[Beowulf] OT: speculation regarding the 22nm fab process
diep at xs4all.nl
Mon Sep 29 15:39:04 PDT 2008
There is thousands of people working in those companies that produce
those machines, like ASML here does do.
ASML arguably has the best machines of the planet and that can get
proven by the fact that they can deliver to more and
more companies world wide.
I happen to know several who work or worked at ASML, and they all
share 1 thing.
Even if i talk to them, what ASML is doing is covered in total
secrecy. If i'd say there is a total information stop on
*how* they get things done, that is still an understatement.
It is far easier to get military informations, and i am not joking
here, than information about the new upcoming ASML machines,
not to mention how they want to solve problems there.
Software is however a very important component of those machines. It
has a lot of embedded processors.
Really a lot.
Where it is relative easy to figure out information on new upcoming
processors and hardware to support that,
the hardware to produce them is such custom tailored hardware, size
of the machines is simply limited by the
maximum size of an object that a jumbo can carry to give one example,
and the users using those machines aren't
greedy to give out information with respect to those machines either.
Maybe they can produce bigger machines when the A380 gets more popular.
So all replies you're gonna get with respect to this is gonna be
written by guys who do not work at those companies themselves.
On Sep 29, 2008, at 10:08 PM, Peter St. John wrote:
> In catching up on email from a week at the beach (got to meet RGB
> for the first time since we were undergrads) and Slashdot had this
> item http://hardware.slashdot.org/article.pl?sid=08/09/19/0126232
> regarding IBM's 22nm process. The explanation (maybe a week old) is
> that IBM usss mathematics to compensate for a lithographic process
> limited naturally to much less accuracy, say 44 or 34. Slashdot
> complains that "computational scaling" is not a good enough
> explanation and they want to know more, which got me thinking.
> I recall the Hubble flaw; IIRC, the flaw in the mirror could
> (partly) be compensated by mathematical analysis, as if the
> information content were there, but distorted, so they just had to,
> um, re-tort. I imagine something similar, in reverse, possible with
> Imagine building a process at say 44 nm, then measuring it's output
> at 22nm precision. I'm considering the 22nm scale measurement as a
> distortion. Then compute the inverse; apply the inverse to your
> design; and feed the distored, or as it were encoded, design to the
> input of the process; it's (measured, not built) effect could be to
> produce a correct feature at 22nm.
> Does that make physics sense? it does rather taste like cheating.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
More information about the Beowulf