Deep neural network-assisted radio frequency inductively coupled plasma discharge simulation

Abstract Since the increasing structural complexity of plasma systems and the growing degree of multi-parameter coupling, numerical simulations encounter challenges like high model complexity, convergence difficulties, and substantial computational costs. This study develops a deep neural network (DNN), as an auxiliary prediction tool for plasma discharge simulation, to efficiently and accurately investigate the radio frequency inductively coupled plasma discharge. The simulation data was used to create the training dataset for the DNN. Validation showed the DNN predictions matched simulation results, achieving a relative error as low as 0.01%. The prediction results indicate that the DNN model can provide satisfactory prediction results like plasma discharge characteristics (e.g., species densities, ionization rate, electron power absorption rate, and electron mean energy) and plasma chemical reaction mechanisms under multi-input parameter coupling conditions (e.g., power, pressure, and O2 ratio) based on the target output parameters. Compared with time-intensive traditional fluid simulations, the DNN model can efficiently and accurately provide predictions closely matching simulation results within 1 s, significantly improving computational efficiency. This research confirms the feasibility of DNN-assisted plasma discharge simulation, providing an efficient and accurate auxiliary computing method for plasma studies.