Industrial process fault detection using principal component analysis

Abstract

A method and system for use in monitoring/evaluating industrial processes such as, for example, plasma processes are provided. In one embodiment, a plasma process fault detection module () includes multiple sub-modules. A data selection sub-module () obtains selected optical emissions spectra (OES) data for each wafer that is processed. A model building/updating sub-module () constructs multiple models from the OES data for a number of wafers. A principal component analysis (PCA) analysis sub-module () utilizes PCA techniques to determine whether the OES data for a particular wafer differs significantly from an expected normal wafer as represented by the models. A model maintenance sub-module () saves and retrieves models for different processes, associating the current wafer with the correct process. A wafer categorization sub-module () categorizes each wafer based on a scalar metric characterizing the residual spectrum vector. A data output sub-module () outputs the results to a user.