A Physics-Informed Neural Networks Algorithm for Simulating Semiconductor Devices

This paper proposes a physics-informed neural networks (PINNs) algorithm for simulating semiconductor devices. It is built on drift-diffusion model (DDM) framework and aiming at describing the behavior of carriers in semiconductor devices. Such algorithm is useful for the development of semiconductor devices with high performance and reliability. In our proposed PINNs algorithm, a surrogate model for coupled drift-diffusion model with artificial neural network (ANN) is built, with backpropagation algorithm used for error computation, and Adam algorithm for parameter updating. Our numerical experiments validate its accuracy as well as convergence rate. In addition, its capability for dealing with local nonlinear problems is addressed, which is a significant potential for its further extension and application.