Search papers, labs, and topics across Lattice.
This paper introduces SePArate, a weakly supervised method for segmenting wafer defects in semiconductor manufacturing, addressing the limitations of manual inspection and existing automated methods. By utilizing only image-level annotations, SePArate achieves pixel-level defect segmentation through a novel three-phase training process that includes encoder pretraining and knowledge transfer. Experimental results show that SePArate significantly outperforms traditional baseline methods, enhancing root-cause analysis and process optimization in wafer manufacturing.
Weak supervision allows SePArate to achieve pixel-level segmentation of wafer defects without the need for extensive labeled data, outperforming existing methods.
In semiconductor manufacturing, defect analysis is essential, but manual inspection cannot scale. However, existing automated inspection methods remain insufficient for root-cause analysis and process optimization. To this end, we present SePArate, a weakly supervised wafer defect segmentation method. SePArate enables pixel-level separation of patterns by leveraging only image-level annotations. It consists of a three-phase training: encoder pretraining, knowledge transfer to learn spatial cues, and training on synthetic mixed-defect data for accurate segmentation. Experiments demonstrate that SePArate outperforms the baselines.
