Some thoughts by Mr.Moore on... "MANUFACTURING LIMITS"
PC MAGAZINE: Let's talk about manufacturing and all that stuff and when
you do the Moore's Law charts and you have all of this wonderful doubling
every 18 months or whatever. How long do you see that going on?
GORDON MOORE: Until I retire!
PC MAGAZINE: When are you retiring?
GORDON MOORE: Well, I was 68 last week [the week of January 1,
1997], so I don't know which will drive what. [The following week Intel
announced that Andy Grove will be replacing Gordon Moore as chairman,
and Moore will become chairman emeritus.]
Actually, there's still quite a bit we can squeeze out of the
technology, and I'm amazed at how effectively people have been able to
continue spinning the next versions. There's kind of been a generation of
technology every three years, where essentially we double the density
every three years. The minimum dimension multiplies by about .7; .7
squared is .49; you get the same thing in half the area. And now, for Intel,
our leading production technology now is .35-micron and .25 is moving into
production as soon as we can get it cleaned up. That'll be the workhorse
in a couple of years. And that's still straight optical lithography. .18, which
is kind of the next step, looks like it can be done optically without any
dramatic changes in what we're doing.
The step after that to me looks a little tougher but the guys that
have to do it don't seem to be intimidated by it at the moment. The reason
is that we'll probably do it with the so-called 193-nanometer light source,
the excimer laser, and to do .18 with that, the wavelength and the
minimum dimension are about the same optically that's not too bad.
To do .13 in the 193 is a much tougher deal, because you're
operating quite a bit below the wavelength. So you probably have to do all
of the tricks available--you know, phase shift masks and this kind of stuff,
multi-layer resists--the ideas that exist in the industry but are not a lot of
fun. You know they're not that well developed yet, so there may be another
light source that lets us operate at a shorter wavelength.
The trouble is, if you go to a shorter wavelength, you essentially run
out of transparent materials, so all the optics have to be reflecting. At 193
you've still got fluoride and a few silicas that are transparent, but if you get
a wavelength around that .13, .12, I don't know, I don't think anything is
transparent, so you really get into a different regime. Things like masks:
We do masks now on fused silica, we always shine the light through the
mask. We can do that at .18 with the 193-nanometer light, but if we go to
.12 we've got to reflect the light off the mask instead. So changing
wavelength is not easy there, and I don't know which is going to be more
difficult--to use the 193 at two-thirds of the wavelength or to, you know,
come up with an all-reflective system and a new light source. There's a lot
of work to do, and fortunately we've got several years to do it.
As long as we've got all the research in place to get it done now.
Beyond that, I guess when we get to an all reflective system then we can
go quite a bit further. But we make a change from the kind of lithography
we've been doing for the last dozen years when we abandon transparent
optical materials. That's a change, and it's going to be an expensive
change at least. Then, you eventually get to some kind of a physical limit,
and the industry has argued where that is for some time. I think the
consensus is it's someplace between .05 and .1 micron minimum
dimension, and like most of these limits it keeps pushing further away as
we get closer.
We're getting down to the point essentially where the atomic nature
of matter starts to be a real limit when you get down there. That's a fairly
fundamental limit. So that carries us well into the next century, and at that
time we'll be able to put, I don't know, several hundred million or a billion
transistors on a logic chip. That leaves phenomenal room for the designers
to innovate in how they're going to use those, so I don't see this as being
the stopping of innovation in the industry or anything, I just see it focusing
more innovation in other directions, so things will advance for a long time.
PC MAGAZINE: Are physical limits the constraint there or is it really just
the cost of building all these fabs? I mean obviously fabs are very
expensive these days, and getting more so.
GORDON MOORE: Well, the atomic nature of matter is really a physical
limit. The devices start behaving differently. The leakage currents get up,
and when the leakage currents get comparable to your signal currents,
you're really in trouble. And that happens someplace in that range. And I
have a feeling that it may bite you a bit before that, statistically.
You know we depend on being able to dope semiconductors, for
example, by putting impurity atoms in. You make everything smaller, the
number of impurity atoms in the active part of the device is dropping and
dropping and dropping. And if you assume they're randomly distributed --
and that's the model people usually use--you're going to get fairly
significant fluctuations in those. And if you expect a circuit to have a billion
transistors that all behave properly, then that's the 8 or 10 sigma [standard
deviations from the mean] in the distribution. You might get to the point
where just statistically you have a few transistors don't work that's all it
takes of course to wipe out one of these things. So we may actually get
bitten statistically a ways away from this limit people are looking at. I think
research is going on on that, I'm not up to date exactly on what's been
done. But it could be an intriguing problem. We certainly haven't seen any
evidence of it yet. That's been an amazing thing about this technology, I
mean, usually something comes up and bites you when you weren't
expecting it to. The only place that's happened so far is the soft error
problem in DRAMs. And everything else is just working beautifully.
INSTRUCTION SETS
THE PENTIUM BUG
PCMag Mar25/97
a manufacturing wall. |
