Primitive Concept Identification In A Given Set Of Wafer Maps

Ahmed Wahba, Chuanhe Jay Shan, Li-C. Wang, Nik Sumikawa

Abstract: In production test data analytics, wafer maps are often analyzed to search for certain conceptual patterns. A wafer concept recognizer is a learning model for this type of pattern recognition. Learning a recognizer can be based on Generative Adversarial Networks (GANs) with a small set of wafer maps. However, such learning depends on how complex the learning problem is as given by the set of wafer maps. A novel approach is presented to assess such learning complexity, which is based on Tucker Decomposition. Wafer maps obtained from an automotive product line are used in the experiments to illustrate the approach and its potential benefits.