Holographic multiplexing metasurface with twisted diffractive neural network

Zhixiang Fan, Chao Qian, Yuetian Jia, Yiming Feng, Haoliang Qian, Er‐Ping Li, Romain Fleury, Hongsheng ChenNature Communications 2024 · DOI

Abstract

As the cornerstone of AI generated content, data drives human-machine interaction and is essential for developing sophisticated deep learning agents. Nevertheless, the associated data storage poses a formidable challenge from conventional energy-intensive planar storage, high maintenance cost, and the susceptibility to electromagnetic interference. In this work, we introduce the concept of metasurface disk, meta-disk, to expand the capacity limits of optical holographic storage by leveraging uncorrelated structural twist. We develop a physical twisted neural network to describe the optical behavior of the meta-disk and conduct a comprehensive lateral error analysis, where the meta-disk stores large volumes of information through internal structural multiplexing. Two-layer 640 µm x 640 µm meta-disk is sufficient to store over hundreds of high-fidelity images with SSIM of 0.8. By harnessing advanced three-dimensional (3D) printing technology, optical holographic storage is experimentally demonstrated with Pancharatnam-Berry metasurfaces. Our technology provides essential backing for the next generation of optical storage, display, encryption, and multifunctional optical analog computing.

Introduction

The global explosion of data driven by AI-generated content and Industry 4.0 has pushed current storage solutions to their limits. Conventional magnetic and electronic disks are increasingly burdened by high energy consumption, maintenance costs, and vulnerability to electromagnetic interference. While optical holographic storage offers a high-capacity, volumetric alternative, its practical implementation has been hindered by complex experimental setups and limited degrees of freedom for data multiplexing.

This research introduces the meta-disk, a novel approach that merges nanophotonics with deep learning. By treating physical layers of metasurfaces as a "Twisted Diffractive Neural Network" (TDNN), the researchers have unlocked a method to store massive amounts of information in a microscopic footprint, controlled through structural rotation.

The Meta-Disk and TDNN Framework

The core of this technology is the meta-disk, which consists of a stack of two or more "neuro-metasurfaces." Unlike standard neural networks where weights are numerical values, the TDNN utilizes the physical twist angles between metasurface layers as its "physical weights."

Conceptual diagram of the meta-disk and twisted diffractive neural network (TDNN) architecture.

Rotating one layer relative to another creates "twin" diffraction domains. These domains are isomorphic yet functionally distinct, allowing the system to switch between different stored images simply by adjusting the rotation angle ($\alpha$). This approach treats the physical orientation of the structure itself as a primary, uncorrelated degree of freedom, exponentially increasing the available design space for multiplexing.

Optimization and Fabrication

To achieve high-fidelity storage, the researchers employed an error backpropagation algorithm to optimize the phase distribution of "neuro-atoms" (individual unit cells on the metasurface). A key finding was that bilinear interpolation provided superior accuracy for the physical model compared to nearest-neighbor methods.

The physical disks were fabricated using advanced 3D laser printing (two-photon polymerization). These Pancharatnam-Berry (PB) metasurfaces consist of rectangular polymer nanopillars on an $SiO_2$ substrate, allowing for precise sub-wavelength control of light-matter interaction.

Results and Performance

The system was evaluated using various benchmarks, including cinematic sequences (e.g., "blooming flowers"), lunar phase analogies, and over 70 distinct Orbital Angular Momentum (OAM) modes. A two-layer meta-disk measuring only 640 µm x 640 µm successfully stored and retrieved hundreds of images.

Reconstructed holographic image at a specific twist angle (alpha = 260 degrees) showing high correlation (CC = 0.95).

The reconstructed images achieved a Structural Similarity Index (SSIM) of 0.8, with some individual frames reaching correlation coefficients (CC) as high as 0.95. For OAM multiplexing, the system maintained a purity exceeding 80% across 70 modes, demonstrating robust handling of orthogonal image multiplexing.

Conclusion and Future Outlook

The TDNN approach provides a universal method for handling complex, high-resolution scenes, moving beyond the limitations of traditional Moiré metasurfaces. While the technology faces a trade-off between the number of stored images and crosstalk, it remains resilient to fabrication noise.

Future work will explore hybrid multiplexing—combining structural twist with wavelength and polarization—and the transition to active materials for real-time, reconfigurable optical computing. This technology provides essential backing for the next generation of optical storage, display, and encryption.

Summary updated

Abstract

As the cornerstone of AI generated content, data drives human-machine interaction and is essential for developing sophisticated deep learning agents. Nevertheless, the associated data storage poses a formidable challenge from conventional energy-intensive planar storage, high maintenance cost, and the susceptibility to electromagnetic interference. In this work, we introduce the concept of metasurface disk, meta-disk, to expand the capacity limits of optical holographic storage by leveraging uncorrelated structural twist. We develop a physical twisted neural network to describe the optical behavior of the meta-disk and conduct a comprehensive lateral error analysis, where the meta-disk stores large volumes of information through internal structural multiplexing. Two-layer 640 µm x 640 µm meta-disk is sufficient to store over hundreds of high-fidelity images with SSIM of 0.8. By harnessing advanced three-dimensional (3D) printing technology, optical holographic storage is experimentally demonstrated with Pancharatnam-Berry metasurfaces. Our technology provides essential backing for the next generation of optical storage, display, encryption, and multifunctional optical analog computing.
BibTeX
@inproceedings{fan2024holographic,
  author = {Zhixiang Fan and Chao Qian and Yuetian Jia and Yiming Feng and Haoliang Qian and Er‐Ping Li and Romain Fleury and Hongsheng Chen},
  title = {{Holographic multiplexing metasurface with twisted diffractive neural network}},
  booktitle = {Nature Communications},
  year = {2024},
  doi = {10.1038/s41467-024-53749-6},
}
Figure