Design of Artificial Neurons of Memristive Neuromorphic Networks Based on Biological Neural Dynamics and Structures

Abstract

Memristive neuromorphic networks have great potentialand advantage in both technology and computationalprotocols for artificial intelligence. Efficient hardware design ofbiological neuron models forms the core of research problems inneuromorphic networks. However, most of the existing researchhas been based on logic or integrated circuit principles, limitedto replicating simple integrate-and-fire behaviors, while morecomplex firing characteristics have relied on the inherent propertiesof the devices themselves, without support from biologicalprinciples. This paper proposes a memristor-based neuron circuitsystem (MNCS) according to the microdynamics of neuronsand complex neural cell structures. It leverages the nonlinearityand non-volatile characteristics of memristors to simulate thebiological functions of various ion channels. It is designed basedon the Hodgkin-Huxley (HH) model circuit, and the parametersare adjusted according to each neuronal firing mechanism. BothPSpice simulations and practical experiments have demonstratedthat MNCS can replicate 24 types of repeating biological neuronalbehaviors. Furthermore, the results from the Joint Inter-spikeInterval(JISI) experiment indicate that as the background noiseincreases, MNCS exhibits pulse emission characteristics similarto those of biological neurons.