Principal curve-based monitoring chart for anomaly detection of non-linear process signals

This study proposes a monitoring chart for anomaly detection of non-linear process signals generated by semiconductor manufacturing processes. In these manufacturing processes, fault detection and classification (FDC) and statistical process control (SPC) have been established as fundamental techniques to improve production efficiency and yield. Non-linear process signals are collected through automatic sensing during each operation cycle of a manufacturing process. As these cyclic signals non-linearly vary on the process state, the usage of the prevalent SPC chart is limited. Therefore, we propose a more efficient monitoring chart considering non-linear and time-variant characteristics. Using the principal curve, a non-linear smoothing algorithm, we construct a time-variant centerline that represents the standard pattern of the process. Then, control limits are calculated with time-variant variances over the course of the process. To evaluate performance, the proposed method was applied to industrial data for chemical vapor deposition (CVD), a semiconductor manufacturing process. We employed the misdetection ratio of signals to evaluate the performance. The proposed method demonstrated superior performance compared to other existing methods.