Advancing Data Privacy in Semiconductor Manufacturing: Thumbnail Preserving Encryption for CycleGAN-Processed Wafer Images

Given the limited availability of defective wafer images on actual production lines, there are challenges associated with time constraints in their collection. To address the data imbalance issue concerning defect modes in wafer images, we utilize CycleGAN deep learning networks to devise a defect mode generation solution. Due to the risk of hardware damage, we store the generated wafer images in the cloud and employ an scheme similar to thumbnail preserving encryption to achieve a balance between privacy and security by presenting ciphertext in a low-resolution version of plaintext. This approach aims to provide user accessibility while enhancing automated production efficiency through an increased number of training samples for defect modes. The obtained results indicate a high resemblance between the generated defective wafer images and authentic defective wafer images. Furthermore, we compute the Peak Signal-to-Noise Ratio of the generated defective wafer images under various TPE schemes to assess their security. Experimental results demonstrate that the generated wafer images exhibit lower PSNR values and higher security under the R-LSB scheme.