Design and encryption application of multi-scroll chain-loop memristive neural networks with initial-boosting coexisting attractors

Memristor has been universally accepted as the perfect device for simulating neuromorphic synapse, sparking recent and widespread interest in the study of memristive neural networks (MNNs). This paper reports an interesting multi-scroll memristive neural network (MMNN) by configuring newly designed memristor with scalable memductance as the only self-excited synapse to chain-loop Hopfield neural network (HNN). The remarkable feature of the proposed MMNN is that it can yield controllable multi-scroll chaotic attractors, initial-boosting coexisting attractors and large-scale amplitude modulation. As parameter conditions change, the multiscroll chaotic attractors can be split into different numbers of single-scroll chaotic and periodic attractors with initial-boosting along the internal variable of memristive synapse. The comprehensive study of the MMNN on dynamic behaviors, circuit implementation and NIST test is presented. An image encryption algorithm with pixel-level minimal matrix permutation and diffusion processes that shows high security and strong resistance to attacks is constructed by recurring to the MMNN for partially illustrating its availability and superiority in engineering applications.