The desire for power electronic systems with increasingly excellent performances has been the catalyst for developing GaN-based power devices, in which the vertical GaN-based heterostructure field effect transistor (GaN-based V-HFET), as a kind of high-power switch device, has been attracting increasing attention. Compared with its lateral counterpart (GaN-based HEMT),^[1–6] the GaN-based V-HFET has the superiorities, i.e., the reduced chip area, the packaging convenience and the suppression of current collapse,^[7] which makes it significantly advantageous and perhaps the only topology choice for power conversions beyond 15 kW.^[8] In a high power system, it has been one of the most important focuses to minimize the system power loss. Therefore, the reduction of the specific on-resistance in GaN-based V-HFET is strongly required to reduce the power device loss even at high frequency operation,^[9] which depends on a better trade-off characteristic between breakdown voltage (BV) and specific on-resistance (R_onA).

Recently, a great deal of effort has been made to develop the device structures and fabrication process for GaN-based V-HFETs. Many novel V-HFET structures have been proposed to further improve the trade-off between BV and R_onA,^[10–13] of which the GaN-based V-HFET with uniform doping superjunction (un-SJ HFET) could be considered as the most effective and promising candidate.^[11] Furthermore, the successful applications of the etching technology, the regrowth technology and the implantation technology in the fabrication of GaN-based V-HFETs^{[7,8,14–16]} have enhanced the possibility of realizing various novel vertical device structures. These advances are a great encouragement to the further study of GaN-based V-HFETs.

In this paper, a novel GaN-based V-HFET with nonuniform doping superjunction (non-SJ HFET) is presented and investigated. The study of this device emphasizes the further reduction in R_onA with the same BV as that of the conventional un-SJ HFET. Compared with un-SJ HFET, the non-SJ HFET has an obvious feature that is the nonuniform doping superjunction (non-SJ) comprising nonuniform doping n- and p-pillars. Based on this, the specific on-resistance could be remarkably reduced without any loss of BV. Two-dimensional numerical simulation with Silvaco-ATLAS software is performed to reveal the intrinsic operation mechanism for the non-SJ by using the potential distributions, electric field distributions and current density distributions. The relationship among the R_onA, the BV and the structure parameters of the non-SJ is studied systematically, aiming at optimizing the minimized R_onA. These results are of importance for designing and further developing the GaN-based V-HFETs.

Figure 1 shows the schematic cross-sections of the proposed GaN-based non-SJ HFET and the conventional GaN-based un-SJ HFET. Except for the superjunction, structures for both devices are the same, whose main parameters are given in Table 1. In GaN-based non-SJ HFET, both n- and p-pillars are divided into two parts. The same thickness H₁ and doping concentration N₁ are used in both the top part of the n-pillar and the bottom part P of the p-pillar. The same thickness H₂ and doping concentration N₂ are used in both the bottom part of the n-pillar and the top part of the p-pillar. The doping concentration increment, namely minus N₁, is defined as . In addition, the doping concentration of n- and p-pillars in un-SJ HFET is always equal to that of the top part of the n-pillar in non-SJ HFET. For convenience and simplicity, typical N₁ concentrations, namely 0.5 × 10¹⁶ cm⁻³, 1 × 10¹⁶ cm⁻³, 1.5 × 10¹⁶ cm⁻³ or 2 × 10¹⁶ cm⁻³, are utilized in simulations, respectively.

Fig. 1.

	Figure Option ViewDownloadNew Window
	Fig. 1. (color online) Schematic structures for (a) non-SJ HFET and (b) un-SJ HFET. Lines A1A2 or B1B2 denote the vertical direction at the n-pillar/p-pillar junction or the middle of the device, respectively.

Table 1.

Table 1.

Table 1. Main structure specifications for non-SJ HFET and un-SJ HFET. . Symbol Description Values /nm -GaN cap thickness 100 -GaN cap doping concentration 1 × 1018 /nm unintentionally doped Al0.15Ga0.85N thickness 30 /nm GaN channel thickness 120 GaN channel doping concentration 2 × 1016 current blocking layer (CBL) thickness 1 CBL doping concentration 1 × 1018 H/ GaN drift layer thickness 10 H1/ top (bottom) part thickness of n(p)-pillar innon-SJ HFET 1∼10 N1/cm−3 top (bottom) part doping concentration of n(p)-pillar in non-SJ HFET or doping concentration of n- and p-pillars in un-SJ HFET H2 bottom (top) part thickness of n(p)-pillar in non-SJ HFET H minus H1 doping concentration increment in non-SJ HFET N2 bottom (top) part doping concentration of n(p)-pillar in non-SJ HFET n-pillar (p-pillar) width ( 5

Table 1. Main structure specifications for non-SJ HFET and un-SJ HFET. .

Two-dimensional device simulations are performed by using the Silvaco-ATLAS. The polarization effect is considered at each of the P⁺-GaN cap/Al^0.15Ga^0.85N and Al^0.15Ga^0.85N/GaN channel interfaces, according to the experimental reports.^[17] Caughey–Thomas mobility model is used based on the experimental measurements.^[18] Carrier generation and recombination models, including the impact ionization model,^[19] the SRH recombination model and the Auger recombination model are also included in the simulation. Other material physical parameters for GaN and Al^0.15Ga^0.85N are adopted according to the reports.^[20,21]

Figure 2 shows the on-state I–V characteristics for non-SJ HFET and un-SJ HFET. In Fig. 2(a), both devices exhibit the same threshold voltage of about 0.6 V. The drain current of the proposed non-SJ HFET is greater than that of un-SJ HFET when the gate voltage is beyond the threshold voltage, indicating a smaller on-state resistance in non-SJ HFET. In addition, comparing with un-SJ HFET, the obvious improvement of the slope of the I–V curve in the linear region could be seen from Fig. 2(b) in non-SJ HFET at the same bias. This results in a smaller R_onA of in non-SJ HFET than that of in un-SJ HFET, thus leading to more than 31% decrease in R_onA. From the off-state I–V performance in Fig. 3, both devices exhibit the same breakdown voltage of 1800 V. Therefore, the significant reduction in R_onA could be realized by means of nonuniform doping superjunctions at no expense of BV, which demonstrates a better trade-off between BV and R_onA in non-SJ HFET.

Fig. 2.

	Figure Option ViewDownloadNew Window
	Fig. 2. (color online) On-state I–V characteristics for two devices, showing (a) transfer characteristics, (b) output characteristics.

Fig. 3.

	Figure Option ViewDownloadNew Window
	Fig. 3. (color online) Off-state I–V characteristics corresponding to the devices in Fig. 2.

A physical insight into the improvement in non-SJ HFET could be achieved according to on-state current density distributions as well as off-state equipotential lines and electric field distributions. At on-state for both devices in Fig. 4(a), only the n-pillar can conduct current because the substrate/p-pillar junctions and the n-pillar/p-pillar junctions are under the reversed bias condition. The current path width along the horizontal direction in n-pillar will be shrunk by the depletion effect to a certain degree. As shown, a similar current path width could be observed in the top part of the n-pillar in non-SJ HFET and the corresponding position in un-SJ HFET. Comparing with un-SJ HFET, the current path in the bottom part of the n-pillar in the non-SJ HFET is obviously widened, which is attributed to the higher doping concentration in the bottom part of the n-pillar. From Fig. 4(b), a greater current density could be seen in the n-pillar of the non-SJ HFET. Therefore, both the wider current path and the greater current density in non-SJ HFET are beneficial to the current conduction and then the reduction in R_onA.

Fig. 4.

	Figure Option ViewDownloadNew Window
	Fig. 4. (color online) On-state current density distributions corresponding to the devices in Fig. 2, showing (a) two-dimensional distributions, and (b) distributions along line B1B2 as shown in Fig. 1.

The breakdown performances in non-SJ HFET and un-SJ HFET are illustrated in Fig. 5. Comparing with un-SJ HFET, in non-SJ HFET the equipotential lines densities near the middle of n- and p-pillars increase slightly while those near the bottom of n- and p-pillars decrease slightly in Fig. 5(a). This implies that the electric field strength near the middle is enhanced while that near the bottom is degraded in n- and p-pillars, which could be demonstrated by the vertical electric field distributions along the line A1A2 in Fig. 5(b). As shown, the area beneath the contour of each electric field distribution can be used to estimate the BV of each device approximately.^[4] Equal areas could be observed in both devices, and thus indicating the same BV.

Fig. 5.

	Figure Option ViewDownloadNew Window
	Fig. 5. (color online) Off-state performances in two devices at breakdown voltage corresponding to Fig. 3. (a) Equipotential lines distribution, and (b) electric field distribution along line A1A2 as shown in Fig. 1.

Figure 6 shows the plots of breakdown voltage and specific on-resistance versus thickness H₂ at different values of N₁ for a given in non-SJ HFET. It is noted that the structure of non-SJ HFET will be changed into that of un-SJ HFET when . As shown, for each concentration N₁, the R_onA decreases continuously with H₂ increasing. This results from the increase in the average doping concentration of n-pillar. Besides, there exists a peak BV at for each N₁. For a given H₂, both BV and R_onA decrease with N₁ increasing. The variations of BV and R_onA with N₁ in un-SJ HFET are also consistent with the reported results in Ref. [11]. The optimized H₂ is determined by minimizing R_onA on the basis of the same BV as that of un-SJ HFET, which means the maximum reduction in R_onA at no expense of BV. It could be found that the optimized H₂ appears to be nearly and will decrease slightly with the increase in N₁. If H₂ is beyond the optimal value, the BV of non-SJ HFET will be lower than that of un-SJ HFET for a given N₁. When varies from to , the optimized R_onA will drop from to , indicating a corresponding reduction from 48.8% to 15.5% compared with the scenario of un-SJ HFET. Figure 7 shows the vertical electric field distributions along the line A1A2 in non-SJ HFETs with different values of H₂. As shown in the figure, the electric field strength around the middle of pillar is raised obviously for , leading to the largest area beneath the contour of electric field distribution and thus the greatest BV. In the same way, the lowest BV appears in the case of because of the smallest area. These features of electric field distributions could provide an appropriate explanation for the variations of BV with in Fig. 6.

Fig. 6.

	Figure Option ViewDownloadNew Window
	Fig. 6. (color online) Plots of breakdown voltage and specific on-resistance versus H2 at different values of for a given in non-SJ HFET.

Fig. 7.

	Figure Option ViewDownloadNew Window
	Fig. 7. (color online) Vertical electric field distributions along the line A1A2 in corresponding non-SJ HFETs with different values of H2 in Fig. 6.

Figure 8 shows the plots of breakdown voltage and specific on-resistance versus H₂ at different values of for a given N₁. For each , the R_onA will drop as increases. Further, it could be seen clearly that the decrement in R_onA will be raised with increasing for a given H₂. The variation of BV with for each is similar to that in Fig. 6 and the peak BV appears at . On the basis of the same BV as that of un-SJ HFET, the optimized H₂ for each is obtained to minimize R_onA. As shown, this optimized H₂ will increase a little with increasing. According to the minimum R_onA corresponding to each optimized in the non-SJ HFET, the reduction of R_onA in the non-SJ HFET compared with the un-SJ HFET, could be calculated and depicted in Fig. 9. The reductions in R_onA for other optimized results at different values of N₁ achieved by the method in Fig. 8 are also shown in Fig. 9, and the values of BV in non-SJ HFETs with different values of N₁ are also equal to those in un-SJ HFETs with the same N₁. For each

Fig. 8.

	Figure Option ViewDownloadNew Window
	Fig. 8. (color online) Plots of breakdown voltage and specific on-resistance versus H2 at different values of for a given in non-SJ HFET.

Fig. 9.

	Figure Option ViewDownloadNew Window
	Fig. 9. (color online) BV and reductions in RonA for optimized results at different values of N1 and .

N₁, the reduction in R_onA could be improved by increasing . The remarkable reduction in R_onA could be obtained in the case of lower and higher . The maximum reduction of 51.1% in R_onA could be realized for and . Even when and , the reduction of 9.8% could still be achieved. These improvements further demonstrate the superiority and effectiveness of non-SJ HFET in minimizing R_onA at no expense of BV.

Nowadays, thick in situ doping GaN layer growth technologies,^[22,23] GaN-based selective area growth technologies (SAG),^[14,24] and GaN etching technologies have become more and more mature.^[25,26] The reported highest GaN pillar height by SAG and the deepest etching trench by plasma etching technology have been more than ,^[27,28] which could pave the way for fabricating the non-SJ HFET. Therefore, the proposed non-SJ HFET may be realized by SAG technology, during which the n-pillar and p-pillar will be grown alternately. This fabrication proposal has also been considered as a possible way to realize the conventional un-SJ HFET in Ref. [11].

As an example, the following gives a feasible process approach of the proposed non-SJ HFET based on the methods mentioned in Refs. [14] and [24], as illustrated in Fig. 10. Firstly, a p-type GaN layer with N₁ doping and H₁ thickness is epitaxially grown by metal–organic chemical vapor deposition (MOCVD) on a bulk GaN substrate. Then trench-1 with width and H₁ thickness is formed by the Cl-based inductively coupled plasma (ICP) etching process. The rest of the p-type GaN layer is masked and an n-type GaN layer with N₂ doping and H₁ thickness is regrown by SAG to fill trench-1. Secondly, a p-type GaN layer with N₂ doping and (H₂ minus thickness is epitaxially grown by MOCVD followed by the trench-2 etching and the SAG regrowth of n-type GaN layer with N₂ doping in the trench. Thirdly, a p-type GaN layer with N₂ doping and H₁ thickness and the 1- CBL layer are epitaxially grown by MOCVD followed by the trench-3 etching and the SAG regrowth of n-type GaN layer with N₁ doping in the trench. The final main procedures comprise the regrowth of the GaN channel and Al_0.15Ga_0.85N, the growth and etching of -GaN cap, the source/drain/gate contact metal and passivation layer deposition and the isolation deep etching of GaN, after which the completed device is obtained and shown in Fig. 10(j).

Fig. 10.

Figure Option ViewDownloadNew Window

Fig. 10. (color online) A feasible process approach of the proposed non-SJ HFET, showing (a) (N1 doping and H1 thickness) p-type GaN layer growth, (b) trench-1 etching, (c) n-type GaN layer regrowth in trench-1, (d) (N2 doping and H2–H1 thickness) p-type GaN layer growth, (e) trench-2 etching, (f) n-type GaN layer regrowth in trench-2, (g) (N2 doping and H1 thickness) ptype GaN layer and CBL growth, (h) trench-3 etching, (i) n-type GaN layer regrowth in trench-3, and (j) completed device.

In this paper, a novel non-SJ HFET based on the nonuniform doping superjunctions (non-SJs) is presented by two-dimensional simulations, with emphasis on the minimized R_onA at no expense of BV. Systematical studies and analyses of the device on- and off-state performance, device physical mechanisms as well as device optimization designs are conducted. The simulation results exhibit that the remarkable reduction in R_onA with the same BV as that of un-SJ HFET, could be realized by modulating the doping concentration and thickness of the light and heavy doping parts in non-SJ. These results demonstrate the potential and advantage of non-SJ HFET in high power electronic applications.