Is it always the first few that are good, with worse progression over
time? If so, you need to check things that may change/age. Assuming your
hotplates and exposure lamp are constantly on, the most obvious thing is
that your developer may be affected by the SU-8 it is developing. Have
your tried refresh the developer for each wafer? Do you clean your
coverglass? Do you clean the glassware the developer is in?


From: 

Hi All,
   I am currently working on fabricating microfluidic channels on
coverglass using SU8-2025 and using the following protocol:
   1. Spin SU8 2025 on No. 1 coverglass.
       500 rpm -> 2000 rpm. Coverglass was blown with nitrogen before
spin coating.
    2. Pre-bake.
       65 degree C for  1 min and 95 degree C for 3 min.
    3. Exposure.
        1 min @ 4.5 mW/cm^2
    4. Post Bake
        65 degree C for 1 min and 95 degree C for 3 min.
    5. Development
        SU8 developer. No IPA rinsing.
    At first, the feature came out nicely, but recently, the success
rate dropped aggressively. I could see bright edges along the channels
and sometimes it is really difficult to develop out the un-crosslinked
polymer in the un-exposed area. It seems that the resist film in
unexposed area is shattered into small pieces.
    If you could kindly have a look at the following image
(http://base.google.com/base/items?oid=10375407859124322045), you may
find the Y shape channel (width 200 micron) surrounded by some
colorful bright edges. I suspect this bright edge is due to the
peel-off of SU8 from coverglass, but cannot figure out why it happens.
What puzzles me most is that by the same procedure, I can obtain
perfect channel if lucky enough. Unfortunately, I don't know what
"luck" means here. So the result changes from trial to trial without
an obvious consistent pattern. For example, I tried 10 samples today.
The first two were perfect. But the following were worse and worse.
For the last one, it takes barely 10 second to develop such edges
after I started development, but the unexposed resist wasn't desolved
even after 1 min.
    I tried different combination of process configuration before and
it seems that the above protocol is the best for me. But, as the
result is not reproducible, I doubt ...