Automatic equipment fault fingerprint extraction for the fault diagnostic on the batch process data

Abstract Equipment condition monitoring in semiconductor manufacturing requires prompt, accurate, and sensitive detection and classification of equipment and process faults. Efficient and effective fault diagnostic is essential to minimizing scrapped wafers, reducing unscheduled equipment downtime, and consequently maintaining high production throughput and product yields. Through analyzing the equipment sensor signals as the batch process data, i.e., process timestamp × sensor × wafer, this paper firstly applies the well-known Support Vector Machine (SVM) classifier to detect the abnormal observations. In the second stage, the normal process dynamics are decomposed into different clusters by K-Means clustering. Each part of the process dynamics is further modelled by Principal Component Analysis (PCA). Fault fingerprints then can be extracted by consolidating the out of control scenarios after projecting the abnormal observations into the PCA models. An empirical study is conducted in collaboration with a local IC maker in France to validate the methodology. The result shows that the proposed approach can effectively detect abnormal observations as well as automatically classify the proper fault fingerprints to give evident guidelines in explaining the known faults.