A neural processing unit for self-organizing maps
Résumé
Face to the limitations of the classical computation
model, neuromorphic systems are envisaged as an alternative.
Interesting properties of the neural systems are its highly parallel
computation, and its self-organizing capabilities. In this paper,
we study how to bring distribution in the hardware computation
of these neural models, especially in the case of self-organizing
maps. We propose a bio-inspired hardware substrate in which a
grid of neural processing elements will support a set of neurocognitive
processes. We describe an original distributed artificial
neural network adapted to hardware scalability and provide the
results of its implementation as a set of NPUs onto FPGA devices.