• 목록

In semiconductor manufacturing, defect inspection is essential for quality assurance and process stability, but full inspection is often impractical because of high inspection cost. Sampling inspection has therefore been widely adopted; however, it inevitably causes loss of spatial information in defect maps and degrade defect identification performance. To address this issue, this study proposes a defect map reconstruction framework that reconstructs the entire wafer from partially inspected wafers. The reconstructed wafers show clear defect patterns compared to partially inspected wafers; thus, it is more informative and effective to detect process anomalies. Missing regions are restored using a Vision Transformer-based Masked Autoencoder. Experiments on the WM-811k dataset under various masking scenarios show that the proposed method effectively reconstructs defect patterns and significantly recovers classification performance degraded by masking.