Small Photoresist Defect Samples Augmentation Based on Generative Adversarial Network

Abstract

Photoresist coating technology is an important part of the surface treatment of semiconductor wafers. The presence of bubbles in photoresist can seriously affect the quality of wafers, however, the lack of sufficient bubble samples makes intelligent automatic detection techniques impossible. To solve such a problem, we propose B-GAN based on adversarial generative network, and design a mapping network for potential encoding mapping and a synthetic network for high-quality bubble image generation, so that defective bubble samples can be automatically generated. Experiments prove that our method achieves excellent results.