Grouping Complex Wafer Defect Patterns Into Meaningful Clusters (Oregon State Univ., Micron)

Researchers from Oregon State University and Micron Technology published a technical paper introducing a new framework called DECOR for clustering complex defect patterns in semiconductor manufacturing. The framework aims to group defect patterns from wafer maps into consistent clusters, addressing challenges like unlabeled, imbalanced data with multiple defects. DECOR utilizes deep embedding clustering with orientation robustness to ensure reliable clustering of defects regardless of rotation or alignment. The method outperforms existing clustering methods, offering a scalable solution for automated visual inspection systems. The paper evaluates DECOR using the MixedWM38 dataset, showcasing its ability to discover clusters without manual tuning.