Mimir - Detector Characterizations & Tests

This page summarizes detector tests and characteristics. At present, there are two summaries, reflecting the major detector operating changes instituted during the Nov/Dec 2005 detector tuning run. Future continued tuning of the detector, and commissioning of new operating modes is planned. As these occur, new values for detector characteristics will be added to this page.

  1. For data obtained after 12/1/2005
  2. For data obtained before 12/1/2005

For Data Obtained since 12/1/2005

Conversion Gain: ~8.21 e/ADU (2006/06/27 - dpc)

Read Noise: 17.8 e/read (2006/06/27 - dpc)

Well Depth: about 7,500 ADU (this is about the lowest seen on the detector)

Dark Current: ~10 e/s/pixel (but not linear with time)

Operating Temperature: 33.5K (currently set to 35K)

Linearity Correction: 4th order

Plot of Read Noise and Dark Current versus Detector Temperature. The strong rise in dark current beyond 34K, but the continuously decreasing read noise with temperature led us to set the detector temperature at 33.5K. Click on the image to see the full sized version of the plot.
Plot of pixel well depth, dark current, and read noise versus detector bias. Until 12/2005, we had been operating with about 0.6V of reverse bias, not much above starvation. Two new operating points were identified, corresponding to the blue and red vertical arrows. The blue arrow, with bias of 0.875V has the lowest dark current and hot pixel count and well depth of about 11-12,000 ADU. This is the ideal operating point for most JHK imaging and spectroscopy. A second operating point, at a bias of 1.35V has even deeper pixel wells, nearly 22,000 ADU, but much higher dark current and hot pixel counts. This mode could be employed for LM imaging and LM spectroscopy, where the shorter integration times are able to offset the dark current and hot pixel problems. Click on the image to see the full sized version of the plot.



For Data Obtained prior to 12/1/2005

Detector Conversion Gain

A series of multiple exposures in the dome flat-field screeen were obtained for a range of integration times, with and without the flat screen illuminator lights on. All images were linearity-corrected. Difference and sum images were created at each integration time and the results plotted as variance of the differenced images versus mean counts in the summed images. The data were linearly fit for each quadrant, yielding a mean conversion gain (inverse of the slope term) of 9.7 electrons per analog to digital unit (ADU) (for the 20050605 and 20050615 data sets, and a somewhat higher 10.3 value for the 20050517 data set, obtained before a detector operating voltage change).

Detector Read Noise

From the y-intercept of the conversion gain plot, adjusted for the contributions from the eight readouts (four each for each of the two images differenced), the rms value for one read, averaged over the four quadrants, was found to be 21.1 electrons per pixel (20050605 & 20050615, for the 20050517 data the value is 23.8 electrons per read).

Detector Pixel Well Depth

In the mean vs variance plot, the variance deviates beyond some point, signaling that many of the pixels had reached saturation. The saturation value depends on quadrant and even/odd row, which is embodied in one of the the linearity correction images. Here, the quadrant averaged value was found to be 30,200 electrons (20050605 & 20050615, and 31,500 for 20050517).

Detector Dark Current

At a detector temperature of 30 K, the measured dark current is about 1 ADU/s/pixel, or roughly 10 electrons per second per pixel, though this varies with exposure time (there is a "prompt" dark current contribution that decays with time). Observers are urged to obtain darks with exposure times identical to their science images, rather than scaling from long darks.

Linearity Correction

InSb arrays are inherently non-linear in their response to light. Mimir data are corrected for this effect via application of a linearity correction algorithm, applied on a pixel-by-pixel basis. The data used to calculate the corrections are drawn from the same detector calibration data described above. The form of the correction is a quartic correction, applied after correcting both first and second reads back to the plateau (reset release) time for each exposure.

Additional, extensive testing of pixel timing and detector operation was performed during Engineering Run #3 at the Perkins telescope:

Detector (pixel) Timing Diagrams

Aug 2004 (initial timing)

Dec 2004 - three different pixel timing models: "Working" (HTML, Excel); "Trophy" (HTML, Excel); "Dream" (HTML, Excel)

AladdinIIIwaveforms.s timing files: "Working" (?) (txt)

Detector Temperature Sweep

The detector temperature was varied from about 20 K to 40 K and bias frames taken to measure read noise and dark current versus temperature. We found the expected rise in read noise for the lower temperatures and the rise in dark current for the high detector temperatures. There appears to be a region from about 30-34 K where the read noise is stable and the dark current is low. We also noted that the ghosting increased dramatically as the temperature was lowered, implying that operating at the warm end of the acceptable region might further reduce ghosting or permit faster pixel read times. (Excel file).

Thumb of Detector Heater Power vs Detector Temperature - click to see full plot

Thumb of Read Noise and Dark Current vs Detector Temperature - click to see full plot

Thumb of portion of an image showing increased ghosting of the crack at low detector temperatures (this image was when the detector was at 24 K) - click to see full size plot

Pixel Integration Time Sweep

The ARC A/D boards have, as part of their video signal processing chain, an integrator circuit. This test was designed to find the minimum integration time needed to achieve low-noise operation. Because the coversion gain depends linearly on this pixel integration time, the read noise measured in these short bias frames was divided by the pixel integration time to yield the dependent figure of merit plotted along the Y-axis versus the pixel integration time (X-axis). Clicking on the thumbnail image below brings up the full plot, which shows a "knee" at about 600 ns. The polynomial fit has no physical meaning but conveys some of the sense of the drop in RN from low integration times to a plateau beyond 600 ns. (Excel file)

Thumb of pixel integration time plot - click to see full plot