Optical neural computing: from fourier optics to integrated photonic artificial intelligence

The growing computational demands of artificial intelligence have accelerated interest in optical computing as a high-speed and energy-efficient alternative to conventional electronic hardware. Fourier optical computing provides an important foundation for this field because operations including convolution, correlation, spectral transformation, and certain classes of linear transformation can be implemented through diffraction, interference, and Fourier-domain processing. This review traces the development of optical neural computing from classical 4-f processors and free-space diffractive systems to programmable optical neural networks, integrated photonic accelerators, reservoir-computing platforms, and field-programmable photonic architectures. The underlying physical principles, computational mechanisms, and hardware implementations are systematically examined, with particular attention to reconfigurable photonics, metasurfaces, diffractive deep learning, photonic integrated circuits, and hybrid optical–electronic systems. Emerging applications in computational imaging, autonomous sensing, edge intelligence, high-speed communications, and quantum-enabled information processing are critically discussed alongside challenges related to scalability, programmability, nonlinear computation, training, and system integration. By positioning Fourier optics as a foundational branch within the broader landscape of optical neural computing, this review highlights the convergence of free-space, integrated, programmable, and hybrid photonic technologies toward next-generation artificial-intelligence hardware.