The Reset Anomaly in HAWAII-1,  1024 x 1024 Arrays

  Craig D. Mackay, 16 Nov 1999

  1. Introduction
  
  The raw data that are read out from a HAWAII -1 infrared array always
  show a ramp at the beginning of the frame in that the mean signal level
  on the first few rows is different from the mean signal level over much
  of the image.  Generally the ramp is in the sense that the first rows
  are darker relative to the remainder of the data.  Using reset
  subtraction we find that nearly all of this effect goes except that the
  first loop continues to show a residual effect, even after subtraction.
  This note is intended to give a brief explanation of what might be
  happening.

  2. CMOS operation
  
  Hybrid arrays such as the HAWAII -1 devices consist of a sandwich of an
  infrared detector layer and the silicon readout multiplexor.  The
  silicon multiplexor is made using CMOS technology (exactly the same as
  is used to make microprocessors such as the Pentium family).  These CMOS
  devices work with a technology that relies on very high degrees of
  installation between the electronic components and the silicon substrate
  of the device.  The substrate needs to be at a fixed potential and all
  the components are effectively working in a capacitively coupled mode.
  Whenever a CMOS device is operated, electrical potentials have to be
  established between the electrodes being clocked and the substrate.
  Although all these electrodes are insulated from the substrate, the act
  of clocking them causes significant displacement currents to flow in the
  substrate from the ground potential to which it must be connected.
  These currents are therefore some kind of average response to all the
  different levels being changed in the device.  It is clear from this,
  however, that there will be a significant change in the substrate
  potential depending on whether the device is being clocked or operated
  or is in a dormant state.  However because the substrate is only
  capacitively coupled to the electrodes what you actually see on the
  substrate potential is the change in activity.  Once the level of
  activity is established then the substrate potential will return to the
  level it has when the device is dormant.  The problems are therefore
  ones of start-up rather than a fixed, permanent offset.
  
  3. HAWAII -1 Reset Anomaly

  Within the HAWAII -1 silicon multiplexor, the output signal comes via
  transistors that are bias referenced to the potential of the substrate.
  As a result if there is any change in the potential of the substrate
  then this will be propagated through to the analogue voltage that is
  measured as the output of the device.  The way the HAWAII-1 device is
  structured does not allow any reference potentials to be derived and
  therefore their is no way of measuring the substrate potential and using
  that level to offset the analogue signal.
  
  This means that any change in the level of activity of the HAWAII -1
  device is going to lead to a short-term glitch in its substrate
  potential and therefore a short-term glitch in the output level.  In
  theory this should be a simple exponential capacitor charging function,
  and indeed in principle it would be possible to correct for it by taking
  all the affected pixels, knowing exactly the times and which each was
  read out, fixing the correct exponential curve to the data and
  subtracting it from it.  A similar glitch will occur when the device
  clocking stops.  At this stage we are no longer reading the device and
  so it does not matter.  However any of these effects take very long time
  to die away and therefore their may be residual effect is still present
  when the next read starts.

  This may be the only way it is possible to remove the reset anomaly in
  its entirety.

  4. Operational Constraints on Removing the Reset Anomaly

  Given that this sort of reset anomaly is an inevitable consequence of
  the sort of design that the HAWAII -1 device is then we can see that
  accurate reset subtraction will only be possible if the precise timing
  of the clocking of a device is exactly repeatable, loop after loop.  It
  is inevitable that the first loop of a readout is going to be different
  from all subsequent loops, and this is very likely the main reason why
  we have found it so difficult to use the first loop.  In principle it
  should be possible to remove the reset effect as detailed above,
  including from loop one.

  It is also inevitable that much better reset anomaly removal will happen
  in non-destructive read mode that happens in read-reset-read mode
  because in the latter mode the timing of the read image is different
  from the timing of the reset image (because each row of the reset image
  is read out after the same row of the read image).

  5. Possible operational changes to improve the reset anomaly

  The problem involved in the first loop being unusable could be overcome
  by putting the system into continuous readout mode, exactly the mode
  that is used for normal readout but rather than start it and stop it
  each time arranged at the computer only takes data from the data stream
  when it is wanted an otherwise dumps it.  This would mean that chips
  were being continually read out in the same pattern all the time and so
  no first loop anomaly could possibly be present.  This change would
  require a significant modification of the PixCel readout software, but
  it should certainly not be impossible to do.

  A second modification would be to change the system to have 16 parallel
  channels of output.  This is something that could clearly improve the
  readout speed dramatically while also allow us to reduce the readout
  rate per channel, reducing the smearing that is noticed at a low-level
  when reading out at high speed.  This would also require some changes to
  the PixCel software as well as changes to the SAM microcode but it would
  be possible.