A-B clocks: spikes and slopes in dark

[roger smith x294]

This message will not make much sense until you have seen Simin Tulloch's
image of an anti-blooming-clocked dark frame at

	(http://www.ast.cam.ac.uk/~apo/docs/probs.html)

Simon,

I know you would really like answers but the best I can do is suggest
some ideas for consideration. You quote the brightest spots in the
image to be 5500e- in 900s with 700 Hz anti-blooming clocking.  That's
equivalent to a probabilty of producing a sopurious electron of 0.009
per shift.  Compare this to the flat spurious charge which we normally
measure which is somewhere in the 0.0002 to 0.0004 range when it is
behaving itself.

I estimate that the density is only one in about 1000 pixels for
middling to bright ones (just eyeballing your image).  So it does not
appear that the spurious charge seen in a normal readout is simply the
average effect of isolated uniformly diistributed sites.  Either there
are two effects or (my guess) a broad distribution of spurious charge
generation probabilities.

The bright spots you see wouldn't show up in a normal readout.  If you
averaged all the lines, you would only get a contribution from about
one bright spot per column. ie. 0.009 e- compared to 0.0004*1024= 0.4
e- contributed by a uniform low level probability at most pixels of
generating a spurious electron.  Thus far I see no evidence that these
are distinct effects rather than variability in the same effect.  Would
you agree?

The excess dark current near the boundary between upper and lower
halves is interesting.  I can think of two possibilities you could
explore: localized heating and clock smoothing by the distributed RC.
The combination of 700 Hz clocking, a few hundred nF image area
capacitance and a hundred ohms or so of total electrode resistance
can't pump much more than a few milliwatts into the image area - much
less than the radiative load, so the thermal explanation seems very
unlikely.  You could test it by seeing if it goes away when cooled.

We did some spice simulations a few years back to look at clock
crosstalk and shape variation from edge to center.  Your excess dark
current is restricted to fewer lines at the center and is perhaps being
improved by the slower rise times seen in the CCD interior due to the
distributed R and C of the polysilicon electrodes.

Crosstalk increases towards the center.  You can reduce it at the edges
by reducing the source impedance (at high frequencies) of your clock
drivers.  The center is less influenced by this since the resistance of
the CCD electrode is interposed, but it can be reduced by having a
slower edge to start with, acording to our spice simulations.  I have
never understood precisely how crosstalk affects performance.  Hand
waving arguments point to loss of well depth.  Maybe there is some
effect on spurious charge generation too.  Can any readers comment on
this?

Maybe the parallel clock edges become softer as the clock signal
propagates down the parallel clock bus at the edge of the chip.  What
is the resistance of the metalization?   ... a lot lower than
polysilicon but it carries a larger capacitive load.  If there was
this sort of slowing of the edges it might explain why the spurious
charge problem diminishes further from the bond pads.

Simon, is the dark frame stable enough to subtract out well?  ie. Is
the issue excess shot noise in the affected pixels, or a residual of
the lumpy dark image.


Roger Smith, CTIO