Deep Neural Network Enabled Complex‐Amplitude Metasurfaces for Simultaneous Nanoprinting and Holography

Abstract

Metasurfaces offer a compact and scalable platform for multidimensional optical field manipulation and engineering. By judiciously designing subwavelength scatterers to tailor the complex transmitted field, a single metasurface can encode a prescribed near‐field distribution at its exit plane while simultaneously generating a desired far‐field response after propagation, thereby significantly increasing the degrees of freedom for high‐dimensional information encoding. However, achieving these near‐/far‐field functional responses across wavelength‐ and polarization‐multiplexed channels in a single device remains challenging. Here, we present an end‐to‐end data‐driven framework for multiwavelength and multipolarization metasurface design. A deep neural network is embedded in a gradient‐based optimization loop to design coherent superpixels, enabling simultaneous nanoprinting and holography across multiple channels. Using this framework, we experimentally demonstrate a single‐layer metasurface that supports nanoprinting and holography at three wavelengths and in three polarization channels, with multiplane holographic reconstruction. This strategy provides a versatile platform for high‐dimensional optical‐field manipulation, multifunctional optical displays, and integrated nanophotonic systems.