Case study in machine learning for predicting moderate pressure plasma behavior

Abstract

Modeling and forecasting the dynamics of complex systems, such as moderate pressure capacitively coupled plasma (CCP) systems, remains a challenge due to the interactions of physical and chemical processes across multiple scales. Historically, optimization for a given application would be accomplished via a design of experiment (DOE) study across the various external control parameters. Machine learning (ML) techniques show the potential to “forecast” process conditions not tested in a traditional DOE study and thereby allow better optimization and control of a plasma tool. In this article, we have used standard DOE as well as ML predictions to analyze I-V data in a moderate-pressure CCP system. We have demonstrated that supervised regression ML techniques can be a useful tool for extrapolating data even when a plasma system is undergoing a transition in the heating mode, in this case from the alpha to gamma mode. Classification analysis of control parameters is another possible application of ML techniques that can be deployed for system control. Here, we show that given a large set of measured data, the models can identify the gas ratio in the feed gas as well as correctly identify the operating pressure and electrode gap in almost all the cases.