Neuro-Metasurfaces for Customizing Far-Field Pattern Without Iteration

Research in metasurfaces have led to significant breakthroughs, enabling a wide range of novel physical phenomena and advanced devices. However, most existing metasurfaces are either static or rely on trial-and-error methods, limiting their efficiency and applications. In this work, we introduce the concept of neuro-metasurfaces that can automatically respond to user demands and external stimuli without iteration. We develop a generative network for inverse design of global metasurfaces, achieving over 90% accuracy using simulated far-field data. Our approach is validated through a preliminary proof-of-concept application. This work presents non-iterative active metasurfaces, opening new possibilities for practical-oriented applications. It represents a significant advancement, revolutionizing the field and paving the way for further development.