Real-Time Fault Detection in Semiconductor Using One-Class Support Vector Machines

In this paper, we propose a real-time fault detection system for the semiconductor domain, which aims to detect abnormal wafers from a recent history of electrical measurements. It is based on a dynamic model which uses our filter method as feature selection approach, and one-class support vector machines algorithm for classification task. The dynamicity of the model is ensured by updating the database through a temporal moving window. Two scenarios for updating the moving window are proposed. In order to prove the efficiency of our system, we compare it to an alternative detection system based on the Hotelling's T 2 test. Experiments are conducted on two real-world semiconductor datasets. Results show that our system outperforms the alternative system, and can provide an efficient way for real-time fault detection.