Understanding Rhythmic Synchronization of Oscillatory Neural Networks Based on NbOx Artificial Neurons for Edge Detection

Oscillatory neural networks (ONNs) directly emulate signal communication between biological neurons in the human brain by encoding the data in phase domain, enabling energy-efficient associative memory. An oscillation neuron (ON) element that generates continuous voltage spikes with a specific frequency needs to be designed for hardware implementation. Thus, we systematically investigate the role of the ON in edge detection in ONN systems through simulation. First, a threshold switch is experimentally fabricated for the ON using niobium oxide (NbO
$_{x}{)}$
material, and voltage oscillation is realized in HSPICE and MATLAB. Subsequently, we examine how each voltage oscillation in a coupled-ON system, in which two NbOx-based ONs are connected with a coupling resistance, is mutually synchronized. Simulation results reveal that a small (or large) coupling resistance strengthens the in-phase (or out-of-phase) synchronization of the two independent oscillations. The synchronized phase expressed in the form of period is found to be adjusted by tuning various components. As two clearly distinguishable phases are obtained, ONN systems, where multiple ONs are cross-coupled, can be utilized for edge detection during image processing. Patterns are trained using Hebbian learning rule in an ONN system comprising ten ONs, and a feature of the handwritten digit image is accurately extracted. Moreover, the feasibility of accelerating the edge detection step is further explored through various engineering approaches to change the characteristics of the NbOx-based ONs.