On the time series analysis of resistive switching devices

Abstract

Resistive switching (RS) based memory or memristive devices have emerged as promising candidates for resistive random-access memory (RRAM) and neuromorphic computing applications. However, the integration of RS devices into commercial production faces significant challenges due to substantial variations in RS parameters, which include cycle-to-cycle (C2C) and device-to-device (D2D) fluctuations. In this context, we propose a multivariate time series analysis framework to investigate the variability exhibited by RS devices. We present a detailed description of the statistical methodology and procedures for conducting both univariate and multivariate time series analysis, along with recommended tests and protocols. Specifically, we focus on utilizing Ti₃C₂ MXene oxide-based RS devices as a case study for this analysis. Our findings reveal that employing the multivariate method yields superior prediction results compared to the univariate approach. This conclusion is based on our observation that the Vector Autoregressive Moving Average (VARMA) model, which concurrently considers multiple variables (V_SET and V_RESET), more effectively explains a larger portion of the variability in the data compared to the univariate model. This underscores the importance of considering multiple factors simultaneously, as it provides a more comprehensive understanding of the underlying patterns within the dataset, thereby enhancing the accuracy of predictions. Consequently, we advocate for adopting the multivariate approach due to its ability to capture the complexity and interactions inherent in the dataset, resulting in enhanced model performance. The proposed model demonstrated superior performance in capturing the variability present in V_SET and V_RESET data, thereby producing the most optimal outcomes.