Numerical Investigation of Short-Channel Effects and RF Performance in Top-Gate In2O3 Thin-Film Transistors Authors: Hanbo Xu, Mingyang Zhu, Zeen Fang, Lei Zhang Published: 2026 ABSTRACT Indium oxide (In2O3) has recently emerged as a promising semiconductor for advanced electronics due to its high electron mobility and wide bandgap. In this article, the lateral scaling characteristics of top-gate In2O3 thin-film transistors (TFTs) featuring a 1.5 nm thick channel and a 7 nm thick HfO2 gate dielectric are investigated by two-dimensional device simulation. The analysis covers short-channel effects, DC characteristics, transconductance behavior, and small-signal radio frequency (RF) metrics across a gate-length (LG) range of 20 nm to 700 nm. Simulation results identify a critical gate length near 100 nm for the transition from long-channel to short-channel behavior. For LG ≤ 100 nm, pronounced short-channel effects emerge, featuring a significant negative VTH shift and a drain-induced barrier lowering (DIBL) coefficient up to ~130 mV/V. A non-classical gm scaling behavior is observed, where gm_max initially increases with LG, then remains within a narrow range and eventually evolves toward the conventional long-channel trend. Further analysis of the lateral electric field distribution, field-dependent mobility, and transconductance efficiency indicates that this behavior originates from a crossover between short-channel field-assisted transport and gate-controlled channel modulation. The devices show strong RF potential, with fT and fmax reaching 124.32 GHz and 157.64 GHz, respectively, at LG = 20 nm. The high-mobility In2O3 channel leads to a less distinct fT scaling transition from the classical 1/L2G dependence to the short-channel 1/LG dependence, while fmax scaling evolves through different regimes governed by capacitance-related limitations, intrinsic transport enhancement, and short-channel non-idealities. This work provides physical insight into the lateral scaling behavior of ultrathin top-gate In2O3 TFTs and highlights their potential for high-frequency and power-dense applications. 1. Introduction Metal oxide semiconductors have attracted extensive interest for next-generation electronics because they combine several attractive features, including wide bandgap, process compatibility with large-area and low-temperature fabrication, and the potential for high electron transport in ultrathin channels [1,2]. Compared with conventional semiconductor platforms, oxide semiconductors are also well-suited for transparent, flexible, and back-end-of-line-compatible device applications. Within this material family, In2O3 is particularly attractive because it can provide relatively high electron mobility and high current-driving capability even in atomically thin channels, making it a strong candidate for aggressively scaled thin-film transistors and high-frequency devices [3,4]. With the rapid development of atomic layer deposition (ALD), atomically thin In2O3 channels have become increasingly feasible, which has stimulated growing interest in ultrathin In2O3 FETs for both logic-oriented and radio frequency (RF)-oriented applications [5,6]. Currently, ALD In2O3 back-gate FETs stand out due to their excellent conformality and uniformity on three-dimensional structures enabled by the ALD process, as well as high electron mobility, high on-current, and remarkable reliability [7]. Back-gate-structured In2O3 field-effect transistors have achieved significant progress. Notably, back-gate In2O3 FETs have achieved record-high drain currents exceeding 2000 mA/mm with aggressive channel length scaling down to 40 nm [8] and have demonstrated high-performance operation even with ultrathin channel thicknesses below 1 nm [9]. However, from the perspective of practical device integration, top-gate In2O3 FETs have attracted increasing attention, although early top-gate devices suffered from high off-state current, negative threshold voltage, and difficulty in achieving complete turn-off. Nevertheless, recent studies have demonstrated that the inherent defects in top-gate In2O3 FETs can be significantly mitigated through appropriate interface engineering. For instance, a recent work successfully developed high-reliability In2O3 top-gate thin-film transistors via in situ surface energy engineering in the ALD process, revealing that the performance limitations previously associated with the top-gate structure can be markedly improved [10]. Significantly, In2O3 TFTs subjected to an optimized oxygen plasma treatment duration of 3 s exhibited significant improvements in electrical characteristics, including a high-field-effect mobility of 84.3 cm2/V·s, a steep subthreshold swing of 76.8 mV/dec, etc. [10]. From these perspectives, top-gate In2O3 devices are of particular interest because they combine the intrinsic transport advantages of ultrathin In2O3 with a gate-stack configuration more relevant to practical device integration. Owing to its high output current density, the top-gate In2O3 transistor is highly suitable for radio frequency (RF) power devices. Recent demonstrations of top T-gate BEOL-compatible In2O3 FETs have achieved a record fₘₐₓ of 70 GHz, along with high current density and transconductance, showcasing their potential for future RF and millimeter-wave applications [11]. Furthermore, top-gate In2O3 transistors show excellent scalability and high current drive, highlighting their potential for high-speed applications [12]. However, in the study of RF power transistors, as device dimensions scale down, short-channel effects become more pronounced. Therefore, understanding short-channel behavior in top-gate In2O3 transistors has become increasingly important. Herein, motivated by recent advances in In2O3 top-gate TFTs, a systematic numerical study on the short-channel effects in these devices is presented to demonstrate their potential for high-frequency operation. This study is conducted using a two-dimensional device simulation tool (Silvaco Atlas 2020), and the short-channel effects as well as the relationship between the fT and the LG are analyzed. 2. Simulation 2.1. Calibration To establish experimentally grounded material and transport parameters for the TCAD simulation, the In2O3-related parameter set was first calibrated using a previously reported back-gate In2O3 FET [13]. This calibration step was not intended to equate the back-gate architecture with the target top-gate device studied here. Instead, it served as an experimentally accessible starting point for determining the effective In2O3 transport and interface-related parameters used in the subsequent top-gate simulation. In the present work, the term “calibration” refers to the extraction of effective material, transport, and interface-related parameters for the TCAD framework by matching the simulated output and transfer characteristics to those of a reported experimental In2O3 device. The device geometry and stack information available from the reference were fixed first, while a limited set of parameters, including the low-field mobility, effective channel doping level, interface-trap-related quantities, fixed interface charge, and saturation velocity, were adjusted within physically reasonable ranges. The fitting targets were the drain current magnitude, threshold voltage trend, and overall shapes of the output and transfer characteristics. The calibrated parameter set was then transferred to the top-gate device platform for the scaling analysis in this work. The top-gate device investigated in this work was then constructed independently using the gate-stack geometry, interface conditions, and contact configuration defined in this study. The reported back-gate device was selected as the primary calibration anchor because its short-channel dimensions are more relevant to the present scaling-oriented analysis than the currently available long-channel top-gate experimental reports. Although recent top-gate In2O3 TFT studies provide valuable experimental guidance, the available top-gate dataset used in Ref. [10] corresponds to a much longer channel length (LG = 20 μm) and was therefore not adopted as the primary calibration target for the present short-channel scaling study. The parameters listed in Table 1 were adjusted to obtain the calibration results shown in Figure 1c,d. Compared with the calibration plots in Ref. [13], although an exact point-by-point match is not expected in a device-level TCAD framework, the calibrated model reproduces the main current level, threshold trend, and overall output and transfer characteristics of the reference device. The extracted parameter set (Table 1) was subsequently applied to the top-gate device platform used in this work for scaling analysis. 2.2. Device Structure The three-dimensional structure of the simulated device is depicted in Figure 1a. For the target top-gate device platform studied in this work, a sapphire substrate was selected as an electrically insulating supporting substrate. This choice was made to reduce substrate-induced parasitic coupling and to provide a stable platform for analyzing the intrinsic influence of gate-length scaling on short-channel electrostatics and RF metrics [14,15]. The purpose of the present study is not to reproduce a specific previously reported substrate stack one-to-one [10,13], but to investigate the scaling behavior of a clearly defined top-gate In2O3 device platform. For comparison, an additional substrate sensitivity result is provided in the Supplementary Materials [10,13]. Owing to its multiple advantages as a gate dielectric for In2O3 FETs, including a high dielectric constant (high-κ) for enhanced electrostatic control and a reported low interface trap density in HfO2/In2O3 gate stacks [16,17], a 7 nm HfO2 layer was employed as the gate oxide. The gate, source, and drain electrodes were formed of nickel, with a work function of 5.1 eV. The source and drain were defined as ohmic contacts. Figure 2a presents resistance as a function of the electrode spacing deduced from a transmission line model (TLM) test structure simulated by our top-gate In2O3 FET with the gate removed. As depicted in Figure 2a, a low contact resistance (Rc) of 0.15 Ω·mm was derived by TLM. The ultra-low contact resistance of 0.15 Ω·mm indicates well-configured ohmic contact parameters. For devices of all gate lengths in this work, the source-to-gate and drain-to-gate edge distances (i.e., the gate–source spacing LGS and gate–drain spacing LGD) vary with gate length. This is because the source-to-drain distance LSD is fixed at 1 μm for all devices, so varying the gate length inevitably alters LGS and LGD, representing a non-standard scaling approach. Since the gate is centered, the relationship can be derived as LGS = LGD = (LSD − LG)/2. 2.3. Physical Model For the top-gate In2O3 FET, the non-ideal HfO2/In2O3 gate interface was modeled by introducing acceptor-like interface traps and fixed positive charges. The trap level was set at approximately 0.5 eV below the conduction band minimum, with a trap density of 3 × 1011 cm−2, electron and hole capture cross-sections of 1 × 10−14 cm2, and a degeneracy factor of 2. In addition, a fixed positive interfacial charge density of 1 × 1010 cm−2 was included. These quantities were used as calibration-guided effective interface parameters based on literature-reported experimental characteristics [18]. By contrast, the In2O3/sapphire bottom interface was treated as electrically passive in the present top-gate framework, because the gate electrostatics are primarily governed by the HfO2/In2O3 top interface. Owing to the ultra-high channel doping concentration (2.5 × 1020 cm−3) and the ultrathin 1.5 nm In2O3 body, the channel operates in a highly degenerate carrier concentration regime. Therefore, Fermi–Dirac carrier statistics were enabled in the Atlas simulation. The parallel field transport was described using the Caughey–Thomas field-dependent mobility (FLDMOB) formulation, because it provides an effective and physically interpretable representation of mobility degradation and carrier-velocity saturation under the strong lateral electric fields relevant to the present scaling study. In the FLDMOB model, expressed in Formula (1), μn(E) is the field-dependent electron mobility, the carrier drift velocity is defined as vn = μn(E)∙E, and E is the parallel electric field:(1)μn(E)=μ0⋅vsatvsat+μ0⋅E here, the adopted low-field mobility (μ0 = 120 cm2/V·s) and electron saturation velocity (vsat = 1 × 107 cm/s) were selected within experimentally reported ranges for In2O3-based devices [19] and were further refined during the calibration procedure. In this sense, the FLDMOB-based transport description is used here as a transport representation for trend analysis rather than as a complete non-equilibrium transport formalism. Considering that the 1.5 nm thick In2O3 channel is already atomically thin, quantum confinement is expected to affect the electrostatic and carrier-distribution behavior. Therefore, a density-gradient-based quantum correction was included in the TCAD framework as an approximate quantum confinement treatment on top of the field-dependent mobility model. Figure 3a,b demonstrate the transfer and output characteristics of the top-gate In2O3 TFT in this work with and without quantum correction at LG = 120 nm. It can be observed that when the quantum confinement effect is included, the drain current decreases in both the linear and saturation regions, leading to an overall downward shift of the curves. In addition, the slope in the linear region is reduced. These changes are mainly attributed to the fact that in nanoscale devices, the quantum confinement effect shifts the peak of the carrier distribution away from the interface and causes an effective band broadening, thereby modifying the electrical behavior of the device. The present transport simulation remains based on a drift-diffusion-type framework with field-dependent mobility and quantum correction. Although quasi-ballistic effects may become relevant in atomically thin In2O3 films under aggressive scaling, they were not explicitly included in the present TCAD framework and are therefore regarded as a limitation of this work. In addition, impact ionization was included through the Selberherr model to avoid neglecting high-field carrier-generation effects under large VDS and short-channel conditions. Its influence is considered as a supporting high-field correction in this work rather than as an independent focus of this study. The SRH carrier lifetime was set to 10 ns as a calibration-guided simulation parameter used to represent defect-assisted recombination within the present TCAD framework. The selected lifetime mainly serves as an effective recombination parameter within the calibrated model rather than a directly measured material lifetime [20]. The transport analysis in this work focuses on electron-dominated conduction in n-type In2O3 TFTs and is based on a drift-diffusion-type formulation. Direct source-to-drain tunneling, gate leakage, hydrodynamic transport, and self-heating effects were not explicitly included. These effects are expected to become more relevant under more aggressive scaling or higher-field operating conditions and are therefore identified as limitations of the current framework and important topics for future work. After all physical models have been applied, the device band diagram is exported from the software. The extracted corresponding energy band diagram and the electron density profile under the gate are presented in Figure 1b and are consistent with the In2O3/HfO2 interface band characteristics experimentally characterized by Tung-Yuan Yu et al. [21]. Specifically, the output characteristics of the simulated device with LGS:LG:LGD = 440:120:440 (units in nm) are presented in Figure 2b, which are in good agreement with the typical output characteristics of a conventional MOS-FET. 3. Simulation Results and Discussion 3.1. Direct Current Characteristics Figure 4a presents the output characteristics at different gate lengths (VGS = 0 V) and Figure 4b shows the transfer characteristics in the saturation region (VDS = 6 V). As device dimensions scale down, the effective channel length decreases, leading to an increase in output current that no longer exhibits saturation. As the gate length (LG) decreases, the on-resistance (Ron) is reduced due to the lateral scaling of the device. Under a constant gate bias, the electric field strength in the short channel becomes higher compared to that in devices with longer LG. This enhanced electric field accelerates the drift velocity of electrons in the channel, consequently leading to an increase in the saturation drain current (IDsat). This finding is consistent with previous numerical studies on H-terminated diamond field-effect transistors (H-diamond FETs) reported in [20]. It is noteworthy, however, that the Ron of the top-gate In2O3 FET is evidently lower than that of the H-terminated diamond FET (H-diamond FET). This is attributed to the relatively high electron mobility (~120–140 cm2/(V·s)) in ultrathin indium oxide, which is significantly higher than the hole mobility in H-diamond. The high electron mobility renders the intrinsic saturation drain voltage (VDsat) of the top-gate In2O3 FET responsive to the increase in lateral electric field upon device scaling. As observed in Figure 4a, the position of the knee voltage varies with gate length and does not align vertically. This misalignment becomes particularly pronounced when entering the short-channel regime (LG < 100 nm), where a further reduction in gate length leads to significant shifts in the knee voltage. Compared with the intrinsic VDsat of 3–4 V for devices with LG > 100 nm, that of short-channel devices even exceeds 6 V. The saturation voltage of short-channel devices warrants reexamination, as the continual increase in drain current with Vd rising from 4 to 6 V constitutes a manifest indication of short-channel effects. On the other hand, the total contact resistance, comprising the source resistance (RS) and drain resistance (RD), is approximately 0.30 Ω·mm. This value is significantly lower than that reported in Ref. [20], and thus the contact resistance does not substantially degrade the output current. For short-channel top-gate In2O3 FETs, while a relatively large intrinsic VDsat may pose challenges, it can also be indicative of certain favorable device characteristics under specific conditions [22]. The high electron density of In2O3 is an advantage, facilitating high transconductance (gm) and cutoff frequency (fT) [23], yet it is accompanied by a relatively low breakdown field strength for the device [24]. The simulation predicts an extremely high intrinsic VDsat; however, in practical applications, self-heating and breakdown limitations must be considered, which will be the focus of future work. The top-gate In2O3 FET with a gate length of 120 nm was biased in the on-state. Figure 4c shows the linear region (VDS = 0.1 V) transfer characteristics along with the extracted gm curve. From this, the linear threshold voltage (VTH_lin) is extracted based on the formula VTH_lin = VTH − 0.5|VDS|, where VTH is determined as the gate-voltage intercept of the tangent line drawn at the point of maximum slope on the transfer curve. Figure 4d displays the saturation region (VDS = 6 V) transfer characteristics and the extracted gm curve, from which the saturation threshold voltage (VTH_sat) is determined. Here, VTH_sat is defined as the gate-voltage intercept of the tangent line drawn at the point of maximum slope on the gm curve. As shown in Figure 5a, with a gate oxide thickness of 7 nm and gate length greater than 150 nm, the threshold voltage in the linear region remains nearly stable, ranging from −1.9 V to −1.8 V, while in the saturation region, it stabilizes between approximately −2.1 V and −2.0 V. A clear trend observed in Figure 5a is that progressively decreasing LG below 150 nm results in a correspondingly larger shift in VTH. Notably, when the gate length is scaled down below 100 nm, the absolute shift in threshold voltage (|ΔVTH|) for both the linear and saturation regions exhibits a near-linear dependence on the inverse gate length (1/LG). This dependence is consistently more pronounced in the saturation region than in the linear region. Figure 5b elucidates several potential mechanisms underlying this behavior. Drain-induced barrier lowering (DIBL) refers to the phenomenon in short-channel devices where a high VDS effectively lowers the potential barrier between the source and the channel, enabling the device to turn on at a lower gate voltage. Quantitatively, it is defined as the ratio of the threshold voltage shift to the drain voltage difference (DIBL = |ΔVTH|/|ΔVDS|). In long-channel devices, the channel potential is governed exclusively by the gate electrode through the vertical electric field. As the channel length shortens, the lateral electric field generated by the high drain voltage significantly affects the barrier height at the source end [25]. Therefore, for gate lengths greater than 150 nm, the channel is sufficiently long such that the influence of the drain voltage is negligible. Consequently, the VTH is dominated by and remains stable with respect to the gate voltage. In the saturation region (VDS = 6 V), the high drain voltage establishes a strong electric field at the drain terminal. The field lines extend toward the source, effectively lowering the gate voltage required for turn-on. Compared to the linear region (VDS = 0.1 V), this mechanism manifests as a negative shift in the VTH. In shorter channels, the influence of the drain on the source becomes stronger. Consequently, a more negative gate voltage is required to compensate for this drain-induced assistance to achieve device turn-off, leading to an increased magnitude of the negative VTH shift. From another perspective, the drain voltage in the linear region is very low (VDS = 0.1 V), resulting in a weak lateral electric field. In contrast, the high drain voltage in the saturation region (VDS = 6 V) generates a much stronger lateral field, whose effect on lowering the source-end potential barrier is substantially greater than that in the linear region. When the channel becomes extremely short (LG < 100 nm), the influence of the drain becomes predominant. Its dependence on the gate length converges to a stable, linear trend, indicating that the device has entered a pronounced short-channel regime where the VTH is strongly modulated by the drain voltage. As observed in Figure 5b, when the gate length is scaled down below 100 nm, the DIBL effect becomes significant, with its coefficient increasing from approximately 6 mV/V in the long-channel regime to about 130 mV/V. Figure 6 presents the variation in saturation region gm with VGS for a fixed source-to-drain spacing (LSD). Following the maximum slope method for VTH_sat extraction described previously, it is observed that the absolute value of the saturation threshold voltage (|VTH_sat|) decreases progressively with increasing LG. Consequently, a less negative gate voltage is required to turn on the device. This is because a longer gate length increases the gate area, thereby enhancing the control of the channel potential by the vertical electric field. Furthermore, as analyzed above, the reduction in |VTH_sat| is directly linked to the significant attenuation of the DIBL effect. Although the simulated devices exhibit strong current-driving capability, the transconductance does not increase indefinitely with current density. This is because gm is determined not only by the drain-current level but also by how efficiently the gate voltage modulates the channel charge and carrier velocity. In the present top-gate In2O3 TFTs, the gm behavior is jointly influenced by the electrostatic coupling of the 7 nm HfO2 gate stack, the highly doped ultrathin channel, short-channel drain field effects, and velocity-saturation-related transport. As a result, a high current density does not automatically translate into a proportionally high gm. This indicates that the relatively moderate gm is mainly limited by the efficiency of gate-controlled charge modulation and high-field-transport degradation, rather than by the available carrier density alone. From a device design perspective, several routes can be expected to further improve gm: reducing the effective oxide thickness to strengthen gate control and increase channel-charge modulation efficiency; improving the HfO2/In2O3 interface quality and reducing interface traps to enhance gate coupling and suppress parasitic degradation of transconductance; and lowering source/drain-access resistance while optimizing contact/gate alignment to further improve the effective gm observed in RF operation. Hence, the present results suggest that gm in top-gate In2O3 TFTs remains closely tied to electrostatic design and interface/contact engineering, rather than being determined by carrier density alone. To evaluate the possible gm improvement through enhanced gate coupling, the 4 nm HfO2 case was selected as a representative thinner-dielectric condition. As discussed previously, Figure 7 presents the gm–VGS characteristics of top-gate In2O3 TFTs with different gate lengths after the HfO2 thickness is reduced from 7 nm to 4 nm. Compared with the 7 nm HfO2 case (Figure 6), the transconductance of the short-gate-length devices increases from approximately 340 mS/mm to 375 mS/mm. This indicates that adopting a thinner gate dielectric can enhance gate control capability and thereby increase gm. However, further reduction of the gate dielectric thickness may also increase gate leakage and electric field stress, so the improvement in gm should be balanced against reliability considerations. Concurrently, it is noteworthy that the trend presented in Figure 8a, which is derived from the maximum transconductance values of Figure 6, differs from that of conventional silicon-based MOSFETs:(2)gm∝(μ0CG)(VGS−VTH)/LG According to the well-established long-channel relationship, the maximum transconductance (gm_max) is generally expected to decrease with increasing LG [26]. However, the trend extracted from Figure 8a for the present top-gate In2O3 FETs is clearly non-classical: gm_max first increases with LG, then remains within a relatively narrow range of about 320–340 mS/mm over an intermediate gate-length range, and only at larger LG does it begin to recover the conventional decreasing tendency. This behavior suggests that the gm_max scaling in the ultrathin In2O3 channel is better interpreted as a crossover between different electrostatic and transport regimes rather than by the classical 1/LG scaling law alone. In the very-short-channel regime, the channel behavior is strongly affected by drain-field-assisted transport and short-channel electrostatics, which weakens the direct applicability of the long-channel transconductance scaling picture. As LG increases from this regime, the gate gains stronger control over the channel potential and the channel-charge modulation becomes more effective, so gm_max can initially increase instead of decreasing. Over an intermediate LG range, the combined influence of strong gate coupling, high carrier concentration, and velocity-saturation-related transport causes gm_max to exhibit a plateau-like behavior rather than a pronounced geometrical scaling dependence. When LG becomes sufficiently large, the device gradually approaches a more classical drift-diffusion-dominated regime, and the expected reduction of gm_max with increasing gate length becomes more evident. Therefore, the non-monotonic/weakly saturated gm_max evolution observed in Figure 8a is more appropriately understood as a transport/electrostatic crossover specific to the highly doped ultrathin In2O3 channel rather than as a violation of MOSFET scaling physics. Figure 8b shows that the VGS position of gm_max shifts progressively toward more positive values as LG increases. This trend is consistent with the evolution from stronger short-channel drain field assistance to stronger gate-controlled channel modulation. In shorter devices, the high drain field contributes more strongly to carrier injection and channel transport, so the condition for reaching peak transconductance can be achieved at a relatively less positive gate bias. As LG increases and the DIBL-related drain assistance is reduced, a larger gate overdrive is required to establish the charge modulation condition corresponding to maximum transconductance, leading to the observed positive shift of the VGS position of gm_max. Taken together, Figure 8a,b indicate that channel length plays a key role in governing the transition from short-channel field-assisted transport behavior toward a more classical long-channel transconductance regime. This interpretation is further supported by the lateral electric field distribution, field-dependent mobility, and transconductance efficiency analysis presented in Figure 9. To further clarify the non-classical evolution of gm_max with gate length, the lateral electric field distribution Ex(x) and the corresponding field-dependent electron mobility μn(|Ex|) were examined for representative devices with LG = 40, 120, and 500 nm under VGS = 2 V and VDS = 6 V, as shown in Figure 9a–c. In all cases, a pronounced high-field region is observed near the drain-side gate edge, accompanied by a clear local reduction in μn, confirming that high-field transport and field-induced mobility degradation play an important role under saturation operation. The position of the mobility minimum follows the position of the maximum magnitude of the lateral electric field, indicating that the local carrier transport is strongly modulated by the lateral field distribution along the channel. In addition, the transconductance efficiency gm/ID was evaluated as a function of overdrive voltage for different gate lengths, as shown in Figure 9d. The gm/ID curves decrease monotonically with increasing overdrive voltage for all devices, while the overall efficiency follows the trend of 500 nm > 120 nm > 40 nm. This indicates that longer-channel devices maintain more efficient gate control over channel charge, whereas the ultrashort-channel device suffers from stronger short-channel electrostatic degradation and drain field interference, which reduce the charge modulation efficiency. Therefore, the evolution of gm_max with LG should be understood as the result of a crossover between local high-field transport effects and gate-controlled channel modulation rather than by the classical 1/LG scaling law alone. In particular, the reduced gm/ID of the 40 nm device suggests that the increased current drive in the ultrashort-channel regime is partially offset by degraded transconductance efficiency. 3.2. Small-Signal Frequency Characteristics To analyze the RF behavior of the simulated top-gate In2O3 TFTs under a controlled TCAD framework, a simplified high-frequency small-signal equivalent circuit was adopted, as shown in Figure 10. The devices with different gate lengths were biased in the saturation region at VDS = 6 V. The purpose of this equivalent-circuit analysis is to compare the intrinsic scaling trends of the capacitance network and transconductance rather than to provide a fully extrinsic circuit-level prediction. Accordingly, several extrinsic non-ideal elements, including RS, RD, RDS, and CDS, were intentionally omitted in the present RF analysis so that the dependence of the intrinsic small-signal quantities on gate-length scaling could be examined more transparently. In this simplified circuit, CgsT denotes the total gate–source capacitance, CgdT denotes the total gate–drain capacitance, and RL is the load resistance, which was taken as 50 Ω for the high-frequency analysis. Based on the TCAD-extracted capacitance network, the relevant capacitance components were obtained for each gate length. Taking the device with LG = 120 nm as an example, the extracted CgsT and CgdT are shown in Figure 11a. Combined with the transconductance data extracted from Figure 4d, the corresponding fT was then calculated according to the following formulas. Therefore, the RF discussion presented below should be understood as an intrinsic trend analysis within the present TCAD framework, while additional extrinsic parasitic and thermal effects remain beyond the scope of this work. A more comprehensive extrinsic RF treatment including additional resistive, thermal, and non-equilibrium effects will be valuable for future quantitative circuit-level studies. According to KCL, we have:(3)Ii = jωCgsTVgs + jωCgdTVgs − Vd Similarly, the sum of currents at the output drain terminal is given by:(4)VdRL+gmVgs+jωCgdT(Vd−Vgs)=0 From the above two equations, we obtain:(5)Ii=jω[CgsT+CgdT(1+gmRL1+jωRLCgdT)]Vgs Since ωRLCgdT is typically much less than 1, the term jωRLCgdT can be neglected. Thus, the above equation simplifies to:(6)Ii = jωCgsT + CgdT1 + gmRLVgs The ideal load current is given by:(7)Id=gmVgs The current gain is then given by:(8)IdIt=gm2πfCgsT+CgdT1+gmRL When the current gain reaches unity, we obtain:(9)fT=gm2πCgsT+CgdT1+gmRL The cutoff frequency (fT) is extracted from the simulation as the frequency increases to the 0 dB point of the current gain. The resulting fT of the laterally down-scaling devices at the VGS voltage of gm_max (see Figure 8b) is shown in Figure 12a. When the device with LG = 120 nm is biased at VDS = 6 V, its current gain cutoff frequency reaches 39.92 GHz at VGS = 1.5 V (where the gm_max occurs). A gradual variation in the slope of the fT versus 1/LG relation is observed in the semi-log plot [see Figure 12a], which appears anomalous initially. This observation contrasts with the classical long-channel theoretical prediction for semiconductor FETs, where the cutoff frequency typically follows the relation:(10)fT=μ0(VGS−VTH)2πL2G In the short-channel limit, the scaling relation transitions to fT∝1/LG [27]. Consequently, during the lateral scaling of FETs, a critical transition point—where the dependence shifts from fT ∝ 1/L2G to fT ∝ 1/LG—is expected, demarcating the long-channel and short-channel regimes. Notably, in contrast to the relatively ambiguous transition observed in the fT versus 1/LG plot, the VTH versus LG relation exhibits a distinct phase change at an LG of approximately 100 nm. The characteristic high carrier mobility of In2O3 fundamentally modifies the scaling behavior of its cutoff frequency. In the long-channel regime, the theoretical fT ∝ 1/L2G dependence stems from a constant mobility. However, for high-mobility materials like In2O3, carriers can more readily achieve and maintain velocity saturation under high lateral fields in shortened channels. This effect mitigates the degradation of carrier velocity with scaling, thereby blunting the expected transition to a pronounced fT ∝ 1/LG regime. Consequently, the long-channel scaling behavior persists observably into shorter gate lengths, resulting in a less distinct transition point in the fT versus 1/LG plot compared to materials with lower mobility. The maximum oscillation frequency (fmax) in this work is determined as the frequency at which the simulated unilateral power gain drops to 0 dB. Figure 12b presents the fmax for the laterally scaled devices at the gate voltage corresponding to gm_max. For the device with LG = 120 nm and VDS = 6 V, its fmax reaches 73.13 GHz at VGS = 1.5 V, where the gm_max occurs. As shown in Figure 12b, the fmax exhibits a non-monotonic variation with respect to 1/LG. In the regime of relatively long LG, fmax exhibits a sluggish increase, which can be attributed within the present simplified framework to the relatively larger influence of the capacitance network and residual parasitic limitations. As LG enters the medium range, the device transitions into an approximate ideal scaling regime, where the enhancement of the intrinsic fT becomes the dominant factor, and fmax thus approximates a linear increase:(11)fmax≈fT2(Rg+Rs+RchgmRs)(gds+2πfTCgd) When LG is further scaled down to the deep sub-micron regime, effects such as velocity saturation, the pronounced DIBL effect, and the relatively increasing influence of parasitic elements collectively act to slow down the growth rate of fmax once again. This evolutionary process vividly illustrates the transition in dominance governing the high-frequency performance of heavily doped In2O3 FETs during lateral scaling: shifting first from parasitic limitations to intrinsic physical properties and finally to short-channel non-idealities. Lastly, Figure 13 presents the relationship between the fT and the VGS for top-gate In2O3 FETs with scaled gate lengths at VDS = 6 V. The corresponding gm versus VGS is also included for reference. It can be observed that regardless of the LG, the variation trend of the fT with VGS closely resembles that of the gm with VGS:(12)fT=gm2πCG As evident from the Formula (12), given that the gate capacitance (CG) remains relatively constant, the variation in the fT is predominantly governed by the gm. Consequently, the trend of gm versus VGS is directly mirrored in the fT characteristic. For a top-gate In2O3 FET on the on-state, the gate capacitance is dominated by the gate oxide capacitance, while the channel capacitance exhibits minimal variation with VGS. Hence, after the device turns on, CG shows insignificant change with VGS, causing the variation in fT to primarily follow that of gm. Even with variations in LG, the Formula (12) remains valid provided that the device is on-state. Consequently, the curves depicting fT versus VGS and gm versus VGS are qualitatively similar across both long-channel and short-channel devices. However, it is noteworthy that although the trends are similar, the VGS corresponding to the maximum fT (fT_max) and that corresponding to the gm_max may differ slightly. Several plausible reasons account for this discrepancy. Specifically, fT exhibits a stronger dependence on the effective gate overdrive voltage (Vov = VGS − VTH), whereas gm is more directly influenced by the drain current [28]. Parasitic resistances, such as source and drain resistances (RS, RD), also affect fT and gm to different extents [29,30]. Moreover, under pronounced short-channel effects, current saturation or velocity saturation can lead to a misalignment in the maximum positions of gm and fT [31]. 4. Conclusions This study investigated the lateral scaling characteristics of top-gate In2O3 TFTs with a 7 nm thick HfO2 gate dielectric and a centered gate configuration for LG ranging from 20 nm to 700 nm using a device simulator (Silvaco Atlas). This study focused on analyzing its short-channel effects, DC performance, and small-signal frequency response. The results indicate that the critical LG for the transition from long-channel to short-channel behavior is approximately 100 nm. When LG ≤ 100 nm, short-channel effects cause a significant negative shift in the VTH, accompanied by a pronounced increase in the DIBL coefficient. When the LG is scaled below 40 nm, the VDsat of the device even exceeds 8 V. For LG = 120 nm, the saturation drain current continues to increase with drain voltage rather than remaining constant, indicating the onset of evident short-channel non-idealities. This study further reveals that for a device with LG = 40 nm, the simulated fT reaches 90.24 GHz and the fmax reaches 122.73 GHz. When LG is scaled to 20 nm, these values rise to 124.32 GHz for fT and 157.64 GHz for fmax. Overall, this work demonstrates through simulation the RF potential of top-gate In2O3 TFTs in the radio frequency domain, with fT/fmax values comparable to early-stage simulation or experimental results reported in the literature for other advanced RF technologies, such as GaN and Ga2O3 [32,33]. It should be noted that the present conclusions are drawn within a drift-diffusion-based TCAD framework with calibrated effective interface/transport parameters and a simplified intrinsic RF equivalent circuit. More comprehensive treatment of substrate dependence, tunneling, self-heating, and additional extrinsic parasitics will be valuable in future work.