Exploring the potential of TiO2/ZrO2 memristors for neuromorphic computing: Annealing strategy and synaptic characteristics

Abstract

Artificial Neural Networks (ANNs) have reshaped computing paradigms, transcending traditional methods. Leveraging oxide-based bilayer RRAM memristors, specifically TiO2/ZrO2 deposited via sputtering, offers remarkable potential for RS memory and neuromorphic computing. This study pioneers an extensive annealing approach to counteract variability challenges in LRS and HRS during endurance tests. The Pt/TiO2/ZrO2/Pt memristor device's structural aspects are validated through cross-sectional high resolution transmission electron microscopy (HRTEM) analysis. Systematic XPS examination investigates the impact of annealing on oxygen vacancies. Successful bipolar resistive switching is unveiled through I-V characteristics, with 550°C annealing optimizing stable endurance cycling (1000 dc cycles). Conduction mechanisms during set/reset are illuminated, corroborated by Schottky emission fitting. Synaptic behavior emulation, Spike-Timing-Dependent Plasticity (STDP), and theoretical simulations with a 28×28 MNIST dataset underscore the ANN's 84.6% average recognition rate. The amalgamation of MNIST-based artificial learning and the innovative annealing strategy holds exciting potential for memory applications and advanced neuromorphic explorations. © 2024 Elsevier B.V.