Wafer fault detection and key step identification for semiconductor manufacturing using principal component analysis, AdaBoost and decision tree

In this paper, a data mining approach is presented to identify the key parameters and the key steps in the manufacture process. For key parameters, a principal component analysis (PCA) algorithm is first used to filter data before classification models of fault detection are established by using SVM and AdaBoost algorithms. A decision tree is then used to locate the key step for root cause identification. In the preliminary study in terms of a set of real wafer fabrication profile data in semiconductor manufacturing, the AdaBoost classifier with PCA has been shown the most effective in identifying key parameters in fault detection. Subsequently, these key parameters along with associated reading values at different timings were used to build a decision tree for the set of empirical rules to best identify problematic timing. It has been further verified that the critical timing among this set of empirical rules had occurred in the same manufacturing phase.