This article introduces an optical neural chip (ONC) designed to implement complex-valued neural networks, showcasing their advantages over real-valued counterparts in optical computing. The ONC integrates input generation, weight multiplication, and output generation on a single photonic chip, leveraging optical interference for efficient complex-valued arithmetic. Benchmarking across various tasks—Boolean operations, Iris species classification, nonlinear dataset classification (Circle and Spiral), and handwriting recognition—demonstrates the ONC's strong learning capabilities, including high accuracy, fast convergence, and the ability to construct nonlinear decision boundaries, surpassing its real-valued counterpart.

H. Zhang, M. Gu, X. D. Jiang, J. Thompson, H. Cai, S. Paesani, R. Santagati, A. Laing, Y. Zhang, M. H. Yung, Y. Z. Shi, F. K. Muhammad, G. Q. Lo, X. S. Luo, B. Dong, D. L. Kwong, L. C. Kwek, A. Q. Liu