Silver/copper-filament-based resistive switching memory relies on the formation and disruption of a metallic conductive filament (CF) with relatively large surface-to-volume ratio. The nanoscale CF can spontaneously break after formation, with a lifetime ranging from few microseconds to several months, or even years. Controlling and predicting the CF lifetime enables device engineering for a wide range of applications, such as non-volatile memory for data storage, tunable short/long term memory for synaptic neuromorphic computing, and fast selection devices for crosspoint arrays. However, conflictive explanations for the CF retention process are being proposed. Here we show that the CF lifetime can be described by a universal surface-limited self-diffusion mechanism of disruption of the metallic CF. The surface diffusion process provides a new perspective of ion transport mechanism at the nanoscale, explaining the broad range of reported lifetimes, and paving the way for material engineering of resistive switching device for memory and computing applications.

Machine learning algorithms such as linear regression are trained in one step with cross-point resistive memory arrays.

In-memory computing (IMC) with cross-point resistive memory arrays has been shown to accelerate data-centric computations, such as the training and inference of deep neural networks, due to the high parallelism endowed by physical rules in the electrical circuits. By connecting cross-point arrays with negative feedback amplifiers, it is possible to solve linear algebraic problems, such as linear systems and matrix eigenvectors in just one step. Based on the theory of feedback circuits, we study the dynamics of the solution of linear systems within a memory array, showing that the time complexity of the solution is free of any direct dependence on the problem size N, rather it is governed by the minimal eigenvalue of an associated matrix of the coefficient matrix. We show that when the linear system is modeled by a covariance matrix, the time complexity is O(logN) or O(1). In the case of sparse positive-definite linear systems, the time complexity is solely determined by the minimal eigenvalue of the coefficient matrix. These results demonstrate the high speed of the circuit for solving linear systems in a wide range of applications, thus supporting IMC as a strong candidate for future big data and machine learning accelerators.

In-memory computing with cross-point arrays of resistive memory is a promising technique for typical tasks, such as the training and inference of deep learning. Recently, it has been shown that a cross-point array of resistive switching memory (RRAM) with a feedback configuration can be used to solve linear systems, compute eigenvectors, and rank webpages in just one step. Here, we demonstrate the PageRank with a real data set (the Harvard500) and an eight-level RRAM model, describing the conductance update, the standard deviation of each level, and the conductance ratio. By carefully placing each memory conductance value via a program-verify technique, we show that an accuracy of 95% can be achieved for the ranking result. The equivalent throughput of the eigenvector circuit for PageRank is estimated to be 0.183 tera-operations per second (TOPS), while the energy efficiency is 362 TOPS/W. This article supports the feasibility of in-memory PageRank with significant improvements in speed and energy efficiency for practical big-data tasks.