Overcoming Small Sample Size Challenges In Lithography Process Window Analysis Through DCGAN-Based Augmentation

Developing an efficient and accurate automated process window analysis method for the lithography process is a challenging small sample size problem due to the difficulty of acquiring sufficient training data. To address this, we propose using deep convolutional generative adversarial networks to augment and expand the lithography process window dataset. With this approach, we can train a lithography process window analysis tool with a smaller amount of original data while achieving performance metrics comparable to those obtained by training with a large amount of raw data. The experiments show that the required size of raw data for training can be reduced by approximately 92%.