Learning from single-defect wafer maps to classify mixed-defect wafer maps

Jaewoong Shim, Seokho Kang

Abstract: Highlights•CNN is trained without mixed-defect wafer maps to classify mixed-defect wafer maps.•Wafer maps are synthesized using mixup, random rotation, and noise filtering.•Mixup combines single-defect wafer maps to create synthetic mixed-defect wafer map.•Random rotation increases diversity of synthetic wafer maps.•Noise filtering prevents accumulation of random noise on synthetic wafer maps.