Normally-off β-Ga2O3 MOSFET with an Epitaxial Drift Layer

A normally-off β-Ga2O3 metal-oxide-semiconductor field-effect transistor (MOSFET) is proposed using a technology computer-aided design (TCAD) device simulation, which employs an epitaxial drift layer grown on an n-type low-doped body layer. The low-doped body layer under the MOS gate enabled normally-off operation, whereas the epitaxial drift layer determined the on-resistance and breakdown characteristics. The effects of the doping concentration of each layer and thickness of the drift channel layer on the device characteristics were investigated to design a device with a breakdown voltage of 1 kV. A threshold voltage of 1.5 V and a breakdown voltage of 1 kV were achieved by an n-type body layer with a doping concentration of 1 × 1015 cm−3 and an n-type drift layer with a doping concentration of 3 × 1017 cm−3, a thickness of 150 nm, and a gate-to-drain distance of 9.5 μm; resulting in an on-resistance of 25 mΩ·cm2.

Although several studies have reported the normally-off operation of β-Ga 2 O 3 field-effect transistors (FET), the experimental results are still far off from the theoretical limits of the material [15][16][17][18].
Chabak et al., demonstrated enhancement-mode FETs using a wrap-gate fin structure in 2016 [15] and a gate recess process in 2018 [16]. In 2017, Wong et al., reported that the utilization of an unintentionally doped β-Ga 2 O 3 channel in MOSFET was able to completely deplete the channel electrons at a gate voltage (V GS ) of 0 V, resulting in a positive threshold voltage [19]. In 2019, Singh et al., proposed a T-shaped recessed gate β-Ga 2 O 3 MOSFET to achieve a normally-off operation [16]. The T-shaped recessed gate depleted the channel at a gate bias of 0 V, where the gate oxide (Al 2 O 3 ) thickness was 20 nm, gate recess depth was 250 nm, and thickness of the active channel under the recess region was 30 nm [17]. The maximum drain current was 40 mA/mm at V GS = +8 V due to the limited channel thickness required to achieve a positive threshold voltage [17].
In this study, we propose a recessed β-Ga 2 O 3 MOSFET with an epitaxial drift layer on top of a low-doped body layer to overcome the trade-off relationship between the threshold voltage and on-current density. The proposed structure does not require precise control of recess depth. Moreover, the threshold voltage could be independently controlled by the drift layer. The output and transfer characteristics of the proposed device were validated using Silvaco ATLAS technology computer-aided design (TCAD) simulation. After investigating the effects of doping concentration on the body layer and additional design parameters of the drift layer, a normally-off MOSFET structure was designed to achieve a breakdown voltage of 1 kV.

Simulation and Device Structure
Two-dimensional (2D) device simulations were performed in a Silvaco ATLAS TCAD environment using several physical models, including a drift-diffusion transport model, Fermi-Dirac statistics, concentration and temperature-dependent analytical mobility model, Shockley-Read-Hall recombination model, and an impact ionization model [20][21][22][23][24][25][26]. The material and physical model parameters used in the TCAD simulations are presented in Table 1. Although the simulation process could have been further optimized by employing more comprehensive models [27], the classical models provided by TCAD are sufficient to validate the proposed concept.

Mobility Model
The mobility model used in the simulation included concentration and temperaturedependent relationships based on an analytical function of Caughey-Thomas' work [25], which is given by: where α, β, γ, and δ are material-dependent coefficients [20], N D is the impurity concentration, and T L is the temperature in Kelvins. Using the experimental data [28,29], these parameters were determined to be α = 0, β = 0, γ = 0, δ = 0.8, N ref = 1.0 × 10 18 cm −3 , and T L = 300 K.

Impact Ionization Model
Selberherr's model, which is a modification of Chynoweth's law, has been widely used to predict the breakdown characteristics of wide-bandgap semiconductors [20,21]. The impact ionization coefficient (α n ) is given by where A N and B N are the material coefficients and E is the electric field. In this study, A N = 2.16 × 10 6 cm −1 and B N = 1.77 × 10 7 V/cm were used while considering the crystal direction of β-Ga 2 O 3 in the [010] direction and a critical electric field of approximately 5 MV/cm [11,20,21].

Shockley-Read-Hall Recombination
In our simulations, the recombination rate was obtained using the Shockley-Read-Hall recombination model [26]: where n, p, and n ie are the electron, hole, and intrinsic carrier concentrations, respectively, and k and T L are the Boltzmann constant and lattice temperature, respectively. E trap is the difference between the trap energy level and the intrinsic Fermi level, and τ n0 and τ p0 are the electron and hole lifetimes, respectively, which are used as 1.2 × 10 −8 s. Figure 1 shows a cross-sectional schematic of the β-Ga 2 O 3 MOSFET proposed in this study. The epitaxial structure consisted of a 20 nm thick ohmic contact layer with an n-type doping concentration of 1 × 10 20 cm −3 , an n-type drift layer, a 300 nm thick low-doped n-type body layer, and a 1 µm thick buffer layer with an n-type doping concentration of 1 × 10 12 cm −3 . In this study, a highly doped ohmic contact layer was employed instead of an ion-implantation process. The structural variables investigated in this study were the thickness (t DRIFT ) and doping concentration (N D.DRIFT ) of the drift layer and the doping concentration (N D.BODY ) of the body layer. A highly doped ohmic contact layer is etched between the source and drain contacts. The gate region was etched down to the body layer to achieve normally-off characteristics. A 20 nm-thick gate oxide (Al 2 O 3 ) layer was used, and its interface charges were considered during the simulation.

Device Structure
Micromachines 2022, 13, x FOR PEER REVIEW 3 of 11

Impact Ionization Model
Selberherr's model, which is a modification of Chynoweth's law, has been widely used to predict the breakdown characteristics of wide-bandgap semiconductors [20,21]. The impact ionization coefficient (αn) is given by α = A exp − (2) where AN and BN are the material coefficients and E is the electric field. In this study, AN = 2.16 × 10 6 cm −1 and BN = 1.77 × 10 7 V/cm were used while considering the crystal direction of β-Ga2O3 in the [010] direction and a critical electric field of approximately 5 MV/cm [11,20,21].

Shockley-Read-Hall Recombination
In our simulations, the recombination rate was obtained using the Shockley-Read-Hall recombination model [26]: where n, p, and nie are the electron, hole, and intrinsic carrier concentrations, respectively, and k and TL are the Boltzmann constant and lattice temperature, respectively. Etrap is the difference between the trap energy level and the intrinsic Fermi level, and τn0 and τp0 are the electron and hole lifetimes, respectively, which are used as 1.2 × 10 −8 s. Figure 1 shows a cross-sectional schematic of the β-Ga2O3 MOSFET proposed in this study. The epitaxial structure consisted of a 20 nm thick ohmic contact layer with an ntype doping concentration of 1 × 10 20 cm −3 , an n-type drift layer, a 300 nm thick low-doped n-type body layer, and a 1 μm thick buffer layer with an n-type doping concentration of 1 × 10 12 cm −3 . In this study, a highly doped ohmic contact layer was employed instead of an ion-implantation process. The structural variables investigated in this study were the thickness (tDRIFT) and doping concentration (ND.DRIFT) of the drift layer and the doping concentration (ND.BODY) of the body layer. A highly doped ohmic contact layer is etched between the source and drain contacts. The gate region was etched down to the body layer to achieve normally-off characteristics. A 20 nm-thick gate oxide (Al2O3) layer was used, and its interface charges were considered during the simulation.   the electron density distributions at V GS = 0 V and +3 V, respectively, which were simulated using the variables N D.DRIFT = 3 × 10 17 cm −3 , t DRIFT = 300 nm, and N D.BODY = 1 × 10 15 cm −3 . For V GS = 0 V, the electrons in the region under the gate were completely depleted, which blocked the flow of current, confirming the normally-off characteristics. For V GS = +3 V, the depletion region under the gate disappeared, creating an electron accumulation channel layer and allowing for current flow.

Device Structure
Micromachines 2022, 13, x FOR PEER REVIEW 4 of 11 Figure 2a,b show the electron density distributions at VGS = 0 V and +3 V, respectively, which were simulated using the variables ND.DRIFT = 3 × 10 17 cm −3 , tDRIFT = 300 nm, and ND.BODY = 1 × 10 15 cm −3 . For VGS = 0 V, the electrons in the region under the gate were completely depleted, which blocked the flow of current, confirming the normally-off characteristics. For VGS = +3 V, the depletion region under the gate disappeared, creating an electron accumulation channel layer and allowing for current flow.

Effects of Al2O3/β-Ga2O3 Interface Charge
Previous studies have reported the presence of negative interface charges at the Al2O3/β-Ga2O3 interface with a density in the range of 1 × 10 12 to 4 × 10 12 cm −2 [18,21,23,30,31]. In this section, the effects of charge density at the Al2O3/β-Ga2O3 interface are investigated, where the negative interface charge density varied from 0 to 2 × 10 12 cm −2 . The transfer characteristics simulated at a drain voltage (VDS) of 5 V as a function of the interface charge density are shown in Figure 3. A positive shift in the threshold voltage was observed with a reduction in drain current density as the negative interface charge density increased. Therefore, based on these prior experimental reports [18,20], a negative interface charge density of 1 × 10 12 cm −2 was selected for the simulations.

Effects of Al 2 O 3 /β-Ga 2 O 3 Interface Charge
Previous studies have reported the presence of negative interface charges at the Al 2 O 3 /β-Ga 2 O 3 interface with a density in the range of 1 × 10 12 to 4 × 10 12 cm −2 [18,21,23,30,31]. In this section, the effects of charge density at the Al 2 O 3 /β-Ga 2 O 3 interface are investigated, where the negative interface charge density varied from 0 to 2 × 10 12 cm −2 . The transfer characteristics simulated at a drain voltage (V DS ) of 5 V as a function of the interface charge density are shown in Figure 3. A positive shift in the threshold voltage was observed with a reduction in drain current density as the negative interface charge density increased. Therefore, based on these prior experimental reports [18,20], a negative interface charge density of 1 × 10 12 cm −2 was selected for the simulations. Figure 2a,b show the electron density distributions at VGS = 0 V and +3 V, respectively which were simulated using the variables ND.DRIFT = 3 × 10 17 cm −3 , tDRIFT = 300 nm, and ND.BODY = 1 × 10 15 cm −3 . For VGS = 0 V, the electrons in the region under the gate were com pletely depleted, which blocked the flow of current, confirming the normally-off charac teristics. For VGS = +3 V, the depletion region under the gate disappeared, creating an elec tron accumulation channel layer and allowing for current flow.

Effects of Al2O3/β-Ga2O3 Interface Charge
Previous studies have reported the presence of negative interface charges at the Al2O3/β-Ga2O3 interface with a density in the range of 1 × 10 12 to 4 × 10 12 cm − [18,21,23,30,31]. In this section, the effects of charge density at the Al2O3/β-Ga2O3 interface are investigated, where the negative interface charge density varied from 0 to 2 × 10 12 cm −2 The transfer characteristics simulated at a drain voltage (VDS) of 5 V as a function of the interface charge density are shown in Figure 3. A positive shift in the threshold voltage was observed with a reduction in drain current density as the negative interface charge density increased. Therefore, based on these prior experimental reports [18,20], a negative interface charge density of 1 × 10 12 cm −2 was selected for the simulations.

Effects of Doping Concentrations in Body and Drift Layer
Initially, the effects of the doping concentration of the body layer (N D.BODY ) on the threshold voltage were investigated. N D.BODY varied from 1 × 10 13 cm −3 to 1 × 10 17 cm −3 , while the drift layer had a thickness of t DRIFT = 300 nm and a doping concentration of N D.DRIFT = 3 × 10 17 cm −3 . Figure 4a,b show the linear and logarithmic transfer characteristics at V DS = 5 V as a function of N D.BODY , respectively. A significant negative shift in the threshold voltage was observed when N D.BODY was equal to or greater than 1 × 10 16 cm −3 , resulting in normally-on characteristics, whereas only a negligible difference was observed when N D.BODY was equal to or less than 1 × 10 15 cm −3 . Therefore, to design a normally-off device, N D.BODY = 1 × 10 15 cm −3 was selected for the simulations.
R PEER REVIEW 5 of 11

Effects of Doping Concentrations in Body and Drift Layer
Initially, the effects of the doping concentration of the body layer (ND.BODY) on the threshold voltage were investigated. ND.BODY varied from 1 × 10 13 cm −3 to 1 × 10 17 cm −3 , while the drift layer had a thickness of tDRIFT = 300 nm and a doping concentration of ND.DRIFT = 3 × 10 17 cm −3 . Figure 4a,b show the linear and logarithmic transfer characteristics at VDS = 5 V as a function of ND.BODY, respectively. A significant negative shift in the threshold voltage was observed when ND.BODY was equal to or greater than 1 × 10 16 cm −3 , resulting in normally-on characteristics, whereas only a negligible difference was observed when ND.BODY was equal to or less than 1 × 10 15 cm −3 . Therefore, to design a normally-off device, ND.BODY = 1 × 10 15 cm −3 was selected for the simulations. Additionally, the effects of the doping concentration of the drift layer (ND.DRIFT) on the drain current density were investigated, where ND.DRIFT varied from 1 × 10 17 cm −3 to 9 × 10 17 cm −3 with a fixed body doping concentration of ND.BODY = 1 × 10 15 cm −3 . The drift layer thickness was tDRIFT 300 nm. The transfer characteristics as a function of ND.DRIFT are shown in Figure 5a,b. It is evident that the drain current density increases with an increase in ND.DRIFT, whereas the threshold voltage remains the same because it is determined by the recessed MOS region on the body layer. The normally-off characteristics were maintained even at ND.DRIFT = 9 × 10 17 cm −3 . The threshold voltage was 0.8 V at 1 μA/mm and 1.5 V at 1 mA/mm. Figure 5c shows the conduction band energy diagrams as a function of ND.DRIFT along the vertical direction below the gate metal, and it is obvious that increasing ND.DRIFT does not change the conduction band energy such that the threshold voltage remains the same regardless of ND.DRIFT. On the other hand, Figure 5d shows the conduction band energy diagrams as a function of ND.DRIFT along the vertical direction in the region between the gate and drain. It can be seen that the depletion width in the β-Ga2O3 drift layer is reduced when increasing the ND.DRIFT, leading to a higher drain current. Additionally, the effects of the doping concentration of the drift layer (N D.DRIFT ) on the drain current density were investigated, where N D.DRIFT varied from 1 × 10 17 cm −3 to 9 × 10 17 cm −3 with a fixed body doping concentration of N D.BODY = 1 × 10 15 cm −3 . The drift layer thickness was t DRIFT 300 nm. The transfer characteristics as a function of N D.DRIFT are shown in Figure 5a,b. It is evident that the drain current density increases with an increase in N D.DRIFT , whereas the threshold voltage remains the same because it is determined by the recessed MOS region on the body layer. The normally-off characteristics were maintained even at N D.DRIFT = 9 × 10 17 cm −3 . The threshold voltage was 0.8 V at 1 µA/mm and 1.5 V at 1 mA/mm. Figure 5c shows the conduction band energy diagrams as a function of N D.DRIFT along the vertical direction below the gate metal, and it is obvious that increasing N D.DRIFT does not change the conduction band energy such that the threshold voltage remains the same regardless of N D.DRIFT . On the other hand, Figure 5d shows the conduction band energy diagrams as a function of N D.DRIFT along the vertical direction in the region between the gate and drain. It can be seen that the depletion width in the β-Ga 2 O 3 drift layer is reduced when increasing the N D.DRIFT , leading to a higher drain current.
3, x FOR PEER REVIEW 6 of 11

Transfer and Output Characteristics
To investigate the effects of the thickness of the drift layer (tDRIFT), the doping concentrations of the body and drift channel layers were fixed as ND.BODY = 1 × 10 15 cm −3 and ND.DRIFT = 3 × 10 17 cm −3 , respectively. The tDRIFT was varied to 75, 150, and 300 nm. As shown in Figure 6, the drain on-current density decreased with a decrease in tDRIFT, while the threshold voltage remained constant as the series resistance of the drift layer increased with a decrease in the thickness. The output current-voltage characteristics are compared in Figure 7. The maximum drain current density (ID.MAX) and on-resistance (Ron) for the thicknesses of tDRIFT = 300, 150, and 75 nm were ID.MAX = 190, 136, and 80 mA/mm, respectively, and Ron = 12.7, 25, and 61.7 mΩ cm 2 , respectively.

Transfer and Output Characteristics
To investigate the effects of the thickness of the drift layer (t DRIFT ), the doping concentrations of the body and drift channel layers were fixed as N D.BODY = 1 × 10 15 cm −3 and N D.DRIFT = 3 × 10 17 cm −3 , respectively. The t DRIFT was varied to 75, 150, and 300 nm. As shown in Figure 6, the drain on-current density decreased with a decrease in t DRIFT , while the threshold voltage remained constant as the series resistance of the drift layer increased with a decrease in the thickness. The output current-voltage characteristics are compared in Figure 7. The maximum drain current density (I D.MAX ) and on-resistance (R on ) for the thicknesses of t DRIFT = 300, 150, and 75 nm were I D.MAX = 190, 136, and 80 mA/mm, respectively, and R on = 12.7, 25, and 61.7 mΩ cm 2 , respectively. Micromachines 2022, 13, x FOR PEER REVIEW 7 of 11

Breakdown Characteristics
Breakdown characteristics with different drift layer thicknesses were simulated at VGS = 0 V, and the results are compared in Figure 8. The catastrophic breakdown voltages were 680, 1012, and 1380 V for the thickness values of tDRIFT = 300, 150, and 75 nm, respectively. With the same doping concentration of the drift layer, the breakdown voltage exhibited a significant dependence on tDRIFT. The electron density and electric field distributions for different tDRIFT values were examined to investigate the reasons for this. Figures  9 and 10 show the electron density and electric field distributions simulated at VDS = 600 V for different tDRIFT values, and the electron concentration and electric field distributions along the cutline from a to a' are plotted in Figures 9d and 10d, respectively. As shown in Figures 9d and 10d, the depletion region extended towards the drain side with decreasing thickness, resulting in a lower peak electric field near the gate. This is because the total number of electrons depleted by a given gate voltage is the same for all cases. Therefore, the thinner drift layer had a longer depletion edge. Consequently, a higher breakdown voltage can be achieved with a thinner drift layer. The tradeoff relationship between Ron and the breakdown voltage as a function of the drift layer thickness is shown in Figure 11.

Breakdown Characteristics
Breakdown characteristics with different drift layer thicknesses were simulated VGS = 0 V, and the results are compared in Figure 8. The catastrophic breakdown voltag were 680, 1012, and 1380 V for the thickness values of tDRIFT = 300, 150, and 75 nm, resp tively. With the same doping concentration of the drift layer, the breakdown voltage hibited a significant dependence on tDRIFT. The electron density and electric field distrib tions for different tDRIFT values were examined to investigate the reasons for this. Figu 9 and 10 show the electron density and electric field distributions simulated at VDS = 6 V for different tDRIFT values, and the electron concentration and electric field distributio along the cutline from a to a' are plotted in Figures 9d and 10d, respectively. As shown Figures 9d and 10d, the depletion region extended towards the drain side with decreasi thickness, resulting in a lower peak electric field near the gate. This is because the to number of electrons depleted by a given gate voltage is the same for all cases. Therefo the thinner drift layer had a longer depletion edge. Consequently, a higher breakdow voltage can be achieved with a thinner drift layer. The tradeoff relationship between and the breakdown voltage as a function of the drift layer thickness is shown in Figure   Figure 7. Output current-voltage characteristics for a drift channel layer thickness of (a) 300 nm, (b) 150 nm, and (c) 75 nm. The drift channel layer has a doping concentration of 3 × 10 17 cm −3 .

Breakdown Characteristics
Breakdown characteristics with different drift layer thicknesses were simulated at V GS = 0 V, and the results are compared in Figure 8. The catastrophic breakdown voltages were 680, 1012, and 1380 V for the thickness values of t DRIFT = 300, 150, and 75 nm, respectively. With the same doping concentration of the drift layer, the breakdown voltage exhibited a significant dependence on t DRIFT . The electron density and electric field distributions for different t DRIFT values were examined to investigate the reasons for this. Figures 9 and 10 show the electron density and electric field distributions simulated at V DS = 600 V for different t DRIFT values, and the electron concentration and electric field distributions along the cutline from a to a' are plotted in Figures 9d and 10d, respectively. As shown in Figures 9d and 10d, the depletion region extended towards the drain side with decreasing thickness, resulting in a lower peak electric field near the gate. This is because the total number of electrons depleted by a given gate voltage is the same for all cases. Therefore, the thinner drift layer had a longer depletion edge. Consequently, a higher breakdown voltage can be achieved with a thinner drift layer. The tradeoff relationship between R on and the breakdown voltage as a function of the drift layer thickness is shown in Figure 11.      In summary, using a body layer with a doping concentration of 1 × 10 15 cm −3 and a drift layer with a doping concentration of 3 × 10 17 cm −3 , a thickness of 150 nm, and a gateto-drain distance of 9.5 μm resulted in a threshold voltage of 0.8 V at 1 μA/mm, a breakdown voltage of ~1 kV, and an on-resistance of 25 mΩ·cm 2 .  In summary, using a body layer with a doping concentration of 1 × 10 15 cm −3 and a drift layer with a doping concentration of 3 × 10 17 cm −3 , a thickness of 150 nm, and a gateto-drain distance of 9.5 μm resulted in a threshold voltage of 0.8 V at 1 μA/mm, a breakdown voltage of ~1 kV, and an on-resistance of 25 mΩ·cm 2 . In summary, using a body layer with a doping concentration of 1 × 10 15 cm −3 and a drift layer with a doping concentration of 3 × 10 17 cm −3 , a thickness of 150 nm, and a gate-to-drain distance of 9.5 µm resulted in a threshold voltage of 0.8 V at 1 µA/mm, a breakdown voltage of~1 kV, and an on-resistance of 25 mΩ·cm 2 .

Conclusions
A normally-off β-Ga 2 O 3 MOSFET structure was proposed, which employed an epitaxial drift layer in conjunction with a recessed MOS gate. A positive threshold voltage was achieved by employing a low-doped n-type body layer, which led to the formation of an electron-accumulation channel layer. An additional drift layer grown on top of the body layer is crucial for determining the on-resistance and breakdown voltage characteristics. The proposed dual epitaxial structure enables normally-off operation without employing an ion implantation process. Considering the difficulty of p-type ion implantation or epitaxial growth with β-Ga 2 O 3 , the proposed structure is a promising candidate for the implementation of a normally-off β-Ga 2 O 3 FET.