High-Performance Conductive Bridging Memristor via van der Waals Interface Engineering for Neuromorphic Computing

Conductive bridging random access memory (CBRAM) is a promising candidate for next-generation memory and neuromorphic computing. However, the stochastic nature of conductive filament rupture and high off-state leakage current often severely limit the on/off current ratio and device reliability. Here, we report a high-performance Ag/SiO₂/Graphene/Pt memristor (Gr-CBMEM) that exploits weak van der Waals (vdW) interaction at the electrode interface to control filament dynamics. Density functional theory (DFT) calculations reveal that the binding energy between the Ag filament and graphene surface is merely 0.48 eV, significantly lower than the strong metallic bonding (6.25 eV) at the Ag/Pt interface and the Ag-Ag cohesive energy (1.94 eV). This weak interfacial bonding facilitates clean and complete rupture of the filament during the RESET process, resulting in an ultralow off-state current of 100 fA and a remarkably high on/off ratio 10⁹. By optimizing SiO₂ thickness to 40 nm to suppress tunneling leakage, the device demonstrates stable bipolar switching and excellent endurance without the stuck-set phenomenon. Furthermore, hardware-based convolutional neural network (CNN) simulations demonstrate that the Gr-CBMEM achieves recognition accuracy of >91% on the CIFAR-10 image dataset. This work suggests that vdW interface engineering is a viable strategy for developing energy-efficient, high-reliability neuromorphic hardware.