Design of multi-dimensional holographic meta-devices based on all-optical diffractive deep neural networks

To meet the demand for the coordinated manipulation of multi-dimensional electromagnetic degrees of freedom in holographic imaging systems, we propose a design method for multi-dimensional holographic meta-devices based on all-optical diffractive deep neural networks (D²NNs). The method combines the multi-layer spatial cascading mechanism of all-optical D²NNs with the angular spectrum theory, and constructs a multi-layer spatially cascaded structure composed of broadband pure-phase metasurfaces and linearly polarized metasurfaces, enabling the coordinated manipulation of multiple electromagnetic degrees of freedom, including frequency, polarization, spatial position, and transmission direction. Simulation shows that the designed meta-devices can achieve holographic imaging with high contrast and low crosstalk under multiple frequencies and polarization states, and feature multi-function integration, large storage capacity, providing a new design method for fields such as dynamic holographic display, optical encryption, and high-capacity information storage.