Since it runs faster than a neuron, with fast context-switching it could be possible to model a human brain using such a device using fewer than 100 billion - perhaps as few as 100 millions? That leaves the 1000-connections-per-neuron issue - a switch 100 million long and N wide to shunt simulated impulses. So that work remains.
1 comment
[ 3.4 ms ] story [ 9.6 ms ] thread