TiO₂ thin-film-on-graphene memristors: a high-uniformity, multilevel-capable platform for ultralow-energy...

Memristors that combine metal-oxide thin films with two-dimensional graphene electrodes hold immense promise for high-density, energy-efficient neuromorphic hardware. Here, we demonstrate a vertical memristor based on a single-crystalline rutile TiO₂ thin film (~150 nm) sandwiched between a chemical-vapor-deposited monolayer graphene bottom electrode and an Ag top electrode. The atomically flat graphene/TiO₂ interface yields exceptional resistive switching uniformity, ultralow leakage currents, and a high resistance ratio. The devices exhibit forming-free bipolar switching with an ultralow set voltage of +0.45 V and reset voltage of -0.35 V, a high ON/OFF resistance ratio exceeding 2 × 10⁵, and a read endurance beyond 10⁷ cycles without degradation. High-temperature retention measurements confirm stable resistance states for over 10⁵ s at 85 °C with an Arrhenius activation energy of 1.1 eV, and cycle-to-cycle and device-todevice variability in both high-and low-resistance states remains below 3.2% (n = 20 devices, 100 cycles each). By precisely controlling the compliance current, we achieve 64 distinct, linearly spaced conductance levels, enabling 6-bit-per-cell multilevel storage. Energy consumption per switching event is calculated as 87 fJ (E = V × I × τ = 0.45 V × 3.9 μA × 50 ns), positioning this platform among the most energy-efficient memristive devices reported. Field-emission scanning electron microscopy (FESEM) reveals a uniform TiO₂ thin film morphology on graphene, and Xray photoelectron spectroscopy (XPS) confirms a Ti:O stoichiometry with 15% oxygen vacancy concentration. The high resistance state (~20 GΩ) is attributed to the atomically flat graphene/TiO₂ interface suppressing leakage, nanoconfinement of the switching volume, and a Schottky barrier at the Ag/TiO₂ interface (εᵣ ≈ 86). These metrics position the TiO₂ thin film/graphene memristor as a robust, scalable building block for next-generation in-memory computing and neuromorphic processors.