Search

We develop a SPICE-compatible neural network-based compact model to accurately capture the temperature dependence and self-heating effects in Field Effect Transistors (FETs). The model is based on artificial neural networks with no semi-empirical temperature equations. The transfer and activation functions are optimized to improve the accuracy of the model. A new temperature relaxation model is proposed, which allows training the model using ambient temperature data without iteratively extracting the self-heating parameters. The proposed method can simply generate the ambient and dynamic self-heating characteristics for circuit simulations. The model can accurately reproduce the current-voltage (IV), capacitance-voltage (CV), and transient characteristics of FETs across a broad temperature range with a speed advantage of up to 12X versus BSIM-CMG.

This article introduces a compact model for single-gate and multigate thin-film transistors (TFTs) based on the Berkeley short-channel IGFET model-common multiple gate (BSIM-CMG) framework, valid in all regions of TFT operation. We develop a model for the channel charge density, including the impact of traps in the amorphous/polycrystalline channel of TFT, based on a linear potential profile approximation along the gate-normal direction. We further develop explicit, charge-based compact models for current and terminal charges. Unlike previous models, this model implicitly includes the impact of trap-limited conduction mechanism. In addition, we include the impact of Schottky source/drain contact resistances and temperature dependencies. The model is carefully validated with published experimental data of TFTs using diverse channel materials, including oxide semiconductors (OSs) and poly-Si, and various geometries, spanning applications in displays and emerging 3-D back-end-of-line (BEOL) integration scenarios.

In this work, we present a compact temperature-dependent model for bulk MOSFETs with applicability down to the cryogenic range of temperatures and validate it against the experimental data. The long-channel and low-threshold voltage MOSFET device is characterized by multiple bias conditions and temperatures. With decreasing temperature, the characteristics exhibit a positive temperature shift in threshold voltage, an increase in the low-field carrier mobility, higher mobility degradation, and subthreshold swing (SS) saturation. Furthermore, the transconductance shows a nonmonotonic behavior with respect to temperature. The observed temperature-dependent variations in the device characteristics have been thoroughly analyzed, incorporated into the temperature-dependent parameters of the BSIM-BULK model, and validated using our characterized data and experimental data from other recent publications. The accuracy of the model and physical correctness are effectively demonstrated by carrying out the Gummel symmetry test (GST), the slope ratio (SR) test, and the tree top test. This study also analyzes the behavior of the results of the SR and tree top tests at cryogenic temperature, presenting these findings for the first time.

Space charge modulation (SCM), that is, the increase of charge carriers in the drift region, modulates carrier concentration in the drift region, which results in conductivity modulation (also known as the expansion effect) of laterally double-diffused metal oxide semiconductor (LDMOS) transistors. In this work: 1) a compact model for conductivity modulation is presented. Since conductivity modulation also causes a change in internal node voltage ( ${d}_{i}$ ), therefore physics in the impact ionization model of drift region which accurately captures the change in node voltage is also included; 2) an improved model of impact ionization for the intrinsic region of LDMOS transistors ( ${d}_{i}$ ) is discussed and validated across a high drain voltage for different gate voltages; 3) the topology of the substrate current flow has been modified, in accordance with the device physics; and 4) finally, this work also includes the improved model of the impact ionization to capture the body bias effect in LDMOS transistors. The new model is extensively validated with the experimental and TCAD data. The research aims to enhance the industry-standard Berkeley short-channel IGFET model-bulk (BSIM-BULK) model, ensuring compatibility with SPICE simulators.

In this work, parameter extraction methodology for the BSIM industry standard high-voltage (HV) device compact model is presented. An extraction procedure covers entire range of operation, including weak inversion to strong inversion region, low and high drain biases, and multiple body biases with temperature effects. The extraction procedure is validated with experimental and numerical simulation data for several configurations of HV devices, including geometric variations.

We present a new compact model of a perpendicular spin-transfer-torque (STT) magnetic tunnel junction (MTJ). Previous studies on STT-MTJs have either focused on solving the Landau–Lifshitz–Gilbert (LLG) equation or utilizing critical current-based macro models. However, the LLG approaches are too complex for large circuit simulations, while the macro models fail to capture the underlying magnetization physics. In this work, we propose a semiphysical and computationally efficient compact model that accurately represents the time-dependent magnet moment and resistance. To validate our model, we compare it with various experimental data and LLG-based STT-MTJ model. The model demonstrates geometry dependence and temperature dependence. Furthermore, we develop a continuous switching probability model to effectively track the probabilities of states under arbitrary waveforms.

Event cameras offer promising properties, such as high temporal resolution and high dynamic range. These benefits have been used into many machine vision tasks, especially optical flow estimation. Currently, most existing event-based works use deep learning to estimate optical flow. However, their networks have not fully exploited prior hidden states and motion flows. In addition, their supervision strategy has not fully leveraged the geometric constraints of event data to unlock the potential of networks. In this article, we propose EV-MGRFlowNet, an unsupervised event-based optical flow estimation pipeline with motion-guided recurrent networks using a hybrid motion-compensation loss (HMC-Loss). First, we propose a feature-enhanced recurrent encoder (FER-Encoder) which fully uses prior hidden states to obtain multilevel motion features. Then, we propose a flow-guided decoder (FG-Decoder) to integrate prior motion flows. Finally, we design an HMC-Loss to strengthen geometric constraints for the more accurate alignment of events. Experimental results show that our method outperforms the current state-of-the-art (SOTA) method on the multivehicle stereo event camera (MVSEC) dataset, with an average reduction of approximately 22.71% in average endpoint error (AEE). To our knowledge, our method ranks first among unsupervised learning-based methods.