Universal Reconfigurable Diffractive Photonic Chip for Multi‐Modal Optical Computing

Optical computing represents a transformative hardware paradigm for the post‐Moore era, exploiting its inherent advantages of high speed, low power consumption, and intrinsic parallelism. Nonetheless, state‐of‐the‐art integrated optical computing systems suffer from critical limitations, including fixed functionality, single input modality, and restricted input scalability, which severely constrain their adaptability in dynamic application scenarios and their capability for high‐dimensional data loading. In this work, we present a universal reconfigurable diffractive photonic chip (URDPC) that enables adaptive processing of on‐chip transverse electric modes, free‐space structured light fields, electrical logic signals, and high‐dimensional task reasoning. Our architecture supports dynamic optical mode routing and sorting, as well as both homogeneous and heterogeneous parallel logic operations. By implementing an on‐chip data loading scheme that directly encodes high‐dimensional data into the reconfigurable modulation units of the chip, we demonstrate 84% classification accuracy on a four‐class MNIST task, effectively alleviating the inherent trade‐off between input dimensionality and data volume. Our results establish a promising technical route toward building high‐performance, highly flexible integrated intelligent photonic computing systems for multi‐modal information processing.