-GaN HEMT to improve the BV and V TH stability, which features a partially-recessed p-GaN layer. At high drain bias, the two-dimensional In this work, we proposed a novel p-GaN HEMT to improve the BV and VTH stability, which features a partially-recessed p-GaN layer. At high drain bias, the two-dimensional electron gas (2DEG) channel under the partially-recessed p-GaN layer can withstand the high drain voltage to achieve higher BV and more stable VTH for the short-channel p-GaN HEMTs. The paper is organized as follows: the device structure and operation mechanism of the proposed p-GaN HEMT are presented in Section 2; the simulation results and discussions are shown in Section 3; the conclusions are drawn in Section 4.

Device Structure and Mechanism
The schematic structure of the proposed p-GaN HEMT (P-HEMT) is shown in Figure  1b. Compared with conventional p-GaN HEMT (C-HEMT), the P-HEMT features a partially recessed layer (PR p-GaN layer). To illustrate the mechanism of improving BV and VTH stability for the P-HEMT, we employ one equivalent model with two series HEMTs, which are defined as high threshold voltage HEMT1 and low threshold voltage HEMT2, as shown in Figure 2a. As the threshold voltage of HEMT1 (VTH1) is larger than the threshold voltage of HEMT2 (VTH2), HEMT2 has been turned on when the gate to source voltage (VGS) is larger than VTH1. Therefore, the threshold voltage VTH of P-HEMT is mainly determined by VTH1, namely, VTH ≈ VTH1. The potential is defined as VC at the connection node, which is also shown in Figure 1b. When 0 < VC < VGS-VTH2 (i.e., VGS-VC > VTH2), HEMT2 is at on-state. Therefore, VC increases with VDS at low drain bias. When VC > VGS-VTH2 (i.e., VGS-VC < VTH2), HEMT2 is in an off-state and the 2DEG channel under the PR p-GaN layer is depleted to withstand VDS voltage. Therefore, VC is clamped and does not increase with VDS at high drain bias, as the blue dash line shown in Figure 2b. As a result, the barrier height for electrons injecting from source to drain will be hardly influenced by high drain bias, which makes VTH more stable. In addition, the stable barrier height leads to decreased electrons flowing from source to drain compared with C-HEMT at high drain bias, which induces delayed occurrence of avalanche breakdown, namely, improves breakdown voltage.    In this work, we proposed a novel p-GaN HEMT to improve the BV and VTH stability, which features a partially-recessed p-GaN layer. At high drain bias, the two-dimensional electron gas (2DEG) channel under the partially-recessed p-GaN layer can withstand the high drain voltage to achieve higher BV and more stable VTH for the short-channel p-GaN HEMTs. The paper is organized as follows: the device structure and operation mechanism of the proposed p-GaN HEMT are presented in Section 2; the simulation results and discussions are shown in Section 3; the conclusions are drawn in Section 4.

Device Structure and Mechanism
The schematic structure of the proposed p-GaN HEMT (P-HEMT) is shown in Figure  1b. Compared with conventional p-GaN HEMT (C-HEMT), the P-HEMT features a partially recessed layer (PR p-GaN layer). To illustrate the mechanism of improving BV and VTH stability for the P-HEMT, we employ one equivalent model with two series HEMTs, which are defined as high threshold voltage HEMT1 and low threshold voltage HEMT2, as shown in Figure 2a. As the threshold voltage of HEMT1 (VTH1) is larger than the threshold voltage of HEMT2 (VTH2), HEMT2 has been turned on when the gate to source voltage (VGS) is larger than VTH1. Therefore, the threshold voltage VTH of P-HEMT is mainly determined by VTH1, namely, VTH ≈ VTH1. The potential is defined as VC at the connection node, which is also shown in Figure 1b. When 0 < VC < VGS-VTH2 (i.e., VGS-VC > VTH2), HEMT2 is at on-state. Therefore, VC increases with VDS at low drain bias. When VC > VGS-VTH2 (i.e., VGS-VC < VTH2), HEMT2 is in an off-state and the 2DEG channel under the PR p-GaN layer is depleted to withstand VDS voltage. Therefore, VC is clamped and does not increase with VDS at high drain bias, as the blue dash line shown in Figure 2b. As a result, the barrier height for electrons injecting from source to drain will be hardly influenced by high drain bias, which makes VTH more stable. In addition, the stable barrier height leads to decreased electrons flowing from source to drain compared with C-HEMT at high drain bias, which induces delayed occurrence of avalanche breakdown, namely, improves breakdown voltage.

Results and Discussions
In this section, the current-voltage and capacitance-voltage characteristics of P-HEMT are investigated by Sentaurus TCAD simulation software [16], and the design considerations are also discussed. In the simulation, the optimized device parameters are as listed in Table 1 unless otherwise specified, which is also based on our previous calibrated work [17]. In particular, the structure parameters of C-HEMT are designed according to the dissected cross-sectional scanning electron microscope (SEM) images. The x-and y-coordinates and the epitaxial structures of the two devices are illustrated in Figure 1 [18]. For C-HEMT, an ionized acceptor concentration N p-GaN = 3.5 × 10 17 cm −3 is induced in the p-GaN layer with the t p-GaN = 50 nm, which contributes to V TH and on-state current calibrations for the C-HEMT. In addition, the deep acceptor traps and self-compensating donor traps [19] are also considered in the AlGaN buffer layer with an activation energy of E V + 0.9 eV and E C -0.11 eV [20], and the trap density is 3 × 10 16 cm −3 and 1.3 × 10 15 cm −3 respectively [21]. Typically, the G L of the PR layer on the source side is only set to 0.1 µm considering the deviation of the fabrication process, and it should be as small as possible to reduce the negative influence on input capacitance in practical application. The G R of the PR layer on the drain side is an adjustable parameter as it makes obvious significance on the improvement of BV and V TH stability. In this paper, the gate length L G of C-HEMT is the same as the length of the thicker p-GaN layer of P-HEMT for achieving the same on-resistance, and the length of the partially-recessed p-GaN layer is not included in the nominal gate length L G . Table 1 shows the calibrated results of the 100 V enhancementmode p-GaN HEMT [22], and it can be seen that the results are in good agreement with the datasheet as shown in Figure 3. Typically, the BV characteristic considering the avalanche model [23] coincides well with the testing result.
tions for the C-HEMT. In addition, the deep acceptor traps and self-compensating donor traps [19] are also considered in the AlGaN buffer layer with an activation energy of EV + 0.9 eV and EC-0.11 eV [20], and the trap density is 3 × 10 16 cm −3 and 1.3 × 10 15 cm −3 respectively [21]. Typically, the GL of the PR layer on the source side is only set to 0.1 μm considering the deviation of the fabrication process, and it should be as small as possible to reduce the negative influence on input capacitance in practical application. The GR of the PR layer on the drain side is an adjustable parameter as it makes obvious significance on the improvement of BV and VTH stability. In this paper, the gate length LG of C-HEMT is the same as the length of the thicker p-GaN layer of P-HEMT for achieving the same onresistance, and the length of the partially-recessed p-GaN layer is not included in the nominal gate length LG. Table 1 shows the calibrated results of the 100 V enhancement-mode p-GaN HEMT [22], and it can be seen that the results are in good agreement with the datasheet as shown in Figure 3. Typically, the BV characteristic considering the avalanche model [23] coincides well with the testing result.  Figure 4 shows the forward-blocking and output characteristics of the P-HEMT. As shown in Figure 4a, it can be seen that the BV (I DS_off = 100 µA) for the P-HEMT is increased by 120% compared with the 100 V C-HEMT, which mainly results from the delayed occurrence of avalanche breakdown. As shown in Figure 4b, it can be observed that impact ionization at the drain-side source field plate is decreased at the same 150 V drain bias, which results from the decreased electrons flowing from source to drain. In addition, as shown in Figure 4c, the conduction energy E C level at the drain-side of the p-GaN layer for P-HEMT is clamped, which results from the clamped V C as stated in section II. As shown in Figure 4d, for typical gate operation voltage V GS = 5 V, the output curves of C-HEMT and P-HEMT are coincident well, which indicates that the PR p-GaN layer makes a negligible impact on the on-state resistance. For V GS = 2 V, the I DS for C-HEMT is slightly higher than P-HEMT, which results from the partial depletion of the 2DEG channel under the PR p-GaN layer. Figure 5 shows the transfer characteristics of the P-HEMT. At low drain bias (such as V DS = 1 V), the transfer curves of C-HEMT and P-HEMT are coincident well and the threshold voltage difference is less than 0.05 V. However, with the increasing of V DS , the V TH of C-HEMT decreases obviously while V TH of P-HEMT slightly reduced. Typically, the V TH decrease from V DS = 1 V to V DS = 50 V is 0.59 V for C-HEMT and 0.1 V for P-HEMT, as shown in Figure 5b. The significantly decreased V TH for C-HEMT will lead to false turn-on at high drain bias (typically, from off-state to on-state), which is not acceptable for practical application. However, from the results, it can be deduced that the P-HEMT with more stable V TH can be contributed to alleviating this problem very well.   Figure 5 shows the transfer characteristics of the P-HEMT. At low drain bias (such as VDS = 1 V), the transfer curves of C-HEMT and P-HEMT are coincident well and the threshold voltage difference is less than 0.05 V. However, with the increasing of VDS, the VTH of C-HEMT decreases obviously while VTH of P-HEMT slightly reduced. Typically, the VTH decrease from VDS = 1 V to VDS = 50 V is 0.59 V for C-HEMT and 0.1 V for P-HEMT, as shown in Figure 5b. The significantly decreased VTH for C-HEMT will lead to false turnon at high drain bias (typically, from off-state to on-state), which is not acceptable for practical application. However, from the results, it can be deduced that the P-HEMT with more stable VTH can be contributed to alleviating this problem very well. To illustrate the impact of the PR p-GaN layer on transient behavior, the simulation using a double pulse circuit is carried out, as shown in Figure 6. Compared with C-HEMT, the calculated turn-on loss and turn-off loss of P-HEMT are increased by 0.09 μJ and 0.02 μJ at 500 kHz respectively, and the total switching loss is increased by less than 7.8%. It can be inferred the increased switching loss mainly results from the increase of the input capacitance CISS. As shown in Figure 7, it can be seen that the off-state and on-state input capacitance CISS is increased by 18.9% and 47.2%, respectively. In addition, as shown in Figure 7a, the COSS at high-drain bias (VDS > 15 V) is the same as C-HEMT, and the output capacitance COSS at a low-drain bias (VDS < 15 V) is decreased by 16.7%, which mainly results from the depletion of 2DEG channel under the PR layer as stated in section II. The decrease of COSS at VDS < 15 V is contributed to reducing the increment of switching loss.   Figure 5 shows the transfer characteristics of the P-HEMT. At low drain bias (such a VDS = 1 V), the transfer curves of C-HEMT and P-HEMT are coincident well and the thresh old voltage difference is less than 0.05 V. However, with the increasing of VDS, the VTH o C-HEMT decreases obviously while VTH of P-HEMT slightly reduced. Typically, the VT decrease from VDS = 1 V to VDS = 50 V is 0.59 V for C-HEMT and 0.1 V for P-HEMT, a shown in Figure 5b. The significantly decreased VTH for C-HEMT will lead to false turn on at high drain bias (typically, from off-state to on-state), which is not acceptable fo practical application. However, from the results, it can be deduced that the P-HEMT with more stable VTH can be contributed to alleviating this problem very well. To illustrate the impact of the PR p-GaN layer on transient behavior, the simulation using a double pulse circuit is carried out, as shown in Figure 6. Compared with C-HEMT the calculated turn-on loss and turn-off loss of P-HEMT are increased by 0.09 μJ and 0.0 μJ at 500 kHz respectively, and the total switching loss is increased by less than 7.8%. I can be inferred the increased switching loss mainly results from the increase of the inpu capacitance CISS. As shown in Figure 7, it can be seen that the off-state and on-state inpu capacitance CISS is increased by 18.9% and 47.2%, respectively. In addition, as shown in Figure 7a, the COSS at high-drain bias (VDS > 15 V) is the same as C-HEMT, and the outpu capacitance COSS at a low-drain bias (VDS < 15 V) is decreased by 16.7%, which mainly re sults from the depletion of 2DEG channel under the PR layer as stated in section II. Th decrease of COSS at VDS < 15 V is contributed to reducing the increment of switching loss. To illustrate the impact of the PR p-GaN layer on transient behavior, the simulation using a double pulse circuit is carried out, as shown in Figure 6. Compared with C-HEMT, the calculated turn-on loss and turn-off loss of P-HEMT are increased by 0.09 µJ and 0.02 µJ at 500 kHz respectively, and the total switching loss is increased by less than 7.8%. It can be inferred the increased switching loss mainly results from the increase of the input capacitance C ISS . As shown in Figure 7, it can be seen that the off-state and on-state input capacitance C ISS is increased by 18.9% and 47.2%, respectively. In addition, as shown in Figure 7a, the C OSS at high-drain bias (V DS > 15 V) is the same as C-HEMT, and the output capacitance C OSS at a low-drain bias (V DS < 15 V) is decreased by 16.7%, which mainly results from the depletion of 2DEG channel under the PR layer as stated in section II. The decrease of C OSS at V DS < 15 V is contributed to reducing the increment of switching loss.

Design Considerations of P-HEMT
This section mainly discusses the impact of PR p-GaN layers' thickness, length, and doping concentration on the BV and V TH stability. Figure 8 shows the V TH and BV results for different thicknesses of the PR p-GaN layer. As shown in Figure 8a, it can be seen that the DIBL value is increased with PR p-GaN layer thickness. The DIBL parameter is defined as (V High TH − V Low TH )/(V High DS − V Low DS ) to represent the V TH stability, and the smaller value symbolizes the more stable V TH . As shown in Figure 8b, it can be seen that the V TH for different thickness PR layers from V DS = 1 V to V DS = 50 V decreases, but the difference is all less than 0.1 V, which indicates the high stable V TH for P-HEMT. The log-scale transfer characteristics are as shown in Figure 8c-e. For the same drain bias, the V TH slightly increases (≤0.05 V) with the thickness of the PR p-GaN layer, which mainly results from the 2DEG depletion under the PR p-GaN layer. In addition, it can be observed that the BV is all larger than 320 V, which indicates the impact of the PR p-GaN layer's thickness on BV is negligible. However, for a smaller thickness PR p-GaN layer, the gate-to-source breakdown voltage can be reduced. Figure 9a shows the I GS -V GS characteristics for 20/30/40 nm PR p-GaN layer, and it can be seen that the I GS for T p2 = 20 nm abruptly increases when V GS is larger than 5.1 V. To explore the origin of the abrupt I GS , the current distribution of the three thickness PR p-GaN layer devices are plotted, as shown in Figure 9b-d. For the device with T p2 = 20 nm, the current density from the PR p-GaN layer to the 2DEG channel is larger than the normal thickness p-GaN layer. This indicates the high gate current mainly results from the breakdown of the PR p-GaN layer. As a comparison, the gate current density for T p2 = 30/40 nm is very small. Based on the above analysis, it can be deduced that the PR p-GaN layer thickness should be taken into careful consideration in the design to avoid gate breakdown.

Design Considerations of P-HEMT
This section mainly discusses the impact of PR p-GaN layers' thickness, length, and doping concentration on the BV and VTH stability. Figure 8 shows the VTH and BV results for different thicknesses of the PR p-GaN layer. As shown in Figure 8a, it can be seen that the DIBL value is increased with PR p-GaN layer thickness. The DIBL parameter is defined as (V DS ) to represent the VTH stability, and the smaller value symbolizes the more stable VTH. As shown in Figure  8b, it can be seen that the VTH for different thickness PR layers from VDS = 1 V to VDS = 50 V decreases, but the difference is all less than 0.1 V, which indicates the high stable VTH for P-HEMT. The log-scale transfer characteristics are as shown in Figure 8c-e. For the same drain bias, the VTH slightly increases (≤0.05 V) with the thickness of the PR p-GaN layer, which mainly results from the 2DEG depletion under the PR p-GaN layer. In addition, it can be observed that the BV is all larger than 320 V, which indicates the impact of the PR

Design Considerations of P-HEMT
This section mainly discusses the impact of PR p-GaN layers' thickness, length, and doping concentration on the BV and VTH stability. Figure 8 shows the VTH and BV results for different thicknesses of the PR p-GaN layer As shown in Figure 8a, it can be seen that the DIBL value is increased with PR p-GaN layer thickness. The DIBL parameter is defined as (V  Figure 10 shows the impact of PR p-GaN layer length on the BV and V TH characteristics. It can be seen that DIBL decreases with G r , which indicates the V TH stability is increased. However, the DIBL tends to be stable and the BV decreases when G r is larger than 0.5 µm. The decrease of the BV mainly results from the high electric field at the drain-side of the PR p-GaN layer, as shown in Figure 11. Therefore, it can be deduced that the PR p-GaN layer length should be in a reasonable range to get a good trade-off for V TH stability and high BV. For the 100 V p-GaN HEMT discussed in this paper, the 0.3~0.5 µm PR p-GaN layer is recommended. As a comparison, the gate current density for Tp2 = 30/40 nm is very small. Based on the above analysis, it can be deduced that the PR p-GaN layer thickness should be taken into careful consideration in the design to avoid gate breakdown.   Figure 10 shows the impact of PR p-GaN layer length on the BV and VTH characteristics. It can be seen that DIBL decreases with Gr, which indicates the VTH stability is increased. However, the DIBL tends to be stable and the BV decreases when Gr is larger than 0.5 μm. The decrease of the BV mainly results from the high electric field at the drain-side of the PR p-GaN layer, as shown in Figure 11. Therefore, it can be deduced that the PR p-GaN layer length should be in a reasonable range to get a good trade-off for VTH stability and high BV. For the 100 V p-GaN HEMT discussed in this paper, the 0.3~0.5 μm PR p-GaN layer is recommended. As a comparison, the gate current density for Tp2 = 30/40 nm is very small. Based on the above analysis, it can be deduced that the PR p-GaN layer thickness should be taken into careful consideration in the design to avoid gate breakdown.   Figure 10 shows the impact of PR p-GaN layer length on the BV and VTH characteristics. It can be seen that DIBL decreases with Gr, which indicates the VTH stability is increased. However, the DIBL tends to be stable and the BV decreases when Gr is larger than 0.5 μm. The decrease of the BV mainly results from the high electric field at the drain-side of the PR p-GaN layer, as shown in Figure 11. Therefore, it can be deduced that the PR p-GaN layer length should be in a reasonable range to get a good trade-off for VTH stability and high BV. For the 100 V p-GaN HEMT discussed in this paper, the 0.3~0.5 μm PR p-GaN layer is recommended.  Figure 12 shows the impact of p-GaN doping density on the V TH and BV. It can be observed that the p-GaN doping density mainly determines the magnitude of V TH , and it makes a negligible effect on BV and DIBL. Figure 13 shows the V TH and BV characteristics of P-HEMT with different gate lengths L g . It can be seen that longer gate length induces higher V TH , lower DIBL, which means longer gate length is contributed to making V TH more stable. In addition, longer gate length induces higher BV, which mainly results from the electric field modulation. However, longer gate length will induce higher on-resistance. Therefore, the gate length should be taken into careful consideration to get a better trade-off for V TH stability, BV, and R ON .   Figure 12 shows the impact of p-GaN doping density on the VTH and BV. It can be observed that the p-GaN doping density mainly determines the magnitude of VTH, and it makes a negligible effect on BV and DIBL. Figure 13 shows the VTH and BV characteristics of P-HEMT with different gate lengths Lg. It can be seen that longer gate length induces higher VTH, lower DIBL, which means longer gate length is contributed to making VTH more stable. In addition, longer gate length induces higher BV, which mainly results from the electric field modulation. However, longer gate length will induce higher on-resistance. Therefore, the gate length should be taken into careful consideration to get a better trade-off for VTH stability, BV, and RON.   Figure 12 shows the impact of p-GaN doping density on the VTH and BV. It ca observed that the p-GaN doping density mainly determines the magnitude of VTH, an makes a negligible effect on BV and DIBL. Figure 13 shows the VTH and BV characteri of P-HEMT with different gate lengths Lg. It can be seen that longer gate length ind higher VTH, lower DIBL, which means longer gate length is contributed to making more stable. In addition, longer gate length induces higher BV, which mainly results f the electric field modulation. However, longer gate length will induce higher on sistance. Therefore, the gate length should be taken into careful consideration to get a ter trade-off for VTH stability, BV, and RON.

Conclusions
This paper proposes a novel p-GaN HEMT with a PR p-GaN layer to improve BV and VTH stability. The device features a PR p-GaN layer compared with conventional p-GaN HEMT. At high drain bias, the two-dimensional electron gas channel under the PR p-GaN layer is depleted to withstand VDS, thereby contributing to improving the BV and VTH stability. Compared with the C-HEMT, simulation results show that the breakdown voltage is improved by 120%, and the VTH stability changing with VDS is increased by 490% (the decrease of VTH at 50 V for P-HEMT and C-HEMT are 0.1 V and 0.59 V respectively). The static transfer and output characteristics are the same as the C-HEMT, and the total switching loss at 500 kHz is increased less than 7.8%. In addition, we investigated the impact of the PR layers' length, thickness, doping density on the performance.

Conclusions
This paper proposes a novel p-GaN HEMT with a PR p-GaN layer to improve BV and V TH stability. The device features a PR p-GaN layer compared with conventional p-GaN HEMT. At high drain bias, the two-dimensional electron gas channel under the PR p-GaN layer is depleted to withstand V DS , thereby contributing to improving the BV and V TH stability. Compared with the C-HEMT, simulation results show that the breakdown voltage is improved by 120%, and the V TH stability changing with V DS is increased by 490% (the decrease of V TH at 50 V for P-HEMT and C-HEMT are 0.1 V and 0.59 V respectively). The static transfer and output characteristics are the same as the C-HEMT, and the total switching loss at 500 kHz is increased less than 7.8%. In addition, we investigated the impact of the PR layers' length, thickness, doping density on the performance.