Harnessing Dynamics of Van Der Pol Oscillators for Phase Computing with Oscillatory Neural Networks

Abstract An oscillatory neural network (ONN) is a novel physics-based computing architecture that leverages the collective dynamics of coupled oscillators with sinusoidal Kuramoto oscillators being typically chosen. However, van der Pol (vdP) oscillators introduce a higher nonlinearity, which we explore for computations. This work presents the first in-depth study of ONNs with vdP oscillators, focusing on their capabilities in regard to associative memory tasks. We derive the Phase Transition Function (PTF), a crucial feature for ONN computing. Furthermore, we detail the circuit design and simulation results for implementing a vdP-ONN architecture. Analytical and circuit-level simulations demonstrate the benefits of vdP oscillators such as increased synaptic resolution, faster time-to-settle, larger coupling range, and superior noise robustness in comparison to standard sine wave oscillators.