Application of machine learning to manufacturing: results from metal etch

With the increasing availability of huge quantities of manufacturing data, and the pressures of continuous process improvement and scrap reduction, engineers are beginning to use machine learning techniques along with traditional statistical methods. In this paper, we discuss the application of standard machine learning techniques to analyze, classify, and predict the quality of metal etch using RIE. Three types of data were used to characterize a metal etch: in-process sensor data from the etch chamber, metrology data for critical dimension measurements before and after etch, and metal resistance measurements from probe tests. Three machine learning paradigms were applied: neural networks, induction learning, and case-based reasoning. This paper describes the techniques used, the results obtained, and the conclusions drawn.