The Gauss neuron is a nonlinear input signal converter required for the implementation of neural networks based on radial basis functions. It was previously shown that such an element can be realized as a two-junction Josephson interferometer shunted by an additional inductance. For certain values of the inductances, the transfer function can be described by a Gaussian distribution, which determines the name of the proposed interferometers. This work derives new analytical expressions for the transfer function (TF) of the Gauss neuron, providing a deeper understanding of its structure as well as of the dependence of the TF shape on the inductive parameters of the device.
Razorenov et al. (Mon,) studied this question.