Abstract: We propose a UNet-based foundation model and its self-supervised learning method to address two key challenges: 1)lack of qualified annotated analog layout data, and 2)excessive variety in ...