High-Indium-Composition, Ultra-Low-Power GaAsSb/InGaAs Heterojunction Tunnel Field-Effect Transistors

Abstract

In this work, we report the first systematic examination of how the In composition in the intrinsic In_xGa_1-xAs layer and the p-type doping concentration in the p-type GaAsSb layer affect the device performance of side-gate p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As TFETs, using the technology computer-aided design (TCAD) simulations with a non-local band-to-band model. By tuning these two parameters, a moderate staggered alignment is achieved, enabling self-off operation at zero gate bias while maintaining high on-current. This tunability is an intrinsic and significant advantage of the p-GaAsSb/i-In_xGa_1-xAs heterojunction that has not been previously explored. It is found that the best device performance does not occur in the TFET with an In composition of 0.53 in the intrinsic layer, which is lattice-matched to the InP substrate, but rather occurs in the device with a higher In composition of around 0.59 in the InGaAs layer, which has been verified by experimental data to some extent. Optimal parameter combinations yield a minimum subthreshold swing of 13.51 mV/dec and an ON-state current of 35.39 μA/μm at V_DS = V_GS = 0.5 V due to the enhanced tunneling capability.

1. Introduction

For decades, the scaling of MOSFETs has driven continuous improvements in integrated circuit performance. With the approaching end of Moore’s law, power dissipation has become a primary constraint in further improving transistor density and computing performance. Conventional CMOS devices are fundamentally limited by the Boltzmann distribution of carriers, resulting in a minimum subthreshold swing (SS) of 60 mV/dec at room temperature [1]. To overcome this limitation, several steep-slope transistor concepts have been proposed, such as negative-capacitance field-effect transistors (NC-FETs) [2], Dirac-source FETs (DS-FETs) [3], and tunnel FETs (TFETs) [4,5,6,7,8]. While NC-FETs and DS-FETs rely on emerging materials like ferroelectrics or two-dimensional materials that remain challenging for large-scale fabrication and integration, TFETs exploit the band-to-band tunneling (BTBT) within conventional semiconductor systems, offering better compatibility with mature epitaxy and fabrication technologies [9]. The reduced SS of TFETs grants them a higher transconductance to current ratio than traditional FETs [10]. The higher output resistance offered by TFET-based designs significantly higher intrinsic voltage gain and higher maximum-oscillation frequency at low current levels [11]. Furthermore, studies have demonstrated that TFETs show great promise for various applications, including energy harvesting systems [12], biosensors [13], and Internet of Things (IoT) devices [14].

Among various TFET material candidates, the GaAsSb/InGaAs heterojunction is particularly promising because of its flexibly tunable type-II band alignment and high carrier transport properties [15,16]. By carefully adjusting the In composition in InGaAs, the conduction- and valence-band offsets can be engineered to obtain a moderate staggered alignment. This alignment provides a sufficient but not overly extreme offset to enhance tunneling probability. As a result, the device maintains strong on-current while remaining off at zero gate bias, avoiding the normally-on behavior. Meanwhile, the p-type doping in GaAsSb source controls the depletion width and further optimizes the tunneling probability, providing an additional degree of freedom for device design. This tunability is the true advantage of the GaAsSb/InGaAs TFETs; however, most of the previous simulations and experiments based on this heterojunction system focus on InP lattice-matched GaAs_0.5Sb_0.5/In_0.53Ga_0.47As only [15,17,18,19,20,21,22,23,24], which thus motivates us to investigate it thoroughly.

In this work, we perform comprehensive technology computer-aided design (TCAD) simulations using a non-local BTBT model on the side-gate p-GaAsSb/i-InGaAs/n-InGaAs p-i-n TFET structure, where the p-type GaAsSb serves as the source, the intrinsic InGaAs as the channel, and the n-type InGaAs as the drain. The influences of In composition and source doping are investigated to quantify their effects on the band alignment, tunneling current, and subthreshold characteristics. Furthermore, the p-GaAsSb/i-InGaAs/n-InGaAs TFETs have been fabricated and their device performance has been characterized. The results provide important guidance for optimizing III-V heterojunction TFETs toward manufacturable low-power applications.

2. Device Structure and Model Construction

The micrometer-sized device structure of the simulated p-GaAsSb/i-InGaAs/n-InGaAs TFET is illustrated in Figure 1a. On top of the semi-insulating InP substrate, the drain layer is 300 nm n-type In_0.53Ga_0.47As with a doping concentration of 1 × 10¹⁹ cm⁻³, followed by a 100 nm intrinsic In_xGa_1-xAs as the channel region. The source layer is a 60 nm p-type GaAs_0.5Sb_0.5. Both In_0.53Ga_0.47As and GaAs_0.5Sb_0.5 exhibit good lattice matching with the InP substrate [21], which helps to reduce interface defects and improve device performance. An Al₂O₃ interfacial oxide layer with a high dielectric constant of 9.3, as well as a sidewall gate structure, has been applied to enhance the electrostatic control over the p-GaAsSb/i-InGaAs tunneling junction. The band structure of this TFET, as illustrated in Figure 1b, will be carefully optimized to a moderate staggered alignment by adjusting the In composition in InGaAs. This optimization ensures that the device will maintain strong on-current while remaining off at zero gate bias.

The p-GaAsSb/i-InGaAs/n-InGaAs TFET model investigated in this work was constructed using commercial Silvaco TCAD software (Silvaco Victory Device V1.24.0). For the physical models, the non-local BTBT model is implemented to account for both forward and backward tunneling currents, as well as the continuous fluctuation of the bands of energy in the degenerately doped source TFET. The net current density for electrons with a longitudinal energy $E_L$ and a transverse energy $E_T$ can be given by

$$J={\displaystyle \frac{q}{\pi ћ}}\int \int T(E_L)[f_l-f_r]\rho d{E}_{T}d{E}_L$$

(1)

where $f_l$ and $f_r$ are the Fermi–Dirac function using the quasi-Fermi level from the left and right side of the junction, respectively. $\rho$ is the 2D density of states. The tunneling probability $T(E_L)$ can be calculated by using the WKB approximation:

$$T(E_L)=\exp \left(-2{\displaystyle {\int }_{x_{\mathrm{start}}}^{x_{\mathrm{end}}}k(x)dx}\right)$$

(2)

where $x_{start}$ and $x_{end}$ are the starting and ending points of the tunneling paths, respectively. The $k(x)$ can be expressed by

$$k(x)={\displaystyle \frac{k_e{k}_h}{\sqrt{k_e^2{k}_h^2}}}$$

(3)

where

$$k_e(x)={\displaystyle \frac{1}{iћ}}\sqrt{2m_0{m}_{e,tun}\left[E_L-E_C\left(x\right)\right]}$$

(4)

and

$$k_h(x)={\displaystyle \frac{1}{iћ}}\sqrt{2m_0{m}_{h,tun}\left[E_V\left(x\right)-E_L\right]}$$

(5)

where $ћ$ is the reduced Planck’s constant, $m_0$ is the rest mass of the electron, and $m_{e,tun}$ and $m_{h,tun}$ are the relative tunneling mass for electrons and holes, respectively. E_C(x) and E_V(x) are the conduction band energy and valence band energy, respectively.

Other physical models, such as Shockley–Read–Hall (SRH) and Auger recombination, energy bandgap narrowing (BGN), Fermi–Dirac Statistics, and electric-field-dependent and concentration-dependent mobility models, are also applied. The SRH recombination rate can be achieved as

$$R_{\mathrm{SRH}}={\displaystyle \frac{np-n_{\mathrm{ie}}^2}{{\tau }_{p}n+{\tau }_{n}p}}$$

(6)

where ${\tau }_n$ and ${\tau }_p$ are the electron and hole lifetime, respectively. $n$ and $p$ are the electron and hole concentrations, and $n_{\mathrm{ie}}$ is the effective intrinsic concentration. Accounting for low-field and transverse-field effects on the mobility, the carrier mobility $\mu$ is calculated by Matthiessen’s rule. Moreover, as carriers are accelerated in an electric field, their velocity will begin to saturate when the electric field magnitude becomes significant. The field-dependent carrier mobility can be defined by Caughey and Thomas’ expressions [25], which provide a smooth transition between low-field and high-field behavior

$$\mu ={\mu }_0{\left[1+{\left({\displaystyle \frac{{\mu }_0\left|E_{\parallel }\right|}{V_{\mathrm{sat}}}}\right)}^{\beta }\right]}^{-1/\beta }$$

(7)

where $\left|E_{\parallel }\right|$ is the magnitude of the parallel electric field, $V_{\mathrm{sat}}$ is the carrier saturation velocity, ${\mu }_0$ is the low-field mobility, and the low-field electron and hole mobilities of In_xGa_1-xAs can be given by the following interpolation formulas

$${\mu }_{n0}=\left[33000x+8000\times \left(1-x\right)\right]{\left({\displaystyle \frac{T_L}{300\mathrm{K}}}\right)}^{-3/2}$$

(8)

and

$${\mu }_{p0}=\left[150x+400\times \left(1-x\right)\right]{\left({\displaystyle \frac{T_L}{300\mathrm{K}}}\right)}^{-3/2}$$

(9)

Specifically, for GaAsSb and InGaAs, the main material parameters are already included in the database of the software, and the CUBIC35 Model is considered, which allows the consistent modeling of the energy bandgaps, electron affinity, conduction band density of states, valence band density of states, and relative dielectric permittivity of III-V binary and ternary compounds using interpolation formulae. The dependence on lattice temperature is also included. Although the interface traps are known to impact the device performance of TFETs, the interface states only near the conduction band in the channel layer can affect the electrical characteristics of n-TFETs (electron tunneling dominates). Furthermore, their effects on the TFETs are smaller than those on the normal MOSFETs [17]. Consequently, to simplify the discussion in this work, only the tunneling mechanism is incorporated into the simulation, while other defect-associated currents and interface states are not considered.

3. Results and Discussion

3.1. Simulation Results

Figure 2 shows the room-temperature transfer characteristics (I_DS-V_GS) and device performance of the p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As TFETs with different In compositions, under a fixed source doping level of 2 × 10¹⁹ cm⁻³. To align with next-generation logic technology, a low V_DS voltage of 0.5 V was set for conducting the device evaluation [15]. It can be found that the ON-state current (I_ON) and OFF-state current (I_OFF) increase with the In composition increases from 0.49 to 0.61. The increase in I_ON is mainly due to the bandgap shrinkage of i-In_xGa_1-xAs with a higher In composition [23]. A smaller bandgap directly reduces the offset between the p-GaAs_0.5Sb_0.5 valence-band and i-In_xGa_1-xAs conduction-band, and thus lowers the tunneling barrier, as can be observed in Figure 2c.

Table 1. Variation in main parameter values with In composition.

Position (μm)	In Composition of i-InxGa1-xAs	Conduction Band Energy of i-InxGa1-xAs (eV)	Valence Band Energy of p-GaAs0.5Sb0.5 (eV)	Tunneling Barrier (eV)
−0.4	0.49	0.250	0.083	0.167
	0.51	0.233		0.150
	0.53	0.217		0.134
	0.55	0.200		0.117
	0.57	0.185		0.102
	0.59	0.169		0.086
	0.595	0.165		0.082
	0.6	0.161		0.078
	0.61	0.153		0.070

also lists the variation in main parameter values with In composition. This leads to an exponential increase in tunneling probability, which is described by Equations (2)–(5). Meanwhile, the increasing tunneling probability also impacts the undesirable tunneling under the Off-state bias, elevating the I_OFF. In Figure 2b, when the In composition exceeds 0.59, the I_ON increases at a decelerated rate and tends to saturate, while the I_OFF rapidly increases. This leads to an abrupt switching transition, resulting in a degradation of the SS, which can also be observed in Figure 2b. When the In composition is less than 0.59, the SS of the studied TFETs is below 25 mV/dec with minor fluctuations and gradually decreases as the In composition increases from 0.53 to 0.59. Beyond this point, the SS rapidly rises to over 97 mV/dec. Regarding the optimal parameter, the TFET achieves the minimum SS of 13.51 mV/dec and an I_ON of 35.39 μA/μm at V_GS = 0.5 V, when the In composition is 0.59.

Figure 3 presents the room-temperature transfer characteristics (I_DS–V_GS) and device performance of the p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As TFETs with different p-type doping concentrations of the GaAs_0.5Sb_0.5 layer at V_DS of 0.5 V. In Figure 3, as the doping concentration of GaAs_0.5Sb_0.5 varies from 9 × 10¹⁸ cm⁻³ to 2 × 10¹⁹ cm⁻³, I_ON continues to increase with doping concentration. This is because a higher doping level induces a steeper band profile at the tunneling junction, directly narrowing the tunneling barrier and thereby increasing the tunneling probability. When the doping reaches a higher level, such as 2.5 × 10¹⁹ cm⁻³, carrier degeneracy and electrostatic screening suppress further band bending, while bandgap narrowing reduces the built-in potential. As a result, the local electric field no longer improves, and the tunneling path reaches its minimum thickness, causing the I_ON to saturate. With regard to I_OFF, as the doping concentration increases, it remains almost unchanged up to 2 × 10¹⁹ cm⁻³ and then degrades rapidly over 2.25 × 10¹⁹ cm⁻³. This behavior arises because, in the OFF-state, the band offset is large and the tunneling probability stays low even though the barrier becomes narrower. However, another effect occurs simultaneously: increasing the source doping also raises the valence-band edge and induces the bandgap narrowing. These two effects jointly reduce the tunneling barrier width and height. Once the doping reaches 2.25 × 10¹⁹ cm⁻³, this lowering effect becomes significant, leading to a rapid increase in band-to-band tunneling leakage and a pronounced rise in I_OFF. Furthermore, in Figure 3b, due to the aforementioned variations in I_ON and I_OFF, the SS of p-GaAs_0.5Sb_0.5/i-In_0.59Ga₄₁As/n-In_0.53Ga_0.47As TFETs gradually decreases with the doping concentration of the GaAs_0.5Sb_0.5 layer increasing from 9 × 10¹⁸ cm⁻³ to 2 × 10¹⁹ cm⁻³, and then steeply rises to 174.99 mV/dec when the doping concentration increases to 2.5 × 10¹⁹ cm⁻³.

3.2. Experiment Validation

Figure 4 shows the simulated minimum SS and I_ON of the p-GaAs_0.5Sb_0.5/i-In_0.59Ga_0.41As/n-In_0.53Ga_0.47As TFET with the doping concentration of the GaAsSb layer of 2 × 10¹⁹ cm⁻³, and the reported simulation results and our experimental data are also given [21,26]. The reported minimum SS of GaAs_0.5Sb_0.5/In_0.53Ga_0.47As TFETs is in close agreement with our simulated results in Figure 4a, and the reported I_ON has an acceptable discrepancy relative to our simulated data in Figure 4b, which confirms the validity of our simulation. Moreover, previous works mostly choose the InP lattice-matched In_0.53Ga_0.47As as the channel material [15,17,18,19,20,21,22,23,24], because the lattice mismatch between different epitaxial layers and the substrate introduces strain [27], which relaxes by forming crystal defects at the interface and within the material itself. These defects act as trapping centers that severely degrade device performance. Although the lattice mismatch between In_0.59Ga_0.41As and InP is approximately 0.4% [28], the experimental results show that for a lattice mismatch of ±1% or less, the crystalline quality of epitaxial layers of InGaAs remains high with a thickness of 100 nm [29]. Therefore, the lattice-mismatch effects on the device performance are almost negligible. Our simulation results show that the p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As TFET achieves better device performance when the In composition is 0.59. Compared to the InP lattice-matched p-GaAs_0.5Sb_0.5/i-In_0.53Ga_0.47As/n-In_0.53Ga_0.47As TFET under the same doping density of GaAsSb layer, the minimum SS of p-GaAs_0.5Sb_0.5/i-In_0.59Ga_0.41As/n-In_0.53Ga_0.47As TFET is reduced by about 45.74%, and the I_ON (V_GS = 0.5 V) is increased by about 261.49%.

To further confirm this trend in the device’s performance, the epitaxial p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As structure was grown by molecular beam epitaxy (MBE) on semi-insulating InP substrate, and then fabricated into TFETs. The fabrication process starts with lithography and ICP-RIE etching to form the mesa. After that, an atomic-layer-deposition (ALD) Al₂O₃ was deposited as the gate oxide, followed by BOE etching of the oxide and source-/drain- metal deposition. Then the source and drain electrodes were patterned using the lift-off technique. At last, gate metal was deposited to complete the process. The layout structure, optical image, and transfer characteristics of actual devices are shown in Figure 5. The device structure of the simulation model is identical to that of the actual device. All the fabricated TFETs were measured from negative to positive bias under the same conditions. In our experiments, we intentionally selected one In composition lattice matched to the InP substrate and one In composition lattice mismatched to the InP to verify the validity of our model. The transfer characteristics of p-GaAs_0.5Sb_0.5/i-In_xGa_1-xAs/n-In_0.53Ga_0.47As TFETs are presented in Figure 5c; the I_ON of the device with higher In composition (0.58) is clearly better than that of the devices with lower Indium composition (0.53). Moreover, in Figure 4, according to the measurement results of several representative devices, the experimental minimum SS and I_ON of the p-GaAs_0.5Sb_0.5/i-In_0.58Ga_0.42As/n-In_0.53Ga_0.47As TFET are both better than those of the InP lattice-matched p-GaAs_0.5Sb_0.5/i-In_0.53Ga_0.47As/n-In_0.53Ga_0.47As TFET. Both the experimental and simulation results demonstrate that within the In composition range of 0.53 to 0.58, TFETs with a higher In composition exhibit superior device performance. This provides evidence that the moderate staggered alignment of the energy band in the p-GaAsSb/i-InGaAs heterojunction is likely to dominate the device performance, and the lattice mismatch does not constitute a primary factor influencing the minimum SS and I_ON of GaAsSb/InGaAs TFETs. Owing to the current limited experimental conditions, the experimental dataset will be expanded in our subsequent work to include results for a wider range of In compositions. In addition, a deviation can be observed between the simulation results and the experimental data. This is mainly because the idealized 2D device model utilized in the simulation is unable to capture the full complexity of the fabricated 3D structure. The material parameters used in the simulation do not perfectly match the properties of the actual fabricated materials, and the inherent process variations in fabrication also contribute to this deviation.

4. Conclusions

In summary, this work systematically investigates how In composition in intrinsic InGaAs and p-type doping concentration in GaAsSb affect the performance of side-gate p-GaAs_0.5Sb_0.5/i-In_0.59Ga_0.41As/n-In_0.53Ga_0.47As TFETs through TCAD simulations. By tuning the In composition, a moderate staggered band alignment is achieved, enabling self-off operation at zero gate bias while maintaining strong tunneling conduction. The optimized device performance is obtained at an In composition of approximately 0.59 and a GaAsSb p-type doping level of about 2 × 10¹⁹ cm⁻³, yielding a minimum SS of approximately 13.51 mV/dec, and an I_ON of 35.39 μA/μm at V_DS = V_GS = 0.5 V. Moreover, the device performance of the p-GaAs_0.5Sb_0.5/i-In_0.59Ga_0.41As/n-In_0.53Ga_0.47As TFET is notably superior to that of the InP lattice-matched p-GaAs_0.5Sb_0.5/i-In_0.53Ga_0.47As/n-In_0.53Ga_0.47As TFET, which has also been validated by experiments to a certain extent. These results offer the most promising and practically manufacturable material system for high-performance III-V TFETs, a clear design guideline. It is an important step toward large-scale, ultra-low-power electronic applications.