DEVELOPING A DATA MINING METHOD FOR WAFER BINMAP CLUSTERING AND AN EMPIRICAL STUDY IN A SEMICONDUCTOR MANUFACTURING FAB

ABSTRACT

Mining potentially useful information from large database becomes increasingly important in both research and application in many fields. Because of the complex fabrication processes and the cost resulted from defects, it is critical to identify possible faults through examining the failure patterns of wafer bin-maps. However, little research has been done to develop methods for clustering and classifying wafer bin-maps. We first used spatial statistics to examine the independence among failed dies and thus classified the bin maps into four categories: random failure pattern, systematic repulsion failure pattern, systematic attraction failure pattern, and others. For WBM with systematic attraction failure patterns, we developed an ART neural network for clustering to assist the product engineers in narrowing possible causes of manufacturing defects. We used empirical data from a fab for demonstration and the results showed the practical viability of this approach for its consistency and efficiency. This study concludes with discussions and remarks on future research.

Keywords:

• wafer bin-map
• neural network
• data mining
• defect diagnosis
• decision analysis
• semiconductor manufacturing