X-Message-Number: 25998
From: 
Date: Mon, 11 Apr 2005 02:33:42 EDT
Subject: Uploading (3.i.0)

Uploading (3.i.0)
 
Here I start to look at possible technologies for uploading. I am only 
interested in present day solutions. There seems to be three possibilities :
- Digital, all purpose computers.
- Field Programmable Gate Array circuits (FPGA)
- Analogic "FPGA" or TRAC : Totally Reconfigurable Analogic Circuits.
 
Computers are largely available but have a basic structure far from the one 
found in the brain. The consequence is that they are very inefficients  in 

modeling biochemical function. On the other hand, they can take any problem and
are readily programmable.
 
FPGA contain up to tens of thousand logic cells, each with complex boolean 
operators and memory. FPGA can be program revired on the fly tens of thousand 
time a second. They run at up to 300 MHz and more.
 
TRAC are relatively slow, in the MHz range, but can model readily 

differential equations, a must for Langevin's based models. A TRAC contains a 
number of 
cells, each has an operational amplifier and 8 ways to connect its inputs and 
output. A programmed binary number from 000 to 111 defines what function is 
implemented in a cell at a given time. The TRAC can be revired from cycle to 
cycle, or one million time per second.
 
TRAC is no longer in production by Zetex, but can be emulated by discrete 
components. A big TRAC circuit can then be ordered as an ASIC product when the 
function is debunked.
 
None of these elements display a full set of characteristics making it a 
choice for an "electronics brain". Taken together, they fulfil yet all 

requirements. The uploading support must so be a mix of all these devices. It 
could be 
said that these electronics elements form the "space" where the brain is built.
 
Because neurons display both, rapid ion channel based activity and slower 
processes such Long Term Potentiation or Depression of different kinds (LTPs, 

LTD), it seems a first split must be done between these activities. The fast ion
channel activity seems always forward, with no back reaction. It seems well 
suited to the FPGA technology.
 
The LTPs, LTD have some back reactions from dendrite spine to axon terminal, 
they are very complex and slow, at least 1,000 time slower than the ion 

channel processes. Computers are best at that level, at least if the model is 
not 
too complex. Some computing "chunks" would have to be "predigested" by TRACs, 
particularly when dealling with Langevin's equation.
 
The first conculsion reached here is that a biological brain would have to be 
mapped as two brains copies : One implemented on FPGA for fast processing and 
one on computer for slower ones.
 
Yvan Bozzonetti.


 Content-Type: text/html; charset="US-ASCII"

[ AUTOMATICALLY SKIPPING HTML ENCODING! ] 

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=25998