FGMOSFET circuit for Neuron Activation Function and its derivative

This paper presents floating gate MOSFET (FGMOSFET) circuit for generating neuron activation function (NAF) and derivative of neuron activation function (DNAF) with programmable characteristics. NAF represents output of a single neuron. DNAF is needed in the training phase of neural networks. This circuit uses only FGMOSFETs to realize the NAF and DNAF. This circuit is designed to operate both in saturation region and sub-threshold region. The performance of this circuit is compared with that of CMOS based NAF DNAF circuit. The comparative study includes linear range, temperature dependence, power dissipation, supply voltage and output resistance. The simulations are done using the tanner tool. The layout of the circuit is also presented. This paper also presents analytical study of proposed circuit and FGMOSFET current mirror in saturation region.