Realization of artificial neural networks on the multinuclear processor SEAforth

 

A. S. Anisimov¹, A.V. Kalachev²

 

¹FGUP BSKB "Vostok", Barnaul,
²Physical-Engineering faculty of Altai state univesity

Received September 4, 2010

 

Abstract. The subject of investigation is the realization of artificial neural networks on the matrix multinuclear processors. The processor SEAforth40 is examined as hardware platform. Characteristics of the processor are given. Variants of building of neurons and neural networks are analysed with reference to the features of processor's architecture. The neuron with threshold function (perceptron) is studied in details: the way of it's representation and realization, variants of arrangement and interaction of several perceptrons. Method of designing and operation of such network is shown. The Hopfield neural network realization is proposed as an example.

 

Keywords: artificial neural networks, multinuclear processors, perceptron, data stream diagram, modularity.