Take a Memo, or Two

So from the previous post we see that the odd/even effect is not a simple offset, but includes a gain variation between even and odd rows.  In one case, when the VRowOn line was clocked opposite the reset pulse, the gain was pretty close for all pixels.

Looking into this a bit further, it turns out that the MIRI folks have addressed a similar issue at least twice, but the prescription they provide differs a bit.

In a MIRI internal document provided by Mike Ressler, a change to the clocking code is described that minimized the odd/even effect and mitigates the 'droop' problem observed at JPL (basically a long-lived transient signal accompanying the transition from destructive to non-destructive reads).  Raytheon's suggested clocking pattern involves clocking the ØRowOn line exactly opposite the ØRstOn line, effectively disabling the row enable during application of the reset pulse.  Makes sense, because if a row is enabled by the circuitry at the time of reset, the paths available for the charge to redistribute are different than for a row which is not enabled.  This would hopefully cancel any artifact introduced by the decision to reset row-pairs simultaneously.

The change outlined in the MIRI internal document is to clock ØRowOn whether it's a destructive frame or not.  Mike concluded from testing that this did the trick, and attributes the improvement to clocking the array in a stable manner so that the reset pulse is the only difference between destructive and non-destructive reads.

However, in another document ( http://ircamera.as.arizona.edu/MIRI/miridetectorperf.pdf , SPIE Vol. 7021  70210O-12), Mike says the solution is to never clock ØRowOn:

"The MIRI detectors are reset by connecting a pair of rows of pixels to a reset voltage. This occurs 512 times during a frame readout. It had been recommended that the rows be disconnected from the column bus while the reset pulse was applied. Given that the readouts do not like abrupt changes at this temperature, we decided to change the clocking so as not to perform this disconnection. The result was spectacular. The droop disappeared entirely and now shows textbook behavior. We have not observed any change in the data quality or power dissipation, so, at least for the MIRI detectors, this disconnection is not desirable." 

Might just be a 'loose' description, but taken literally it's a different approach.

Because the eBox simply masks the reset bit during non-destructive reads, our scheme was identical to what's laid out in the MIRI-internal document as best practice.  Alternatively, if we set both rails of the ØRowOn line to the 'enable' voltage, we achieve the result in the SPIE paper.

Turns out that the devil is in the details.....

That's Odd

Finally getting around to looking at some of the data taken with the science array last week, trying to get a better idea of what drives the odd/even effect.

Bob used a trick to get a long series of consecutive non-destructive scans.  Using a 1-1-4-4 SUR opstable with a script with 50 sequential 'go' commands, we get 200 consecutive scans saved to disk.  Bob manually masked out the reset pulse by dialing down the level shifter after the first reset scan, resulting in 196 consecutive scans saved to disk without a reset.

This lets us map the debias curve and experiment with relinearization, and to look at the even- and odd-pixel subpopulations independently.

Starting from our 'nominal' operating point, and using a ctdelay of 110 to give ~3s scan times (to match earlier data from the detector lab), we get the following plot at T = 7K (temp not quite stable):

Mean values for all, pixels, all odd rows, and all even rows v. time.

T=7K, VRowOn clocked 0V/-3.9V.

Odds and evens differ by about 2kADU, or about 70K electrons.  The slope of each line is about 1.1ADU/s, or ~38 electrons/sec...this is way too high to be a thermal dark current at this temperature, but there may be a light leak or something warm inside the dewar.  It might also be a statistical artifact from a few bombing bright leakers, since the bimodal distribution of the pixels pushes me to use the mean rather than the median here.  We can tape off the array socket on the next cold cycle as a check on this.

Finding the slope in odds and evens to be the same is encouraging...if the odd/even effect is a pure offset it should subtract out in sky data, yes?  But sadly, no.

The data above were collected with VRowOn clocked opposite the rest pulse.  Bob tried setting both rails of the VRowOn level shifter to -3.9V, netting a DC level.  A similar dataset collected this way is presented here:

Mean values for all, pixels, all odd rows, and all even rows v. time.

T=7K, VRowOn @-3.9VDC.

Here the offset between even and odd is much smaller, but the slopes differ by about 25%.  The shallowest slope (the odd rows) is little changed from the first set.

Glow from the row mux was pretty evident in this data, and Bob reduced it by shifting VpRow/VpCol from -0.8VDC to -1.5VDC.  Retake data:

Mean values for all, pixels, all odd rows, and all even rows v. time.

T=7K, VRowOn =-3.9V. VpRow/VpCol=-1.5V.

Here again the odd rows are little changed, but the change in VpRow/VpCol has shifted the even rows away.  We now have both a gain and an offset discrepancy.

Bob repeated the test at different temperatures.  We consistently see more dark current in the even rows than in the odd rows, and the discrepancy seems to increase at higher temps (to 40% by 8.5K).  Here is an Arrhenius-style plot of log DkI v. inverse temperature (using the array average pixel):

Dark current v. inverse temperature (whole-array average)

Note the departure from the thermal activation line below 8K...leak or glow?

All this looks to me like a different net bias is applied to the even and odd rows, and that the difference may be affected by clock/bias details and possibly by temperature.