Silicon Photonics for Short‐Reach Interconnects: Progress in High‐Bandwidth and Energy‐Efficient Optical I/O

Abstract

Artificial intelligence (AI) is driving unprecedented demands on computing infrastructure, where increasing data movement between accelerators, memory, and networking components has become a critical challenge for system scalability. Silicon‐photonic optical interconnects have emerged as a promising technology to overcome the bandwidth density, energy efficiency, and scalability limitations of conventional electrical links in next‐generation AI clusters. Meeting these demands requires coordinated optimization across optical devices, electronic circuits, packaging technologies, and system architectures. This review analyzes the development of silicon‐photonic optical I/O technologies, highlighting their performance trade‐offs, deployment challenges, and future opportunities for scalable, energy‐efficient optical interconnect architectures for AI computing.