Demonstration of an oscillatory neuron using SiOx-based memristive switches

Artificial neural networks (ANN) and spiking neural networks represent the two mainstream approaches in artificial intelligence (AI) models. Oscillatory neural networks (ONN) offer a paradigm shift from both since ONNs can compute in phase space by utilizing phase differences in coupled oscillators [1] . This approach facilitates effective computation and contributes to energy efficiency, as the signal phase is prioritized over the amplitude. Few works have explored the design space of oscillatory neural networks using memristive elements [2] [3] . VO 2 -based volatile memristive switches have been previously employed to develop oscillatory neurons [4] , [5] . In this work, we have designed an oscillatory neuron circuit utilizing SiO x -based volatile as well as non-volatile (NV) memristive devices. Earlier works on memristive oscillators have used a fixed resistor between the supply voltage and the volatile memristor, thereby fixing the oscillation frequency. We employ an innovative yet straightforward approach of replacing the fixed resistor with an NV SiO x memristor. The system can be tuned to different oscillation frequencies depending on the different on-state resistance values of the NV device. The volatile SiO x memristors exhibit lower switching voltages and greater near-threshold non-linearity (switching-on slope of 0.3 mV/decade) in comparison to VO 2 memristors [6] . Our approach also offers excellent CMOS compatibility and the promise to integrate oscillatory neural networks in a single material platform.