Recomposable Layered Metasurfaces for Wavelength‐Multiplexed Optical Encryption via Modular Diffractive Deep Neural Networks
Cherry Park, Youngsun Jeon, Seokho Lee, Yeseul Kim, Junsuk Rho
Abstract
Abstract Metasurfaces, composed of arrays of subwavelength scale meta‐atoms, enable precise control of light, paving the way for advanced holographic functionalities in compact, planar devices. Here, a modular diffractive deep neural network (MD 2 NN) architecture is proposed, where each metasurface layer operates as an independent hologram while jointly performing tasks such as optical encryption. The MD 2 NN architecture allows for holographic encoding across all possible combinations of metasurface layers. For m wavelengths and N metasurface layers, the system can achieve up to m (2 N − 1) distinct channels. A metasurface holography device is experimentally demonstrated capable of reconstructing distinct holograms such as identification codes, QR codes, and encrypted passwords, depending on the incident wavelengths and the configuration of the layers. The password hologram is accessible only under specific interlayer spacing, which serves as a physical security key. This work establishes a new paradigm for multifunctional and secure optical devices by integrating principles of diffractive deep neural network (D 2 NN) with advanced metasurface design.