Energy-efficient neural network inference with microcavity exciton-polaritons

TL;DR

This paper introduces an all-optical neural network design using microcavity exciton-polaritons, achieving extremely high energy efficiency and performance density by leveraging optical nonlinearity without electronic conversion.

Contribution

It presents a novel neural network architecture based on microcavity exciton-polaritons, demonstrating ultra-low energy consumption and two types of nonlinear binarized nodes for optical computing.

Findings

Energy cost estimated at attojoules per bit, far below current hardware.

Design utilizes properties of photons and electrons for seamless optical nonlinearity.

Two nonlinear binarized node types proposed: phase shifts/interferometry and spin rotations.

Abstract

We propose all-optical neural networks characterized by very high energy efficiency and performance density of inference. We argue that the use of microcavity exciton-polaritons allows to take advantage of the properties of both photons and electrons in a seamless manner. This results in strong optical nonlinearity without the use of optoelectronic conversion. We propose a design of a realistic neural network and estimate energy cost to be at the level of attojoules per bit, also when including the optoelectronic conversion at the input and output of the network, several orders of magnitude below state-of-the-art hardware implementations. We propose two kinds of nonlinear binarized nodes based either on optical phase shifts and interferometry or on polariton spin rotations.