| Abstract | Silicon photonic devices often exhibit fabrication-induced variations such as over-etching, under-etching, and corner rounding, which can significantly alter device performance. These variations are nonuniform, influenced by feature size and shape. Accurate digital twins are needed to predict the range of possible fabricated outcomes for a given design. In this paper, we introduce Gen-Fab, a conditional generative adversarial network (cGAN) based on the Pix2Pix framework, to predict and model uncertainty in photonic fabrication outcomes. The proposed method takes a design layout (in GDS format) as input and produces diverse high-resolution predictions similar to scanning electron microscope (SEM) images of fabricated devices, capturing the range of process variations at the nanometer scale. To enable one-to-many mapping, we inject a latent noise vector at the model’s bottleneck. We compare Gen-Fab against three baselines: (1) a deterministic U-Net predictor, (2) an inference-time Monte Carlo Dropout U-Net, and (3) an ensemble of varied U-Nets. Evaluations on an out-of-distribution dataset of fabricated photonic test structures demonstrate that Gen-Fab outperforms all baselines in accuracy and uncertainty modeling, and an additional distribution-shift analysis confirms its strong generalization to unseen fabrication geometries. Gen-Fab achieves the highest intersection-over-union (IoU) score of 89.8%, outperforming the deterministic U-Net (85.3%), the MC-Dropout U-Net (83.4%), and Varying U-Nets (85.8%), and better aligns with the distribution of real fabrication outcomes, attaining lower Kullback–Leibler divergence and Wasserstein distance. |
|---|
