Anomaly Detection of Semiconductor Processing Data Based on DTW-LOF Algorithm
Yong Wang, Guiyun Mao, Xu Chen, Wei Zhengying · 2022 China Semiconductor Technology International Conference (CSTIC) · 2022
Fault detection and classification (FDC) of equipment sensor data is of great significance in semiconductor wafer manufacturing to monitor equipment and the processing wafer condition. The early detections of anomalies of sensor data facilitate later process adjustments and avoid further economic loss. Sensor data varies greatly among different processing wafer lots because of machine status changes during the maintenance cycles. This paper combines the Dynamic Time Warping (DTW) and Local Outlier Factor (LOF) algorithms to achieve stable anomaly detection of sensor data under unstable machine conditions. The DTW-LOF model shows good anomaly detection accuracy in different chip processing technologies with a small amount of data. Since the anomalies of semiconductor processing were usually reflected as comprehensive abnormality of multi-sensor in same time ranges, the normalized DTW distance of these sensors could effectively identify the processing abnormal stage and sensors.