【International Papers】Self-Rectifying Dynamic Memristor Based on 2D-Graphene/V-doped Ga₂O₃ for Neuromorphic Processing

      Researchers from the Ajou University have published a dissertation titled "Self-Rectifying Dynamic Memristor Based on 2D-Graphene/V-doped Ga₂O₃ for Neuromorphic Processing" in Materials Today Physics.

Background

      Traditional digital computers follow the von Neumann paradigm, which separates memory and computation and creates a data-transfer bottleneck that limits throughput and energy efficiency-especially for AI tasks that demand real-time pattern recognition, continual learning, adaptation, and massive parallelism. Neuromorphic computing based on resistive random-access memory (RRAM) has therefore become a major direction for advancing AI and reservoir computing. RRAM devices are attractive for memory, logic, neural networks, and sensing because of their simple structures, high integration density, low power operation, and fast switching, while enabling in-memory processing that minimizes data movement. In metal-oxide RRAM, electric-field-induced SET/RESET produces multi-level, history-dependent states that emulate synaptic plasticity and support networks of artificial neurons and synapses. Importantly for wearable and conformal electronics, RRAM can be fabricated on flexible substrates (e.g., polyimide) with transparent or ultrathin electrodes (graphene, Ag-nanowires, ITO), sustaining reliable switching under small bending radii and repeated flexing, which makes them well-suited for flexible, real-time neuromorphic systems.

Abstract

      Memristors adjust their internal resistance in response to external stimuli, enabling random-access memory, neuromorphic computing, sensing, and in-memory processing functions that are difficult to achieve with conventional devices. These capabilities arise from diverse switching mechanisms, making mechanistic understanding and the discovery of new memristive operation modes crucial for reliable implementation and advanced applications. In this work, we present a flexible Au/V-doped a-Ga₂O₃/single-layer graphene (SL-Gr)/Cu memristor platform, benchmarked against a reference device without graphene, in which mechanical strain directly modulates the switching mechanism. The SL-Gr interlayer regulates Cu⁺ injection and diffusion so that Cu-rich nanoclusters nucleate predominantly at graphene defect and edge sites (ECM contribution), while concurrent oxygen-vacancy (V_o) enrichment and field-driven redox in the V-doped a-Ga₂O₃ contribute to a valence-change pathway (VCM contribution), yielding a controllable hybrid-filament system. Notably, under stable and flexible operation, the device exhibits key neuromorphic functions, including Pavlovian associative learning, image memorization in a 3 × 3 array of letter “C”, “F”, “I”, and “L”, 4-bit state encoding, and digit recognition within a reservoir computing (RC) framework, underscoring its potential as a promising candidate for next-generation artificial intelligent (AI) hardware.

Highlights

      A flexible, forming-free Au/V-doped amorphous (a)-Ga₂O₃/SL-Graphene/Cu memristor is realized, delivering stable self-rectifying and compliance-free resistive switching.

      Graphene interlayer precisely controls Cu ⁺ injection and enables a strain-tunable hybrid ECM-VCM mechanism with robust filament dynamics.

      The device emulates essential neuromorphic behaviors, including symmetric LTP/LTD, PPF, STM-LTM transition, Pavlovian learning, and 3 × 3 image memorization.

      A memristor-based reservoir computing system achieves deterministic multilevel states, 4-bit encoding, and accurate digit recognition for compact AI hardware.

Conclusion

      In summary, we demonstrate a self-rectifying, forming-free, and stable Au/V-doped a-Ga₂O₃/single-layer graphene (SL-Gr)/Cu memristor that unifies reliable resistive switching, rich neuromorphic plasticity, and hardware-efficient reservoir computing within a single flexible platform. Comprehensive structural, spectroscopic, and DFT analyses confirm the formation of amorphous V-doped a-Ga₂O₃ with mixed-valence V⁴⁺/V⁵⁺ states and engineered defect levels, enabling controlled electronic transport and providing a rational basis for the observed device characteristics. The insertion of an atomically thin graphene interlayer between Cu and V-doped a-Ga₂O₃ suppresses uncontrolled Cu penetration, owing to hybrid ECM-VCM soft-filament mechanism in which Cu-rich nanoclusters and oxygen-vacancy-enriched pathways co-operatively govern conduction. This architecture delivers self-limited, compliance-free switching with a large memory window, excellent cycle-to-cycle and device-to-device uniformity, and stable operation under bending, evidencing strain-tunable yet robust filament dynamics ideally suited for flexible electronics.

      Beyond memory function, the device operates as an artificial synapse capable of emulating key biological learning rules, including gradual and symmetric LTP/LTD, interval-dependent nonlinearity tuning, paired-pulse facilitation (PPF), STM-to-LTM transition, Pavlovian associative learning, and image memorization in a 3 × 3 array. These rich, tunable dynamics are further exploited in a proof-of-concept reservoir computing framework, where deterministic multilevel conductance states and transient responses enable robust 4-bit state encoding and digit recognition using a minimal array, underscoring the scalability and computational relevance of the proposed system. Collectively, our results position the 2D-single-layer graphene/V-doped Ga₂O₃ flexible memristor as a promising building block for next-generation flexible in-memory computing, neuromorphic systems, and compact, AI hardware.