Deep-Learning Enabled Multicolor Meta-Holography

Multicolor holography, which can store and reconstruct wavefront information of optical waves at multiple wavelength channels, is demonstrated as a powerful platform for colorful image display. Recently, interleaved and segmented metasurfaces have emerged as appealing alternatives to realize the multicolor holography. However, the crosstalk among different wavelength channels can severely lower their performance. How to obtain the nanostructures with on-demand resonance wavelength, bandwidth, and phase delay is the key to overcome this challenge. Here, a hybrid framework composed of a neural network and an evolutionary strategy is proposed to implement the inverse design of nanostructures with desired resonance wavelength, bandwidth, and phase delay. With the proposed hybrid framework, the crosstalk between different wavelength channels can be eliminated by precisely controlling the resonance wavelength and the bandwidth of every nanostructure. As a proof of concept, a multicolor meta-holography for linear polarized light is experimentally and theoretically validated. The proposed hybrid framework provides a powerful platform for the design of metasurfaces for multi-frequency optical manipulation and multiplexing.

Automated summary

This study proposes a hybrid framework based on a neural network and an evolutionary strategy for the inverse design of nanostructures with desired characteristics. By precisely controlling the resonance wavelength and bandwidth of the nanostructures, crosstalk between different wavelength channels can be eliminated, achieving the concept validation of multicolor meta-holography.