Anomaly Detection for Semiconductor Tools Using Stacked Autoencoder Learning

Anomaly detection for semiconductor tools deals with the problems of finding patterns in process and equipment data that do not conform to expected behaviors. Due to the complexity and unknown correlation of data, machine learning is promising for anomaly detection for semiconductor tools. This paper proposes a Stacked Autoencoder Learning for Anomaly Detection (SALAD) framework that enables anomaly detection in realtime by using a multidimensional time-frequency analysis of sensory data from fab tools. We adopt the Chemical Vapor Deposition (CVD) tool as our study vehicle to demonstrate the feasibility and effectiveness of the developed SALAD framework for anomaly detection of semiconductor tools.