Alt-BEAM Archive
Message #03410
To: beam@corp.sgi.com
From: Jean auBois aubois@trail.com
Date: Wed, 19 May 1999 18:39:35 -0600
Subject: [alt-beam] Re: Answer.Tricore info question
At 8:10 PM -0600 5/19/99, Dennison wrote:
:Good question! The answer, as I understand it to be, is really infact quite
:simple yet interesting. SEe, when we think about digital logic components we
:tend to think of everything as being in a "perfect world" and as we know,
:nothing is. Nor is digital logic, we would like to think that when the
:voltage on the input of an inverter goes high the output instantly goes low.
:But it doesn't.
etc.
However, Nv neurons are differentiators -- it is really the edges that
count, not how long the pulse lasts (the derivative of a constant being
zero and all that.) The amount of time spent during the rise/fall time is
quite a few orders of magnitude smaller than the delay time of a given
neuron. In a way, there are two interpretations of what a "process" is --
either it is the time that an LED connected to the output of a neuron is
'on' (i.e. low, because the other end is usually connected to Vcc) or it is
the edge itself (which occurs as the output of a 'typical' Nv neuron has a
rising edge.) Personally, I think that the "lit LED" interpretation is
sort of confusing.
An Nv loop that has an odd number (n) of neurons can have floor(n/2)
processes in it. As a result, a tricore can either be quiet (i.e. 0,0,0)
or saturated (1,0,0 or some variation.) As Sean remarks, the "third state"
can be very useful -- you can get some "dead time" in between two "active"
states (although I personally think that using a microcore and getting dead
times interleaved between the active states is somewhat more useful.)
And if none of this makes no sense, I'm sure that someone else can explain
it far better than I.
jab
------------------------------------------------------------------------
eGroups.com home: http://www.egroups.com/group/alt-beam
http://www.egroups.com
- Simplifying group communications
Home