Reliable Bi-Layer HfO2/Zr-Doped ZnO Memristive Synaptic Devices

Abstract

Oxide-based memristive synaptic devices are promising hardware elements for neuromorphic computing, but stochastic resistive switching caused by uncontrolled defect distributions and localized conductive paths remains a critical challenge. Here, we demonstrate Zr-doped ZnO memristive synaptic devices fabricated using an atomic layer deposition (ALD) super-cycle process to regulate both defect states and surface morphology. X-ray photoelectron spectroscopy revealed a decrease in the V_O-related O 1s component from 34.9% to 23.2%, while atomic force microscopy showed a reduction in RMS surface roughness from 0.7532 to 0.2603 nm with less distinguishable surface-grain features. These results suggest that Zr incorporation suppresses V_O-related defects and reduces grain-boundary-assisted localized conduction. Compared with the undoped device, the Zr-doped device exhibited improved switching uniformity, reduced resistance-state variation, stable pulse endurance, and retention for over <inline-formula> <tex-math notation="LaTeX">${10}^{{4}}$ </tex-math></inline-formula> s at 25, 50, and <inline-formula> <tex-math notation="LaTeX">$75~^{\circ }$ </tex-math></inline-formula>C. The regulated defect distribution and surface morphology enabled gradual and reproducible conductance modulation during long-term potentiation and depression, resulting in improved synaptic linearity. Neural network simulations using experimentally extracted LTP/LTD nonlinearity parameters further demonstrated enhanced recognition accuracy in ANN and CNN classification tasks.