Exploring statistical approaches for accessing the reliability of Y2O3-based memristive devices

Abstract

Memristive devices have emerged as promising alternatives to traditional complementary metal-oxide semiconductor (CMOS)-based circuits in the field of neuromorphic systems. These two-terminal electronic devices, known for their non-volatile memory properties, can emulate synaptic behavior within artificial neural networks, offering remarkable advantages, including scalability, energy efficiency, rapid operation, compact size, and ease of fabrication. They hold the potential to serve as fundamental components for artificial neurons, revolutionizing neuromorphic computing systems by closely mimicking biological neurons. However, the integration of resistive random-access memory (RRAM) into commercial production faces challenges due to substantial variations in resistive switching (RS) parameters, which include cycle-to-cycle (C2C) and device-to-device (D2D) fluctuations. These variations are rooted in the stochastic nature of RS, linked to physical mechanisms like diffusion and redox reactions. Nonetheless, limitations exist in the current analytical approaches, emphasizing the need for more standardized tools to assess memristive device reliability consistently. Weibull distribution is widely used to analyze RRAM variability and many further studies are based on it. However, this distribution may not work well for some memristive devices. In such cases, one can use other statistical distributions available in the literature. In the present work, statistical distributions, namely Weibull, Exponential, Log-Normal, Gamma, and Logistic distributions, are employed to scrutinize memristive devices device parameters, shedding light on their performance and reliability. Also, analytical methods namely maximum likelihood estimates for parameter estimation and Kolmogorov-Smirnov test for assessing goodness of fit of the distributions are used. This study aims to provide an approach with a deeper understanding of memristive device parameters and analysis techniques. © 2024 Elsevier B.V.