Mantra project - Part A: Design, implementation and test of a multi-model systolic neural network accelerator
Thierry Cornu, Paolo Ienne, Dagmar Niebur, Patrick Thiran, Marc A. Viredaz · 1994
. A multi-model neural network computer has been designed and realized. A compute intensive application using the Kohonen self-organizing map, in the field of power system monitoring, has been ported onto the neural machine. The system is described and its performance is evaluated and discussed. 1. INTRODUCTION This paper overviews the first phase of the MANTRA project, now reaching the final stage. A digital neuro-computer has been developed to fulfill the computational requirements of demanding neural network applications. The hardware has been designed to support several popular neural network algorithms. Section 2 describes the basic principles of the systolic processing element used in the computations and of the machine that has been built around that chip. A few other artificial neural systems exist today and display features not too dissimilar from those of the neuro-computer presented here. Each implies specific modifications to the algorithms to suit the implemented hardware...