Fault detection based on Sensitive Marginal Fisher Analysis for class imbalance

Abstract

The class imbalance problem has made researchers pay intensive attention in recent years. In the modern semiconductor industry, the class imbalance makes fault detection, which aims at constructing a decision tool to maintain high process yields, quite challenging. Marginal Fisher Analysis (MFA) is a popular method which can analyze the discriminant relationship between data points from different classes, and has been applied in fault detection. However, the performance of MFA is poor if it is applied in the data having imbalance distributing between classes. This paper analyzes the challenge of class imbalance and a improved approach of MFA named Sensitive Margin Fisher Analysis (SMFA) was proposed for the problem. The proposed fault detection method has been verified by applying it in the semiconductor wafer fabrication process. The experiment results confirm the new method improves the fault detection performance.