PROCEEDINGS OF THE WORLD CONGRESS ON NEUROINFORMATICS, VIENNA, AUSTRIA, SEPTEMBER 2001, PP. 443-445.
A COMPETITIVE NETWORK OF SPIKING VLSI NEURONS
INDIVERI, G., HORIUCHI, T., NEIBUR, E., DOUGLAS, R. J.
We present a VLSI implementation of a competitive network of integrate and fire neurons, internally configured with local, recurrent excitation and with global feedback inhibition. The VLSI device employs a communication framework for receiving and sending spikes off-chip, implemented using hybrid analog/digital circuits. We characterized the network’s main properties by stimulating the neurons with Poisson distributed spike trains. Using the mean-rate interpretation, we demonstrate how the system is capable of exhibiting winner-take-all behaviors. We also show how the VLSI network is able to enhance correlations in the output signals and filter signals on the basis of input correlation, by exploiting the timing of spikes.
Keywords: