In this paper we describe three digital hardware concepts of the modified H
ebbian learning algorithm and compare them with regard to speed, area consu
mption and I/O-performance. The application of the hardware in a sensory-mo
tor application is presented. A small vehicle that drives on a slot-racetra
ck points out the boundary conditions the neural hardware has to fulfill du
e to the speed and the small dimensions of this setup. (C) 1999 Elsevier Sc
ience B.V. All rights reserved.