The exceptional resistive switching characteristics of complex oxides make them promising for use in memristors to mimic biological synapses in neuromorphic computing architectures in an energy-efficient manner. Specifically, the electrical conductance of complex oxides can be controllable switched across multiple orders of magnitude by either (a) electroforming a conduction channel (e.g., in tungsten oxide), or (b) inducing Mott-Hubbard transition (e.g., in rare-earth nickelates)– both via controlled migration of defects (such as oxygen vacancies) under applied bias. However, development of accurate neuromorphic computers (i.e., low variability) remains extremely challenging due to a lack of fundamental understanding of the atomic-scale processes that underlie migration and spatiotemporal evolution of oxygen vacancies (and other extended defects) over nano-to-mesoscopic length/timescales under applied electric field. Here, we demonstrate that a synergistic integration of density functional theory (DFT) calculation, ab initio/classical molecular dynamics (AIMD/CMD) simulations, precision synthesis, and multi-modal X-ray imaging experiments provides an effective route to address this knowledge gap. First, we use a combination of multi-modal X-ray imaging (with chemical/structural sensitivity) and CMD simulations to show that the intrinsic randomness of electroforming process in tungsten-oxide memristors can be minimized by careful introduction of sharp protrusions in the electrode gap. We find that sharp protrusions confine the electric field, which yields a reproducible distribution of oxygen vacancies over several switching cycles; and reliably form conduction channel in a specific location in the memristor. Next, we employed a large dataset of data from DFT calculations and short AIMD trajectories to develop deep neural network (DNN) models to accurately capture energetics, atomic forces, atomic charges, and maximally localized Wannier centers in rare-earth nickelates (RNO). CMD simulations based on these newly developed DNNs show that metal-insulator transition (MIT) in RNO occurs via a local structural distortion in the extended nickel structure, which facilitates the electronic instability and causes a sharp concomitant Ni-O bond and Ni-charge disproportion with decreasing temperature; this MIT is also strongly influenced by applied pressure and rare-earth ion. Furthermore, DFT calculations reveal the correlations the oxygen stoichiometry, OV distribution (including different NiOx motifs), OV transport, and electron-lattice coupling in perovskite nickelates. These results will be discussed in the context of controlling defect-induced transitions to design accurate and reliable memristive devices for neuromorphic computing platforms that are vital for artificial intelligence technologies.
Read full abstract