Characteristics of multi-level resistance states in hafnium oxide-based nonvolatile memristors

Abstract HfO 2 -based nonvolatile memristors have attracted broad research interest in neuromorphic computing. However, the stochastic nature of conductive filament formation poses significant challenges, including high operating voltages, low on/off ratios, limited endurance, and high device variability, which severely limits their applications in high-precision neuromorphic computing. To overcome these limitations and explore the potential of embedded metal nano‐island (NIs) arrays for performance enhancement, this study fabricated Pt/TaO x /HfO 2 /TiN, Pt/Ti/Al 2 O 3 /HfO 2 /TiN devices, and a Ta NIs‐embedded device with the structure Pt/Ti/Al 2 O 3 /HfO 2 /Ta NI/TiN, and investigated their corresponding resistive switching characteristics. The experimental results demonstrate that during multi‐level conductance state cycling tests, the Pt/TaO x /HfO 2 /TiN structure-benefiting from TaO x as a thermal enhancement layer-exhibits a relatively low coefficient of variation (CV), with its conductance states showing completely non‐overlapping distributions. Notably, compared to the Pt/Ti/Al 2 O 3 /HfO 2 /TiN structure, the Pt/Ti/Al 2 O 3 /HfO 2 /Ta NI/TiN device with embedded Ta NIs achieves three core improvements: a notable decrease in operating voltages, greatly improved cycling consistency with significantly reduced CV of multi-level conductance, and effectively suppressed resistance drift in retention tests. Furthermore, the device shows excellent conductance state distinguishability in analog computing: its 8 independent conductance states remain fully separable under multiple pulse stimulation, with no overlap between adjacent states. These characteristics of HfO 2 -based nonvolatile memristors can provide supportive assurance for the overall accuracy of neuromorphic computing systems.