This study proposes a model that combines deep learning (DL) techniques with plasma simulations to efficiently investigate optimal process conditions. The DL model was trained using data obtained from an Ar/O₂ inductively coupled plasma discharge simulation. Plasma discharge parameters such as the O₂ ratio, pressure, and power were trained as input data to predict the electron density, electron temperature, potential, and the densities of Ar⁺, O²⁺, O⁻, and O⁺. The performance of the DL model was verified by comparing the results of interpolation, which predicted a constant pattern within the range of the trained data, and extrapolation, which predicted a pattern beyond the trained data range, with the ground truth to verify the low error rate. The proposed deep neural network model can significantly reduce the necessity for trial and error when adjusting the process conditions. This model is expected to be an effective tool for narrowing the process window during the early stages of equipment and process development.

Keywords: Deep learning, Plasma simulation, Interpolation, Extrapolation, Plasma process