Modeling of an inductively-coupled Cl2/Ar plasma using neural network

Abstract

In this study, a neural network model for an inductively-coupled Cl2/Ar plasma (ICP) process has been proposed. Plasma experiments were performed in a planar inductively coupled plasma reactor. The input parameters considered for plasma modeling were the gas mixing ratio, and the source and bias powers, which were varied in the ranges 0–100% Cl2/Ar, 500–800W, and 50–300W, respectively. Plasma diagnostics were performed by double Langmuir probe measurements. Analysis of voltage–current curves in order to obtain electron temperature and total positive ion density was carried out. A back propagation neural network model with a pre-processor was constructed. The prediction errors for the proposed model were shown to be very small compared to those of a conventional neural network model. In the experiments conducted, it was found that an increase in Ar mixing ratio or input source power resulted in an increase in the total positive ion density. An increase in the input bias power, however, corresponded to a decrease in the total positive ion density. The electron temperature increased as the Ar fraction, input source power, or input bias power increased.