X-Message-Number: 26012
From: 
Date: Thu, 14 Apr 2005 02:02:43 EDT
Subject: Re: CryoNet #26005 To T Donaldson

T. Donaldson:
> I honestly don't know just what books
> would be best for you toread if you want to learn what we know now about how
> brainswork. I would suggest THE NEURON, by IB Levitan and LK Kaczmarek;the
> most recent edition of Shephard, NEUROBIOLOGY; and FROMMOLECULES TO NETWORKS
> edited by JH Byrne and JL Roberts. Youmay also want to consult a book on
> biochemistry 

I have read Biochemistry from L. Stryer (W.H. Freeman ed.) nearly 950 p. 

Molecular Biology of the CELL from Albert et al. 1146 p. From Molecule to 
Networks 
is half that.

>Your TRACs may be able to change 8connections (is that the small number they
>have?) but lots of neuronsconnect to many more than 8 others. To be fair,
>that's not true for ALLneurons, so the TRACs might sometimes fit.

I have not described the TRAC concept yet. There is a preview:
A TRAC contains a number of logical cells, each cell has one operational 

amplifier, this one can be configured along 8 different ways, this is sufficient
to implement all basic mathematical fuctions: identity, inversion, adding, 
multiply, derivate, integrate and so on.

Assume we want to simulate two linked dendrite section to form an "Y", using 
the cable equation. This equation is a sum with 3 terms: two are partial 

second order derivatives and the last is a simple potential. There would be 2 
cells 
configured as derivative function for the first and second terms. One cell 

would add them and a last one would add the potential. The total is 6 cells for
one cylindrical element. Because the "Y" has 3 sections, there will be 6 x 3 = 
18 cells used here. The junction performs an integration, it is done by one 
more cell. This simple configuration use so 19 TRAC logical cells.

What is interesting holds in that: You type the equation in a software window 
and the computer defines how much cells must be used, how to connect them and 
what functions they must perform. The basic TRAC has 20 cells, a test board 

has four TRACs or 80 cells. A real neuron may have 4000 dendrite "Y"s. but they
display all the same configuration, so this structure is simple to enter in 
the computer and a big TRAC may accept it. The trick is that the repetitive 
structure of TRAC simplifies the mask making of the circuit and one generic 

component fits all the needs. It is possible to build TRACs in the million of 
cells 
range. In the example above one neuron would be simulated with 80,000 TRAC 
cells. 12 neurons would fit at the same time on a single IC used .1% of the 

time. In time sharing mode, the integrated circuit would support 12,000 neurons.

Current TRAC speed is very low, in the MHz range, it could be multiplied by 100
or more with current technology. There would be then more than one million 
neurons on a single chip. Multiply that by 10 with comming HEMT transistor 
technology on IC.

And that is not the best use for TRACs, I'll come back on the subject later.

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=26012