Abstract
The article presents the results of the development and study of a combined circuitry (compact) model of thin metal oxide films based memristive elements, which makes it possible to simulate both bipolar switching processes and multilevel tuning of the memristor conductivity taking into account the statistical variability of parameters for both device-to-device and cycle-to-cycle switching. The equivalent circuit of the memristive element and the equation system of the proposed model are considered. The software implementation of the model in the MATLAB has been made. The results of modeling static current-voltage characteristics and transient processes during bipolar switching and multilevel turning of the conductivity of memristive elements are obtained. A good agreement between the simulation results and the measured current-voltage characteristics of memristors based on TiOx films (30 nm) and bilayer TiO2/Al2O3 structures (60 nm/5 nm) is demonstrated.
Highlights
Compact Model for Currently, all over the world, there is a significant expansion of research aimed at the development and hardware implementation of artificial neural networks (ANN), using the basic principles of the functioning of brain neurons [1,2]
The algorithms for the functioning of the ANN have a number of advantages that distinguish artificial neural networks favorably from classical computing systems based on the von Neumann architecture: the ability to learn and adapt to the working environment and process, high performance with a significant reduction in energy consumption due to asynchronous parallel data processing [3,4,5]
The purpose of this study is to develop a combined compact model of memristive elements based on thin oxide films, which makes it possible to simulate both bipolar switching processes caused by the filling and release of trap energy levels by electrons and multilevel conductivity tuning determined by the transport of oxygen vacancies in metal oxide films
Summary
Compact Model for Currently, all over the world, there is a significant expansion of research aimed at the development and hardware implementation of artificial neural networks (ANN), using the basic principles of the functioning of brain neurons [1,2]. The algorithms for the functioning of the ANN have a number of advantages that distinguish artificial neural networks favorably from classical computing systems based on the von Neumann architecture: the ability to learn and adapt to the working environment and process, high performance with a significant reduction in energy consumption due to asynchronous parallel data processing [3,4,5]. The elements of nonvolatile resistive memory, memristors, predicted by. Chua in 1971 [6] and for the first time manufactured in 2008 [7], are considered as the most promising candidates for the role of electronic equivalents of synapses in hardwareimplemented neuromorphic systems, due to the possibility multilevel conductivity tuning, which, in combination with small topological dimensions of memristive elements (up to 2 nm [8,9]), provides a high information recording density (up to 0.7 TB/cm2 [10]) with low power consumption (switching energy less than 10 fJ [11]) and possibility integration into cross-bar arrays.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
