mtang at stanford.edu
Tue Sep 14 22:41:14 PDT 2010
Hi J --
Excellent question, which is basically "how good is good?" I'm no
expert and will defer to others on this committee who have more
experience than I. However, I have tried to learn and have had long
discussions, mostly with Jim McV and Baylor Triplett about this. Baylor
once showed us a lot of data from Intel, which spent a lot of time
trying to figure this out. Basically, the data showed that you can
think of contamination as a defect density -- then "good" or "bad" is
based on the area and complexity of your device. Counterintuitively,
this means if you are making only a few nanoscale transistors (a few
masks, devices a few microns in size) contamination is much less of a
problem than if you are making conventional CMOS chip.
So, "good" is relative. If you are making detectors, it's a different
answer than if you are making an EE412 chip or some nanotransistors.
The criterion that I believe we've been using is a sort of "do no harm"
kind of approach. Basically, if contaminant appears at or below the
detection level of the Evans TXRF system, then it's not considered a
problem. That said, it is also important to consider the possible
contaminants. At Evans, typically the sources used for TXRF are
molybdenum and tungsten. The W source has better sensitivity to lower
atomic weight (like Fe) whereas the Mo source is better for higher
atomic weights (like Au.) When in doubt, people run both -- although in
many cases, it's a matter of looking at the possible sources of
contaminants and then looking for them. So, if it's a metal deposition
system where a lot of gold is deposited, Au should be one of the
elements covered. For the Evans systems, the typical limit of detection
of K is around 10^10 atoms per cm2. If your W has 20 ppm K and silicon
is about 10^16 atoms/cm2, this is about 10^10 which strikes me as being
at or around the limit of detection. This may very well be how they can
promise <20 ppm of alkali metals. Please double-check these
back-of-the-envelope numbers with Evans or other SpecMat members.
As for contamination, Jim always says "look for the mechanism of
transfer." The best mechanism of transfer is shared chemicals at a wet
station. Unless contamination is really gross, hard surfaces like wafer
chucks or quartzware do not typically transfer contamination very
effectively. I suspect ALD will not have quite the range of concerns
for contamination transfer that other stations are scrutinized for.
We've talked about stretching our contamination policy for years, but
have faced an uphill battle with the lab community on established
tools. It should be a lot easier to establish different rules with new
tools and a new community of users. With ALD, there is a lot of
opportunity to stretch these contamination definitions.
Sorry, I don't think I answered your question but had a good rant.
On 9/14/2010 9:51 PM, J Provine wrote:
> hi specmat,
> what are the numbers associated with clean and semi-clean status in
> the snf. ie what density of trace contaminants are allowed? i tried
> to find this information on the snf website but to no success. i
> remember some of these numbers being presented at a clean-tamination
> meeting last year.
> for instance, two questions have arisen recently concerning the savannah:
> 1) clean wafer goes in...how can we verify that it is clean or
> semi-clean coming out? what number do we need to hit with TXRF or
> other methods.
> 2) a user can purchase W with "low alkali content" from a company.
> they promise <20ppm of alkalis in the W filament (most of the
> contamination is likely K). is that semi-clean?
More information about the specmat