Plasma-Chemical Etching Process Behavioral Models Based on Tree Ensembles and Neural Network

In the modern semiconductor manufacturing technology it is essential to control the result of a wafer processing to ensure stability and high production yield. One of the promising techniques, which can provide the information about the result of a process, is predictive modeling based on machine learning models. In this paper, the possibilities of using Tree Ensembles and Artificial Neural Networks for modeling the plasma-chemical process of deep trench etching in the silicon substrate are considered. Mathematical background for machine learning techniques used for modeling is discussed, principles of regression trees generation are presented and formal descriptive algorithm of composing several regression trees in an ensemble is demonstrated. The developed predictive models were tested on physical-technological model of the plasma-chemical etching process. The results have shown that accurate and robust models based on Tree Ensembles and Artificial Neural Networks were developed in order to predict the trench depth.