Billy Brown writes:
> William Kitchen wrote:
> > I also wonder to what extent their greater speed can compensate
> > for their lesser capabilities. The future is ours to create.
Sigh. So are great legends born. These are simulated neurons. They
dwell as 3d patterns in RAM. The FPGA reads from, mangles, and writes
RAM contents back. Because the FPGA is not integrated into the RAM
die, it can only access a tiny fraction of it simultaneusly. In a
sense, we're emulating computronium here. This is an FPGA accelerated
cellular automaton machine, the rules implemented as reconfigurable
logic. It is largely sequential, and hence there is no greater speed
than what biology can deliver. You have essentially no limits to
connectivity, but have to sacrifice speed. Carver Mead's approach is
exactly the other way round: low connectivity, high speed.
If the average length a spike has to traverse is several thousands
cells, you can compute the time necessary for computation of the
result by multiplying time for a single iteration times path length.
> My understanding is that they are already exploiting this. Their hardware
> only instantiates a few thousand virtual neurons at a time, and it does time
> sharing between sets of neurons to reach that 'x million' figure.
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:02:43 MDT