Polymorphic Butterfly Attractors in a Bi-Magnetized Tabu Learning Neural Network and Its Analog Circuit Implementation

Understanding complex dynamical behaviors in neural systems is important for further developing neuromorphic computing and brain-inspired information processing technologies. In most cases, conventional Tabu learning neural networks are capable of exhibiting chaos but usually cannot catch the impact of multiple external electromagnetic stimuli. This paper proposes a bi-magnetized Tabu learning neural network with two improved memristor models, aiming at modeling induction currents produced by dual electromagnetic radiations. We build the mathematical model and study its dynamical evolution by numerical tools such as bifurcation diagrams, Lyapunov spectra, Poincaré sections, and the 0-1 test. A variety of polymorphic butterfly attractors, including single-wing, double-wing, and four-wing chaotic and hyperchaotic forms, as well as multistable phenomena in which infinitely many coexisting attractors may appear under different initial states, are revealed in this investigation. Besides, an analogue circuit is designed and experimentally demonstrated in Multisim. The experimental results are consistent with the numerical predictions, which confirm the physical realizability of the model. This work may provide a feasible platform to explore complex neural dynamics and may also have potential applications in secure communications and neuromorphic hardware.