Fast and Accurate Deep Learning-Based Poisson Solver for Single Event Simulations

We investigated the possibilities of enhancement of the random-walk drift-diffusion (RWDD) simulations of charge transport in single-events phenomena by modeling electrostatic effects. We first developed a fast and versatile physical Poisson solver based on a simple finite-difference formulation, which was then evaluated, especially in the context of particle simulations, by comparison with technology computer-aided design (TCAD) and irradiation experiments (alpha particles and heavy ions). Capitalizing on this physical solver, we then developed an artificial intelligence (AI)-based solver, using a convolutional neural network (CNN), to further accelerate solving electrostatics and make it usable at large scale in industrial simulators.