A Bilayer Rare-Earth/High-κ Oxide Memristor for Energy-Efficient Neuromorphic Intelligence

The growing demand for brain-inspired computing systems has intensified research into energy-efficient, scalable, and adaptive hardware that mimics biological synaptic behavior. Neuromorphic memristor devices, which integrate memory and processing functionalities within a single nanoscale unit, are emerging as promising building blocks for next-generation artificial intelligence systems. In this work, we demonstrate a CMOS-compatible Ag/Gd₂O₃/HfO₂/Pt bilayer memristor engineered with atomically sharp interfaces and optimized defect landscapes to achieve stable and efficient resistive switching behavior. The device exhibits excellent performance, including an ON/OFF current ratio exceeding 10⁷, retention beyond 10⁴ s, a sub-microsecond switching transition time (350 ns), and low programming energy of just 13.6 pJ. Interface engineering effectively stabilizes multilevel conductance states, suppresses stochastic filament growth, and supports analog long-term potentiation and depression. Incorporating the experimentally measured synaptic plasticity into convolutional neural network simulations yields a 78% classification accuracy on the Fashion-MNIST dataset, along with robust color recognition. These results demonstrate, as a device-level proof of concept, that bilayer rare-earth/high-κ oxide memristors can inform the development of future non-volatile memory and low-power edge neuromorphic systems.