Analysis of neural network connections based on memristors and their multiple offset phenomena

Abstract

The article suggests a construction method of a magnetron memristor connecting a three-dimensional Hopfield neural network and a Rulkov neuron in order to build a more complex and achieve more bio-like neural network properties, which has rarely been proposed before. It is discovered that the dynamical behavior of this high-dimensional neural network system is rich, and that the system exhibits many dynamical behaviors depending on the parameter changes. It is possible to change the attractor’s amplitude and its offset boosting behavior by varying various parameters. Changing the system parameters and modifying the system’s initial value may result in initial offset boosting behavior. Combining nonlinear dynamics research methodologies, such as phase diagram, bifurcation diagram, Lyapunov exponential spectrum, and time series diagram, demonstrates the system’s complex dynamical behavior. By analyzing the system complexity and random sequence test, we found that the system has the characteristics of large complexity and strong pseudo-randomness. Eventually, the hardware realizability is proved by the construction of the DSP platform.