A multivariate parameter trace analysis for online fault detection in a semiconductor etch tool

The objective of this paper is to develop a wafer-by-wafer fault detection model for a semiconductor etch tool operating in a worksite situation in which the tool parameter traces are correlated and drift slowly from an initial recipe setting. Process drift is a common occurrence in many processes because of the aging of tool components. The proposed fault detection model compares the entire trace structures of the tool parameters with reference templates by using an improved DTW (dynamic time warping) algorithm, and it performs a T 2 -based multivariate analysis with the structure similarity scores created by the improved DTW. In addition, to adapt to the process drift, a recursive T 2 update procedure with an optimal correction factor is incorporated in the model. The optimal correction factor is derived using the Kalman filtering technique. Experiments using the data collected from a worksite reactive ion etching process demonstrate that the performance of the proposed fault detection model is very encouraging.