Multiscale Modeling for Application-Oriented Optimization of Resistive Random-Access Memory

Abstract

Memristor-based neuromorphic systems have been proposed as a promising alternative to von Neumann computing architectures, which are currently challenged by the ever-increasing computational power required by modern artificial intelligence (AI) algorithms. The design and optimization of memristive devices for specific AI applications is thus of paramount importance, but still extremely complex, as many different physical mechanisms and their interactions have to be accounted for, which are, in many cases, not fully understood. The high complexity of the physical mechanisms involved and their partial comprehension are currently hampering the development of memristive devices and preventing their optimization. In this work, we tackle the application-oriented optimization of Resistive Random-Access Memory (RRAM) devices using a multiscale modeling platform. The considered platform includes all the involved physical mechanisms (i.e., charge transport and trapping, and ion generation, diffusion, and recombination) and accounts for the 3D electric and temperature field in the device. Thanks to its multiscale nature, the modeling platform allows RRAM devices to be simulated and the microscopic physical mechanisms involved to be investigated, the device performance to be connected to the material’s microscopic properties and geometries, the device electrical characteristics to be predicted, the effect of the forming conditions (i.e., temperature, compliance current, and voltage stress) on the device’s performance and variability to be evaluated, the analog resistance switching to be optimized, and the device’s reliability and failure causes to be investigated. The discussion of the presented simulation results provides useful insights for supporting the application-oriented optimization of RRAM technology according to specific AI applications, for the implementation of either non-volatile memories, deep neural networks, or spiking neural networks.

Memristors for Neuromorphic Circuits and Artificial Intelligence Applications

“…Moreover, these devices are to be used in different applications, such as embedded non-volatile memories, SNNs and DNNs, and the device requirements change for each application. Thus, La Torraca et al present a multi-scale simulation platform which includes all physical mechanisms such as charge transport, charge trapping, ion generation, diffusion, drift and recombination in an environment that considers the 3D distribution of temperature and electric field [27]. This multiscale approach allows simulating the performance of RRAM devices connecting their electrical properties to the underlying microscopic mechanisms, optimizing their analog switching performance as synapses, determining the role of electroforming and studying variability and reliability.…”

Modelling the switching effect in graphene oxide-based memristors

“…where k is rate (in V s −1 ). Introducing this function in equation (12) and integrating, we obtain:…”

“…Memristors have attracted much attention because they are promising candidates to replace the CMOS-based non-volatile memories, which have reached their practical limits due to leakage currents, power consumption and switching speed issues that affect the device performance in the miniaturization process [6][7][8]. Resistive random access memories (RRAM), based on the metal-insulator-metal (MIM) system, are said to be the most viable substitute of the current technology in terms of low cost, high performance and compatibility with the standard microelectronic industry processes [9][10][11][12]. The transport mechanisms in RRAM devices strongly depend not only on the material used in the active layer of the MIM structure, but also on the electrode material.…”

Ultra-Low Power Logic in Memory with Commercial Grade Memristors and Fpga-Based Smart-Imply Architecture

“…Among them, a particular device that is increasingly gaining interest in the development of these new paradigms is the memristor [20] (such as Resistive Random Access Memory (RRAM) [21,22]), a 2-terminal NVM considered an optimum candidate to work both in ultra-low power analog and digital Neural Networks (NN) [23,24], bio-inspired architectures for associative learning [25][26][27], and Logic-in-Memory (LiM) circuits [28][29][30]. Although the requirements for implementing digital LiM architectures are far more attainable with respect to Neural Networks (NN) [31], the experimental validation of this approach is still poorly investigated, especially with commercial-grade RRAMs or memristors. In fact, evaluating the performances of memristors available on the public market can be seen as an indicator of the actual large scale implementation readiness of the investigated approach.…”