In-Line Predictive Monitoring Framework

Abstract

Process monitoring, which is used to ensure the product quality in semiconductor manufacturing, develops a control chart and alerts engineers whenever a control limit is exceeded. However, a late alarm could generate defects or scraps, and thus, a prealarm is urgent to be developed. This study proposes an in-line predictive monitoring (ILPM) framework that uses process parameter monitoring (PPM) in the first phase and equipment parameter monitoring (EPM) in the second phase. PPM includes off-line training and in-line prediction, where we collect the first half of time-series data and predict the second half for in-line quality control. To maintain robustness and accuracy, the concept drift is used to update the ILPM model in real time; specifically, the EPM detects the loss of prediction by cumulative sum (CUSUM) control chart or detects the change of equipment parameters by Bayesian approach for developing retraining mechanism. An empirical study of a semiconductor manufacturer indicates that the proposed ILPM framework improved both the quality control and production capacity. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —Although some in-line monitoring approaches have been proposed for specific conditions, little research has been done to develop a prealarm method; in particular, this study proposes predictive monitoring by using data from multiple sensors (or status variable identifications, SVIDs) to predict one sensor in one process step of semiconductor manufacturer. The prealarm benefits the early troubleshooting and equipment capacity. This study also applies concept drift and equipment parameter monitoring (EPM) to identify the prediction model misalignment or equipment misalignment. In practice, root cause identification of the prediction error is critical and benefits the model retraining and equipment maintenance.