Predictive performance of clustered feature-weighting case-based reasoning

Because many factors are complexly involved in the production of semiconductors, semiconductor manufacturers can hardly manage yield precisely. We present a hybrid machine learning system, i.e., a clustered feature-weighting case-based reasoning, to detect high-yield or low-yield lots in semiconductor manufacturing. The system uses self-organizing map neural networks to identify similar patterns in the process parameters. The trained back-propagation neural networks determine feature weights of case-based reasoning. Based on the clustered feature-weighting case-based reasoning, the hybrid system predicts the yield level of a new manufacturing lot. To validate the effectiveness of our approach, we apply the hybrid system to real data of a semiconductor company.