Memristor-based Coupling Connection Characteristics And Synaptic Circuit Design

With the dramatic increase in the greater amount of information, and higher information processing requirement, more intelligent and more diminutive information processing system are expected urgently. The emergence of memristor makes the circuit theory more complete and has great potential in applications, so it provides a feasible solution for these expectations. Memristor is a new nanoscale electronic element with low power consumption and high integration. Memory effect of memristors is similar to the function of some synapses in neuromorphic systems. It is expected to change the way of modern intelligent information processing completely.In this paper, numerical simulations and circuit experiments of memristor are firstly carried out, and typical features are analyzed. Secondly, we discuss the coupled condition of two flux-controlled memristors in series and parallel connections, and the detailed theoretical analysis is illustrated. Thirdly, based on the improved model of memristor, a switching synaptic circuit consists of two memristors which are serially connected with opposite polarities, and this synapse circuit is introduced to memristive crossbar array to store image. Fourthly, memristive synapse is used to achieve the fusion of image storage and operation. Lastly, the circuitry implementation of a memristive Hopfield neural network is discussed. The main contents of this paper include the following four parts:(1) Based on the coupled flux-controlled memristors, the mathematical model of coupled memristor is deduced in detail. The different polarity connection and coupling strength are considered. By building the Pspice simulator of coupled memristors, the coupling effect of memristor is confirmed by circuit simulations.(2) Based on the improved model of ion migration memristor, a switching synaptic circuit of two memristors is addressed which are connected in series with opposite polarities. This synapse circuit is introduced to memristive crossbar array to store image. We optimize storage solutions, discuss the impact of the noise voltage of image storage and conduct a comparison of numerical analysis and analog simulation. Experimental results show that the proposed storage solution is more reliable and robust than a single memristor cross-array storage.(3) According to the electrical characterization of Ag-chalcogenide memristor, we could use this memristor as a synapse to mimic the STDP and SRDP learning rules in neuromorphic systems. Furthermore, we propose a novel crossbar array structure whose information processing and storing are set in a single unit. The synapse plasticity is used to realize the basic operation in image processing. LTspice experimental results verify the feasibility of the proposed scheme.(4) An improved memristor bridge circuit is employed to realize synaptic operation for neuron which better performs zero, negative and positive synaptic weights without any switches and inverters. In addition, based on memristor bridge synapse, a novel memristive Hopfield neural network circuit is constructed, and we demonstrate the associative memory capability via the cases of single-associative memory and multiassociative memory.