Use of neural network to model X-ray photoelectron spectroscopy data for diagnosis of plasma etch equipment

Abstract

X-ray photoelectron spectroscopy (XPS) is widely used to analyze chemical states on plasma-processed film surfaces. For XPS data to be utilized for plasma monitoring, a prediction model is in demand. This type of model can further be used for optimization of chemical compositions on plasma processes film surfaces. In this study, a prediction model is constructed by combining XPS and backpropagation neural network (BPNN). Genetic algorithm (GA) was used to improve the BPNN prediction performance. The experimental data were collected during the plasma etching of silicon carbide films in a NH 4–CF 4 inductively coupled plasma. For a systematic modeling, the etching process was characterized by means of face-centered Box Wilson experiment. Four major peaks modeled include C1s, O1s, N1s, and Si2p. For comparisons, other conventional and statistical models were also constructed. For all peaks, GA-BPNN models yielded an improved prediction with respect to conventional BPNN and statistical regression models. The improvements were even more than 30% over BPNN models for C1s and Si2p and over statistical regression models for O1s and Si2p.