Localized Electric Field Tailoring to Balance Voltage Reliability, Current Density, and High-Frequency Performance of AlGaN/GaN HEMTs

Abstract

Emerging applications including advanced industrial manufacturing, cutting-edge scientific research and medical equipment demand AlGaN/GaN HEMTs possessing both high-frequency and high-voltage characteristics. However, a persistent trade-off remains between the frequency characteristics and breakdown characteristics of these devices. In this study, we employed localized electric field tailoring (LEFT) by introducing materials with different dielectric constants to construct a non-uniform composite gate dielectric layer, aiming to balance the breakdown voltage and cut-off frequency of the device. Device models were developed using APSYS-2018 software and their reliability was experimentally validated. Research data indicates that, compared to traditional uniform high-k (typically with dielectric constants k > 10, such as HfO₂ and HfZrO) gate dielectrics, the non-uniform composite gate dielectric structure demonstrates superior transconductance, saturation current density and cut-off frequency, with minimal degradation in breakdown voltage. Specifically, relative to HfO₂ and HfZrO uniform devices, the Al₂O₃/HfO₂ and Al₂O₃/HfZrO non-uniform HEMTs achieved 20.0% and 35.2% increases in cut-off frequency, respectively. Meanwhile, breakdown voltage remained above 97% of their uniform counterparts, saturation current density and transconductance increased by approximately 5%. Therefore, this non-uniform composite gate dielectric layer structure of AlGaN/GaN HEMT with LEFT holds great potential for industrial plasma generators, magnetic resonance imaging systems and biomedical radiofrequency hyperthermia devices.

1. Introduction

AlGaN/GaN heterojunction exhibits stronger spontaneous and piezoelectric polarization effects [1,2] compared to other III-V heterojunctions, generating a two-dimensional electron gas (2DEG) [3] at the interface. Consequently, AlGaN/GaN HEMTs demonstrate superior electrical characteristics, including high carrier concentration, mobility, and breakdown field [4,5,6], coupled with low on-resistance [7] and high switching speed [8]. Given the aforementioned advantages of AlGaN/GaN HEMTs and the inefficiency of Si-based devices at MHz frequencies, HEMTs integrating both high-frequency and high-voltage capabilities therefore demonstrate application prospects in emerging fields, including industrial plasma generators (e.g., PECVD devices), medical and scientific equipment (e.g., magnetic resonance imaging systems), and biomedical radiofrequency hyperthermia (e.g., 13.56 MHz tumor ablation systems).

However, the high-frequency performance and voltage withstand characteristics of AlGaN/GaN HEMTs are inherently trade-offs, posing significant challenges for co-optimization. Current breakdown voltage enhancement strategies include gate-drain spacing extension [9], gate lengthening [10], and MOS-structure integration [11]. Since dimensional scaling substantially increases on-resistance (R_on), degrading power density and high-frequency characteristics, we implement MOS architectures for balanced performance. This approach deposits high-k gate dielectrics between metal and semiconductor interfaces [12], including hafnium-based oxides (e.g., HfO₂ [13], HfSiO_x [14], HfZrO_x [15]), aluminum-based oxides (Al₂O₃ [16]), and zirconium-based oxides (ZrO₂ [17]). In this context, “high-k” refers to dielectric materials with relative permittivity significantly higher than that of SiO₂ (k ≈ 3.9), typically exceeding 10, such as HfO₂ (k ≈ 25) and HfZrO (k ≈ 30). Al₂O₃ (k ≈ 9) is considered a moderate-to-high-k material. Although improving breakdown robustness, these high-k layers inevitably impose cut-off frequency penalties [18]. Notably, breakdown is typically localized, whereas frequency characteristics are uniform and holistic. This difference arises from non-uniform electric field distribution within the device: electric field concentration at the gate edge amplifies impact ionization, ultimately triggering localized avalanche breakdown. Conversely, the homogeneity of frequency characteristics relies on holistic carrier transport throughout the channel, which is governed by the statistically averaged properties of carriers (e.g., density, mobility, and velocity distribution) rather than local electric field peaks. Therefore, we propose localized reinforcement to enhance the device’s cut-off frequency while maintaining high breakdown voltage. Localized reinforcement refers to depositing the high-k dielectric layer at the drain-side gate edge where electric field crowding induces breakdown, while lower-k materials occupy low-field regions to minimize equivalent gate capacitance.

This study implements localized electric field tailoring (LEFT) through non-uniform composite gate dielectrics. High-k material at the gate-drain side enhances breakdown voltage, while lower-k material on the gate-source side improves cut-off frequency. This structure balances breakdown and frequency characteristics. Conventional uniform Al₂O₃ devices were fabricated for experimental validation, confirming model reliability. Additional devices incorporating HfO₂ and HfZrO were systematically characterized. Compared to uniform high-k gate dielectrics, the proposed non-uniform composite structure effectively optimized electric field distribution while maintaining comparable breakdown voltage. More impressively, the cut-off frequency was significantly enhanced. These results demonstrate LEFT-enabled non-uniform gate dielectrics’ strong potential for practical applications.

2. Materials and Methods

Device models were developed using the APSYS platform, and corresponding devices were fabricated. Figure 1a illustrates the cross-sectional structure of the simulated device, with key geometric parameters provided in

Table 1. Specific geometric parameters of the AlGaN/GaN MOS-HEMT.

Parameters	Value (µm)
LG	4
LS	70
LD	70
Lgs	3
Lgd	14
WG	100
WS	100
WD	100
Oxide dielectric layer	0.02
GaN cap layer	0.002
Al0.3Ga0.7N barrier layer	0.03
AlN spacer layer	0.001
GaN channel layer	0.46
Al0.5Ga0.5N buffer layer	3.4
AlN nucleation layer	0.1
Si substrate layer	5

.

After device modeling, appropriate physical models and simulation parameters were applied to predict device behavior. Self-heating effects were simulated via an integrated thermal module, and breakdown characteristics were evaluated using an impact ionization model. The AlGaN/GaN heterojunction interface was treated with a polarization charge model. Besides the aforementioned physical models, the modified transferred-electron mobility model was also employed to accurately describe the current transport process. This model effectively hybridizes MTE and Canali models [19], as shown in Equations (1)–(3):

$$v\left(E\right)={sin}^2\left({\displaystyle \frac{\pi }{2x}}\right)·v_{MTE}\left(E\right)+{cos}^2\left({\displaystyle \frac{\pi }{2x}}\right)·v_C\left(E\right),$$

(1)

$$v_{MTE}\left(E\right)={\displaystyle \frac{F\left(E\right)+v_{sat}{(E/E_{MT})}^{{\beta }_T}}{1+{(E/E_{MT})}^{{\beta }_T}}},$$

(2)

$$v_C\left(E\right)={\displaystyle \frac{{\mu }_{low}E}{{(1+{({\mu }_{low}E/v_{sat})}^{{\beta }_C})}^{1/{\beta }_C}}},$$

(3)

where V(E) denotes the electron drift velocity, while V_MTE(E) and V_C(E) represent the electron drift velocities described by the MTE and Canali models, respectively. As shown, Equation (1) reduces to the MTE model when x = 1, and approaches the Canali model as x tends to 0. For the intermediate case of x = 0.5, Equation (1) incorporates an equal weighting of 50% from both the MTE and Canali models. The function F(E) is to imitate the kink in the low-field region, V_sat denotes the saturation velocity, E represents the electric field, E_MT is a fixed field value, μ_low is the low-field mobility, and β_T and β_C correspond to the values of parameters beta_mte and beta_n in the model.

Another important effect is the impurity dependence of the low field mobility, as shown in Equations (4) and (5):

$${\mu }_{0n}={\mu }_{1n}+{\displaystyle \frac{({\mu }_{2n}-{\mu }_{1n})}{1+{(\frac{N_D+N_A+{\sum }_j{N}_{tj}}{N_{rn}})}^{{\alpha }_n}}},$$

(4)

$${\mu }_{0p}={\mu }_{1p}+{\displaystyle \frac{({\mu }_{2p}-{\mu }_{1p})}{1+{(\frac{N_D+N_A+{\sum }_j{N}_{tj}}{N_{rp}})}^{{\alpha }_p}}},$$

(5)

where μ_0n and μ_0p are the low-field electron and hole mobilities, respectively; μ_1n and μ_1p are the carrier mobilities at high impurity concentration; μ_2n and μ_2p denote the carrier mobilities at low impurity concentration; N_D and N_A are the donor and acceptor concentrations; Σ_jN_tj is the total density of trap/defect levels summed over all considered trap types j; N_rn and N_rp are the empirical reference impurity concentrations; and α_n and α_p are empirical fitting exponents that determine how rapidly the mobility transitions with increasing total charged/neutral defect concentration for electrons and holes.

A drift-diffusion model was used to govern the electrical behavior of the device. Fermi–Dirac statistics were employed to accurately calculate the carrier distribution of the high-concentration 2DEG, as shown in Equations (6)–(8):

$$n=N_c{F}_{1/2}({\displaystyle \frac{E_{fn}-E_c}{kT}}),$$

(6)

$$p=N_v{F}_{1/2}({\displaystyle \frac{E_v-E_{fp}}{kT}}),$$

(7)

$$F_{1/2}(x)\approx {(e^{-x}+\xi (x\left)\right)}^{-1},$$

(8)

where n and p are the electron and hole concentrations; N_C and N_V are the effective density of states in the conduction and valence bands; E_fn and E_fp are the quasi-Fermi levels for electrons and holes; E_C and E_V are the conduction and valence band edges; k is the Boltzmann constant; T is the absolute temperature; F_1/2 is the Fermi integral of order one-half; and ξ(x) is a correction function used in the approximate form of the Fermi integral.

The Shockley–Read–Hall recombination model was utilized to simulate the carrier recombination process via defect levels in the material:

$$R_n^{tj}=c_{nj}n{N}_{tj}\left(1-f_{tj}\right)-c_{nj}{n}_{1j}{N}_{tj}{f}_{tj},$$

(9)

$$R_p^{tj}=c_{pj}p{N}_{tj}{f}_{tj}-c_{pj}{p}_{1j}{N}_{tj}(1-f_{tj}),$$

(10)

where $R_n^{tj}$ and $R_p^{tj}$ denote the electron and hole recombination rates via the j-th trap level; c_nj and c_pj are the capture coefficients for electrons and holes; n and p represent the free electron and hole concentrations; n_1j and p_1j are the carrier concentrations when the trap level equals the Fermi level; N_tj is the trap density; and J is the trap occupancy probability.

The capture coefficients can be further expressed as follows:

$$c_{nj}={\sigma }_{nj}{\overline{v}}_n={\sigma }_{nj}\sqrt{{\displaystyle \frac{8kT}{\pi m_n}}},$$

(11)

$$c_{pj}={\sigma }_{pj}{\overline{v}}_p={\sigma }_{pj}\sqrt{{\displaystyle \frac{8kT}{\pi m_p}}},$$

(12)

where σ_nj and σ_pj are the capture cross-sections for electrons and holes; ${\overline{v}}_n$ and ${\overline{v}}_p$ are the average thermal velocities of electrons and holes; m_n and m_p represent the effective masses of electrons and holes. The trap or recombination center is completely specified by its density N_tj, capture cross-sections σ_nj and σ_pj, and energy level E_tj. The numerical values of the relevant physical parameters in Equations (1)–(12) are presented in

Table 2. Relevant physical parameters in Equations (1)–(12).

Parameters	Units	Value
Vsat	cm/s	1.91 × 107
EMT	kV/cm	257
μlow	cm2/V·s	1000
βT	-	5.7
βC	-	1.7
μ0n	cm2/V·s	1200
k	eV/K	8.617 × 10−5
T	K	300
Ntj	cm−3	1.85 × 1016
σnj	cm2	1 × 10−15

. Figure 1b shows the energy band diagram of the MOS structure in Al₂O₃ MOS-HEMT. The conduction band of GaN on the left side contacts the Fermi level due to strong spontaneous and piezoelectric polarization effects. These effects lead to a high electron concentration in the channel layer, driving it into a degenerate state and forming a conductive channel. A two-dimensional electron gas is present at the interface between the Fermi level and the triangular quantum well. Figure 1c shows the two-dimensional electric field distribution under the gate in Al₂O₃ MOS-HEMT. The electric field is uniformly distributed, and the scale is calculated based on the actual electric field value of the formula 20log₁₀.

To validate the simulation model, conventional Al₂O₃ gate dielectric HEMTs were fabricated. The process began with mesa isolation by ICP etching using a Cl₂/BCl₃/Ar mixture (10/25/5 sccm). Source and drain electrodes were formed via electron-beam deposition of Ti/Al/Ni/Au (20/120/45/55 nm), followed by rapid thermal annealing at 850 °C for 30 s in N₂ to achieve ohmic contacts. A 20 nm Al₂O₃ gate dielectric was deposited by ALD at 160 °C, and Ni/Au (80/50 nm) was evaporated to form the Schottky gate. A micrograph of the completed Al₂O₃ MOS-HEMT is shown in Figure 1d.

3. Results

3.1. Matching Simulated and Measured Data for the MOS-HEMT

Following fabrication, the DC output performance and transfer characteristics of the devices were characterized using a semiconductor analyzer and compared with simulation data for accuracy validation. As shown in Figure 2a, the measured and simulated maximum drain current density (I_ds,max) values were 292.6 and 290.6 mA/mm, respectively, representing a difference of 0.7%. Figure 2b presents the transfer characteristics, with maximum transconductance (g_m,max) values of 65.32 and 65.12 mS/mm, and threshold voltage (V_th) values of −4.80 and −4.77 V, showing a 0.3% and 0.6% discrepancy, respectively. To quantitatively verify the simulation accuracy,

Table 3. Comparison between simulated and measured characteristics.

Parameters	Measurement	Simulation	Error (%)
Ids,max	292.6 mA/mm	290.6 mA/mm	0.7
gm,max	65.32 mS/mm	65.12 mS/mm	0.3
Vth	−4.80 V	−4.77 V	0.6

compares the electrical characteristics extracted from the simulation and experiment, showing excellent consistency with deviations below 1%. These results demonstrate good agreement between measurement and simulation, confirming the reliability of the models.

3.2. Mechanism of Localized Electric Field Tailoring by Non-Uniform Composite Gate Dielectrics

The schematic diagram of the non-uniform composite gate dielectric structure and the electric field distribution of Al₂O₃, HfO₂, and Al₂O₃/HfO₂ MOS-HEMTs are shown in Figure 3. Figure 3a presents the schematic of the non-uniform composite gate structure, incorporating high-k dielectrics (e.g., HfO₂ and HfZrO) on the gate-drain side and a lower-k dielectric (e.g., Al₂O₃) on the gate-source side. Figure 3b displays the two-dimensional electric field distribution under the gate in the Al₂O₃/HfO₂ MOS-HEMT, showing a peak at the 2 μm interface between the two dielectrics. Correspondingly, Figure 3c illustrates the transverse electric field profiles under the gate for Al₂O₃, HfO₂, and Al₂O₃/HfO₂ MOS-HEMTs. While the fields are uniform in the single-dielectric devices, a peak occurs at the dielectric interface in the composite structure, consistent with the observation in Figure 3b. This field discontinuity arises from the permittivity difference between HfO₂ and Al₂O₃; the potential is continuous in the direction perpendicular to the interface, leading to a discontinuous jump in electric field strength. According to Gauss’s law, the electric fields perpendicular to the interface satisfy [20]

$${\epsilon }_1{E}_1={\epsilon }_2{E}_2,$$

(13)

where ε₁ and ε₂ denote the dielectric constants of HfO₂ and Al₂O₃, respectively, and E₁ and E₂ are the corresponding perpendicular electric fields. This field jump introduces a transverse electric field peak, enabling LEFT via regional control of dielectric constants. Figure 3d compares the transverse electric field at the drain-side gate edge for the three configurations, with values of 4.33, 4.30, and 4.24 MV/cm, respectively. The reduced field in the composite structure alleviates drain-edge field crowding and suppresses the hot electron effect, thereby enhancing reliability.

3.3. Analysis of Simulation Results

The DC output characteristics of the MOS-HEMTs are shown in Figure 4. As shown in Figure 4a, the Al₂O₃ MOS-HEMT operates under a gate voltage ranging from −5 V to 3 V and achieves an I_ds,max of 291 mA/mm. Beyond the saturation region, a discernible current reduction occurs, attributed to self-heating that elevates the heterojunction temperature and degrades output performance [21]. In contrast, the HfO₂ MOS-HEMT (Figure 4b) is biased from −1 V to 6 V due to its higher V_th and positive bias stability, delivering an I_ds,max of 354 mA/mm without significant current decrease. This represents a 22% improvement over the Al₂O₃ device, resulting from the higher dielectric constant of HfO₂, which reduces the depletion region width and source–drain resistance, thereby enhancing channel conductivity [22]. Moreover, analysis based on interface and trap models indicates that the increased permittivity also mitigates surface scattering and carrier trapping [23] at the metal/AlGaN junction, which in turn increases channel carrier concentration [24]. Figure 4c shows the Al₂O₃/HfO₂ MOS-HEMT, gated from −2 V to 5 V with V_th and positive bias stability intermediate between the single-dielectric devices. It exhibits an I_ds,max of 371 mA/mm and minimal current degradation. The improved current stability in both HfO₂ and Al₂O₃/HfO₂ stems from the stronger electric field shielding under the gate, which reduces surface field intensity [25] and suppresses dynamic charge trapping/detrapping. The high-k dielectric also alleviates self-heating, enhancing reliability [26]. Finally, as summarized in Figure 4d, the simulated I_ds,max of nine devices confirms that composite gate dielectrics achieve better performance compared to uniform higher-k layers, demonstrating that non-uniform composite gate dielectric devices with LEFT can deliver enhanced output characteristics.

Figure 5a presents the transfer characteristics of the Al₂O₃ MOS-HEMT at V_ds = 10 V and V_g swept from −10 V to 10 V, yielding a g_m,max of 65.1 mS/mm and a V_th of −4.77 V. For the HfO₂ MOS-HEMT (Figure 5b), g_m,max reaches 74.5 mS/mm and V_th is −0.85 V. The 14.4% improvement in g_m,max is attributed to the higher dielectric constant, which enhances gate control capability [27,28]. The positive shift in V_th also results from the increased permittivity [29]. As shown in Figure 5c, the Al₂O₃/HfO₂ MOS-HEMT achieves a g_m,max of 78.7 mS/mm and V_th of −1.96 V. The further 5.6% increase in g_m,max over the HfO₂ MOS-HEMT arises from LEFT, which optimizes field distribution and improves electron transport efficiency. The negative V_th shift relative to the HfO₂ device is due to the reduced unit-area gate capacitance caused by the incorporation of the lower-k material. Based on the parallel-plate capacitance formula [30]:

$$C={\displaystyle \frac{\epsilon S}{4\pi kd}},$$

(14)

where C represents the gate capacitance, ε denotes the absolute permittivity of dielectric layer (ε = ε₀ε_r), S is the gate area, k is the Coulomb’s constant, and d represents the thickness of the gate dielectric layer. The threshold voltage can be expressed as follows [31]:

$$V_T={\displaystyle \frac{\sqrt{2{\epsilon }_{s}q{N}_A\left(2{\psi }_B\right)}}{C_0}}+2{\psi }_B,$$

(15)

where V_T represents the threshold voltage, ε_s denotes the semiconductor permittivity, q is the elementary charge, N_A is the acceptor concentration, ψ_B is the bulk potential, and C₀ is the unit-area gate capacitance derived from Equation (14). The numerical values of the relevant physical parameters in Equations (14) and (15) are presented in

Table 4. Representative physical parameters in Equations (14) and (15).

Parameters	Units	Value
ε0	F/cm	8.85 × 10−14
εr(SiO2)	-	3.9
εr(Al2O3)	-	9
εr(HfO2)	-	25
εr(HfZrO)	-	30
k	N·m2/C2	8.99 × 109
d	nm	20
q	C	1.602 × 10−19

. As the gate capacitance decreases, the absolute value of V_th increases accordingly. Figure 5d,e summarize V_th and g_m,max across nine devices. The composite dielectric devices exhibit slightly higher g_m,max than their uniform high-k counterparts, consistent with previous trends. Figure 5f compares the gate voltage swing (GVS), defined as the V_g variation range where g_m,max remains within 90% of its peak. GVS generally decreases with increasing dielectric constant, though composite dielectrics show improved GVS over uniform high-k layers. A higher GVS indicates enhanced linearity, which helps reduce phase noise, intermodulation distortion, and extends dynamic range in practical applications [32].

The breakdown voltage (V_bd) is defined as the V_ds at which V_bd exceeds 1 mA/mm while V_g is held at −7 V to maintain the off-state. Figure 6a shows the V_bd of four uniform gate dielectric devices. With increasing dielectric constant, V_bd rises significantly [33]. Similarly, Figure 6b presents V_bd for five composite dielectric devices, where V_bd also increases with equivalent permittivity. Figure 6c compares V_bd among Al₂O₃, HfO₂, and Al₂O₃/HfO₂ MOS-HEMTs. The Al₂O₃/HfO₂ MOS-HEMT exhibits V_bd nearly identical to the HfO₂ device, with only a 1.7% reduction, attributable to non-uniform field distribution leading to localized breakdown near the gate-drain side [34], rather than uniform breakdown. Figure 6d summarizes V_bd across all nine devices. V_bd increases consistently with dielectric constant from left to right. Moreover, the breakdown voltages of SiO₂/Al₂O₃, Al₂O₃/HfO₂, and Al₂O₃/HfZrO MOS-HEMTs closely match those of their uniform higher-k counterparts, confirming the feasibility and effectiveness of LEFT in improving breakdown performance and device reliability.

The device’s gate capacitance and cut-off frequency are shown in Figure 7. Figure 7a presents the C_g-V_g characteristics of Al₂O₃, HfO₂, and Al₂O₃/HfO₂ MOS-HEMTs at V_ds = 10 V. As V_g increases, C_g rises slowly initially, increases rapidly in the linear region, and flattens out in the saturation region, reaching a maximum at V_g = 3 V. The C_g values are 396, 474, and 428 pF/mm, respectively, differing mainly due to the dielectric constants of the materials [35]. Figure 7b summarizes the simulated C_g for nine devices, showing a positive correlation with dielectric constant. Figure 7c presents the f_T-V_g curves for the three devices. The Al₂O₃ MOS-HEMT peaks at f_T = 36.5 MHz (V_g = −1.4 V), while the HfO₂ and Al₂O₃/HfO₂ devices peak at 24.5 MHz (V_g = 2.8 V) and 29.4 MHz (V_g = 2.2 V), respectively. Based on the approximation [36]:

$$f_T={\displaystyle \frac{g_m}{2\pi C_g}}$$

(16)

the increase in f_T is mainly due to reduced C_g, with a minor contribution from increased g_m. As shown in Figure 7d, f_T generally decreases with higher dielectric constant [37]. Compared to their uniform higher-k counterparts, the Al₂O₃/HfZrO, SiO₂/Al₂O₃, SiO₂/HfO₂, and SiO₂/HfZrO MOS-HEMTs show f_T improvements of 35.2%, 8.2%, 28.2%, and 44.4%, respectively. These results demonstrate that composite gate dielectrics with LEFT effectively alleviate the f_T degradation typical of high-k devices, enabling balanced frequency performance.

4. Discussion

This study simulates and analyzes the electrical characteristics of various gate dielectric structures. The data indicates that the non-uniform composite gate dielectric device with LEFT exhibits V_bd on par with that of uniform higher-k devices, while simultaneously delivering superior g_m,max, I_ds,max and f_T. Specifically, compared to HfO₂ and HfZrO MOS-HEMTs, the Al₂O₃/HfO₂ and Al₂O₃/HfZrO structures show f_T improvements of 20.0% and 35.2%, respectively, with only marginal V_bd reductions of 1.7% and 2.3%, as summarized in Figure 8.

5. Conclusions

In summary, we propose a LEFT technique and develop corresponding device structures and TCAD models for both conventional uniform and non-uniform composite gate dielectrics. The simulation models were validated against experimental measurements. The LEFT technique retains a high-k dielectric on the gate-drain side while introducing a lower-k material on the gate-source side. This configuration modulates the electric field distribution, improves channel transport efficiency, and achieves an optimal trade-off between the two critical yet competing metrics: breakdown voltage and cut-off frequency. The core advantage of a non-uniform composite gate dielectric structure lies in its functional zoning. This provides a new approach to resolving the inherent contradictions in HEMTs. The AlGaN/GaN MOS-HEMT with a non-uniform composite gate dielectric structure, shows significant potential for high-voltage and high-frequency applications, such as industrial plasma generators, magnetic resonance imaging systems and biomedical radiofrequency hyperthermia devices.