Prognostics of tool failing behavior based on autoassociative Gaussian process regression for semiconductor manufacturing

Predictive maintenance is considered as one of the helpful techniques to improve the manufacturing process, especially in high-mixed semiconductor manufacturing suffered from the frequent tool failing behavior. Prognostics is a good solution to infering the failing occurrence and realizing predictive maintenance. Because of the complicated and unclear tool failing phenomena, it is difficult to model the tool failing behavior by mechanism analysis. The difficulty impedes the development of prognostics in semiconductor manufacturing. As data-driven methods aim at estimating future behaviors without knowledge of the underlying failing phenomena, this article proposes a novel data-driven prognostic method based on auto-associative Gaussian process regression to infer the tool failing behavior in semiconductor manufacturing. Through extracting the failing factors, determining the failing degree and constructing the prognostic model, the tool failing behavior tendency is predicted and the suitable chamber cleaning time is determined to improve the productivity and save the cost of semiconductor manufacturing. The effectiveness and practicality of the proposed method are demonstrated by a practical semiconductor manufacturing factory. The obtained prognostic results can help operators understand the tool failing behavior better and guide decision-makers to make a suitable plan for chamber cleaning in semiconductor manufacturing.