Anomaly Detection of Non-Normal Distribution Wafer Acceptance Test Data Via GMM-Based Method

This paper focuses on techniques for automatically anomaly detection of wafer acceptance test (WAT) data from a production database based on Gaussian Mixture Model (GMM). An integrated analysis scheme is developed to facilitate interpretation of the results. It consists of four phases: Kernel Density Estimation, parameter initialization, GMM iteration until convergence, and anomaly detection. A field data case study shows that this analysis scheme is able to diagnose the non-normal distribution data of WAT which is difficult to detected by system process control (SPC) rule, which is helpful to engineers take sight into the problems of specific manufacturing process, and is also very useful in process development and debugging.