Classification of neural network hardware

Abstract

There is a need for studies to classify Neural Network hardware according to some generally accepted criteria to make it easier to understand the basic properties of newly proposed neurochips. This paper aims at putting forward a new proposal for the classification of Neural Network hardware. For this purpose, first the basic components and architecture of a neurochip are described. Then attributes are selected and outlined for the classification, and possible values they may take are discussed. A number of well-known Neural Network chips are then classified using the suggested method.