Artificial intelligence empowered metasurface design

Metasurfaces enable precise control over electromagnetic (EM) waves through subwavelength structural engineering, offering compact solutions for imaging, sensing, and information processing. However, the design of high-performance metasurfaces remains challenging due to the high-dimensional parameter space and complex, nonlinear EM interactions. To address these challenges, artificial intelligence (AI) has recently emerged as a powerful tool for metasurface inverse design, enabling efficient exploration of large design spaces and facilitating the discovery of non-intuitive structures. This review outlines recent progress in AI-assisted metasurface design, covering the underlying paradigms, algorithm fundamentals, and emerging application scenarios. We discuss the evolution of design strategies from conventional optimization approaches to modern data-driven and physics-informed frameworks, highlighting how these methods improve design efficiency and expand achievable device functionalities. Finally, we discuss the challenges and future perspectives of this emerging research direction, including data availability, model generalization, system-level complexity, and the gap between simulation and fabrication. With the rapid development of large-scale pre-trained models and expanding data resources, AI-driven metasurface design is expected to evolve toward more automated and efficient workflows. These developments may ultimately enable autonomous, scalable, and application-oriented metasurface engineering for next-generation intelligent photonic systems.