On-chip rewritable phase-change metasurface for programmable diffractive deep neural networks

TL;DR

This paper demonstrates an on-chip, rewritable phase-change metasurface for programmable diffractive deep neural networks, enabling rapid, low-power, and nonvolatile photonic neural computing at telecom wavelengths.

Contribution

It introduces a novel on-chip programmable neural network using Sb2Se3 phase-change metasurfaces created by direct laser writing, achieving high accuracy and low power consumption.

Findings

Achieved neural network performance comparable to state-of-the-art methods.

Demonstrated ultralow-loss, rewritable, and nonvolatile metasurfaces on-chip.

Validated on pattern recognition and MNIST digit classification tasks.

Abstract

Photonic neural networks capable of rapid programming are indispensable to realize many functionalities. Phase change technology can provide nonvolatile programmability in photonic neural networks. Integrating direct laser writing technique with phase change material (PCM) can potentially enable programming and in-memory computing for on-chip photonic neural networks. Sb2Se3 is a newly introduced ultralow-loss phase change material with a large refractive index contrast over the telecommunication transmission band. Compact, low-loss, rewritable, and nonvolatile on-chip phase-change metasurfaces can be created by using direct laser writing on a Sb2Se3 thin film. Here, by cascading multiple layers of on-chip phase-change metasurfaces, an ultra-compact on-chip programmable diffractive deep neural network is demonstrated at the wavelength of 1.55um and benchmarked on two machine learning tasks of pattern recognition and MNIST (Modified National Institute of Standards and Technology) handwritten digits classification and accuracies comparable to the state of the art are achieved. The proposed on-chip programmable diffractive deep neural network is also advantageous in terms of power consumption because of the ultralow-loss of the Sb2Se3 and its nonvolatility which requires no constant power supply to maintain its programmed state.