CCD Reset Feedthrough

[roger smith x294]

_________________________________________________________

     A  METHOD  TO  REDUCE  RG  FEEDTHROUGH !

          ......Increase RG low.
_________________________________________________________

Most subscribers to ccd-world are familiar with the pesky
feed-through of the CCD Reset pulse, but for those who aren't... 
It is a +200 mV to +300 mV transient on VOS during CCD Reset,
similar in amplitude and opposite in sign to the signal.  It can
cause signal chain saturation and at the very least carries a
settling time penalty which becomes increasingly relevant as we
read faster.  We have seen all sorts of circuit designs to
accommodate it: diodes across gain setting resistors (a linearity
hazard), clamping switches, and increased supply rails plus
faster settling op amps.   Some fast op amps, notably the Burr
Brown OPA627, can take microseconds or even tens of microseconds
to recover from saturation (depending on which rail), so it is
especially important to deal with reset feedthrough.

Having never heard of a way to reduce it, I assumed that it was
an inescapable feature of CCDs.  I had been thinking of it as a
narrow positive going pulse - the way it looks on the
oscilloscope.  This led me to adjust the high level of Reset
Gate.  This has no effect since the output node is connected to
Reset Drain while RG is high.

With copious hindsight it is obvious that the same doesn't apply
to the low level of RG.  When the Reset Gate falls below the
threshold of the reset FET, the reset switch opens and its 
gate-source capacitance (2-3 femtofarads) can no longer be
charged from Reset Drain.  Charge is shared between it and the
CCD output node.  The voltage change on the output node is
proportional to the difference between the Reset FET's turn-off
threshold and the **low** level of RG.  The feedthrough pulse is
really a high duty cycle low going pulse!

If you set RG_low too high then the reset switch doesn't turn off
fully.  I found that the reset FET's threshold is about 5V for
SITe CCDs (with VRD=14V) and that increasing RG-low from 0 to 3V,
reduced feedthrough to 40% or about 90 mV.  I haven't yet
measured linearity but noise, cosmetics and full-well are not
adversely affected.

I am interested in hearing if anyone else has tried increasing
RG_low, or if you foresee any problems. 


Roger Smith,  CTIO       (rsmith at noao.edu)