Neural Network-Based BSIM Transistor Model Framework: Currents, Charges, Variability, and Circuit Simulation

Deep Learning MOSFET Current-Voltage Model with Symmetric Fourier Features

“…which is the weighted sum of the squared magnitude errors in the Fourier coefficients Á and H. Note that, given kGS and kDS are greater than 1, the costs in predicting derivatives are larger than those in predicting the function itself. This not only explains the observations in previous studies [1][2][3][4], but also underscores the importance of accurate Fourier coefficient prediction in minimizing the overall cost. Before doing so, we will first discuss the IV data collection and spectral bias in neural networks.…”

“…Deep-learning approaches have recently gained significant traction in semiconductor device modeling, particularly for MOSFET models [1][2][3][4][5]. These techniques aim to improve computational efficiency and enhance accuracy.…”

Enhancement and Expansion of the Neural Network-Based Compact Model Using a Binning Method

“…F AST and accurate models of next-generation devices are very important for circuit simulation and design optimization [1]. Industry-standard compact models such as the Berkeley short-channel IGFET model (BSIM) have used physics-based equations to describe electrical characteristics of devices [2], [3].…”

Neural-network-based transfer learning for predicting cryo-CMOS characteristics from small datasets

“…Considering that the predicted results will be used for circuit simulations for future, not only I d values but also derivatives of I d values should be optimized. The optimization of the derivatives can be achieved by customizing the loss function as demonstrated by Tung et al 21) The small discrepancy between the experimental and predicted exhibited when V d = 1 V in the subthreshold region around V g ∼ 0.3 V, is probably due to sample variations in the SOI MOSFETs. At cryogenic temperatures and in the subthreshold region, I d -V g characteristics exhibit large variations.…”