Abstract
In the advanced production ramp up stage, the product is always bringing small sample size, taking ill understood production process and hardly detecting root cause with past engineers' experience. The semiconductor manufacturing is a capital intensive industrial. A fast low yield detection method can reduce the product cost, and increase the business competition. This research proposed a low yield diagnosis framework for advanced semiconductor manufacturing, and illustrated the proposed method with an empirical case study. In the proposed method, we defined the excursions in yield, grouped the collinear process stages by Cramer's V correlation, used logistic regression to final all the possible low yield root causes, and three domain principles to narrow the root cause candidates. In the empirical study, we compare the proposed method with random forest and linear regression. The experiment results show the proposed method can perform priority than random forest and linear regression. Therefore, the proposed method indeed can help engineers, in a short period of time, to find the excursion root cause and improve the production yield.