Nonorthogonal Polarization-Multiplexed and Detachable Diffractive Processors

Polarization multiplexing provides an elegant route for parallel all-optical processing, yet its intrinsic orthogonality fundamentally limits channel capacity and interlayer interaction in diffractive systems. We present a detachable diffractive processor that leverages nonorthogonal polarization multiplexing to overcome these bottlenecks. The processor features a convolutional-like architecture composed of detachable diffractive layers, enabling global and interlayer function multiplexing with accelerated training and expanded versatility. Experimentally, we realize 12 independent functions-4 classifiers and 8 holograms-across 4 polarization-decoupled channels in a dual-layer system and extend to 18 channels supporting 6 classifiers and 36 holograms using a triple-layer configuration. Furthermore, we demonstrate a multi-level optical encryption framework that combines image transformation and holographic generation. This work establishes a scalable framework for multifunctional diffractive computing, enabling compact processors, high-capacity holography, and secure optical information technologies.