Sb2Se3 Based Non-volatile Memory for Photonic Matrix-Vector Multiplications

This work integrates nonvolatile antimony selenide (Sb2Se3) phase shifters onto a Mach-Zehnder interferometer (MZI) for reconfigurable 2 × 2 photonic matrix-vector multiplications (MVMs). The Sb2Se3 phase shifters enable programming via a one-time 500 mW power input, offering significant potential for reducing power consumption when matrix configurations remain fixed. Consequently, it offers promise for the development of energy-efficient photonic accelerators for machine learning applications. Through comprehensive device characterization and genetic algorithm fitting, we achieved a normalized mean square error of 0.0494 between measured data and the ideal MZI model, validating the reliable implementation of programmable matrix multiplications on this platform.