Real Time Electrostatics of Double Gate FETs With Neural Networks

Traditional Technology Computer‐Aided Design (TCAD) tools have played a vital role in improving the understanding of the Metal Oxide Semiconductor Field Effect Transistors (MOSFETs) and the exploration of its design space. The increase in the complexity of the MOSFET's structure due to multiple innovations has correspondingly increased the complexity of the mathematical models used in the simulations. This results in increased computational cost, simulation time, and non‐convergence. Hence, in this work, to the best of the knowledge, for the first time, a Deep Neural Network (DNN) based quantum corrected electrostatic solver is developed to simulate the Double Gate MOSFETs across cross‐section (3 to 40 nm) and gate bias (0.0 to 0.6 V). The DNN is developed to predict the spatially varying quantities that are solutions of Poisson, Schrodinger's equation, and quantum‐corrected electron concentration. The proposed DNN‐based solver has the relative mean absolute percentage error over the entire data set for electrostatic potential, and carrier concentration is 0.27% and 0.079%, respectively, at a significantly lower computational time than the traditional numerical solvers. This work opens up the possibility of performing real‐time simulations of advanced FETs, unlike the conventional simulators with much lower computational cost.