IUBio

Mind.forth Programming Journal: 28 NOV 1998

Eugene Leitl eugene.leitl at lrz.uni-muenchen.de
Sun Nov 29 08:03:00 EST 1998


Donald Tees writes:
 > Forth!  FORTH!
 > You should be doing this in Cobol!!

Actually, the idea of using FORTH is not such a red herring as it may
appear to be. Mentifex' approach of coding an AI application instead
of evolving it a la ALife is clearly doomed. However, if you can't lay
your hands upon de Garis' wet dream, a wafer full of analog FPGA 
evolware, using a massively parallel FORTH system consisting of 
hundreds and thousands of CPUs is at least a step in the right 
direction. Chuck Moore, the inventor of Forth, has been into CPU 
design since mid-80s (Novix 4000, Mup21). At iTv there is now a 
big effort going on to build a Internet set-top box all done in 
Forth. The VLIW FORTH CPU thing is currently 1 mm^2 large, contains 
a few thousand transistors and executes several 100 Forth MIPS for 
a price of only about 1-2 $. Due to structure shrink the final 
CPU will occupy about 1/6th of its current size. Add a few kBytes 
on-die SRAM, some high-speed networking, and you can have a 
self-contained system on a 1 mm^2. Now, if you know something about 
wafer scale integration, the bottleneck is the die size, since the 
bad die yield goes up exponentially with die size. One square 
millimeter is comfortably below that critical threshold, and a 
300 mm wafer has a whole damn lot of these square millimeters.

You can't go WSI with any other language/according CPU architecture.

Now, I would't want to rewrite NEURON or CONICAL in FORTH to fit into
the tiny wafer grains. But why, writing a spiking code kernel for a
learning autonomous robot, that's very possible. Of course that's not
what Mentifex is doing, so he will only embarras himself even more
than he does with his constant spamming to inappropriate usenet
newsgroups <HINT>.

ciao,
'gene



More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net