This problem was observed on bulk micromachined devices at Redwood
Microsystems.  Ultimately, our process was compatible with nitride, so
we ended up using nitride instead of CrAu.  Our pinholes, however, were
lithographic in nature:  the CrAu was on bare Si.

Speculating, then:  the poly has a different grain structure than bare
Si.  The grain structure of your CrAu depends not only on deposition
conditions (temperature, base pressure, Ar sputter clean [if any]), but
also on the substrate structure.  I suspect the underlying structure is
contributing to the nanoscale pinholes, which the HF then attacks.
Since you cannot work on the poly deposition (it is part of the Summit
process), your choice it seems to me is limited to the metallization.
Perhaps Ti would work better than Cr as an adhesion layer?  Cr has a lot
of stress, and I believe (but am not certain) Ti has lower stress, which
may work to your advantage in terms of grain growth in the Au.

Al

---
Albert K. Henning, PhD
Director of MEMS Technology
NanoInk, Inc.
215 E. Hacienda Avenue
Campbell, CA  95008
408-379-9069  ext 101
ahenning@nanoink.net

-----Original Message-----
From: Felix Lu [mailto:felix_lu@yahoo.com]
Sent: Monday, March 01, 2010 7:42 AM
To: General MEMS discussion
Subject: [mems-talk] Survivability of Cr/Au metallization in 49% HF for
30 min?

Hello,

    I was wondering if anyone has any useful tips for Chrome-Gold
metallization on polysilicon that will survive a 20-30 min 49% HF soak.
We have Sandia's SUMMiT V die that we would like to metallize in house
with gold (wirebond pads and reflector plates) on Poly4. The current
design does not allow for a blanket evaporation without electrically
shorting out devices and the TEOS sacrificial oxide they use requires
roughly 30 min to release in 49% HF (as determined by watching the
lateral etch rate progression under a Nomarski scope and from some
failure analysis). We can pattern the wirebond pads and/or the
reflector plates using e-ebeam evaporation of Cr/Au and liftoff (using
Futurexx NR9-3000PY negative resist). The problem, as some of you may
have surmised, is that the metal layer loses its adhesion to the
polySi after the 30 min HF (total time) release, presumably to the HF
attacking the chrome metal through pinholes in the gold and/or
creeping up from the metal-polySi interface.

The process outline is as follows and seems pretty standard:

1. Remove protective photoresist from die by acetone/IPA soak and rinse.
2. "Pre-etch" some sacrificial oxide away  for 10 min  in 49%
HF(leaving 20 min for later). We have determined that 10 min seems to
be a good compromise between making the MEMS devices too fragile and
leaving too much time afterwards for the final HF release. 15 min of
"pre-etching" already leaves the devices too fragile for standard
handling.
3. Photolithography/development using negative resist
4. 1 min oxygen ashing at 100 W to descum. (critical - without this,
the metal pads survive visually, but not during wirebonding)
5. 10 second BOE dip and N2 drying.
6. E-beam evaporation of 200/5000 Angstroms Cr/Au (base pressure <
2e-6 Torr, Cr dep rate = 5 Angstroms/s, Au dep rate = 10 Angstroms/s)
onto an unheated substrate. We have tried higher base pressures and
slower Cr rates but those cases seemed to be worse as they etched
faster in 49% HF -- the HF seems to attack the chrome oxide faster
than pure chrome, which is not unexpected.
7. liftoff in acetone (IPA rinse & N2 blow dry)
8. 20 min final release in 49% HF.
9. Critical point drying
10. wirebonding (this is typically where we find that the wirebond
pads did not adhere to the polySi).


We have tried the same process on a piece of single crystal silicon
and found that the metal adhesion is intact after 30 min in 49% HF,
suggesting that perhaps the culprit lies in the rougher surface of the
polySi allowing an avenue for HF to attack the chrome. However, we
seem to recall from past experience that MEMScAPs polyMUMPS devices
(which are slightly rougher than Sandia's SUMMiT V devices (as
measured by AFM and directly compared)) have Cr/Au wirebond pads that
have been able to survive an HF soak in similar times. From previous
communications with MEMScAP, we believe that our metal thicknesses and
evaporation base pressures are similar to theirs.

The above process does leave about half the wirebond pads intact after
wirebonding, but the location of the intact pads seem to be random.

Any tips/advice/suggestions/comments would be appreciated!

Best regards,

Felix Lu, Ph. D.
felix_lu@yahoo.com
Applied Quantum Technologies
Durham, NC 27707