Reduced Order Models for 1D Argon CCP Self-Bias and Potential Profiles

Simulation-based reduced order models (ROMs) have the potential to enable the exploration of plasma behavior across a wide parameter space at reduced computational cost. In this work, we apply a Machine Learning (ML) method for the formulation of a ROM for the prediction of the potential profile and self-bias of a 1D argon capacitively coupled plasma, one of the most prolific example problems used to explore the basics of CCP physics. We generate an input data set by running simulations sampled from the input parameter space which is composed of power, pressure, and a variable secondary electron emission coefficient of one electrode. The input parameter space is sampled using two methods: Latin hypercube sampling and random sampling. A 1D drift-diffusion fluid simulation is run for each sampled point in the input parameter space and the potential profile, RF voltage, and RF current are used as output data to the model. We utilize state-of-the-art ML-based regression models on these 1D simulated samples to build a reliable ROM. The resulting ROM is able to predict the potential profile and self-bias across the parameter space and captures the variation in the self-bias induced by the CCP asymmetry driven by the variable secondary electron emission coefficient.