Applications and challenges of AI-based photonic device design

With the rapid advancement of information technology, photonic devices have become indispensable for high-speed communications and data processing, yet traditional design approaches face bottlenecks in efficiency and complexity. In recent years, artificial intelligence has emerged as a transformative solution for nanophotonics. This review establishes a unified methodological framework for artificial intelligence-driven photonic design, systematically categorizing existing technologies into generative models, discriminative solvers, underlying neural architectures, and data-efficiency strategies. Moving beyond a descriptive overview, we present a critical comparative analysis of these paradigms, offering specific selection guides to help researchers navigate tradeoffs regarding multi-modality handling, training stability, and inference speed. Furthermore, the review delves into the critical challenges impeding practical engineering adoption: bridging the simulation-to-reality gap through fabrication-aware design and deciphering ‘black-box’ decisions via advanced interpretability techniques. Finally, emerging frontiers such as photonic foundation models and closed-loop autonomous discovery systems are outlined to provide a roadmap for future research.