Enhanced Reliability and Neuromorphic Functionality of a-IGZO-Based Memristors via N2 Annealing-Induced Vacancy Modulation

Amorphous indium-gallium-zinc oxide (a-IGZO)-based memristors have attracted considerable attention for neuromorphic computing applications owing to their structural uniformity, amorphous-phase stability, CMOS compatibility, and oxygen-vacancy-mediated resistive switching characteristics. However, uncontrolled oxygen-vacancy concentrations can lead to unstable charge transport and switching variability, limiting reliable device operation. In this work, post-deposition N2 annealing was employed to regulate the oxygen-vacancy-related defect states of Pd/HfO2/a-IGZO/HfO2/Pd memristors. The a-IGZO layer was annealed in an N2 atmosphere at 300-600 °C, and the resulting changes in material properties, resistive switching behavior, and synaptic characteristics were systematically investigated. XPS and XRD analyses indicated that N2 annealing decreased the relative oxygen-deficient/defect-related O 1s component while maintaining the amorphous nature of the a-IGZO layer. Among the investigated conditions, the 500 °C-annealed device exhibited the most stable switching behavior, with reduced variation, improved retention and endurance characteristics, and reliable analog conductance modulation. Conduction mechanism analysis revealed that the optimized device maintained stable space-charge-limited current conduction, whereas excessive vacancy reduction at 600 °C promoted a premature transition toward the trap-filled-limit regime, resulting in degraded switching stability. Furthermore, the optimized device successfully emulated long-term potentiation/depression and paired-pulse facilitation, and its measured synaptic characteristics were validated through ANN-based MNIST pattern recognition and PPF-driven neuron circuit simulation. These results demonstrate that N2 annealing-induced oxygen-vacancy engineering is an effective strategy for improving the reliability and neuromorphic applicability of a-IGZO memristors.