Neural Network-Assisted End-to-End Design for Full Light Field Control of Meta-Optics

Abstract

Meta-optics, with unique light-matter interactions and extensive design space, underpins versatile and compact optical devices through flexible multi-parameter light field control. However, conventional designs struggle with the intricate interdependencies of nano-structural complex responses across wavelengths and polarizations at a system level, hindering high-performance full-light field control. Here, a neural network-assisted end-to-end design framework that facilitates global, gradient-based optimization of multifunctional meta-optics layouts for full light field control is proposed. Its superiority over separated design is showcased by utilizing the limited design space for multi-wavelength-polarization holography with enhanced performance (e.g., ≈6 × structural similarity index experimentally). By harnessing the dispersive full-parameter Jones matrix, orthogonal tri-polarization multi-wavelength-depth holography is further demonstrated, breaking conventional channel limitations. To highlight its versatility, non-orthogonal polarizations (>3) are showcased for arbitrary polarized-spectral multi-information processing applications in display, imaging, and computing. The comprehensive framework elevates light field control in meta-optics, delivering superior performance, enhanced functionality, and improved reliability, thereby paving the way for next-generation intelligent optical technologies.