Design of An Analog Circuit-Based Artificial Neural Network

Abstract

In this paper, a feed-forward artificial neural network with a single hidden layer has been realized using analog circuit blocks designed in 90 nm UMC technology. The network is capable of solving non-linearly separable problems and successfully realizes the XOR gate, which is one of the most basic and common non-linear classification problems. The inputs and the weights of the network are represented by the amplitudes of the transient signals. The weights have been calculated through the back-propagation (BP) algorithm. The analog circuit-based learning implementation yields accurate results with the expected outputs.