Physics-Embedded autoencoder for solving broadband light field modulation with metasurfaces

In this paper, we propose a novel AI-based method to arrange metasurface’s structures for light field modulation, which is based on an autoencoder embedded with the physics equation. For the problem of light-field modulation, we employ the Rayleigh–Sommerfeld equation within the autoencoder method. We employ this AI method to design a focus lens with a focusing efficiency of 32.5% to 56% within the 1000 to 1250 nm range, and a far-field imaging metasurface spanning the same wavelength range. Our full-wave simulations confirm the correction of our designs. • Propose a physics - embedded autoencoder for broadband metasurface inverse design. • Achieve broadband focusing lens and far-field imaging validated by simulations. • Demonstrate universality of the method by enabling flexible physical equations.