CuAlO2-Based Memristor with Device-Aware Temporal Learning for Neuromorphic Applications.

Abstract

Memristive devices have emerged as promising candidates for neuromorphic computing because of their ability to emulate analog synaptic functionalities via electrical tuning of conductance states. However, achieving stable and reliable switching remains challenging in copper oxide systems due to multiple intrinsic defect-mediated transport processes existing simultaneously. In this work, we demonstrate CuAlO2-based memristors exhibiting stable bipolar resistive switching, analog conductance modulation, and synaptic plasticity suitable for neuromorphic applications. The structurally ordered delafossite CuAlO2 switching layer reduces competing intrinsic defect migration and promotes controlled Ag-mediated electrochemical metallization dynamics. Furthermore, the incorporation of an ultrathin SnO2 interlayer introduces a p-n heterojunction that modifies the local electric-field distribution and improves conductive filament confinement, resulting in improved switching properties. The memristor further exhibits key synaptic functionalities, including paired-pulse facilitation, potentiation-depression behavior, and synaptic plasticity. To evaluate its neuromorphic capability, a device-aware spiking neural network incorporating experimentally measured memristor dynamics was implemented for neuromorphic learning tasks like MNIST handwritten digit classification and physiological signal classification using ECG arrhythmia data. These results establish CuAlO2-based memristors as a promising platform for physically realistic neuromorphic computing and demonstrate the importance of integrating device-level switching dynamics into neural network design.