The exponential growth of data necessitates innovative, high-capacity, low-cost, and energy-efficient storage solutions. Conventional magnetic-electric disks suffer from high energy consumption and vulnerability to electromagnetic interference. Optical holographic storage offers a promising alternative, leveraging the properties of light for volumetric data storage. Existing methods, however, face limitations in achieving high-density storage due to challenges in experimental setup and photonic structural design. This paper proposes a novel approach using a meta-disk with uncorrelated structural twist to overcome these limitations.

Optical holographic storage has been explored using various techniques including multiplexing methods that utilize variations in incident light properties like wavelength, polarization, and angle. While these methods have shown promise, they have not achieved the desired high storage densities due to practical implementation challenges. This research builds upon previous work on metasurface displays and encryption, aiming to significantly increase storage capacity using a novel multiplexing strategy.

The proposed meta-disk leverages a twisted diffractive neural network (TDNN) architecture. Each layer of the meta-disk consists of neuro-metasurfaces (NMs), where the twist angle between layers acts as a physical weight in the TDNN. The TDNN architecture enables control over complex optical behavior through rotated metasurfaces. The researchers used Pancharatnam-Berry metasurfaces as the basic neuron of the TDNN, fabricated using advanced 3D printing techniques. The TDNN training process involves randomly generating initial phase distributions for the NMs, computing diffraction using fast Fourier transforms (FFTs), and iteratively optimizing the transmission spectra using error backpropagation to minimize MSE and maximize correlation with target images. Lateral error analysis, including fabrication imperfections, is performed using nearest-neighbor and bilinear interpolation methods. The Fresnel number is optimized to ensure effective connectivity between neuro-atoms. Gaussian noise is introduced to simulate fabrication errors, evaluating the robustness of the system. The design of Pancharatnam-Berry metasurfaces is detailed, focusing on high polarization conversion efficiency. Numerical simulations are conducted to assess the capacity and scalability of the TDNN architecture. Experimental validation involves fabricating two 640 µm × 640 µm meta-disks using two-photon polymerization 3D printing. A detailed optical setup is described for experimental measurement and holographic image reconstruction, including an analysis of optical path calibration and error sources.

A two-layer 640 µm × 640 µm meta-disk is shown to store over hundreds of high-fidelity images with a structural similarity index (SSIM) of 0.8. The TDNN architecture allows for significant design flexibility and high storage capacity. Bilinear interpolation provides high accuracy and computational efficiency in handling lateral errors. The system demonstrates robustness against fabrication errors, with acceptable image quality even with a standard deviation of up to 30°. High polarization conversion efficiency is achieved with optimized Pancharatnam-Berry metasurfaces. Experimental results demonstrate dynamic focusing and lunar phase analogy holograms. Scalability analysis shows that increasing the number of layers and neuro-atoms improves storage precision. The combination of TDNN with frequency multiplexing is explored, showcasing the potential for further capacity enhancement. The average SSIM and PSNR values for 70 distinct OAM modes generated using a three-layer 512 × 512 neuro-metasurface were 0.4062 and 12.3001, respectively.

The results demonstrate the feasibility of high-capacity holographic storage using the proposed meta-disk and TDNN architecture. The novel structural multiplexing method significantly expands design flexibility compared to traditional multiplexing techniques based solely on varying optical properties. The robustness to fabrication errors makes the approach practical. Future research will focus on improving the meta-disk design by incorporating metasurface coupling effects using graph neural networks, aiming to enhance the quality of holographic images. The broad applicability of TDNN in diverse fields including 6G communication and photonic computing is also highlighted.

This work presents a significant advancement in optical holographic storage through the introduction of the meta-disk based on a TDNN architecture. The high capacity, robustness, and design flexibility demonstrated offer significant potential for next-generation optical storage and processing technologies. Future research directions include improving the system through graph neural network integration and exploring further capacity enhancements using combined multiplexing techniques.

The current experimental demonstration utilizes relatively simple images due to challenges in precise fabrication and experimental calibration. While the system shows robustness to certain levels of fabrication error, further improvements in 3D printing precision could enhance image quality. The analysis of metasurface coupling effects is planned for future work, as this could further improve performance.