Achieving Functional Meta-Devices by Generalized Meta-Atom Model for Metasurfaces and Genetic Algorithm

Built by machine-learning, the surrogate model of metasurfaces reduces the need for a huge number of simulations in the design process, enhancing the efficiency and performance of the designed meta-devices. However, the surrogate model of metasurfaces is often constructed-based on specific physical perspectives or experiences, which limits its versatility. In this study, a generalized surrogate meta-atom model for metasurfaces is introduced. This model can simulate arbitrary meta-atoms and their corresponding electromagnetic responses at any polarization within the full space of pixelated unit cells. Utilizing a genetic algorithm, the model is employed to design various types of meta-devices, automatically generating configurations of meta-atoms with optimal performance for specific application scenarios. Three typical meta-devices, including the reflective linear-circular polarization converter, the metasurface-based absorber, and the asymmetrical transmission meta-slab, are designed and validated through full-wave simulations and/or experiments. This work presents an efficient and flexible approach to model arbitrary metasurfaces, opening new possibilities for metasurface design and applications. The use of a single machine-learning-based meta-atom model is demonstrated to enable the design of various functional meta-devices, verified through simulations and experiments. By pixelating the unit cell of metasurfaces, the meta-atom model is developed to cover the entire pixel space and simulate any potential meta-atoms operating at any polarization. This enables significantly improving the design efficiency for diverse meta-devices.image