Aaron
You got the resistance part wrong. As the trace gets smaller the resistance
increases because the cross-section of the trace gets smaller. Electrical
resistance is bulk phenomenon, more cross-section less resistance. The
advantage of copper is that it has a lower bulk resistance. There is an
additional phenomenon that comes into play at the clock rates we are
talking about. This is impedance (the dynamic equivalent of resistance).
This is part of what slows down the rise of the clock or any other
signal.and also causes signal reflections that fouls signals up and looks
like noise in many ways. This phenomenon is very sensitive to the shape of
trace. A flat sheet and square trace of the same cross section have the
same resistance but quire different impedance. You are right that much of
this verges on Voodoo. When I was working with some of the first digital
designs in the 100-600 MHz range over 25 years ago we referred to it as
plumbing since we had to think in terms of Radio Frequency type design
techniques. I take my hat off to the current generation of chip designers.
They are working on the bleeding edge of digital circuit design.
At 09:53 AM 2/28/1999 , Aaron Blosser wrote in flowing prose:
>> [mailto:[EMAIL PROTECTED]]On Behalf Of Paul Derbyshire
>
>> >...don't expect a speed improvement beyond the obvious boost to
>> >500MHz (and faster once .18 fabs come online).
>>
>> Fabs? What are those? I found a fabs in my float.h once, never did see it
>> again. I think it was for finding absolute values. :-)
>>
>> I guess I have to make a guess here. A .18 fab is...
>>
>> * A nasty hypervelocity rifle with small caliber but potent slugs,
>> hopefully in a Quake type game and not being manufactured in a
>> black market weapons factory in the Middle East.
>> * A version of a program. A very very very very beta version.
>> * A very small something-or-other with which you something-or-other
>> your something-or-other in order to achieve something-or-other.
>>
>> Which of these is correct? :-)
>
>Argh! :-)
>
>None are correct...heh.
>
>Short for fabrication plant. Basically, every time they shrink the pesky
>things, they have to invent a whole new technology to do it. :-) And that
>means time and money.
>
>Usually, they can convert one plant over, but they can't convert 'em all
>because they need to keep making current chips as well, so they'll make a
>new plant or add-on to an existing one.
>
>Then there are the different yields. By increasing the disc size they use
>to cut the silicon, you can get more chips on one die, increasing yield.
>What are they using now...18" dies or something? You can fit a lot of
>PIII's on that.
>
>Anyway, I think the current PIII's will be at the same level as the PII's,
>but when they move to 0.18 micron, the trace width, you can get faster
>speeds.
>
>The problem with speeds are that with resistance, when you apply a square
>wave to it, it doesn't change instantly. Were this not a text based forum,
>I could do a picture, but suffice to say that you get a gradual increase
>going from logical 0 to logical 1 (the PIII uses 1.8 volt methinks, so
>you're going from 0V to 1.8V).
>
>The clock signal changes like that every 2ns at 500MHz. The clock signal
>must barely be able to get up to 1.8V in that time, though reducing the
>voltage is a good way to help, as well as decreasing power consumption.
>
>By making the traces smaller, you get less resistance and you can apply a
>faster clock without screwing up timing signals.
>
>The talk about "copper" is because Intel, as do most chip makers, use
>aluminum traces. By going to copper you'll reduce the resistance even more,
>even with the same trace lengths/widths, and you can go faster that way.
>
>It's just plain voodoo I'm telling ya. :-)
>
>Some chipmakers use copper now to get faster speeds...
>
>IBM uses copper in some of it's ASIC (application specific integrated
>circuit) using .16 micron technology. Those bad boys are fast...up to
>800MHz. And I think IBM fabs some PowerPC's using copper, but curiously
>those are running around 400MHz.
>
>Intel says they can keep using aluminum up to about .13 micron width at
>which time aluminum just don't get much smaller. So then they'll go to
>copper, and I can gaurantee they're already doing research with copper,
>thanks to what an anonymous Intel folk shared with me recently.
>
>Aaron
>
>________________________________________________________________
>Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Terry S. Arnold 2975 B Street San Diego, CA 92102 USA
[EMAIL PROTECTED] (619) 235-8181 (voice) (619) 235-0016 (fax)
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
