Self-Rectifying Dynamic Memristor Circuits for Periodic LIF Refractory Period Emulation and TTFS/Rate Signal Encoding

Abstract

Spiking Neural Networks (SNNs) have gained attention due to their potential to improve computational efficiency compared to traditional artificial neural networks. This study investigates the use of dynamic memristors Ta/IGZO/TaO_x/Pt combined with peripheral circuits to emulate the leaky integrate-and-fire behavior of neurons, focusing on incorporating a refractory period. The refractory period is crucial as it prevents neurons from becoming overactive and ensures precise timing in signal processing. This improvement allows the memristor to mimic biological neuron behavior more accurately. The memristor's transient resistance exhibits nonlinear I–V hysteresis and changes in response to pulses, enabling functions of integration, leakage, and firing. Additionally, the memristor is configured as an encoder, converting external signals into voltage pulse sequences. Using coding methods, including rate coding and time-to-first-spike (TTFS) coding, the encoder demonstrates improved signal processing, with TTFS occurring within 21 to 62 ms and encoder frequencies from 2500 to 9500 Hz. Experimental results show that this approach enhances SNN performance, making it more suitable for real-time applications and complex temporal signal processing tasks. This research highlights the potential of dynamic memristors to bridge the gap between neurons and artificial neurons, paving the way for more efficient neuromorphic computing systems.