Abstract:A novel metasurface platform is proposed and experimentally demonstrated to mimick DNNs with unprecedented capability in massively parallel processing. The designed DNNs enable functionalities in identifying digits/fashions, wavelengths, and spatial sequences between multiple targets. Unlike traditional DNNs with a single function in classifying digits/fashions, this multiplexed DNN architecture opens up new avenues for further integrating versatile functions into multichannel metasurfaces. Benefiting from low power consumption and high processing speed, there is a growing interest in diffraction neural networks (DNNs), which are typically showcased with 3D printing devices, leading to large volumes, high costs, and low levels of integration. Metasurfaces can desirably manipulate wavefronts of electromagnetic waves, providing a compact platform for mimicking DNNs with novel functions. Although multi‐wavelength and multi‐target recognition provides a richer and more detailed understanding of complex environments, existing architectures are primarily trained to classify a single target at a specific wavelength. A metasurface approach is proposed to design multiplexed DNNs that can classify multiple targets and spatial sequences across various wavelengths in multiple channels. To realize multi‐task processing, the dielectric metasurface is designed based on phase and wavelength multiplexing, which can integrate multi‐target DNNs with different tasks such as operating at distinct wavelengths and classifying diverse targets. The efficacy of this method is exemplified through the numerical simulation and experimental demonstration of recognizing a single target with two wavelengths, two targets at a single wavelength, and two targets at dual wavelengths. This compact metasurface approach enables the design of multi‐target and multi‐wavelength DNNs, opening a new window to develop massively parallel processing and versatile artificial intelligence systems.

Matrix Diffractive Deep Neural Networks Merging Polarization into Meta‐Devices

Diffractive Deep Neural Network for Optical Orbital Angular Momentum Multiplexing and Demultiplexing

Adding/dropping Polarization Multiplexed Cylindrical Vector Beams with Local Polarization-Matched Plasmonic Metasurface.

Polarization Multiplexing Bifunctional Metalens Designed by Deep Neural Networks

Polarization multiplexed diffractive computing: all-optical implementation of a group of linear transformations through a polarization-encoded diffractive network

High-Capacity Metasurface at Limits of Polarization and Wavelength Multiplexing

Polarized deep diffractive neural network for classification, generation, multiplexing and de-multiplexing of orbital angular momentum modes

Metasurface Enabled Multi‐Target and Multi‐Wavelength Diffraction Neural Networks

Universal Polarization Transformations: Spatial programming of polarization scattering matrices using a deep learning-designed diffractive polarization transformer

Frequency and Angle Multiplexed Metadevices with Multifunctional Polarization Modulation

Diatomic Metasurface for Efficient Six‐Channel Modulation of Jones Matrix

Leveraging Multiplexed Metasurfaces for Multi-Task Learning with All-Optical Diffractive Processors

Exploring the limits of metasurface polarization multiplexing capability based on deep learning

Metasurface-enabled on-chip multiplexed diffractive neural networks in the visible

Arbitrary Jones matrix on-demand design in metasurface using multiple meta-atoms

Digitized Phase Change Material Heterostack for Diffractive Optical Neural Network

Universal on-chip polarization handling with deep photonic networks

Neural Network-Assisted End-to-End Design for Dispersive Full-Parameter Control of Meta-Optics

All dielectric metasurface based diffractive neural networks for 1-bit adder

Arbitrary Polarization Readout with Dual-Channel Neuro-Metasurfaces

Self-design of Arbitrary Polarization-Control Waveplates Via Deep Neural Networks