Programmable phase-change metasurfaces on waveguides for multimode photonic convolutional neural network

Unknown author(s)

Nature Communications · 2021 · 351 citations · 54 references

DOIFull text

Open access

Concepts

TL;DR

Neuromorphic photonics offers speed and energy advantages for machine learning, and integrated photonic networks excel at analog matrix‑vector multiplication, especially when augmented with phase‑change materials that provide on‑chip programming and in‑memory computing. The study demonstrates a multimode photonic core of programmable phase‑change metasurface mode converters. The converters exploit Ge‑Sb‑Te refractive‑index changes to tune waveguide modes to 64 discrete levels, encoding 6‑bit signed matrix elements for high‑precision matrix‑vector multiplication. The core implements an optical convolutional neural network that accurately processes and classifies images, offering broad bandwidth and a compact footprint that supports scalable high‑throughput photonic neural networks.

Abstract

Neuromorphic photonics has recently emerged as a promising hardware accelerator, with significant potential speed and energy advantages over digital electronics, for machine learning algorithms such as neural networks of various types. Integrated photonic networks are particularly powerful in performing analog computing of matrix-vector multiplication (MVM) as they afford unparalleled speed and bandwidth density for data transmission. Incorporating nonvolatile phase-change materials in integrated photonic devices enables indispensable programming and in-memory computing capabilities for on-chip optical computing. Here, we demonstrate a multimode photonic computing core consisting of an array of programable mode converters based on metasurface made of phase-change materials. The programmable converters utilize the refractive index change of the phase-change material Ge-Sb-Te during phase transition to control the waveguide spatial modes with a very high precision of up 64 levels in modal contrast. This contrast is used to represent the matrix elements, with 6-bit resolution and both positive and negative values, to perform MVM computation in neural network algorithms. We demonstrate an optical convolutional neural network that can perform image processing and classification tasks with high accuracy. With a broad operation bandwidth and a compact device footprint, the demonstrated multimode photonic core is very promising toward a large-scale photonic processor for high-throughput optical neural networks.

References

54