Diffractive Optical Neural Networks: Pathways Toward Intelligent and Scalable Photonic Computing
Yufei Wang, Mingheng Zhong, Kun Liao, Quan Sun, Xiaoyong Hu, Qihuang Gong · Nanophotonics · 2026
ABSTRACT Optical neural networks have witnessed remarkable progress in recent years, driven by the demand for parallel, high efficiency, and ultrafast information processing. Among various optical neural network architectures, diffractive optical neural networks (DONNs) have emerged as a particularly promising approach due to their scalability in the neuron number, ultrahigh throughput, and passive low‐power operation. Although free‐space implementations laid the foundation for DONNs, on‐chip implementations were later developed to overcome integration challenges and to enhance compatibility with photonic circuits, forming together the two primary platforms of DONNs. Across both free‐space and on‐chip implementations, increasing efforts have been devoted to reconfigurability and multidimensional multiplexing, enabling more flexible and sophisticated optical computing functions. Beyond their intrinsic advantages for imaging and sensing, DONNs have been increasingly adopted for broader neural computing tasks, extending toward large model‐level functionalities. This review provides a comprehensive overview of DONNs, covering their theoretical foundations and structural evolution from free‐space systems to integrated on‐chip and hybrid photonic configurations. It further summarizes recent progress in enhancing computational flexibility through multidimensional modulation and reconfigurable control strategies. Additionally, representative applications and emerging directions are discussed. Finally, the review identifies key challenges and outlines prospective pathways toward intelligent and scalable photonic computing systems.