Mach Zehnder Interferometer-Based Photonic Matrix Multiplication for AI Acceleration Devices
Mursal Ayub Hamdani, Umar Farooq, Akash Kumar Pradhan · 2026
The escalating computational demands of artificial intelligence have exposed the limitations of conventional electronic accelerators, which face fundamental bottlenecks in energy efficiency and computational scaling. This work presents a comprehensive analysis of photonic matrix multiplication based on Mach-Zehnder interferometer (MZI) meshes as a transformative approach to AI acceleration. Our investigation demonstrates that MZI-based photonic accelerators achieve better energy efficiency over state-of-the-art GPUs, with theoretical projections exceeding 1000 TOPS/W. Furthermore, our analysis reveals that photonic systems enable multi-terabit bandwidth scaling through wavelength division multiplexing while maintaining orders-of-magnitude better thermal efficiency compared to electronic counterparts. Despite practical limitations including insertion loss, thermal crosstalk, and fabrication variations, photonic accelerators maintain stable performance even for large matrix operations, establishing their potential as the foundation for next-generation AI hardware capable of sustainable scaling beyond Moore&s;s Law limitations.