Failure Detection and Primary Cause Identification of Multivariate Time Series Data in Semiconductor Equipment

Abstract

Downtime caused by equipment failure is the biggest productivity problem in the 24-hour a day operations of the semiconductor industry. Although some equipment failures are inevitable, increases in productivity can be gained if the causes of failures can be detected quickly and repaired, thus reducing downtime. Univariate control charts are commonly used to detect failures. However, because of the complexity of the process and the structural characteristics of the equipment, detection and identification of the causes of failures may be difficult. The purpose of this study is to use correlations of variables to detect failures in semiconductor equipment, to predict the parts to be replaced and to identify the primary causes of failures. The proposed method consists of four steps: (1) conversion of the multivariate time series data of the equipment into signature matrixes, (2) detection of anomalies through a convolutional autoencoder, (3) learning classification models with supervised learning methods that use the residual matrixes of fault sections, and (4) application of an explainable algorithm to interpret the classification model. The effectiveness and applicability of the proposed method are demonstrated by the actual multivariate time series data obtained from 8-inch ashing process equipment that produces semiconductors on 8-inch silicon wafers.