A DSP based board for neural network simulation

Abstract The possibility of the very large scale integration technique and the coming of innovative architectures have led the semiconductor industry to realize very specialized microprocessors, able to obtain better results than similar general purpose circuits with the same clock rate, as far as execution velocity and easiness of use is concerned. Simulation of parallel computing structures is one of the fields that gets a great advantage from this situation. This paper describes a prototype board based on a Motorola digital signal processor, the DSP56001, and interfaced directly with the VME bus, which can be used to perform neural network simulation in an easily and cheap way. This board has a total memory capacity of 192 Kbytes with possibility to execute 5.6 Mconnections per second.

[1]  Jirí Benes,et al.  On neural networks , 1990, Kybernetika.

[2]  J J Hopfield,et al.  Neural networks and physical systems with emergent collective computational abilities. , 1982, Proceedings of the National Academy of Sciences of the United States of America.

[3]  R.K. Jurgen,et al.  Sarnoff Labs: 'still crazy' but coping , 1988, IEEE Spectrum.

[4]  Teuvo Kohonen,et al.  An introduction to neural computing , 1988, Neural Networks.

[5]  A. A. Mullin,et al.  Principles of neurodynamics , 1962 .

[6]  R. Hecht-Nielsen,et al.  Neurocomputing: picking the human brain , 1988, IEEE Spectrum.

[7]  Leslie G. Valiant,et al.  A theory of the learnable , 1984, STOC '84.