International Journal of Numerical Modelling-Electronic Networks Devices and Fields最新文献

A novel approach for optimizing the hyperparameters of a support vector regression (SVR) model is presented for radio frequency (RF) power transistors. In standard SVR models, hyperparameters are enhanced using grid search optimization (GSO), which can be inefficient. In this study, particle swarm optimization (PSO) is introduced as a method for optimizing hyperparameters in a SVR model that increases the model optimization efficiency significantly in comparison with GSO while maintaining a high level of performance. To verify the accuracy and effectiveness of the model, a 10-W GaN power transistor produced by Wolfspeed is used. In comparison to the existing GSO-SVR model, the proposed PSO-SVR model demonstrates superior performance and efficiency.

SiC double gate (DG) junction field effect transistor (JFET) is promising for low-noise and high-temperature electronics. Existing studies indicate that JFETs can be considered a special case of MOSFETs when the oxide layer thickness approaches zero. In this article, we exploited the structural similarity between the DG JFETs and the DG MOSFETs. By obtaining the 2D Poisson's equation for the DG MOSFETs and deriving the limits, we developed a model for calculating the channel current of SiC DG JFETs in the subthreshold region. The model is derived from device physics, requiring no fitting parameters and offering relatively low computational complexity. The results indicate that, whether for enhancement mode or depletion mode JFETs, the calculated values of this model are in good agreement with the 2D numerical analysis results obtained from Silvaco Atlas. Moreover, for enhancement mode JFETs, even when significant short-channel effects occur, the subthreshold current can still be well predicted. In addition, the model displays predictive capability for the depletion-mode JFETs.

In this article, a Schottky barrier β-Ga₂O₃ MOSFET is proposed. It shows improvements in drain saturation current, I_on/I_off ratio, transconductance, and off-state breakdown voltage. The proposed design, which implements the Schottky barrier source and drain contacts, has led to reduced on-state resistance (R_on), reduced forward voltage drops, faster switching speed, higher frequency, and improved efficiency. After device optimization, we determined that a source and drain having a work function of 3.90 eV result in the highest drain saturation current of (I_ds) 264 mA. Additionally, in the transfer characteristics, we demonstrate that increasing the channel doping concentration led to a shift toward depletion mode operation, while decreasing the doping concentration moved the device toward enhancement mode at the cost of drain current. Analysis of lattice temperature and self-heating effects on different substrates has also been performed. Furthermore, introducing a passivation layer of SiO₂ as a gate oxide and an unintentionally doped (UID) layer of 400 nm doping concentration of 1.5 × 10¹⁵ cm⁻³, results in further significant improvements in the drain saturation current (I_ds) of 624 mA and transconductance of 38.09 mS, approximately doubling their values compared with the device without a passivation layer of SiO₂ and an I_on/I_off ratio of 10¹⁵, and the device's performance at various substrate temperatures has been evaluated. In addition, the inclusion of a passivation layer of SiO₂ improves the breakdown voltage to 2385 V, which is significantly high compared with the conventional device. Moreover, the lower specific-on-resistance R_on,sp of 7.6 mΩ/cm² and higher breakdown voltage then the high-power figure of merit (PFOM) (BV²/R_on,sp) of 748 MW/cm² have been achieved.

The characteristics of multilayered microwave absorbing materials are very efficient compared with those of single layer. In this article, a hybrid optimization algorithm (genetic algorithm + pattern search) combined with transmission line matrix method has been presented. The selection of parameters, including the arrangement of layers, thickness of layers, absorption index, and shielding efficiency, forms the foundation of this process. The optimization algorithm was applied to two new multilayered structures. The first structure consists of conductive layers (CLs) of carbon nanotube (CNT) with ceramic layers of zirconium dioxide $��\left({\mathrm{ZrO}}_2\right)�$. The second structure includes CLs of CNT with layers based on magnetite polyaniline nanomaterial (PANI_$��{\mathrm{Fe}}_3�O_4�$). Performances of both structures were evaluated in the X-band frequency range. Simulation results showed that both designs have higher absorption index picks (> 90%) and, low $��S_{11}�$ magnitude value with low layer thickness. This approach offers a solid foundation for future experimental trials in the development of efficient microwave absorbing and shielding structures with tunable electromagnetic performances suitable for X-band applications.

This paper presents a gap coupled triple band slot loaded microstrip patch antenna with parasitic patches. It consist inverted U-shape and inverted T-shape open-ended slots along with a rectangular slot at center of patch. The inverted U-shape open-ended slot generates a driven patch at bottom side and an inverted U-shape parasitic patch at middle side of patch while inverted T-shape open-ended slot generates two rectangular shape parasitic patches of same dimension at top side of patch. The proposed gap coupled antenna covers 2.29 to 2.77 GHz in first band, 3.25 to 3.65 GHz in second band and 4.67 to 5.72 GHz in third band with return losses of −23.2, −19.90, and −38.06 dB, respectively. The proposed antenna resonates at 2.57, 3.48, and 5.37 GHz with fractional bandwidth of 18.97% (480 MHz), 11.59% (400 MHz), and 20.21% (1050 MHz), respectively. The return loss and bandwidth of presented antenna is increases gradually by loading inverted U-shape and inverted T-shape open-ended slots along with a rectangular slot in antenna patch. The proposed antenna exhibits stable peak gain of 4.45, 4.81, and 5.26 dBi and efficiency of 89.5%, 89%, and 90% in three resonating bands. The antenna resonating bands are applicable for WiMAX: 2.5/3.5/5.5 GHz (2.5–2.69, 3.4–3.69, and 5.25–5.85 GHz), WLAN: 2.4/5.2 GHz (2.4–2.484 and 5.15–5.35 GHz) and sub-6 GHz 5G: 3.5 GHz (3.3–3.8 GHz). The size of antenna is 40 mm × 50 mm (0.34 × $��{�0.43�\lambda �}_0^2�$ at frequency 2.57 GHz). The gap coupled antenna geometry is fed by microstrip line feed and simulated by IE3D simulation tool.

FeFET architectures for non-volatile on-chip memory are designed and compared in this investigation study. Because of its inherent non-volatile properties and low power requirements, FeFETs are attracting a lot of interest as prospective candidates for future memory technology. The aim of this paper is to investigate several FeFET designs and assess how well they function in terms of important factors including durability, retention, speed, and endurance. Using device simulations and experimental data, a number of FeFET architectures, such as MFS, MFIS, MFMIS, and MF-ABO₃, are analyzed and contrasted. Comparative study gives light on the advantages and disadvantages of various FeFET architectures; improving our comprehension of how well-suited they are for non-volatile on-chip memory. This work will contribute to the improvement of FeFET devices for upcoming integrated circuits and progress the development of sophisticated FeFET-based memory techniques.

This paper presents an energy-efficient single-transistor leaky integrate-and-fire neuron, based on Suspended Gate-FET (SG-FET), for signal classification and neuromorphic computing applications. By leveraging the SG-FET model, extensive simulations were conducted to demonstrate the device's remarkable neuronal ability. The device faithfully emulated the intricate behaviour of biological neurons, without the need for external circuitry. One of the standout achievements lies in the device's astonishingly low energy consumption of 94.5 aJ per spike. Therefore, it outperforms the previously proposed one-transistor (1-T) neurons, which makes it a potential candidate for energy-efficient neuromorphic computing. To verify the practical viability of the device, an emulation was seamlessly integrated into a spiking neural network framework, allowing for real-time signal classification. In this specific case, the device excelled in the classification of electrocardiogram (ECG) signals, achieving an impressive accuracy rate of 85.6%. This outcome highlights the device's efficacy in handling real-world signal processing tasks with remarkable precision and efficiency.

This work presents a FinFET-based stable, and low-power consuming static random access memory (SRAM) bit-cell that used eight transistors. The performance parameter of proposed feedback-cutting 8T (FC8T) is compared with four pre-published cell circuits, i.e., 6T, read-decoupled 8T(8TRD), Schmitt-trigger based 10T (10TST), and Schmitt-trigger-based modified 10 T (10TMST). The write power in proposed design is reduced by 1.36×/1.32×/1.88×/1.47× compared to 6T/8TRD/10TST/10TMST cells. The write and read stability of proposed design is improved by 1.15×/1.86×/1.28×/1.148× and 2.27×/1×/1.57×/1.11×, respectively. The proposed design also shows the low variability compared to other SRAM bit-cells.

Characteristic mode (CM) analysis serves as a powerful tool for evaluating the radiation and scattering characteristics of objects. CM formulations within the method of moments (MoM) framework are widely favored due to their ability to offer clear physical insights, handle complex shapes, and facilitate straightforward implementation. However, MoM-based CM formulations become inefficient when applied to electrically large objects due to the dense matrices involved. This article introduces a novel approach using a fast low-rank decomposition-based implicitly restarted Arnoldi method (IRAM) to accelerate CM computations. The adaptive cross approximation (ACA) and QR-SVD algorithms are employed to efficiently compute the low-rank decomposition of matrices. The ACA-QR-SVD algorithm offers advantages in matrix filling, LU factorization, and matrix–vector multiplication processes, thereby enhancing efficiency. Numerical simulations on two representative objects demonstrate that the proposed algorithm notably improves computational speed and reduces memory requirements while maintaining high computational accuracy.