Treatment of missing values for association rule-based tool commonality analysis in semiconductor manufacturing

In semiconductor manufacturing, there are hundreds of processing steps with multiple tools at most steps. Association rule-based tool commonality analysis (TCA) is believed to be an effective approach to identify tool excursions for yield enhancement. However, we found that missing values frequently occurred in industry practice will lead to high false positive and false negative rates in association rule-based TCA. Incorrect identification of root cause of yield loss not only loses engineer´s trust on TCA but also delays the process improvement opportunity. A Markov-chain based Missing Value Estimation (MC-MVE) method is proposed in this paper to improve the effectiveness of association rule-based TCA. MC-MVE adopts an iterative approach to estimate the probabilistic state of the missing values, with the state transition probability derived by a Bernoulli distribution, until convergence criteria. The real case study shows that MC-MVE has a better accuracy for recovering missing values to improve the identification accuracy in TCA.