Designing Coupling Elements to Explore Thermal Fluctuations on Oscillatory Neural Networks

Oscillatory neural networks (ONNs) represent a novel, physics-inspired computing paradigm, showing promising advantages for solving NP-hard combinatorial optimization problems. ONN computing relies on the synchronization phenomenon of coupled oscillators, which exhibit different phase dynamics based on the coupling strengths. The two key elements in an ONN are the oscillators and the coupling elements. Coupling elements determine the network’s graph topology and edge density.

In this work, we design analog circuits for coupling elements that harness thermal fluctuations to modulate their coupling strength and, consequently, the phase relations between oscillators. We explore and investigate various ONN topologies, network sizes, and edge densities to understand the impact of thermal fluctuations on oscillator synchronization. We benchmark ONNs by embedding the max-cut problem and report on the influence of thermal fluctuations on problem-solving accuracy and performance.