A Nonvolatile Programmable Photonic Crystal Nanobeam Cavity Based on Sb 2 Se 3 for Photonic Neural Networks
Yingjie Xu, Lidan Lu, Bofei Zhu, Li Yang, Guang Chen, Jinyi Du, Guanghui Ren, Jin Zhang, Zheng You, Lianqing Zhu · ACS Photonics · 2026
Photonic neural networks (PNNs) hold great promise for artificial intelligence (AI) acceleration computing with their high-speed and energy-efficient processing capabilities. However, it is still difficult to obtain compact, nonvolatile, and programmable photonic integrated components. Thus, a configurable photonic crystal nanobeam cavity (PCNC) with nonvolatility is introduced based on Sb 2 Se 3 for PNNs. Furthermore, the unique features of Sb 2 Se 3, such as low loss in the telecommunication band and significant change in the refractive index between crystalline and amorphous phases, are utilized. By integrating Sb 2 Se 3 into a PCNC structure, we achieve nonvolatile programmability of the PCNC. In addition, it is demonstrated that our device can be tuned to different resonant states, corresponding to various logic levels, making it suitable for configuring synaptic weights in PNNs. Using combinatorial selection and adjustment of seven segmented regions of Sb 2 Se 3, we achieve 32 discrete states for synaptic-like programmability. While integrating with a 5-bit quantized photonic convolutional neural network (PCNN), the system attains 98.78% accuracy on the MNIST data set. Therefore, our work presents a novel approach to developing nonvolatile, high-bit quantized, small-footprint, and scalable PNNs, opening up new potential pathways for applications in fields such as edge computing and image processing.