Experimental Robust Spontaneous Synchronizations in Coupled NbOx Oscillation Neurons for Unconventional Computing

In this brief, we demonstrate experimentally achieved oscillation dynamics in coupled NbOx threshold switch (TS)-based oscillation neurons (ONs). We first show the challenges in observing oscillation caused by device variability, which is further validated through HSPICE simulations. To obtain robust oscillation response, a reliable TS developed via atomic layer deposition for conformal NbOx is introduced. By developing coupled NbOx ONs using a coupling resistor ( ${R}_{C}$ ), the out-of-phase synchronization is demonstrated owing to the weak connection. Moreover, we achieve a transition to in-phase synchronization as the coupling strength increases, showing an inverse relationship with the magnitude of ${R}_{C}$ . Based on these results, oscillatory neural networks, where multiple NbOx ONs are fully connected, are examined to execute pattern recognition via MATALB simulations.

I. Introduction

Conventional von Neumann computing, which performs sequential processing, causes latency and energy-consumption challenges in the data-driven era. Thus, various bioinspired systems that enable energy-efficient processing owing to their parallel structures have been explored [1], [2], [3]. In particular, oscillatory neural networks (ONNs), in which artificial oscillation neurons (ONs) are fully connected, have recently attracted considerable attention [4], [5]. ONNs have the ability to emulate realistic brain functions, making them valuable for both neuroscience research and efficient image processing in the field of neuromorphic engineering. For ONN hardware implementation, a threshold switch (TS) based on VO_x or NbO_x materials has been developed for nanoscale ONs [6], [7], [8], [9]. TS, which initially has a high-resistance state (HRS), exhibits a low-resistance state (LRS) as long as the applied voltage exceeds both the threshold voltage () and hold voltage () [10], [11]. This transition enables continuous oscillation in the form of voltage in the TS, serving ON, which is connected in series to a load resistor () [12]. In two coupled ONs connected by a coupling component, such as a resistor or capacitor, the oscillations start interacting and then spontaneously synchronize in an in-phase or out-of-phase fashion, depending on how strongly they are connected [13], [14], [15]. The ONN system utilizes data encoded in the phase domain based on two synchronization methods instead of judging the weighted sum current; thus, energy-efficient processing is expected [7]. To date, synchronization dynamics have predominantly been studied through simulations, and a few experimental observations using VO₂ materials have been performed [6], [8], [14], [16], [17]. More specifically, it has been demonstrated that injecting an external signal during oscillations in VO₂-based TS effectively locks the oscillation frequency, resulting in synchronization [17]. Additionally, adjusting the input time delay made the synchronization in the coupled system with a given coupling strength either in-phase or out-of-phase mode [14], [16]. Note that most of the achieved synchronization behaviors are driven by the external environment. Considering the manner in which biological neurons communicate via spontaneous synchronization, the balance of coupling strengths between ONs plays a vital role in determining the phase, which is known as the Kuramoto model [18].