WFC3 Instrument Handbook

The QE of the flight IR detector, as measured at the Goddard Detector Characterization Lab (DCL), is shown as a solid curve in Figure 5.19. The QE curve demonstrates very high sensitivity of the IR detector for wavelengths longer than 1000 nm. The actual total system throughput of WFC3 depends on many factors including the HST OTA, pick off mirror, filter transmission functions, QE, etc. Based on ground measurements of these quantities, the total system throughput was calculated and compared to the first on-orbit measurements. A 5–20% increase in the total system throughput was discovered, which we attribute to multiple factors. The dashed curve represents the QE under the assumption that the entire flight correction is in the QE. Note, however, that this assumption is unphysical given the realities of anti-reflection coatings and interpixel capacitance.

To avoid the complexity and limited lifetime of a stored-cryogen system, while at the same time providing the low operating temperatures required for dark-current and thermal-background reduction, the WFC3 IR detector is refrigerated with a six-stage TEC to a nominal operating temperature of 145 K. This is an unusually high operating temperature for near-IR detectors, and required tailoring the composition of the HgCdTe material for a long-wavelength cutoff at ~1700 nm. The higher band-gap associated with the short cutoff wavelength effectively limits both the intrinsic detector dark current and its sensitivity to the internal thermal background.

Direct thermal control of the detector (via a sensor integrated in the MUX that controls the 6-stage TEC current) provides typical thermal stability of < 50 mK. Tests made on similar detectors indicate that the residual dark-current variations can be largely calibrated and subtracted out using reference pixels.

WFC3 IR exposures taken with an aluminum blank in place, rather than a filter, provide a measure of the detector dark current. The dark current of the flight array has a skewed distribution, with a mode, median, and mean of 0.045, 0.048, and 0.048 e–/s/pixel respectively. The shifted mode is due to the asymmetry of the dark-current distribution among the pixels, characterized by a long tail of “hot pixels” randomly located across the detector. The mean dark current remained unchanged in the first three years of in-flight operations (WFC3 ISR 2012-11).

The histogram of dark current values, along with the cumulative dark-current distribution, i.e., the fraction of pixels with a dark current lower than a certain level, is shown in Figure 5.20. (see WFC3 ISR 2009-21 for further details on dark current calculations). Improved superdark reference files have been created for all allowed full-frame and subarray modes using data collected during cycles 17, 18, 19, and 20 (WFC3-ISR 2014-06). The signal-to-noise has improved by a factor of 3-11 due to the use of a great deal more data, the use of a non-linearity correction, and the use of persistence masks.

Note that in broad filters, the zodiacal light background is 0.3-1.0 e–/s/pixel, a factor of 10-20 times larger than the dark current. The WFC3 ETC can be used to compute the zodiacal light contribution for a given pointing, in addition to providing thermal and dark current estimates. See Sections 7.9.5 and 9.7.

The IR detector has four independent readout amplifiers, each of which reads a 512â512 pixel quadrant. The four amplifiers generate very similar amounts of read noise. This is illustrated in Figure 5.21 (left), which compares the correlated double sampling (CDS) read noise levels for the four quadrants of the detector. CDS read noise refers to the noise associated with subtracting a single pair of reads. These read noise values were derived from a series of RAPID ramps taken during SMOV testing, providing a measure of the total noise in a difference image. For short ramps, such as these RAPID ramps, the contribution of shot noise due to dark current accumulation is less than 0.01 e–. Figure 5.21 (left) therefore shows that the CDS read noise of the detector is between 20.2–21.4 e–.

By averaging over multiple reads, the effective noise of an IR ramp can be reduced. As shown in Figure 5.21 (right plot), the effective noise in a SPARS200 ramp can be reduced from ~20.0 e– down to ~12.0 e– (2 reads plus zeroth read and 15 reads plus zeroth read, respectively). Similar reductions in noise can be achieved with other sample sequences (WFC3 ISR 2009-23).

For some programs, read noise will not be an issue while for others, such as ultra-low-background observations, the read noise can be a non-negligible component of the noise floor. The relative contribution of read noise to the total noise will depend, of course, on infrared background levels as well (see Section 7.9.5). The contribution to the read noise in WFC3 IR data due to digitization errors associated with the conversion from electrons to data numbers (DN) is negligible.

Before launch, ground-based flats were obtained for the 15 imaging IR filters at a mean S/N of ~500 per pixel using an external optical stimulus (WFC3 ISR 2008-28). On-orbit monitoring using flat fields made with a tungsten lamp shows no evidence of pixel-to-pixel variations in any of the filters (WFC3 ISR 2015-11). The lamp appears to be slowly degrading, with a slight decrease in count rate (~0.3% per year from Oct. 2010 to Dec. 2014).

Because the overall illumination pattern of the ground-based flats did not precisely match the illumination attained on-orbit from the OTA, there are errors in these ground-based flats on large spatial scales. These errors were initially measured by performing stellar photometry on rich stellar fields that were observed using large-scale dither patterns during SMOV and cycle 17. In the SMOV exposures for 4 of the wide (W) filters, the rms difference between the sigma-clipped average magnitude of a star and its magnitude in the first pointing was 1.5%, independent of wavelength (WFC ISR 2009-39). The errors have since been determined more accurately by creating sky flats from thousands of on-orbit exposures, masking out astronomical sources. Flat field reference files corrected using these sky flats were delivered in December 2011. These reference files are expected to support photometry to better than 1% rms accuracy over the full WFC3 IR field of view. A detailed description of their production and accuracy is given in WFC3 ISR 2011-11. Analysis of a grid of observations of a standard star in 3 filters has shown that these flats produce consistent photometry over most of the detector, contributing rms uncertainty ~0.007 mag to photometric measurements (WFC3 ISR 2013-01). See Section 7.9.6 for a discussion of "blob flat fields" that can be used to correct the photometry of blob-impacted stars in crowded stellar fields in WFC3/IR images.

	The latest information about IR flats can be found on the WFC3 website: http://www.stsci.edu/hst/wfc3/analysis/ir_flats

Figure 5.22 shows examples of bias-corrected ground-based flats taken with wide-band filters (left: F110W, right: F160W). Both flats are displayed with an inverse greyscale stretch chosen to highlight features.

The WFC3 IR calibration program shows that the detector response is in fact (slightly) non-linear over the full dynamic range. This behavior is illustrated in Figure 5.23, which presents a plot of average counts as a function of time. The black diamonds are the measured average signal; a linear fit has been made to the signals up to ~25,000 electrons (solid red line). The dashed red line shows this best-fit line extended out to the total exposure time of the ramp. The blue horizontal line marks the level at which the counts deviate by more than 5% from linearity (about 78,000 electrons). For the purposes of non-linearity correction, the 5% nonlinearity level has been defined as “saturation.”

The linearity correction implemented in the WFC3/IR calibration pipeline corrects pixels over the entire dynamic range between zero and saturation. Once the pixel value exceeds the saturation threshold, the pixel is flagged as saturated in the data-quality array within the FITS file and no linearity correction is applied. Pixels driven heavily into saturation can begin to show decreasing readout values, such that their DN values fall back below the defined saturation threshold. To prevent a situation where a pixel is flagged as saturated in one or more readouts, but then not flagged in later readouts, the calibration processing system flags saturated pixels in all subsequent readouts for pixels that are found to be above the saturation threshold in any given readout.

Trials of non-linearity corrections have shown that a third-order fit to the measured linearity versus signal for each pixel provides a better correction than the one currently implemented in the pipeline (WFC3 ISR 2014-17). Photometric results between short and long exposures are more consistent by up to 0.5% when this method is used. We expect to implement it in the pipeline sometime in 2015.

Previous HgCdTe detectors on HST have suffered from a count-rate dependent non-linearity. We are investigating this effect on the WFC3-IR detector. An initial measurement of this effect was made by comparing the photometry of star clusters observed over a wide dynamic range and at overlapping wavelengths in WFC3-IR and NICMOS and/or ACS-WFC. We found a significant detection of a non-linearity in WFC3-IR photometry which is in the same direction but a few times smaller than that of NICMOS. From the stars we measured a non-linearity of WFC3-IR of ~1% per dex over a range of 10 magnitudes (4 dex) which was independent of wavelength. (See WFC3 ISR 2010-07.) This measurement was confirmed using exposures that boosted count rates with Earth limb light (WFC3 ISR 2010-15) and observations of groups of stars observed with 2MASS (WFC3 ISR 2011-15). The impact of this non-linearity is that photometry of faint (i.e., sky dominated) sources calibrated with WFC3-IR zeropoints will appear 0.04 +/-0.01 mag too faint.

The make-up of the WFC3/IR detector’s pixel population includes several flavors of anomalously responsive pixels: hot, cold, unstable, dead, and deviant in the zeroth read. Hot pixels, those showing excess charge, are defined as pixels with more than 100 times the average dark current. Cold pixels are inversely sensitive to incident photons and exhibit a negative slope when measured up the ramp (i.e., pixel value is lower in last frame up the ramp compared to first frame). The anomalous response of a cold pixel could be due to lower intrinsic QE in that pixel or to surface defects. Unstable pixels, as the name implies, are those that behave in an unpredictable fashion; that is, the signal up the ramp does not repeat reliably from ramp to ramp (see Appendix 2, WFC3 ISR 2010-13 for examples). There are dead, or unbonded, pixels which do not respond to light (Figure 5.24). Overlapping the dead pixel population is the population of pixels which have bad zeroth read values, generally due to being short-circuit or unbonded (WFC3 ISR 2003-06).

In addition to randomly-distributed bad pixels, coherent regions of bad pixels exist in the IR detector (Figure 5.24).

Pixels in the lower-right region (dubbed “wagon wheel”) have lower than normal quantum efficiency. There are dead pixels near the detector edge and in the circular “death star” feature near the bottom. Pixels with deviant zeroth read are concentrated in the areas of the death star, the upper corners of the detector, and the quadrant boundaries. (The death star region is marked in the WFC3 FOV in APT to aid in observation planning.) WFC3 ISR 2008-28 describes the characterization of these defects based on ground-testing data; WFC3 ISR 2010-13 describes the various types of populations of bad pixels as observed on-orbit.

The anomalously responsive pixels comprise a small percentage of the science pixel population. The current values of the percentages by type are: 0.4% hot, 1% unstable, 0.4% dead or cold, and 0.5% deviant in the zeroth read. (Some pixels are counted twice, as dead and as deviant in the zeroth read.)

As is common in devices with multiple amplifiers being read out simultaneously, the IR channel exhibits crosstalk: a bright source in one quadrant causing electronic ghosting in another quadrant. In the IR, the crosstalk manifests itself as a very low-level negative mirror image; amplifiers 1 and 2 are coupled (upper left and lower left quadrants; see Figure 5.17) and amplifiers 3 and 4 are coupled (lower right and upper right quadrants). That is, sources in quadrant 1 generate crosstalk in quadrant 2, sources in quadrant 2 generate crosstalk in quadrant 1, and so on.

The level of the IR crosstalk is only ~1e−06 that of the target flux (WFC3 ISR 2010-02); for unsaturated sources, the crosstalk is below the background noise. Once a source saturates, the crosstalk becomes visible at about the level of the background and remains constant as the voltage of the device is pinned.

Image persistence is a common problem in HgCdTe and other types of IR arrays. Persistence manifests itself as ghost images or afterglows from earlier exposures. It was seen in NICMOS, and is also seen in a small but non-negligible fraction of the exposures taken obtained with the Hawaii 1R detector that is the heart of the WFC3 IR channel.

Persistence is caused by traps that exist in the active regions of the reverse-biased diodes that make up the pixels of the detector. Resets, which occur at the end of multi-accum exposures (and during the process of flushing the detector when not observing with the IR channel), maximize the reverse bias of the diodes. Light impinging on the diode creates photo-electrons which cause the reverse bias to decrease. Changing voltages within the diode expose portions of the depletion region to free charge. Dislocations in these newly exposed regions trap charge. More traps are exposed for bright sources than for faint ones. This trapped charge is released in later exposures, resulting in after-images. The greater the saturation of the detector, the greater the number of traps and the greater the afterglow. Smith et al. 2008 (Proc. SPIE, 7021) has provided a very clear description of the physics of persistence and the effects in IR arrays.

The characteristics of persistence vary for different devices and device technologies, reflecting in part how traps are distributed within the diodes. Persistence in the WFC3 channel is primarily a function of the fluence (the total number of photo-electrons released) in an exposure, and secondarily a function of the amount of time the pixel is held at a high fluence level. As discussed by Long et al 2012 (Proc. SPIE, 8442), the amount of persistence in the IR detector on WFC3 is a non-linear function of the fluence. Persistence is observed mainly in situations where fluence levels approach or exceed saturation of the detector.

Several examples of persistence in WFC3 observations and strategies for avoiding persistence are described in Section 7.9.4. A description of a phenomenological model of persistence used to aid in removing the effects of persistence is given in the WFC3 Data Handbook.

Figure 5.25 shows the characteristic shape of persistence versus fluence as observed in a series of darks following an image of Omega Cen which had been deliberately exposed to a level where many stars in the image were saturated. The first dark exposure took place a few minutes after the end of the Omega Cen exposure and the last dark exposure took place about one orbit later. (See WFC3 ISR 2013-07.) The amount of persistence is fairly small until the exposure level reaches about half of full well and saturates near full well exposure. The persistence gradually decays with time from the first dark exposure (highest curve in figure) to the last dark exposure (lowest curve in figure).

Persistence decays roughly as a power law with time, as illustrated in Figure 5.26, which is based on the data displayed in Figure 5.25. The different curves here show the decay for different levels of saturation, as measured in electrons. Persistence at low fluence levels decays more rapidly than persistence at high fluence levels. There are 3 curves for each level corresponding to the 3 times this experiment was repeated. The differences are partially due to the fact that different pixels were illuminated to different levels each time, but may also indicate some intrinsic variability that is not understood. For comparison, the dark current is about 0.015 electrons/sec. If one assumes that a power law describes persistence from 100 to 10,000 seconds after an exposure, then one concludes that about 3% of charge is trapped in an exposure that has a nominal fluence level of 100,000 electrons.

While fluence is the primary factor in determining how much persistence there will be after an observation, the amount of persistence actually depends on the time history of each pixel. Tests show that there is more persistence from a pixel exposed multiple times to the same brightness source (WFC3 ISR 2013-07) and the longer a pixel is held at a fixed flux level (WFC3 ISR 2013-06). This can be understood qualitatively as being due to the fact that traps have finite trapping times. More accurate prediction of persistence has been achieved using an exposure-time dependent power law decay model (WFC3 ISR 2015-15) along with a "correction flat" that takes into account large-scale variations over the detector (WFC3 ISR 2015-16). MAST includes a search form that provides a persistence image for a specified exposure, which has been produced by applying this model to the preceding WFC3/IR exposures. (See the WFC3 Data Handbook.) Since the model is imperfect, the persistence image is intended as a guide to which pixels to flag in an exposure rather than as a reliable indicator of flux corrections.