Analysis of Hyperbolic Tangent Passive Resistive Neuron With CMOS-Memristor Circuit

Sigmoid and hyperbolic tangent functions are the computational elements of neural networks, which are applied very widely. This paper aims to propose a simple design for improving the tanh-like passive resistive-type neuron by introducing memristor. Minimal leakage current and small on-chip area, low power consumption and non-volatile memory are the features that make the memristor promising and powerful tool in circuit design. However since memristive devices are not capable to supply energy to a circuit, they should be coupled with conventional CMOS devices, thus forming hybrid circuit configurations. In the frame of this study, we examine the previously proposed circuit for passive neuron. The elements are replaced by memristor to produce tanh activation function. The most efficient circuit configuration in terms performance metrics is to be determined. Our design proves that replacing CMOS device by memristor element improves the circuit performance by reducing the total power, area of the chip and THD level.