Neuromorphic computing with optomechanical oscillators

TL;DR

This paper explores optomechanical oscillator networks as a novel neuromorphic computing platform, providing a theoretical framework and demonstrating potential for implementing logic gates like XOR.

Contribution

It introduces a theoretical model for optomechanical oscillator networks in neuromorphic computing and discusses their training and physical implementation.

Findings

Modeling of an XOR gate with 5 nodes demonstrates computational capability.

Proposed platform based on drum resonators for physical realization.

Theoretical framework describes dynamics of optomechanical oscillator networks.

Abstract

The increasing resource demands of artificial neural networks have prompted the exploration of novel platforms better suited for machine learning. In this context, phase oscillators represent a promising candidate due to their intrinsic nonlinearity and their ability to exhibit collective synchronization when coupled together. In the present work, we investigate one such implementation: a network of optomechanical oscillators pumped in the blue-detuned regime to achieve self-sustained oscillations. We propose a theoretical framework to describe their dynamics and demonstrate how such systems can be employed for neuromorphic computing. We discuss how they can be trained and analyze a platform, based on drum resonators, that could enable their physical implementation. Ultimately, the theoretical results obtained from modelling an XOR gate using 5 nodes in an all-to-all configuration are discussed.