Is multi-pass exposure really worth the effort?
Our lesson on drift and multipass writing is found in our e-beam tutorial, and so by now you know how to do it. But should you? Multipass writing does not make every pattern magically better. Let's think about it.
If you do three-pass patterning, the lower-left corner should be shifted by a large amount - usually 1/3 of the field size. If you shift just one micron, then you will still get a serious bump at the field boundary. Shifting by tiny bits might help with continuity, but this will not smooth out the field stitching error.
On the other hand IF you don't care about a little bump at the field boundary, then you don't need multipass! Instead, you can force a small overlap in the fields using the utility fieldfudge. Then you can run the exposure with a single pass, with lots of current, and/or with a small beam step. Maybe you don't even need fieldfudge. Think about the consequences of multipass printing. You could be wasting a lot of time.
But perhaps you need multipass to make a smoother waveguide! Well, maybe not. You might be making it worse by using a large beam step.
Here are the exposure shots (as shown by cview) on a three-pass waveguide, using a 50 nm beam step: 
Compare this to the same pattern converted for a single pass, using a 10 nm beam step: 
The multipass pattern has bumps around 20 nm, while the single-pass pattern has bumps around 10 nm. These may or may not become physical bumps in the resist, depending on the resist process.
The moral of this story: Multipass writing is sometimes good and sometimes bad.
A comparison of waveguide writing techniques
This is a picture of HSQ, 6% ~ 100 nm. We printed this waveguide in a variety of ways, to compare the field stitching errors.
This waveguide test pattern was printed with the EBPG 5200 in four ways:
1. Three passes, prepared with gpfmerge/gpfsort, and just to be extra paranoid, the block boundaries include interdigitated fingers from Beamer. Exposure current was 4 nA.
2. Single pass, again with the interdigitated fingers. Exposure current was 12 nA.
3. Single pass, 12 nA, now using fieldfudge to create a 50 nm overlap at field boundaries.
4. Single pass, 12 nA, using spiral-filled polygons
Results:
1. Roughness: They all looked the same. Multiple passes did not improve smoothness, because the roughness is limited by the resist (6% HSQ ~ 100 nm thick). Thicker resist is rougher.
2. Stitching: The pattern uses a block size of 650 um, and was printed on silicon. There were no discernable stitching errors in any of the cases. Even when we included a 50 nm overlap on purpose, the added bumps could not be found. Maybe we did not look hard enough.
3. Beamer's facility for creating interdigitated fingers at block boundaries works fine if the pattern is parallel to the X or Y axis, but it fails if the pattern is at an angle. This is a known bug. Fieldfudge is much more reliable.
4. You do not need multipass writing just to patch up minor stitching errors. Multipass writing severely limits the beam current, and slows down the exposure. This adds MORE drift and runout to the pattern. On the other hand, multipass writing can be useful if the bumps at field boundaries are causing waveguide loss, or are adding high spatial frequency components to gratings.
Suggested thought process
| FIRST | ||
| Do the exposure without any multipass or other fancy stuff | ||
| THEN | ||
| If you see bad stitching errors | ||
| THEN | ||
| expose the pattern on plain silicon. If the errors go away, then the problem was charging. You've got to solve that problem first. | ||
| ELSE | ||
| maybe the ebeam really has a drift problem - which you should report | ||
| SECOND | ||
| If the stitching errors are small and you don't care about a few bumps, | ||
| THEN | ||
| use fieldfudge to create small overlaps at the field boundaries | ||
| ELSE | ||
| If multipass is still sounding good, | ||
| use the classic way to do multipass writing: with large shifts between passes. |