Phase-change nanoclusters embedded in a memristor for simulating synaptic learning

Abstract

A type of memristor with a structure of Pd/Nb : AlNO/Pd was designed and fabricated in this work. Its filaments are embedded by phase-change NbO nanoclusters confirmed by the analysis of cross-sectional profiles. The resistive switching mechanism includes the contribution of oxygen vacancy (V_O) migration and the structural evolution of phase-change nanoclusters. The system experiences two types of kinetics under external stimulations to replicate the critical dynamics in real synapses: V_O migration corresponding to the dynamics of the Ca²⁺ flux and transmitter release at the pre-synapse, and the phase change of the NbO nanoclusters corresponding to the ionic flux modulated by the post-synaptic potential (current). It was found that the memristor can respond to a set of pulse stimulations in a pattern containing a slow linear increase term and a periodic oscillation term, suggesting that the output signals might be encoded. The simulation of long-term plasticity indicates that the memristor is suitable for diverse learning protocols, including spike-rate dependent plasticity and spike-timing-dependent plasticity. Our work proposes an elementary cell that closely approximates biological synapses and is usable for brain-like computing.