- Abstract: There exists a large number of process deviations in semiconductor manufacturing processes. Automated root cause analysis and decision-making help to significantly improve the effectiveness of manufacturing processes. Manufacturing defects reveal typical patterns in wafer measurement data. Spatial patterns recognition in wafermap data improves the efficiency of finding production issues during different process steps, as early as possible. In this paper, we introduce a deep learning approach for recognition and clustering of spatial patterns in wafermap test data in an unsupervised fashion. First, measurement values are pre-processed, then, a deep variational autoencoder is used to extract a low-dimensional representation of the wafermaps. Finally, various structures in the latent space are detected and wafers assign to the extracted clusters. Extensive simulations show that the proposed approach outperforms the best existing methods over a real-world dataset.
- Keywords: Variational Autoencoder, Deep Neural Networks, Pattern Recognition, Clustering, Semiconductor Manufacturing, Wafermap Measurement Data
- TL;DR: In this paper, we introduce a deep learning approach for recognition and clustering of spatial patterns in wafermap test data in an unsupervised fashion.