Аннотация:All-optical diffractive deep neural networks (D2 NNs) offer significantadvantages in processing speed and power consumption, therebyaccelerating the development of optical computing and artificial intelligence(AI). Integrating multiple degrees of freedom (multi-DoF) into D 2 NNs is apivotal role in improving information processing and task-loading capacity, anenormous challenge in current all-optical diffractive computing/processors.Here, a multi-DoF diffractive processor is proposed and experimentallydemonstrated that leverages a metasurfaces-based approach to integratepolarization, distance, and rotation channels for versatile inference tasks andinformation encryption. The approach is validated using three-layermetasurfaces that enable high task-capacity tasks, including single-/dual-digitand single-/dual-fashion-product classification, logic operators, and imagetransformation. Moreover, by mapping large volumes of input data intomulti-DoF channels and encoding the information in Morse code with ourD2 NNs framework, a high-security information transmission system isexperimentally implemented. The integration of polarization, distance, androtation channels into an all-optical diffractive processor with multifunctionalcapabilities paves the way for multifunctional integrated devices andcommunication.