ScholarWorks@동국대학교

초록

This study investigates polycrystalline tungsten (W)-diffused barium titanate (BaTiO₃) memristors, which demonstrate remarkable enhancements in both electrical and neuromorphic performance. Compared to their pure BaTiO₃ counterparts, the W-diffused memristors exhibit reduced forming, set, and reset voltages, thereby enabling energy-efficient operation. The W-diffused BaTiO₃ memristors achieve stable cycle-to-cycle (C2C) endurance over 1200 DC switching cycles, with low power consumption (36.6 pJ for Set and 45.5 pJ for Reset) and robust non-volatile retention exceeding 10⁴ seconds. These devices also support multilevel switching, controlled through precise modulation of current compliance (ICC) and reset-stop voltages within the range of [sbnd]1 V to [sbnd]1.6 V. In addition to their electrical characteristics, the devices exhibit essential neuromorphic features, including long-term potentiation (LTP) and long-term depression (LTD), modulated by pulse parameters such as pulse number (50/50, to 110/110), width (10 µs to 50 µs), and amplitude. Core biological synaptic functionalities such as paired-pulse facilitation (PPF), post-tetanic potentiation (PTP), spike-voltage-dependent plasticity (SVDP), spike-number-dependent plasticity (SNDP), and synaptic metaplasticity were successfully emulated. A multibit neuromorphic system was experimentally realized using an incremental step pulse with verify algorithm (ISPVA), achieving stable 4-bit to 6-bit conductance states for high-density in-memory computing. Furthermore, the memristors exhibited nociceptive responses, enabling simulation of biological pain signals, and demonstrated Pavlovian associative learning behavior. Synaptic weight updates from the W-diffused BaTiO₃ memristors were implemented in a convolutional neural network (CNN) for CIFAR-10 image classification, achieving 91.3 % accuracy—closely matching the 91.7 % software baseline under optimized training conditions. These findings establish W-diffused BaTiO₃ memristors as strong candidates for next-generation, energy-efficient neuromorphic computing systems. © 2025 Elsevier Ltd

키워드