Anomaly Detection in Semiconductor Cleanroom Using Isolation Forest

Abstract

Wafer fabrication in semiconductor companies is frequently afflicted by anomalies that can have a negative impact on the cleanroom environment, leading to wafer defects. This study describes an anomaly detection system for edge devices in a semiconductor cleanroom that uses the Isolation Forest method. The multidimensional dataset, which comprises various sizes of ultrafine PM1 particles, is extremely harmful to wafers and can be detected very effectively using Isolation Forest, with an F 1 -score of 0.9899795 and an AUC of 0.99.