Nanotechnology. 2025 Dec 1. doi: 10.1088/1361-6528/ae2626. Online ahead of print.
ABSTRACT
Memristors with multilevel storage capabilities have emerged as promising candidates for high-density memory and neuromorphic computing systems. In this study, a trilayer-structured memristor with an Al2O3/HfO2/Al2O3 (3/14/3 nm) dielectric stack was fabricated via atomic layer deposition (ALD), sandwiched between Ti and Pt electrodes. The analog switching characteristics of the memristor were systematically investigated through two strategies: adjusting the compliant current (Icc) during the SET process and controlling the RESET-stop voltage (VRESET-stop) in the RESET process. The experimental results indicate that Icc primarily modulates the values of low resistance states (LRSs), whereas VRESET-stop mainly influences the values of high resistance states (HRSs). To validate multilevel storage feasibility, Icc values of 0.5, 1, 2.5, and 5 mA and VRESET-stop voltages of 1.5, 1.7, 2, and 2.3 V were systematically applied. Statistical analysis demonstrated that VRESET-stop modulation yields more stable and repeatable resistance states compared to Icc tuning. Furthermore, the continuous resistance (or conductance) tuning capability of our fabricated memristor emulates neural network weight updates. This allows trained weights to be directly mapped to the memristor’s conductance states, achieving 91.6% accuracy in handwritten digit recognition. This work underscores the significant potential of the Al2O3/HfO2/Al2O3 trilayer-structured memristor for high-performance multilevel storage and neuromorphic computing applications.
PMID:41325628 | DOI:10.1088/1361-6528/ae2626
