Evaluation of electronic artificial neural network implementations

Abstract

Artificial Neural Networks (ANNs) have gained recognition as a valuable tool to aid in the solution of difficult computing problems. Despite an abundance of work in the application and mathematical foundations of ANNs there exists no consensus on how to implement ANNs in hardware. The goal of this paper is to provide an analytical perspective on electronic ANN implementation by addressing issues and design tradeoffs. The issues discussed include the question of analog vs. digital implementation, virtual vs. literal realizations, as well as speed and cost considerations in implementing a network.<>