Study of High-Performance GaN-Based Trench CAVET with Stepped Doping Microstructure

In this article, an innovative GaN-based trench current-aperture vertical electron transistor (CAVET) with a stepped doping microstructure is proposed and studied using Silvaco-ATLAS. According to the simulation and analyzed characteristics, the best performance renders a remarkable Baliga’s figure of merit (FOM) of 4.767 GW·cm2 owing to the modulation of the electric-field distribution. By adjusting the size of the stepped doping microstructure and doping concentration in the GaN drift, the maximum optimized result can achieve a relatively high breakdown voltage (BV) of 2523 V with a very low specific on-resistance (Ron,sp) of 1.34 mΩ·cm2, or the BV can be improved to 3024 V with a specific on-resistance (Ron,sp) of 2.08 mΩ·cm2. Compared with the conventional superjunction GaN-based trench CAVET, the newly demonstrated structure can achieve a 43% reduction in Ron,sp and increase by almost 20% the original BV. These results indicate the superiority of using the stepped doping microstructure in a trench CAVET to improve the BV and decrease Ron,sp, providing a reference for further development of GaN-based CAVETs.


Introduction
Over the past few decades, GaN-based devices went through rapid development, such as GaN MOSFETs, GaN Schottky diodes [1] and GaN photodetectors [2]. In the field of high-power applications, GaN-based power devices with a vertical structure attract much attention for their suitability in the fabrication of high-efficiency power converters [3]. Although typical lateral structures [4,5] dominate the GaN-based high-electron-mobility transistors (HEMTs), vertical devices holding a vertical conduction path draw great interests for their superior ability in achieving high breakdown voltages (BVs) and high current levels without enlarging the chip size or introducing extra thermal management [6,7], such as GaN vertical MOSFETs [8], GaN interlayer-based vertical MOSFETs [9] and GaN-based trench current-aperture vertical electron transistors (CAVETs) [10]. Among these, the trench CAVET deserves attention since it enables the normally off mode to be easily realized, compared with planar-gate CAVETs [11]. In addition, the high-electron-mobility channels formed by polarization-based 2DEG in GaN-based CAVETs make them more suitable for applications in the order of kilowatts [7]. In 2013, Li [10] et al. proposed an enhancementmode CAVET with a superjunction (SJ). The design of a half p-GaN pillar below the current blocking layer (CBL) greatly improved the BV of the device. The best R on,sp of 4.2 mΩ·cm 2 with a BV of 12.4 kV was obtained. In 2016, Shibata [11] et al. reported a CAVET with a regrown p-GaN/AlGaN/GaN trench-gate semi-polar structure. Making possible normally off operation, the CAVET held a specific on-resistance (R on,sp ) of 1.0 mΩ·cm 2 and a BV of 1.7 kV. In 2018, Ji [12] et al. successfully fabricated an MIS trench-gate CAVET with a high BV of 880 V and low R on,sp of 2.7 mΩ·cm 2 . By adopting standard GaN-HEMT epitaxial high temperature, the high-quality in situ Si 3 N 4 gate dielectric brought a low hysteresis of~0.1 V. In 2022, Bai [13] et al. simulated a novel trench-gate CAVET with dual-current aperture. By inserting the p-GaN island between the two p-GaN CBLs, the electric-field distribution was properly modulated to achieve a high BV of 1504 V and low R on,sp of 0.77 mΩ·cm 2 .
In conventional planar MOSFETs, R on,sp has a weak correlation with the BV [14], whereas vertical structures such as SJ-HEMTs and CAVETs establish a linear relationship between R on,sp and the BV [15]. CAVETs are specialized in high-voltage and high-power applications, especially when introducing an SJ, which is a bulk of p-buried layers to deplete the 2DEG in the channel. The geometry, thickness and doping are significant parameters to determine the forward characteristics and balance the tradeoff between R on,sp and the BV. However, while the BV can be improved utilizing the P-buried layer since it can fully deplete the 2DEG in the channel [16], it also increases R on,sp and reduces the maximum saturation current due to the decrease in the 2DEG density of the channel, which degrade the forward characteristics of the device. As per the principle of charge compensation, modulating the drift region in CAVETs using a uniform doping profile is a feasible method to surpass the performance of conventional SJ-CAVETs. Yet, seldom investigations and experiments were carried out on GaN-based CAVETs using p-buried layers or SJs in the drift. Although a few simulations [17,18] designed nonuniform doping profiles in the drift of GaN-based CAVETs, these studies only focused on the half pillar of the SJ and its doping variation in the longitudinal direction. There are no research studies reporting about the alternation in the drift doping region along the plane axis. Thus, the purpose of this paper is to design a reliable GaN-based CAVET model to explore and predict the influence of nonuniform profile doping in the horizontal direction.
In this paper, a GaN-based trench CAVET with stepped doping microstructure is proposed to further improve both the breakdown voltage (BV) and specific on-resistance (R on,sp ). By alternating the size and doping concentration of n-, p-buried regions in the GaN drift layer, the 2DEG can be appropriately depleted in the channel, and the vertical excess electric field from source to drain can be absorbed as an extra current blocking layer (CBL). Meanwhile, the current density in the channel can be appropriately promoted, and the electric field in the vertical path can be boosted. Utilizing this method, the increase in specific on-resistance can be ignored, while the BV can be remarkably enhanced. Thus, the option of modulating the GaN drift layer via stepped doping is a feasible choice to achieve a trade-off performance enhancement. To evaluate the performance and investigate the parameters, two-dimensional numerical simulations are operated and analyzed using Silvaco-Atlas.

TCAD Model Simulation
The half-cell schematic cross-section of a GaN-based trench CAVET with stepped doping microstructure (SDS-Trench CAVET) is illustrated in Figure 1b. In addition, a superjunction GaN-based trench CAVET (SJ-Trench CAVET) is also exhibited in Figure 1a for comparison. With an overall width of 5 µm, the main parameters include a 2 µm thick GaN n+ substrate, a 15 µm thick n-GaN drift layer and a 700 nm thick p+ GaN current blocking layer (CBL). The channel layer is composed of 20 nm thick Al 0.25 Ga 0.75 N and 210 nm thick UID GaN. In addition, 60 nm thick silicon nitride (Si 3 N 4 ) serves as a gate dielectric. The gate length, trench depth and CBL length are set as 2 µm, 1 µm and 4 µm, respectively. The n+ GaN substrate is heavily doped with Si, and the p+ GaN is heavily doped with Mg. On the basis of Ref. [19], the origin CAVET is redesigned, and the other detailed device parameters are represented in Table 1. to that of the n-and p-pillars in the SJ-Trench CAVET. For simplicity, typical N1 conc trations, namely, from 0.5 × 10 16 cm −3 to 2 × 10 16 cm −3 , are utilized in simulations. Top p NN1 and bottom part PP1 keep the same width (W2); thus, bottom part NN2 and top part are given the same width (W1). It is worth noting that W1 should not exceed the length the CBL to avoid excessive depletion of 2DEG in the channel.  To obtain more accurate results in Silvaco TCAD, the concentration depend (CONSRH) and Auger physical models are used as recombination models, while Selb herr's Model (IMPACT SELB) is adopted for impact ionization. In addition, the high-fi saturation and -polarization models are also used in the simulation. According to Ref. [ Li et al. proposed and simulated a classic GaN vertical superjunction HEMT. Thus, model and material parameters could refer to a similar device. The parameter detail low-field-mobility models and Farahmand-Modified Caughey-Thomas (FMCT) mod are listed in Table 2.  As is shown in Figure 1a, half of the GaN drift layer in the SJ-Trench CAVET is divided into an n-pillar and a p-pillar with the same width (1/2W) and the same height (H). In the SDS-Trench CAVET, the GaN drift layer is evenly separated into a top stepped doping microstructure and a bottom stepped doping microstructure. In addition, both the top part and bottom part of the stepped doping microstructure are divided into two parts, the p-doping area below the CBL and the n-doping area adjacent to it. The same doping concentration (N 1 ) is used in both top n-doping part N N1 and bottom p-doping part P P1 . Simultaneously, the same doping concentration (N 2 ) is used in both bottom n-doping part N N2 and top p-doping part P P2 . Doping concentration increment N s is defined as N 2 minus N 1 . In addition, doping concentration N 1 in the SDS-Trench CAVET is initially set equal to that of the n-and p-pillars in the SJ-Trench CAVET. For simplicity, typical N 1 concentrations, namely, from 0.5 × 10 16 cm −3 to 2 × 10 16 cm −3 , are utilized in simulations. Top part N N1 and bottom part P P1 keep the same width (W 2 ); thus, bottom part N N2 and top part P P2 are given the same width (W 1 ). It is worth noting that W 1 should not exceed the length of the CBL to avoid excessive depletion of 2DEG in the channel.
To obtain more accurate results in Silvaco TCAD, the concentration dependent (CON-SRH) and Auger physical models are used as recombination models, while Selberherr's Model (IMPACT SELB) is adopted for impact ionization. In addition, the high-fieldsaturation and -polarization models are also used in the simulation. According to Ref. [10], Li et al. proposed and simulated a classic GaN vertical superjunction HEMT. Thus, the model and material parameters could refer to a similar device. The parameter details of low-field-mobility models and Farahmand-Modified Caughey-Thomas (FMCT) models are listed in Table 2. When the breakdown performance is simulated, the avalanche model with a dependent electric field is adopted in the simulations. The impact-ionization rate is defined by [20]: In Equation (1), n and p refer to the concentrations of electrons and holes, respectively, while ν n and ν p refer to the saturation velocities of electrons and holes, respectively, where α n and α p are the values of the impact-ionization coefficient related to the electric field. Equations (2) and (3) define α n and α p as follows: In Equations (2) and (3) the temperature dependence of the phonon gas against which carriers are accelerated is expressed by parameters γ n and γ p . Coefficients a n , a p , b n and b p are fitting parameters, with modeling parameters a n = 2.81 × 10 8 cm −1 , b n = 3.43 × 10 7 V/cm, a p = 5.41 × 10 6 cm −1 and b p = 1.96 × 10 7 V/cm [21].
In order to examine the validity of the proposed model and parameters, Figure 2 compares the experimental results from Chowdhury [22] et al. and the simulation results when the normally on CAVET is in the on state at V GS = 0, −2 V. The reference is similar to the designed CAVET; thus, these models are considered available in simulations. As shown in Figure 2, the I DS -V DS curve shows a great fit between the experimental data and the simulation results. In addition, R on,sp also obviously indicates a good agreement between them. Thus, the models adopted are proved to be effective and correct.  Figure 3 juxtaposes the transfer characteristics of the two devices with their output characteristics. In Figure 3a, the same threshold voltage (Vth) of 3.1 V is manifested in the SDS-Trench CAVET and SJ-Trench CAVET. However, compared with the SJ-Trench CA-VET, the drain current of the SDS-Trench CAVET prevails when the gate voltage goes beyond Vth, which indicates a smaller resistance in the proposed structure. Furthermore, in Figure 3b, the noticeable enhancement of the on-state I-V curve demonstrates that the Ron,sp value of 2.08 mΩ·cm 2 in the SDS-Trench CAVET is smaller than that of 2.37 mΩ·cm 2 in the SJ-Trench CAVET. To explain the upgraded performance of the SDS-Trench CAVET, the on-state current-density distributions are analyzed in Figure 4. For both devices, only the n-doping area in the drift is conductive, since n+ substrate/p-GaN is under reversed bias as well as lateral n/p connection [10,23]. As shown in Figure 3b,c, the horizontal current paths are exhibited in the SDS-Trench CAVET and SJ-Trench CAVET. Obviously, in the SDS-Trench CAVET, NN1 owns a narrower current path, but NN2 holds a wider current path, owing to the varied width of the stepped doping part. However, the current density in NN1 is  Figure 3 juxtaposes the transfer characteristics of the two devices with their output characteristics. In Figure 3a, the same threshold voltage (V th ) of 3.1 V is manifested in the SDS-Trench CAVET and SJ-Trench CAVET. However, compared with the SJ-Trench CAVET, the drain current of the SDS-Trench CAVET prevails when the gate voltage goes beyond V th , which indicates a smaller resistance in the proposed structure. Furthermore, in Figure 3b, the noticeable enhancement of the on-state I-V curve demonstrates that the R on,sp value of 2.08 mΩ·cm 2 in the SDS-Trench CAVET is smaller than that of 2.37 mΩ·cm 2 in the SJ-Trench CAVET.  Figure 3 juxtaposes the transfer characteristics of the two devices with their output characteristics. In Figure 3a, the same threshold voltage (Vth) of 3.1 V is manifested in the SDS-Trench CAVET and SJ-Trench CAVET. However, compared with the SJ-Trench CA-VET, the drain current of the SDS-Trench CAVET prevails when the gate voltage goes beyond Vth, which indicates a smaller resistance in the proposed structure. Furthermore, in Figure 3b, the noticeable enhancement of the on-state I-V curve demonstrates that the Ron,sp value of 2.08 mΩ·cm 2 in the SDS-Trench CAVET is smaller than that of 2.37 mΩ·cm 2 in the SJ-Trench CAVET. To explain the upgraded performance of the SDS-Trench CAVET, the on-state current-density distributions are analyzed in Figure 4. For both devices, only the n-doping area in the drift is conductive, since n+ substrate/p-GaN is under reversed bias as well as lateral n/p connection [10,23]. As shown in Figure 3b,c, the horizontal current paths are exhibited in the SDS-Trench CAVET and SJ-Trench CAVET. Obviously, in the SDS-Trench CAVET, NN1 owns a narrower current path, but NN2 holds a wider current path, owing to the varied width of the stepped doping part. However, the current density in NN1 is To explain the upgraded performance of the SDS-Trench CAVET, the on-state currentdensity distributions are analyzed in Figure 4. For both devices, only the n-doping area in the drift is conductive, since n+ substrate/p-GaN is under reversed bias as well as lateral n/p connection [10,23]. As shown in Figure 3b,c, the horizontal current paths are exhibited in the SDS-Trench CAVET and SJ-Trench CAVET. Obviously, in the SDS-Trench CAVET, N N1 owns a narrower current path, but N N2 holds a wider current path, owing to the varied width of the stepped doping part. However, the current density in N N1 is greater than that in the corresponding position in the SJ-Trench CAVET, which is attributed to the extended P P1 with a higher doping concentration, making the electric charge accumulate in N N1 . From Figure 4a, a greater current density along line L1L2 can be seen in the n-doping area of the SDS-Trench CAVET. It can be noticed that the current-density curve in N N2 is slightly lower than that in the corresponding position in the SJ-Trench CAVET. As for the reason, the higher doping concentration in N N2 may bring higher current density, but the electrons disperse in the wider N GaN drift. Consequently, with a spacious current direction and a high current density, the proposed device represents the potential to modulate current density and promote current conduction.

Results and Discussion
greater than that in the corresponding position in the SJ-Trench CAVET, which is attributed to the extended PP1 with a higher doping concentration, making the electric charge accumulate in NN1. From Figure 4a, a greater current density along line L1L2 can be seen in the n-doping area of the SDS-Trench CAVET. It can be noticed that the currentdensity curve in NN2 is slightly lower than that in the corresponding position in the SJ-Trench CAVET. As for the reason, the higher doping concentration in NN2 may bring higher current density, but the electrons disperse in the wider N GaN drift. Consequently, with a spacious current direction and a high current density, the proposed device represents the potential to modulate current density and promote current conduction.  From the corresponding off-state breakdown characteristics in Figure 5a, the BVs of the SJ-Trench CAVET and SDS-Trench CAVET extracted at ID = 1 × 10 −6 mA/mm are 2104 V and 2786 V, respectively. To investigate the distinction between the two devices and mechanisms accounting for it, the electric field and the corresponding contour plots are illustrated in Figures 5b and 6, respectively. Figure 5b represents the comparison of the electric-field strengths along line L3L4 in the proposed structures. As can be seen, owing to the incomplete depleted 2DEG in the trench, there are two small peaks of electric field in the corresponding position. However, the larger p-GaN region below the CBL in the SDS-Trench CAVET consumes more electric field in the trench corner, resulting in a lower peak at the CBL/GaN drift interface. Additionally, since NN2 is given a higher doping concentration and a larger region, the electric-field strength is enhanced and reaches its max- From the corresponding off-state breakdown characteristics in Figure 5a, the BVs of the SJ-Trench CAVET and SDS-Trench CAVET extracted at I D = 1 × 10 −6 mA/mm are 2104 V and 2786 V, respectively. To investigate the distinction between the two devices and mechanisms accounting for it, the electric field and the corresponding contour plots are illustrated in Figures 5b and 6, respectively. Figure 5b represents the comparison of the electric-field strengths along line L3L4 in the proposed structures. As can be seen, owing to the incomplete depleted 2DEG in the trench, there are two small peaks of electric field in the corresponding position. However, the larger p-GaN region below the CBL in the SDS-Trench CAVET consumes more electric field in the trench corner, resulting in a lower peak at the CBL/GaN drift interface. Additionally, since N N2 is given a higher doping concentration and a larger region, the electric-field strength is enhanced and reaches its maximum at the dividing line. In addition, the electric-field strength approaching the substrate is smaller in the SDS-Trench CAVET, which could be explained by the fact that the electric field is distributed and weakened in the larger N N2 . The greater value of the electric field in the SDS-Trench CAVET results in the tolerance to higher forward voltage [24]. Figure 6 plots more details about the two peaks and the whole distribution in the GaN drift. In the SJ-CAVET, there are only two strong electric fields, which are located in the trench corner and substrate. However, the drift layer in the SDS-CAVET is much brighter than the former, especially at the stepped doping interface. Such a great electric-field strength guarantees the high current levels and high voltage tolerance of the device. strate is smaller in the SDS-Trench CAVET, which could be explained by the fact tha electric field is distributed and weakened in the larger NN2. The greater value of the ele field in the SDS-Trench CAVET results in the tolerance to higher forward voltage Figure 6 plots more details about the two peaks and the whole distribution in the drift. In the SJ-CAVET, there are only two strong electric fields, which are located in trench corner and substrate. However, the drift layer in the SDS-CAVET is much brig than the former, especially at the stepped doping interface. Such a great electricstrength guarantees the high current levels and high voltage tolerance of the device.  Figure 7, the off-state potential equipotential lines show similar slopes in th Trench CAVET and SDS-Trench CAVET, which guarantee device stability in breakd performance. Compared with the SJ-Trench CAVET, in the SDS-Trench CAVET, the potential areas below the CBL move down slightly, and the high-potential areas abov substrate move upward a lot, causing narrower medium-potential areas. The variatio equipotential-line densities verifies the electric-field-distribution result in Figur whereby the electric-field strength near the middle is enhanced, while that near the tom is degraded in n-and p-pillars. The total enlarged high-potential distribution in bottom area is beneficial for breakdown performance, especially during reverse block electric field is distributed and weakened in the larger NN2. The greater value of the elect field in the SDS-Trench CAVET results in the tolerance to higher forward voltage [2 Figure 6 plots more details about the two peaks and the whole distribution in the Ga drift. In the SJ-CAVET, there are only two strong electric fields, which are located in t trench corner and substrate. However, the drift layer in the SDS-CAVET is much brigh than the former, especially at the stepped doping interface. Such a great electric-fie strength guarantees the high current levels and high voltage tolerance of the device.  Figure 7, the off-state potential equipotential lines show similar slopes in the Trench CAVET and SDS-Trench CAVET, which guarantee device stability in breakdow performance. Compared with the SJ-Trench CAVET, in the SDS-Trench CAVET, the lo potential areas below the CBL move down slightly, and the high-potential areas above t substrate move upward a lot, causing narrower medium-potential areas. The variation equipotential-line densities verifies the electric-field-distribution result in Figure  whereby the electric-field strength near the middle is enhanced, while that near the b tom is degraded in n-and p-pillars. The total enlarged high-potential distribution in t bottom area is beneficial for breakdown performance, especially during reverse blockin  Figure 7, the off-state potential equipotential lines show similar slopes in the SJ-Trench CAVET and SDS-Trench CAVET, which guarantee device stability in breakdown performance. Compared with the SJ-Trench CAVET, in the SDS-Trench CAVET, the lowpotential areas below the CBL move down slightly, and the high-potential areas above the substrate move upward a lot, causing narrower medium-potential areas. The variation in equipotential-line densities verifies the electric-field-distribution result in Figure 5, whereby the electric-field strength near the middle is enhanced, while that near the bottom is degraded in n-and p-pillars. The total enlarged high-potential distribution in the bottom area is beneficial for breakdown performance, especially during reverse blocking.  Figures 8-10 show how the four key parameters, N 1 , N s , H 1 and W 1 , influence the breakdown characteristics and R on,sp . Since the stepped doping region is composed of top part and bottom parts, step lengths W 1 and W 2 deserve investigation to explore their modulation effect and are consequently selected as the main variations in the optimization. Figure 8 shows the plots of the breakdown voltage and specific on-resistance versus length W 1 for different values of N 1 for given N s and H 1 in the SDS-Trench CAVET. As shown, R on,sp is at its minimum when W 1 is 2.0 µm or 2.5 µm, while the maximum BV appears when W 1 is 3 µm. Although the largest BV is obtained with N 1 = 0.5×10 16 cm −3 , its R on,sp exceeds the original R on,sp in the SJ-Trench CAVET. Thus, the best BV result, 2786 V, is obtained with N 1 = 1 × 10 16 cm −3 with a slightly reduced R on,sp of 2.08 mΩ·cm 2 , or a lower R on,sp of 1.34 mΩ·cm 2 and a relatively high BV of 2500 V could be acquired when N 1 = 2 × 10 16 cm −3 . The above two trade-offs are both represented with W 1 = 3 µm, which means that such a stepped doping method is effective and significant in enhancing the performance of the devices.  Figures 8-10 show how the four key parameters, N1, Ns, H1 and W1, influence the breakdown characteristics and Ron,sp. Since the stepped doping region is composed of top part and bottom parts, step lengths W1 and W2 deserve investigation to explore their modulation effect and are consequently selected as the main variations in the optimization. Figure 8 shows the plots of the breakdown voltage and specific on-resistance versus length W1 for different values of N1 for given Ns and H1 in the SDS-Trench CAVET. As shown, Ron,sp is at its minimum when W1 is 2.0 μm or 2.5 μm, while the maximum BV appears when W1 is 3 μm. Although the largest BV is obtained with N1 = 0.5×10 16 cm −3 , its Ron,sp exceeds the original Ron,sp in the SJ-Trench CAVET. Thus, the best BV result, 2786 V, is obtained with N1 = 1 × 10 16 cm −3 with a slightly reduced Ron,sp of 2.08 mΩ·cm 2 , or a lower Ron,sp of 1.34 mΩ·cm 2 and a relatively high BV of 2500 V could be acquired when N1 = 2 × 10 16 cm −3 . The above two trade-offs are both represented with W1 = 3 μm, which means that such a stepped doping method is effective and significant in enhancing the performance of the devices.  In Figure 9, for given N1 and H1, the different values of Ns are investigat examine the above results. The peak of the BV emerges when W1 is 3 μm or l the BV increases as Ns declines, the rapidly enlarged Ron,sp cannot be ignore the optimized result is achieved with Ns = 2 × 10 16 cm −3 , since it represents a hi lower Ron,sp than other Ns, especially when W1 is smaller than 3 μm. Based on the above results, Figure 10 explores how the thickness of the s ing structure affects the performance of the SDS-Trench CAVET with given cm −3 and N1 = 1 × 10 16 cm −3 . It seems that Ron,sp follows a regular pattern, whe expanded, Ron,sp first decreases and then increases. In addition, when H1 is en rapidly declines, which could be ascribed to the overconsumption of 2DEG b ened p-doping region. However, in the embedded figure of BV variation, the the same as that of Ron,sp. As is shown, when W1 is extended, the BV first in then drops when H1 is larger than 5 μm, while the BV seldom changes or ev when H1 is smaller than 5 μm. Pursuing the highest BV, 10 μm H1 provid breakdown voltage with a non-deteriorated Ron,sp of 2.08 mΩ·cm 2 , which is ob W1 = 2.5 μm. Such a high breakdown voltage could be attributed to the thicken region beneath the CBL, greatly affecting the reverse-blocking process [25].

Fabrication Process
Recent years witnessed more and more mature GaN-growth technolo technologies for thick in situ doping-GaN-layer growth, GaN-based selective technologies (SAG) and GaN etching technologies. The height of the Ga grown via SAG and the depth of the plasma-etched GaN trench were reporte In Figure 9, for given N 1 and H 1 , the different values of N s are investigated to further examine the above results. The peak of the BV emerges when W 1 is 3 µm or longer. While the BV increases as N s declines, the rapidly enlarged R on,sp cannot be ignored. Therefore, the optimized result is achieved with N s = 2 × 10 16 cm −3 , since it represents a higher BV and lower R on,sp than other N s , especially when W 1 is smaller than 3 µm.
Based on the above results, Figure 10 explores how the thickness of the stepped doping structure affects the performance of the SDS-Trench CAVET with given N s = 2 × 10 16 cm −3 and N 1 = 1 × 10 16 cm −3 . It seems that R on,sp follows a regular pattern, whereby as W 1 is expanded, R on,sp first decreases and then increases. In addition, when H 1 is enlarged, R on,sp rapidly declines, which could be ascribed to the overconsumption of 2DEG by the broadened p-doping region. However, in the embedded figure of BV variation, the trend is not the same as that of R on,sp . As is shown, when W 1 is extended, the BV first increases and then drops when H 1 is larger than 5 µm, while the BV seldom changes or even descends when H1 is smaller than 5 µm. Pursuing the highest BV, 10 µm H 1 provides a 3024 V breakdown voltage with a non-deteriorated R on,sp of 2.08 mΩ·cm 2 , which is obtained when W 1 = 2.5 µm. Such a high breakdown voltage could be attributed to the thickened p-doping region beneath the CBL, greatly affecting the reverse-blocking process [25].

Fabrication Process
Recent years witnessed more and more mature GaN-growth technologies, such as technologies for thick in situ doping-GaN-layer growth, GaN-based selective-area-growth technologies (SAG) and GaN etching technologies. The height of the GaN drift layer grown via SAG and the depth of the plasma-etched GaN trench were reported to be more than 10 µm [26]. Therefore, the fabrication of the proposed GaN-based trench CAVET with a stepped doping microstructure is achievable, since the n-doping region and p-doping region could alternately be grown using the above technologies.
The diagrams in Figure 11 and the following procedures demonstrate an available process to realize the proposed SDS-Trench CAVET. Firstly, a p-GaN layer with doping concentration N1 and height H minus H 1 is epitaxially grown via metal-organic chemical vapor deposition (MOCVD) on a bulk conductive GaN substrate. Mg ions could be implanted into the bottom p-GaN layer at 80 keV [22]. Utilizing the Cl-based inductively coupled plasma (ICP) etching process, trench 1 with width 2W 1 and height H-H 1 is formed. Then, the rest of the p-GaN layer is masked and an n-GaN layer with doping concentration N 2 and height H-H 1 is regrown via SAG in trench 1. Secondly, a p-GaN layer with doping concentration N 2 and height H 1 is epitaxially deposited via MOCVD. The same Mg ionimplantation method is applied to it. Simultaneously, trench 2 is etched via ICP, and the n-GaN layer with doping concentration N 1 and height H 1 is regrown via SAG to fill trench 2. Thirdly, a heavily Mg-implanted doped p-GaN layer is epitaxially grown via MOCVD to form a CBL layer. The current aperture is protected using an implantation mask consisting of Ti/Ni on SiO 2 deposited via PECVD. The implantation mask is later removed with the wet-etching technique using hydrofluoric acid. After wet etching and MOCVD growth on the CBL layer, the final main procedures involve the deposition of the UID-GaN channel and MIS trench-gate structure. After p+GaN activation and HF surface treatment, unintentionally doped (UID) GaN is regrown on the CBL via MOCVD. Using Cl 2 /BCl 3 gases in reactive-ion etching (RIE) to etch the trench, UV-ozone and HF cleaning treatments are performed after trench formation [12]. Subsequently, the AlGaN layers are grown via MOCVD at a high temperature. The Si 3 N 4 dielectric is deposited on the AlGaN barrier layer via PECVD. Then, the trench gate and source are formed via ICP etching, followed by surface treatment. Next, the buried p-GaN and CBL are processed via thermal annealing with the diffusion of hydrogen and Mg activation. Later, the dielectric gate is patterned via ALD. Finally, the Ni/Au stack is taken as the gate, while the Ti/Al stack is used to form the source and backside drains, after which the entire proposed device is obtained.
Ti/Al stack is used to form the source and backside drains, after which the entire proposed device is obtained.

Conclusions
A GaN-based trench CAVET with a stepped doping microstructure is demonstrated and evaluated using two-dimensional simulations to promote the BV and Ron,sp. By modulating the step width, thickness, doping concentration and variation in the GaN drift layer, the structure presents an extremely high BV of 3024 V with Ron,sp of 2.08 mΩ·cm 2 . Defining Baliga's figure of merit (FOM) as BV 2 /Ron,sp, the optimized device achieves a high FOM of 4.396 GW·cm −2 , or the device can manifest an extremely low Ron,sp of 1.34 mΩ·cm 2 with a BV of 2523 V, yielding a remarkable FOM of 4.767 GW·cm −2 . The above two trade-

Conclusions
A GaN-based trench CAVET with a stepped doping microstructure is demonstrated and evaluated using two-dimensional simulations to promote the BV and R on,sp . By modulating the step width, thickness, doping concentration and variation in the GaN drift layer, the structure presents an extremely high BV of 3024 V with R on,sp of 2.08 mΩ·cm 2 . Defining Baliga's figure of merit (FOM) as BV 2 /R on,sp , the optimized device achieves a high FOM of 4.396 GW·cm −2 , or the device can manifest an extremely low R on,sp of 1.34 mΩ·cm 2 with a BV of 2523 V, yielding a remarkable FOM of 4.767 GW·cm −2 . The above two tradeoffs exhibit the potential and superiority of the stepped doping microstructure in enhancing the performance of high-power GaN devices and applications.