A tristable locally active memristor and its application in Hopfield neural network

This paper proposes a kind of nonvolatile locally active memristor. The three fingerprints for distinguishing memristor is verified by the stable and coexisting pinched hysteresis loops, when excited by bipolar periodical signal. The memristor has three stable equilibrium states, which can be mutually switched by injecting suitable voltage pulses. Therefore, it is considered as a three-bit-per-cell memory device. Moreover, the locally active region can be adjusted by memristive parameter. Then, a neural network model composed of three Hopfield neurons is introduced, which is built by replacing one of the connecting synapses with the locally active memristor. It is found that the distribution of system equilibrium points depends on the coupling weight of memristor synapse. The bifurcation diagram reveals the coexistence phenomenon of multiple stable modes. In particular, when there exists a step difference between the natural frequency of the system and the external excitation frequency, complex bursting oscillation will emerge in the neural network. Finally, the equivalent hardware circuit is designed and implemented to confirm the results of numerical analysis, following commercial discrete components.

Automated summary

This paper introduces a nonvolatile locally active memristor with three stable equilibrium states for three-bit-per-cell memory device functionality. A neural network model composed of three Hopfield neurons is presented, showing that the distribution of system equilibrium points depends on the coupling weight of the memristor synapse. The bifurcation diagram reveals the coexistence phenomenon of multiple stable modes, with complex bursting oscillation emerging in the neural network when there is a step difference between the system's natural frequency and the external excitation frequency.