Comparative Study of Bi-Layer and Tri-Layer Oxide Structures in Synaptic Devices

Von Neumann bottleneck in traditional computational architecture has spurred interest in neuromorphic computing, which integrates processing and memory in a single element for efficient data handling. Among neuromorphic hardware devices, oxide-based resistive random-access memory (RRAM) has gained attention due to its non-volatile nature, simple structure, low power consumption, and CMOS compatibility. Additionally, in RRAM, single-layer structures have limited control over conductive filament (CF) behavior, whereas bi-layer structures enhance switching characteristics through interfacial engineering, and tri-layer structures further improve reliability and analog synaptic performance by precisely controlling oxygen vacancy (VO) distribution. In this study, we fabricated bi-layer (Pd/ZrO2/HfO2/Pd) and tri-layer (Pd/ZrO2/ZnO/HfO2/Pd) synaptic devices to investigate the impact of structural design on resistive switching (RS) and neuromorphic performance. Insertion of ZnO, known for its highVOconcentration, into the tri-layer structure enhanced CF formation and rupture dynamics, resulting in improved RS behavior. The tri-layer device exhibited superior endurance, reduced variability, lower SET/RESET voltages, and more linear long-term potentiation (LTP) and long-term depression (LTD) characteristics compared to the bi-layer counterpart. These results demonstrate that tri-layer oxide structures incorporating vacancy-rich materials are promising for stable, low-power, and high-performance neuromorphic systems.