Back to 2021.

r4 CCD show similar vertical structure at the top of the flats(see picture attached).

I would be curious to know whether dithered flats can pick this up. The Optics Lab dewar is the only place we can get real flats like these.

Also, note that the detector mask was trimmed back since 2015-12, so the edges of the detector are now illuminated.

r2 data from feb 2018 show the same structure, especially on the right detector.

r1 data from feb 2020 still show that structure on the left ccd.

But different structure.

Switching the wipe phase VOD from -20V to -5V, as suggested by Kamata-san based on HSC experience, largely corrects the problem. But perhaps not entirely. Even on amps which do not show any obvious tearing, there are not-dissimilar vertical features which look roughly the same in both the -5V and the -20V readouts.

The attached are from stacks of 43 ~9k ADU flats. The individual amps have had per-row medians subtracted to get rid of the worst illumination gradients, but are otherwise raw. Note in particular the left-most amp (#0), or the 0 and 3 amps in the plots. What are the vertical features? How significant are they?

The LBNL erase procedure does appear to work reliably and well. We will need to encode it as a builtin FEE command: the overhead from running it from the BEE side is high (>5s). The basic idea is to raise the parallel clock voltages to a high level (10V for PFS, which is the upper rail for the FEE), ramp the substrate voltage (VBB) down from 30V to 0V thus inverting the two, hold briefly, ramp VBB back up but lowering the parallel clock voltages when VBB gets up past 10V or so. A figure from the LBNL manual is attached.

The LBNL e-purge procedure creates a band of tearing across the middle of the detector.

The HSC wipe mode VOD=-5V procedure is not dependable in a still-confusing way, and leaves residual banding.

According to a recent report from Jim Gunn https://sumire.pbworks.com/w/file/fetch/145332735/Detectors210616.pdf : a provisional solution to the tearing has been found, that also mitigates the effect on biases. But quoting the summary in that report:

[...]A recent short report from the
Rubin/LSST group, however, showed that the effect varies enough from sensor
to sensor in their set that we have to figure out a way to find the effect in R1, 2, and
3 to be sure it works for all our sensors. This is not trivial in a fiber spectrograph, but
I believe that this can be done by taking a single flat exposure during which we move the slit enough to create a special dithered flat, but this must be tested.

As discussed in past Princeton Group Internal telecons, this ticket is considered closed. New tickets should be filed against additional work.