Discrete Memristive Hopfield Neural Network With Multi-Stripe/Wave Hyperchaos
Han Bao, Ruimin Wang, Haigang Tang, Mo Chen, Bocheng Bao
Abstract
The synapse-like properties of memristors make them a popular choice for neuromorphic circuits, particularly in neural networks. However, most existing neural networks capable of generating complex dynamics are continuous-time models with high dimensionality, resulting in significant resource overhead for digital implementation. This study incorporates a memristor with an internal multisegment state function into two-neuron Hopfield neural network (HNN), thereby constructing a novel discrete-time memristive two-neuron HNN (DMT-HNN) capable of emerging complex multi-stripe/wave hyperchaos. DMT-HNN generates multistripe and multiwave hyperchaotic attractors, with the number of stripes/waves continuously expanding as the segment number of the memristor state function increases. By decreasing the memristor scaling factor, the multi-stripe/wave hyperchaotic attractors can be decomposed into varying numbers of coexisting hyperchaotic attractors. An FPGA hardware setup is designed based on DMT-HNN, and the multi-stripe/wave hyperchaotic attractors are successfully captured on an oscilloscope. Besides, hardware pseudorandom level generators are fabricated, and the results are confirmed by NIST randomness tests.