Gen-Fab: A Variation-Aware Generative Model for Predicting Fabrication Variations in Nanophotonic Devices

2026-03-11

AI Summary

This work addresses the critical challenge of non-uniform fabrication imperfections—such as over-etching, under-etching, and corner rounding—in nanophotonic device manufacturing, which severely degrade performance and necessitate high-fidelity digital twin models capable of predicting diverse fabrication outcomes. To this end, the study introduces, for the first time, a variable-output conditional generative adversarial network (cGAN) that takes GDS layout files as conditional input and leverages latent-space noise injection to enable one-to-many mapping, thereby generating high-resolution, realistic SEM images in an end-to-end manner that captures manufacturing uncertainty. The proposed method demonstrates strong generalization to out-of-distribution geometries and significantly outperforms existing deterministic and probabilistic U-Net baselines across multiple metrics, including Intersection over Union (89.8%), KL divergence, and Wasserstein distance.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Silicon photonic devices often exhibit fabrication-induced variations such as over-etching, underetching, and corner rounding, which can significantly alter device performance. These variations are non-uniform and are influenced by feature size and shape. Accurate digital twins are therefore 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 Pix2Pix 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 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 both accuracy and uncertainty modeling. An additional distribution shift analysis further 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%). It also better aligns with the distribution of real fabrication outcomes, achieving lower Kullback-Leibler divergence and Wasserstein distance.

Problem

• fabrication variations
• nanophotonic devices
• process uncertainty
• digital twins
• geometric deviations

Innovation

• conditional generative adversarial network
• fabrication variation modeling
• one-to-many mapping
• nanophotonic device
• uncertainty quantification