Machine Learning-augmented High-efficient TCAD on Accurate Characteristics of Gate-all-around Transistors with Quantum Effect

A machine learning (ML)-augmented TCAD framework is proposed to build an adaptive density gradient (DG) model for the ultra-scaled gate-all-around (GAA) devices. First, to capture the impact of quantum confinement on Silicon nanowire and nanosheet GAA FETs, the multi-subband k•p model is calibrated with the first-principles calculations. In parallel, a fully-connected multi-layer neural network, i.e., Multilayer Perceptron (MLP), is trained to learn the empirical quantum correction potential parameters from drift-diffusion-equation (DD) based Technology Computer-Aided Design (TCAD). Then the MLP is incorporated into the ML-augmented TCAD framework to obtain an adaptive DG model regarding to the prepared k•p results. This ML-augmented adaptive DG model extend the scope of the applications of the DD based TCAD, approaching the level of the subband Schrodinger Poisson solver.