Hybrid Approach for Modeling Memristive Elements

Abstract

Typically, changes in the conductive properties of resistive random‐access memory elements happen due to the movement of ions in an ultra‐thin dielectric layer under the influence of an electric field. In the case of oxides, they often talk about the movement of oxygen vacancies and the formation/destruction of conducting filaments. Such processes are often described by dynamic systems in which the state parameter corresponds to the position of the boundary between regions with low and high concentrations of oxygen vacancies. In this case, the dependence of the current (or resistance) on the state parameter and the voltage applied to the element can be quite complex. In this regard, the work proposes an approach that uses neural networks to approximate the dependence of the current on the state parameter and voltage. Thus, a hybrid model is obtained in which the state parameter is determined using a dynamic system that takes into account the basic physical characteristics of the elements, and the model is finely tuned to the experimental data at the neural network level. A hybrid model for a memristor based on hafnium oxide (HfO2), as well as on nanocomposite (Co–Fe–B)m(LiNbO3)100−m, has been successfully constructed.