Subquantum Semimetal Bi and Oxygen Vacancy Filament Memristors for Neuromorphic Computing

Memristors, as neural synapse devices, have been regarded as excellent candidates for non-von Neumann architecture because of their high scalability. However, the randomness of the filaments of state-of-the-art filamentary memristors leads to high variability and poor reliability. Herein, a semimetal bismuth (Bi)-based memristor with oxygen vacancy (V_O)–Bi filaments was proposed. The Bi-based memristor has a subquantum conductance change, high switching consistency, and controllable weight update linearity. Through spherical aberration-corrected scanning transmission electron microscopy (AC-STEM) and density functional theory (DFT) calculations, the formation mechanism of V_O and Bi clusters in the filaments and the overall switching mechanism of the V_O–Bi filaments were elucidated. Specifically, V_O provides a conductive path while Bi ions migrate, leading to the reduction of Bi clusters in the SiO₂ layer. Furthermore, artificial neural network (ANN) simulations based on back-propagation and reservoir computing (RC) systems achieved large digit recognition accuracies of 95.77 and 94.15%, respectively.