A Simple LTSPICE Memristor Neuron with a Modified Transfer Function

Memristors are state-of-the-art nano-scale electronic components characterized by excellent switching and memorizing capabilities, small energy usage and a good compatibility to CMOS integrated circuits. They are potentially applicable in neural nets, memory crossbars, and different electronic schemes. This paper suggests a plain LTSPICE model of a modified transfer function and a neuron, based on memristors. In neurons’ implementation, synapses are realized with single memristors, leading to a significantly reduced circuit complexity. The scaling and summing circuits are built on utilization of memristors and op-amps. The offered modified tangent-sigmoidal transfer function is realized with memristors and MOS transistors. For the analyses, a simple and high-speed memristor model is proposed. The offered memristor-based neuron is analyzed in LTSPICE and MATLAB. A comparison of the obtained results approves the proper action of the suggested neuron. The realized memristor neuron is an essential step towards engineering of complex neural nets and implementation in ultra-high-density integrated neural chips.