The Influence of Traps on the Self-Heating Effect and THz Response of GaN HEMTs

Abstract

This study systematically investigates the effects of trap concentration on self-heating and terahertz (THz) responses in GaN HEMTs using Sentaurus TCAD. Traps, inherently unavoidable in semiconductors, can be strategically introduced to engineer specific energy levels that establish competitive dynamics between the electron momentum relaxation time and the carrier lifetime. A simulation-based exploration of this mechanism provides significant scientific value for enhancing device performance through self-heating mitigation and THz response optimization. An AlGaN/GaN heterojunction HEMT model was established, with trap concentrations ranging from 0 to $5\times {10}^{17} {\mathrm{c}\mathrm{m}}^{-3}$. The analysis reveals that traps significantly enhance channel current (achieving 3× gain at $1\times 10^{17} {\mathrm{c}\mathrm{m}}^{-3}$) via new energy levels that prolong carrier lifetime. However, elevated trap concentrations (>$1\times {10}^{16} {\mathrm{c}\mathrm{m}}^{-3}$) exacerbate self-heating-induced current collapse, reducing the min-to-max current ratio to 0.9158. In THz response characterization, devices exhibit a distinct DC component (${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$) under non-resonant detection ($\mathsf{\omega }\mathsf{\tau }\ll 1$). At a trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$, ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ peaks at $0.12 \mathrm{V}$ when ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-7.8 \mathrm{V}$. Compared to trap-free devices, a maximum response attenuation of 64.89% occurs at ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-4.9 \mathrm{V}$. Furthermore, ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ demonstrates non-monotonic behavior with concentration, showing local maxima at $4\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ and $7\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$, attributed to plasma wave damping and temperature-gradient-induced electric field variations. This research establishes trap engineering guidelines for GaN HEMTs: a concentration of $4\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ optimally enhances conductivity while minimizing adverse impacts on both self-heating and the THz response, making it particularly suitable for high-sensitivity terahertz detectors.

1. Introduction

The terahertz band offers substantial spectral resources and distinctive absorption properties, presenting broad application potential in communications, remote sensing, bio-medicine, and security screening [1,2,3,4]. In the AI era, escalating demands for higher data transmission rates align precisely with terahertz’s spectral advantages. However, significant atmospheric attenuation challenges terahertz wave propagation. Enhancing receiver responsivity enables the detection of weaker terahertz signals, thereby extending effective communication distances.

GaN-based high-electron-mobility transistors (HEMTs) demonstrate exceptional high-frequency performance, particularly in terahertz applications [5,6,7,8], exhibiting considerable development potential. Under high-frequency operation, GaN HEMTs experience self-heating effects where traps critically influence current degradation [9,10,11,12,13,14,15]. Specifically, when single-pulse voltages stabilize at peak values, localized temperature increases occur within the device, most notably at the two-dimensional electron gas (2DEG) channel. In undoped structures, the positive current gain from intrinsic carrier increase is outweighed by mobility-reduction-induced negative gain [16], resulting in channel current collapse.

Traps intrinsically present in GaN HEMTs—introduced within GaN/AlGaN layers adjacent to the 2DEG channel—significantly modulate channel current [17]. These defects primarily enhance current through new energy levels that bind carriers, thereby extending carrier lifetime [18]. Macroscopically, prolonged carrier lifetime increases channel carrier density, improving device conductivity. Strategic trap introduction thus enables controllable current enhancement. This study employs the Sentaurus platform to systematically investigate the impacts of traps on both self-heating effects and terahertz (THz) response characteristics.

2. Materials and Methods

To investigate the impacts of traps on self-heating effects and the terahertz response in HEMTs, this study implemented a Sentaurus TCAD simulation framework. The SDE module was first employed for device structural modeling, establishing the HEMT configuration illustrated in Figure 1a, with channel-proximate mesh refinement detailed in Figure 1b. Following structural verification, a mesh file was generated.

Variations in aluminum (Al) composition within AlGaN induce spontaneous polarization by displacing positive and negative charge centers. At the AlGaN/GaN heterointerface, lattice mismatch generates piezoelectric polarization during epitaxial growth. The synergistic effect of these mechanisms establishes a built-in electric field within the AlGaN layer, causing electron accumulation and subsequent two-dimensional electron gas (2DEG) formation [19,20].

AlGaN layer thickness critically influences device performance:

• Under-thinned layers (<25 nm) yield weak built-in fields, reducing 2DEG concentration and degrading conduction;
• Over-thickened layers (>25 nm) diminish gate control efficacy, attenuating the terahertz response.

For this simulation, AlGaN thickness was optimized at 25 nm following Zhang et al.’s methodology. The remaining structural parameters are summarized in

Table 1. Parameters and materials of layers in HEMT model.

Material/Contact	Thickness ($\mathsf{\mu }\mathbf{m}$)	Width ($\mathsf{\mu }\mathbf{m}$)
${\mathrm{S}\mathrm{i}}_3{\mathrm{N}}_4$ (S side)	0.025	1.5
${\mathrm{S}\mathrm{i}}_3{\mathrm{N}}_4$ (D side)	0.025	2.2
$\mathrm{A}\mathrm{l}\mathrm{G}\mathrm{a}\mathrm{N}$	0.025	5.9
$\mathrm{G}\mathrm{a}\mathrm{N}$	2	5.9
$\mathrm{S}\mathrm{i}\mathrm{C}$	10	5.9
Electrode S	—	0.5
Electrode G	—	1
Electrode D	—	0.5
thermal	—	5.9

.

Two fundamental electrical contact types exist between electrodes and materials: Ohmic and Schottky contacts. To establish conductive paths between channel and source/drain electrodes, heavy doping creates Ohmic contacts [21]. Conversely, the gate electrode utilizes Schottky contacts to minimize leakage current [22]. The thermal electrode at the substrate interface provides thermal boundary conditions, maintained at 300 K throughout simulations.

Subsequently, Sentaurus TCAD (invoked via SDevice) implemented critical physical models. Carrier mobility, high-field saturation effects, and doping considerations were incorporated. Given that doping was solely employed for source/drain Ohmic contact formation—exhibiting negligible influence on channel electrical properties—and comparative simulations revealed minimal performance differences between mobility models, the computationally efficient Constant Mobility model was selected over the Doping Dependence model.

The mobility of the Constant Mobility model ${\mathsf{\mu }}_{\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{t}}$ only depends on the lattice temperature [23]:

$${\mu }_{const}={\mu }_L{\left({\displaystyle \frac{T}{300K}}\right)}^{-\zeta }$$

(1)

where ${\mu }_L$ is the mobility due to bulk phonon scattering. The default values of ${\mu }_L$ and the exponent ζ are listed in

Table 2. Constant mobility model: default coefficients for silicon.

Symbol	Electrons	Holes	Unit
${\mu }_L$	1417	470.5	${\mathrm{c}\mathrm{m}}^2/\mathrm{V}\mathrm{s}$
ζ	2.5	2.2	1

. When the material is changed to AlGaAs, the default values of ${\mathsf{\mu }}_{\mathrm{L}}$ and the exponent ζ will change too.

Before investigating the self-heating effect and terahertz response, the electrical characteristics of the device should be tested to ensure the correctness of its fundamental electrical properties and operational integrity.

Under zero gate bias, the drain-source voltage was swept from $0 \mathrm{V}$ to $10 \mathrm{V}$ to characterize conduction properties. Maintaining a constant DC source/drain voltage, the gate voltage was subsequently swept from $-8 \mathrm{V}$ to $4 \mathrm{V}$ to evaluate transfer characteristics.

To explore the self-heating effect under different trap concentrations, as shown in Figure 2a, a pulse signal was configured in SDevice with an amplitude of 5 V, a delay time (${\mathrm{T}}_{\mathrm{d}}$) of $1.0\times {10}^{-8} \mathrm{s}$, rise and fall times (${\mathrm{T}}_{\mathrm{r}}={\mathrm{T}}_{\mathrm{f}}$) of $5.0\times {10}^{-7}$ s, a pulse width (${\mathrm{p}}_{\mathrm{w}}$) of $5.0\times {10}^{-7} \mathrm{s}$, and a pulse period of $1.6\times {10}^{-6}$ s. Since Sentaurus simulations may struggle to converge when abrupt boundary condition changes occur at the start, the delay time (${\mathrm{T}}_{\mathrm{d}}$) was introduced to allow the pulse to activate only after a brief simulation warm-up period.

A configured pulse was applied across the device source and drain while varying the trap concentration in the SDevice simulation environment. The analysis of channel current waveforms under different trap concentrations quantified trap concentration’s effects on self-heating, followed by a mechanistic investigation.

Simulation experiments revealed a strong convergence in the HEMT model at a trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$. Consequently, all subsequent investigations of gate voltage effects on the terahertz response employed a consistent trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$. As Figure 2b illustrates, the simulated waveform represents terahertz wave incidence with amplitude ${\mathrm{U}}_{\mathrm{W}}=400\hspace{0.17em}\mathrm{m}\mathrm{V}$, period ${\mathrm{T}}_{\mathrm{w}}=1\times 10^{-11} \mathrm{s}$, and corresponding frequency ${\mathrm{f}}_{\mathrm{w}}=100\hspace{0.17em}\mathrm{G}\mathrm{H}\mathrm{z}$—parameters within the terahertz band (0.1–10 THz). During terahertz radiation exposure, gate voltage sweeps were performed to analyze their impact on device response and underlying physical mechanisms.

3. Results and Discussion

The fundamental electrical characteristics of the device are presented in Figure 3. During source/drain conduction characterization, channel current saturation occurs when the drain/source voltage reaches $10 \mathrm{V}$, consistent with high-field saturation behavior. For transfer characteristic measurements, the saturation voltage determined from conduction analysis ($10 \mathrm{V}$) was applied between the source and the drain. The channel approached cutoff conditions when gate voltage was reduced to $-8 \mathrm{V}$, resulting in zero channel current. Positive gate voltages produced only a modest enhancement of channel current.

The self-heating effect manifests as degraded carrier mobility with increasing channel temperature, causing current collapse within single-pulse operation. Although multiple mitigation strategies exist [24,25,26], this work specifically examines trap-mediated effects. As demonstrated in Figure 4, elevated trap concentrations yield greater channel current enhancement. This improvement originates from trap-induced energy levels that extend carrier lifetime, thereby increasing channel carrier concentration. The capture rate $c_i$ and emission rate $e_i$ of these trap states are governed by

$$e_i={\displaystyle \frac{c_i}{g}}\exp ({\displaystyle \frac{E_{trap}-E_F^i}{k{T}_i}}),$$

(2)

Here, $E_{trap}$ is the energy of the trap, $E_F^i$ is the Fermi energy of the electronic reservoir, $T_i$ is the temperature of the reservoir, and g is the degeneracy factor. When the trap energy level ($E_{trap}$) approaches the electron Fermi level ($E_F^i$), the emission rate of the trap level becomes lower than its capture rate. This leads to electron trapping, thereby increasing the carrier lifetime. When the trap concentration reaches $1\times 10^{17} {\mathrm{c}\mathrm{m}}^{-3}$, the channel current is almost three times that of a device without traps.

Figure 4. Manifestation of self-heating effects. Id curve when the pulse in Figure 2a has an effect on S and D at different trap concentrations.

Figure 4. Manifestation of self-heating effects. Id curve when the pulse in Figure 2a has an effect on S and D at different trap concentrations.

During the pulse width (${\mathrm{p}}_{\mathrm{w}}$) period, devices with varying trap concentrations all exhibit temporally dependent current collapse. It is observed that higher trap concentrations produce more pronounced absolute values of channel current collapse. As summarized in

Table 3. Min-to-max current ratio within the pulse width duration.

Trap Concentration (${\mathbf{c}\mathbf{m}}^{-3}$)	${\mathbf{I}\mathbf{d}}_{\mathbf{m}\mathbf{a}\mathbf{x}}$ ($\mathsf{\mu }\mathbf{A}$)	${\mathbf{I}\mathbf{d}}_{\mathbf{m}\mathbf{i}\mathbf{n}}$ ($\mathsf{\mu }\mathbf{A}$)	${\mathbf{I}\mathbf{d}}_{\mathbf{m}\mathbf{i}\mathbf{n}}$$/{\mathbf{I}\mathbf{d}}_{\mathbf{m}\mathbf{a}\mathbf{x}}$
0	738.7534	732.3348	0.9913
${1\times 10}^{11}$	742.8229	732.3365	0.9859
$1\times {10}^{12}$	742.8384	732.3516	0.9859
$1\times 10^{13}$	742.9932	733.6027	0.9874
$1\times 10^{14}$	744.5444	735.1205	0.9873
5$\times 10^{14}$	751.5934	740.8572	0.9857
$1\times 10^{15}$	756.6317	749.8305	0.9910
$5\times 10^{15}$	848.5516	833.7205	0.9825
$1\times 10^{16}$	953.7343	942.0889	0.9877
$1\times 10^{17}$	2677.808	2613.827	0.9761
5 $\times 10^{17}$	8478.159	7763.949	0.9158

, devices with elevated trap concentrations show greater differentials between ${\mathrm{I}\mathrm{d}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ and ${\mathrm{I}\mathrm{d}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$.

However, the absolute magnitude of current collapse cannot serve as a criterion for determining whether the self-heating effect is mitigated or exacerbated. Here, we define the degree of current collapse as the ratio of ${\mathrm{I}\mathrm{d}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ to ${\mathrm{I}\mathrm{d}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$. A smaller ratio indicates a more severe current collapse. By correlating each obtained ratio with its corresponding trap concentration, we establish the trend shown in Figure 5.

The current collapse exhibits a “three-stage” variation with increasing trap concentration.

Specifically,

• In a low concentration range (0 to $1\times {10}^{14} {\mathrm{c}\mathrm{m}}^{-3}$), the current collapse trend is negligible;
• In an intermediate concentration range ($1\times {10}^{14}$ to $1\times {10}^{16} {\mathrm{c}\mathrm{m}}^{-3}$), the current collapse fluctuates, peaking at a trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$;
• In a high concentration range ($>1\times {10}^{16} {\mathrm{c}\mathrm{m}}^{-3}$), the current collapse intensifies significantly, showing a clear degradation trend.

Simulation results demonstrated a strong convergence for devices with a trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$. Current collapse trend analysis revealed that introducing traps at this concentration produces the least pronounced exacerbation of device self-heating. Consequently, among approaches employing traps to enhance conduction characteristics, a trap concentration of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ yields optimal DC operational efficiency.

Conventional theory dictates that the upper frequency limit of field-effect transistors (FETs) is governed by electron transit time. For the GaN HEMT devices simulated in this work, plasma wave theory within the two-dimensional electron gas (2DEG) framework explains the observed phenomena [27,28,29].

The 2DEG concentration in the device channel is modulated by the gate-to-channel voltage ${\mathrm{V}}_{\mathrm{g}\mathrm{s}}$, expressed as

$$n_s=C{V}_{gs}/q,$$

(3)

where C is the gate-to-channel capacitance per unit area, and q is the elementary charge. The two-dimensional electron gas can be treated as a hydrodynamic fluid model. Consequently, the propagation of plasma waves can be described using the Euler equation for carriers and the current continuity equation [28]:

$${\displaystyle \frac{\partial v}{\partial t}}+v{\displaystyle \frac{\partial v}{\partial x}}+{\displaystyle \frac{q}{m}}{\displaystyle \frac{\partial U}{\partial x}}+{\displaystyle \frac{v}{\tau }}=0,$$

(4)

$${\displaystyle \frac{\partial v}{\partial t}}-{\displaystyle \frac{1}{q}}{\displaystyle \frac{\partial j_n}{\partial x}}=0,$$

(5)

Here, $v$ is the local electron velocity, m is the effective electron mass, $\partial U/\partial x$ is the longitudinal electric field in the channel, τ is the momentum relaxation time, and $j_n$ is the current density.

HEMT devices exhibit two distinct detection modes: resonant and non-resonant detection. Resonant detection occurs when $\mathsf{\omega }\mathsf{\tau }>>1$, whereas non-resonant detection corresponds to a condition in which $\mathsf{\omega }\mathsf{\tau }<<1$, with ω representing the angular frequency of the incident wave. For the 0.1 THz frequency range examined in this study, the non-resonant detection criterion is satisfied.

When an alternating current (AC) signal is applied as the electrical boundary condition (illustrated in Figure 6) and the channel length L satisfies the long-channel condition, plasma waves generate a direct current (DC) component ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ between the source and the drain. This component is expressed as

$$U_{dc}={\displaystyle \frac{U_w^2}{4U_0}}\left[1+{\displaystyle \frac{2\omega \tau }{\sqrt{1+{\left(\omega \tau \right)}^2}}}\right],$$

(6)

The long-channel criterion is given by

$$L>>s\sqrt{\tau /\omega },$$

(7)

$U_w$ is the amplitude of AC signal, and $U_0=V_{gs}-V_{th}$, where $V_{th}$ means the threshold voltage of the device. For a signal frequency of 0.1 THz, devices with a channel length exceeding 600 nm can be classified as long-channel devices.

Following the theoretical framework of this study, the drain was configured with a current boundary condition—one of the fundamental modes for simulating terahertz detection. Trap densities were initialized at $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$, ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-4 \mathrm{V}$, and ${\mathrm{V}}_{\mathrm{g}}\mathrm{A}\mathrm{C}=400 \mathrm{m}\mathrm{V}$, as specified in Figure 6. Figure 7a clearly demonstrates an alternating current oscillating at the terahertz frequency of 100 GHz. Concurrently, Figure 7b reveals a small direct current component (724.13 μV) through Fast Fourier Transform (FFT) analysis, which constitutes the characteristic signature of terahertz detection.

As demonstrated in the preceding analysis, traps exert a profound influence on the thermodynamic framework and self-heating phenomena. Given that terahertz (THz) radiation inherently contributes to power dissipation, traps further exert discernible impacts on THz detection performance. In Figure 8a, the degradation effect of traps in terahertz detection is clearly observed through the wide range of ${\mathrm{V}}_{\mathrm{g}}$. In addition, with the density of traps ($1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$), the relationship between ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ and ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}$ shows the non-monotonic signature which is widely observed in most of the terahertz simulations [29,30,31]. The maximum value of ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ is 0.12 V at ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-7.8 \mathrm{V}$ when traps of $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ are present. It is verified that the trap-free configuration exhibits characteristics similar to the case with $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ traps. Figure 8b reveals a systematic degradation of ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ across the operational range, with peak degradation (64.89% reduction) observed at ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-4.9 \mathrm{V}$ under trap-free conditions.

Equations (2) and (6) demonstrate that the discrepancy in terahertz detection performance between the $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ trap density case and the trap-free scenario arises from variations in the parameter $\tau$ (all other simulation parameters being identical). As Equation (7) indicates, GaN HEMT operation in non-resonant mode results in $\mathsf{\omega }\mathsf{\tau }\ll 1$. Under this condition, traps capture electrons, leading to increased $\mathsf{\omega }\mathsf{\tau }$ and consequently higher DC. However, the simulation results deviate from the theoretical calculations due to externally induced electric fields originating from temperature gradients, which are substantially enhanced in the $1\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ trap density model.

In the next section, the effect of different densities of traps on ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ degradation is simulated. In this section, due to the bigger ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$ near the threshold voltage, ${\mathrm{V}}_{\mathrm{g}}\mathrm{D}\mathrm{C}=-7 \mathrm{V}$ and ${\mathrm{V}}_{\mathrm{g}}\mathrm{A}\mathrm{C}=400 \mathrm{m}\mathrm{V}$ are effectted. In Figure 9, some characteristic has been clearly shown:

• A larger trap density leads to more sever degradation in ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$;
• Two maximum values are observed at trap densities of 4 $\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ and 7 $\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$.

The non-monotonic feature is novel in terahertz detection. Figure 10 shows trap densities of 4 $\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ and 4 $\times {10}^{16} {\mathrm{c}\mathrm{m}}^{-3}$. A larger oscillation voltage is observed in the device with a trap density of 4 $\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ and leads to a larger DC voltage, which is the maximum value in Figure 9.

4. Conclusions

Based on Sentaurus TCAD simulation results, key conclusions regarding the design of GaN-based HEMT power devices were derived. Traps significantly enhance the channel current by extending the carrier lifetime through the introduction of new energy levels, achieving a current gain of up to threefold at a concentration of $1\times {10}^{17} {\mathrm{c}\mathrm{m}}^{-3}$. However, when introducing traps to improve DC conduction performance, their concentration should not exceed optimal levels. At trap concentrations above $1\times {10}^{16}$ cm⁻³, current collapse severity increases sharply and pronounced device self-heating occurs. This leads to thermal accumulation in the channel, consequently degrading the power conversion efficiency.

Regarding terahertz response characteristics, concentration-dependent and gate-bias-dependent trends were observed. Increasing trap concentration generally reduces terahertz response ${\mathrm{U}}_{\mathrm{d}\mathrm{c}}$, though notable fluctuations occur near concentrations of 4 $\times {10}^{15} {\mathrm{c}\mathrm{m}}^{-3}$ and $1\times {10}^{17} {\mathrm{c}\mathrm{m}}^{-3}$. Therefore, where traps are unavoidable in device fabrication, introducing an optimized trap concentration may improve terahertz response performance. Similarly, for trap-affected devices, gate voltage selection should avoid regions of significant terahertz response degradation. When designing operating gate voltages, the simulation methodology presented herein enables modeling the terahertz response versus gate voltage characteristic under actual trap concentrations. This facilitates the characterization of terahertz response degradation across gate voltages, providing critical guidance for determining reference gate bias conditions.