Actually, I think the neural autorouter does have to be trained.  My first
experience with Protel's autorouter was when it was a separate product
(Advanced Route 3.1).  I seem to remember the manual saying the autorouter
had to be trained after the software was installed on the PC.  The training
consisted of autorouting several reference designs.  Maybe now they ship it
pre-trained in the 99SE kernel.

On the hardware side of things, Intel may have abandoned the technology if
they saw it as a threat to their existing products.  Just think how many
Pentiums they would not have sold if they had introduced that neural net
chip!  They will probably introduce the neural net chip again when their
conventional CPU designs have hit the limits of physics and the fabs require
more $ than the U.S. GDP to build.  Which strategy makes more money:

1)  Introduce new technology now.  Your old technology dies almost
immediately.  You make $$$.
2)  Introduce new technology years from now.  In the meanwhile, you can make
$$$ from pumping your old technology.  Then you can make $$$ from the new
tech.  You make $$$ x 2.

It's called "planned obsolesence", and it makes the biz world turn!  It also
makes people wait for the good stuff, and fills up the landfills.

Best regards,
Ivan Baggett
Bagotronix Inc.

----- Original Message -----
From: "Brian Guralnick" <[EMAIL PROTECTED]>
To: "Protel EDA Forum" <[EMAIL PROTECTED]>
Sent: Thursday, September 13, 2001 4:34 PM
Subject: Re: [PEDA] Autorouting Problems

>     The size and scope of the neuro net for the neuro router (if it really
is an
> authentic neuro router) is an infinitetecibly miniature size compared to
what would
> necessary for real intelligence.
>     Not only that, but the router's net would need training & re-training
on design,
> scope, pattern, and so many other things outside of PCB design, I don't
think that
> Protel has the developers to develop such an intelligent router.  I also t
hink that
> with the current PC design architecture, it will not be possible to get a
PC fast
> enough to emulate such a complex neuro net which would rout more than 1
trace a day.
>     Years ago, Intel was designing a neuro net IC which had an analog
core.  It
> learned digitally and operated in an analog environment.  It supposedly
> hundreds of times faster than it was possible to emulate a neuro net with
> software, and it operated thousands of times faster in general.  I wonder
why such
> technology was abandoned, this single IC which did everything was a new
way of
> thinking in IC design.
> ____________
> Brian Guralnick

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* To post a message: mailto:[EMAIL PROTECTED]
* To leave this list visit:
* Contact the list manager:
* Forum Guidelines Rules:
* Browse or Search previous postings:
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Reply via email to