Metasurfaces enabled dual-channel complex-amplitude hologram designed with neural network

Abstract

In this work we present a polarization-multiplexing metasurface hologram designed with neural network, which can modulate the amplitude and phase of incident light with four nanopillars comprised unit structures with high efficiency. Then, to improve image quality by minimizing the difference between the complex amplitude field modulated by the structure and that obtained through the network, we developed an end-to-end framework-assisted multifunctional metasurface design method based on the aforementioned neural network, which can directly map geometric parameters of the metasurface to holographic images. The comparative results indicate that for the end-to-end framework, although the optimized distributions of complex amplitudes are limited by the selected finite number of structures, in the simulation results, higher-quality images can still be obtained compared to the forward design method. Full-wave simulation results show our proposed method can obtain higher quality holographic images compared to the GS algorithm. This work may open new possibilities in holographic displays and multifunctional optical devices.