Photonic Neural Networks And Optical AI Accelerators: A Comprehensive Review Of Architectures, Material Platforms, And System-Level Challenges

The rapid advancement of artificial intelligence (AI), particularly deep learning and large-scale neural networks, has created significant demand for high-performance and energy-efficient computing architectures. Conventional electronic processors such as GPUs and TPUs are increasingly constrained by power consumption, memory bandwidth limitations, and data movement bottlenecks. In response, photonic neural networks and optical AI accelerators have emerged as promising alternatives that exploit the properties of light to perform computation at high speed and low energy consumption. This paper presents a comprehensive systematic narrative review of photonic neural networks and optical AI accelerators, focusing on their architectures, material platforms, and key engineering challenges. The methodology employed involves structured literature collection from recent peer-reviewed studies, thematic classification of photonic architectures (including Mach–Zehnder interferometer meshes, microring resonator networks, and diffractive optical systems), and comparative analysis of material platforms and performance metrics such as energy efficiency, scalability, and computational latency. The results of the review indicate that photonic systems offer significant advantages over electronic computing, particularly in terms of energy per multiply–accumulate operation (femtojoule-level), ultra-high bandwidth (terahertz range), and low-latency computation. However, practical deployment remains limited by challenges in scalability, fabrication variability, noise sensitivity, and the lack of efficient optical training mechanisms. The analysis further shows that hybrid photonic–electronic architectures currently represent the most viable pathway toward near-term implementation, while heterogeneous material integration is essential for achieving fully functional photonic AI systems. The contribution of this research lies in providing a structured and critical synthesis of recent advancements in photonic AI hardware, identifying key technological bottlenecks, and outlining future research directions toward scalable and commercially viable optical computing systems. This work serves as a reference framework for researchers working at the intersection of photonics, electronics, and artificial intelligence.