The quantum efficiencies (QEs) of the two WFC3 CCDs are plotted against wavelength in Figure 5.2
. Here the QE is defined as electrons yielded per incident photon. The solid curves illustrate the QEs as measured at the Detector Characterization Laboratory (DCL) at Goddard Space Flight Center, slightly corrected downward by the TV3 ground tests. The plots demonstrate the high sensitivity of the CCDs in the UV down to 200 nm. On the other hand, the peak QE at ~600 nm is less than that of the ACS/WFC detectors which reach ~85% at their peaks. The QE measurements were made with the detectors perpendicular to the incident light. As installed in WFC3, the CCDs are tilted by 21 degrees with respect to the normal. The nominal change in optical thickness is ~6%, but the QE variations, as measured at the DCL on similar devices, turn out to be negligible.
The integrated system throughput of WFC3 depends on many factors including the HST
OTA, pickoff mirror, filter transmission functions, QE, etc. Based on ground measurements of these components, the intergrated system throughput was calculated and compared to the first on-orbit measurements during SMOV4. A 5 to 20% increase in the integrated system throughput was discovered, likely attributable to multiple components. The dashed curves represent the QE under the assumption that the entire flight correction is in the detector QE. For UV observations, UVIS2 achieves a higher sensitivity than UVIS1.
Because the generation of multiple electrons is a random phenomenon, an extra noise term must be added to account for an observed variance larger than that associated with the normal Poisson distribution of incoming photons. The correction is theoretically about 1.7 e–
/photon at 200 nm, decreasing linearly to 1.0 at 340 nm. Measurements of ground-based data, however, have indicated that the effect in the WFC3 chips is much less, 1.07 e–
/photon at 218 nm and 1.03 e–
/photon at 275 nm in broadband data (WFC3 ISR 2008-47
) as well as monochromatic narrowband data (WFC3 ISR 2010-11
). The cause for this is unclear, but may be due to charge sharing (Janesick, J.R., 2007, “Photon Transfer DM-->λ
“, SPIE, Bellingham, Washington, p 45-48).
Given the low level of quantum yield measured in the WFC3 data, neither the QE curves presented in Figure 5.2
nor the WFC3 Exposure Time Calculator (ETC)
include the effects of quantum yield. The noise distortion from multiple electrons is not large compared to other contributions to the signal-to-noise ratio in the ultraviolet (see Section 9.2
Before launch, ground-based flats were obtained for all UVIS filters at a S/N of ~200 per pixel using an external optical stimulus (WFC3 ISR 2008-12
). 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 measured by performing stellar photometry on rich stellar fields that were observed using large-scale dither patterns during SMOV and cycles 17 and 18. In the SMOV exposures, calibrated with the ground-based flats, the rms difference between the average magnitude of a star and its magnitude in the first pointing varied from 1.5% to 4.5%, from the long to the short wavelengths (WFC ISR 2009-19
). The needed corrections to the ground-based flats are now well understood, including the treatment of window ghosts (see WFC3 ISR 2011-16,
and Section 6.5.3
). New reference files were delivered for all UVIS filters except the QUAD filters in December 2011 (WFC3 ISR 2013-10
). They were expected to support photometry to ~1% accuracy over the full WFC3 UVIS field of view for most of the broadband filters (F336W, F390W, F438W, F555W, F606W, F775W, F814W), and to 2-3% accuracy for the remaining filters, for apertures of radius 0.4 arcsec. To verify the accuracy of these flat fields, a new calibration program observed bright HST standard stars on different portions of the detector in 8 filters to quantify any spatial variability in the photometry. (See WFC3 ISR 2015-18
.) For the UV filters, photometric residuals due to the crosshatch pattern in the flats (see Figure 5.3
) were found to be 1.8% rms and 6.7% peak-to-peak. Color differences between the blue HST standards and the average color of Omega Cen, which was used to compute inflight corrections to the ground flats, were found to account for offsets in the UV photometry of up to 5% between the two UVIS chips. Based on this analysis, an improved set of flat fields was generated and delivered to the HST archive to support the new chip-dependent calibration implemented in calwf3
version 3.3 in February 2016. (See Appendix E:
A complete description of the new chip-dependent flat fields is given in WFC3 ISR 2016-04
. Flats were computed using the TV3 ground flats, corrected for a large internal reflection or flare (WFC3 ISR 2011-06
). Low - frequency differences in the inflight response were derived from dithered observations of stars in the Omega Centauri cluster. The three key differences with respect to the prior (2011) set of flat field reference files are:
During the time between anneals, reductions in sensitivity ~1% to several percent develop in some pixels, especially at the bluer wavelengths (WFC3 ISR 2014-18
). Reductions greater than 2-3% are rare in F438W and F814W (0.1% of the pixels), but occur in 3-10% of the pixels in F225W. About 90% of these pixels recover during an anneal, but the more strongly affected pixels can require more than one anneal to recover.
shows examples of bias-corrected ground-based flats for two wide-band filters. Both are displayed with an inverse greyscale stretch chosen to highlight features; the vignetting in the upper-right corner is not instrument-related but an artifact of the optical stimulus. The crosshatch features in the UV flat field (F336W) are normal, due to the detection-layer structure in the CCDs; the level is typically <5% peak-to-peak compared to the rest of the flat.
An analysis of fringing effects in broadband-illuminated
ground flats longward of 600nm (WFC3 ISR 2010-04
) has shown that F953N has the greatest fringe amplitude (~16%), followed by the quad filters FQ889N, FQ906N, FQ942N, and FQ937N (~10%). Other narrowband and quad filters have fringe amplitudes in the range of 0.5-4.6% (F656N, F658N, FQ672N, F673N, FQ674N, FQ727N, and FQ750N). Although fringing is generally weak at wavelengths shorter than 700 nm, the very narrow H alpha filter (F656N) exhibits a fringe amplitude of up to several percent in flat fields acquired during ground testing (WFC3 ISRs 2008-17
Note, however, that the amplitudes of fringing listed here (and in WFC3 ISR 2010-04
) should be used only as an estimate of the effect in science data. Fringing will be different for sources with spectral energy distributions (SEDs) which differ significantly from the calibration lamp used to generate the ground flat fields. For example, continuum sources in broad filters will effectively smooth out fringing effects but that same filter can show strong fringes when illuminated by sources with strong spectral lines or SEDs much narrower than the filter bandpass. Conversely, for sources with SEDs similar to the calibration lamp, the fringes can be corrected by the flat-fielding process.
The dynamic range of a detector is limited either by the full-well capacity of the device or by the analog-to-digital converter (ADC) and gain setting that are used during readout to convert the accumulated charge into data numbers (DN). At the standard UVIS gain of ~1.5 e−
/DN, saturation always occurs on chip unless binning is used, in which case saturation of the binned readout pixel can occur in the ADC. (See Section 5.4.6
for the numerical limits on DN and electrons for unbinned and binned readouts.) If saturation occurs on chip, photometric information can be recovered with great accuracy, as described below. If the charge accumulated in a given (binned) readout pixel exceeds the ADC maximum, any additional charge does not result in any further increase in the DN and may, in cases of extreme saturation, result in values of zero.
On-orbit observations have shown that on UVIS2, the onset of saturation varies from about 67000 to 72000 electrons per pixel over the CCD, while UVIS1 has a somewhat larger range of 63000 to 71000 electrons per pixel (WFC3 ISR 2010-10
). The distribution of full well depth on the detector is shown in Figure 5.6
. Once the charge exceeds the pixel full-well level, it can escape that pixel and spread into adjacent vertical pixels; as the signal continues to accumulate, these adjacent pixels themselves can accumulate charge up to full-well and leak into further adjacent vertical pixels, resulting in the “blooming,” or charge overflow, effect. The MPP (multi-phased pinned) operation of the detectors, used to minimize surface dark current, constrains the blooming along the detector columns, so the blooming is only vertical and not horizontal. Photometric information well beyond saturation can be recovered for relatively isolated sources in unbinned images by defining a special aperture that encompasses all of the pixels that have been bled into. WFC3 ISR 2010-10
presents an algorithm that can be invoked for UVIS1 to regain full linearity to ~1% up to nearly 7 magnitudes past saturation. UVIS2 is linear with simple summation over the saturated pixels. Some small non-linearities (a few percent) have been detected at the lowest signal levels at greater distances from the amplifiers, but the behavior is consistent with CTE loss (Section 5.4.11
) rather than a true non-linearity in the chips.
Electrons that accumulate in the CCD wells are read out and converted to data numbers (DNs), often called Analog-to-Digital Units (ADUs), by the analog-to-digital converter (ADC). The ADC output is a 16-bit number, so that the maximum DN that can be read out is 216
- 1 or 65,535 for each readout pixel (single or binned detector pixel). A straightforward scheme in which one DN corresponded to one electron would make it impossible to measure signals larger than 65,535 electrons. Hence the conversion gain parameter provides a way of adjusting the scale so that multiple counts correspond to a single DN, allowing larger numbers of electrons to be measured. The conversion gain is defined as the number of electrons per DN.
The maximum full well depth of the pixels on the WFC3/UVIS chips is ~72500 e–
(see Figure 5.6
). For the default gain ~1.5 e–
/DN, this corresponds to ~48000 DN, well below the ADC limit of 65,535 DN. If the pixels are binned 2x2 or 3x3, the binned pixels could reach flux levels of 4 or 9 times 48,000 DN, respectively, which would be truncated to 65,535 DN (~98,300 e–
) during readout.
Although it is possible to operate the WFC3 CCD detector at gains of ~ 1, 1.5, 2, and 4 e–
/DN, only a gain of ~1.5 e–
/DN is supported. For unbinned readouts, this gain permits sampling of the entire dynamic range of the detectors, with negligible impact on the readout noise.
The gains for the WFC3 CCDs measured during Cycle 20 are summarized in Table 5.3
. Uncertainties in the measurements are less than +/-0.01 e–
/DN. Gain measurements made from ground-based data and from data taken in SMOV and cycles 17 to 22 have remained constant to within 1-2%. (See WFC3 ISR 2014-05
, WFC3 ISR 2015-05
, WFC3 ISR 2016-09
, WFC3 ISR 2016-13
The read noise level in the science area pixels of bias frames for all of the amplifiers at the default gain setting was measured during SMOV (WFC3 ISR 2009-26
). Table 5.4
shows the results obtained at the default gain setting of 1.5 e–
/DN. The read noise was found to be stable to 1%, 0.4%, 0.7%, and 0.8%, for amps A,B,C, & D, respectively, based on measurements through the end of August 2009. A more recent study (WFC3 ISR 2015-13
) has shown that the read noise has increased by 0.04 to 0.05 e–
in unbinned readouts from May 2009 to May 2015. This is attributable to radiation damage, increasing numbers of hot pixels, increasing CTE losses, and aging of the instrument.
The WFC3 CCDs, like most large-area scientific CCDs, operate with buried channels. Earlier generations of CCDs worked with surface channels, i.e., storing and transferring charges only along the surface of the semiconductor. In these earlier devices, the Si-SiO2
interface between the detector material Si (p-doped conductor) and the surface layer of SiO2
(isolator) created significant charge traps, which limited both the charge transfer efficiency and the dark current. In buried-channel devices, a shallow (~0.5 micron thick) n-type Si layer is implanted just below the surface between the p-doped Si and the SiO2
surface, to store and transfer the collected signal charge away from the traps at the interface.
Dark current in WFC3 detectors is further reduced using MPP technology. The dark current generated at the Si-SiO2
interface ultimately depends on two factors: the density of interface states and the density of free carriers (holes and electrons) that populate the interface. Electrons can thermally “hop” from the valence band to an interface state (sometimes referred to as a “mid-band state”) and from there to the conduction band, producing a dark electron-hole pair. Free carriers also fill interface states and, if these states were completely populated, can suppress hopping and conduction, reducing the surface dark current at levels comparable to the bulk dark. Unfortunately, normal CCD operations deplete the interface of free carriers, maximizing dark current generation.
In MPP technology, the Si-SiO2
interface is populated with holes that suppress the hopping conduction process. MPP mode is applied to the CCD by significantly biasing the array clocks negatively to invert (push electrons away from) the n-buried channel and “pin” the surface potential beneath each phase to substrate potential (hence the name multi-pinned phase). Biasing the array clocks in this manner causes holes from the p+ channel stops to migrate and populate the Si-SiO2
interface, eliminating surface dark-current generation. Note that it is not possible to invert conventional CCDs in this way, as the sensor's full-well capacity would be annihilated since the potential wells within a pixel all assume the same level. To circumvent this difficulty in MPP CCD technology, an additional implant is included below one of the phases, allowing charge to accumulate in collecting sites when biased into inversion.
The MPP CCD also eliminates residual image, a serious problem that has plagued low-signal-level CCD users for many years. Residual image, also known as quantum-efficiency hysteresis, results when the sensor is either overexposed or first powered up. Under these circumstances, electrons are found trapped at the Si-SiO2
interface that slowly release into the pixel's potential well. Residual charge may take hours or even days before its level falls below the read-noise floor. Inverting the CCD causes holes to recombine immediately with the trapped residual electrons, eliminating remnant image effects during integration as well as readout.
Like all CCDs operated in a low-earth-orbit radiation environment, the WFC3 CCDs are subject to radiation damage by energetic particles trapped in the radiation belts. Ionization damage and displacement damage are two types of damage caused by protons in silicon. The MPP mode is very effective in mitigating the damage due to ionization, such as the generation of surface dark current due to the creation of new trapping states in the Si-SiO2
interface. Although protons lose only a minor fraction of their total energy via non-ionizing energy loss, lattice displacement damage can cause significant performance degradation in CCDs by decreasing the charge transfer efficiency (CTE), increasing the average dark current, and introducing pixels with very high dark current (hot pixels). Displacement damage to the silicon lattice occurs mostly due to the interaction between low-energy protons and silicon atoms. The generation of phosphorous-vacancy centers introduces an extra level of energy between the conduction band and the valence band of the silicon. As described above, new energy levels in the silicon bandgap increase the dark current as they allow thermally generated charges to reach the conduction band. As a consequence, the dark current of CCDs operated in a radiative environment is predicted to increase with time.
Indeed, the WFC3 dark current has been slowly increasing over time. The median dark level of pixels below the hot pixel threshold (<54 e–
/hr) has been increasing by ~0.5 e–/hr/pix/year, and was ~8 e–
/hr/pix as of Oct. 2016 as measured using the non-CTE corrected darks. This can be seen in Figure 5.7
. The rate of change has decreased recently, corresponding to an increase in solar activity. Such a correlation has been documented for the other CCD detectors on HST; e.g., see the discussion of dark current evolution in WFPC2
. The measured dark current changed significantly (on the date shown by the green line in Nov. 2012) when we began measuring darks made with post-flash (WFC3 ISR 2014-04
), which is used to reduce CTE losses by increasing the background (see Section 5.4.11
). Less flux appears to trail out of hot pixels and cosmic ray hits due to delayed release of electrons, thus decreasing the dark current. (There is a corresponding change in the measured number of hot pixels at that time; see Section 5.4.9
As of February 2016, with the introduction of calwf3
version 3.3 into the calibration pipeline, we now offer users the chance to use CTE corrected images and calibration files. The production of CTE-corrected dark files on a daily basis is described in WFC3 ISR 2016-08
. The CTE correction helps return electrons that have trailed out of their original pixel back to the pixel they came from which allows a more accurate measurement of the dark current for each individual pixel. This correction has lead to an overall decrease in the median dark current of the detector over time as seen in Figure 5.8
. We, of course, still see the increase in the median dark levels of pixels over time, and we see the same patterns of change in slope as described above (i.e. a recent decrease due to solar activity). However, we do see that the rate of those changes is slightly smaller than when measured with the non-CTE corrected darks. The median dark levels of pixels below the hot pixel threshold (still <54 e-/hr) can be seen in Figure 5.8
to be increasing at a level of >0.5 e-/hr/pix/year now, and was ~7 e-/hr/pix as of Nov. 2016 as measured using the CTE corrected darks.
On orbit, the number of hot pixels increases with time due to radiation damage, but is periodically reduced by annealing, when the UVIS detector is warmed to ~20°
C. The number of sink pixels also grows over time, presumably because they are radiation-induced. The extent of the sink pixel CTE trail to be flagged is based on the science image background: the lower the background is, the longer the sink pixel trail will be. (See WFC3 ISR 2014-19
For WFC3, we have chosen a limit of 54 e-
/hr (0.015 e-
/s/pix) as a threshold above which we consider a pixel to be “hot,” based on the tail of the dark current distribution as well as a visual examination of 900 sec dark frames taken during Cycle 17. Figure 5.11
shows a histogram of CR-free pixels from 900 sec darks taken at three different times after the April 2010 anneal procedure: about 1 day (red line), 11 days (green line) and 29 days (blue line) later. The increase in hot pixels due to on-orbit radiation damage is apparent. The anneal procedure initially repaired 90% of the hot pixels which accumulate over time; in 2015, the repair rate is 20-30%. The hot pixel cutoff is shown with a vertical line at 54 e-
/hr; at this threshold, the growth rate for WFC3 hot pixels is ~1000 pix/day per chip. (Radiation damage produces an overall higher dark current as well as an increase in the number of individual hot pixels. See Section 5.4.8
shows the number of hot pixels as a function of time since the installation of WFC3 on HST as measured using the non-CTE-corrected darks, while Figure 5.13
shows the same measurement made with the CTE-corrected darks which became available in February 2016. (See WFC3 ISR 2016-08
and Section 5.4.8
.) In both figures, the alternating gray and white regions represent anneal cycles. The red vertical lines indicate the dates of the SIC&DH failures, when WFC3 was safed (prior to Oct. 2009, WFC3 safings warmed the chips to 20°
C, the temperature attained in the annealing procedure). The green vertical line indicates when we began to measure darks made with post-flash in Nov. 2012 (see WFC3 ISR 2014-04)
to reduce CTE losses (see Section 5.4.11
). The increased background level introduced by the flash reduces the trailing of electrons from hot pixels along columns during the readout, so more hot pixels are detected above the threshold, accounting for the discontinuity in the number of hot pixels at that time. (There is a corresponding drop in the measured dark rate at that time; see Section 5.4.8
.) The number of permanent hot pixels, i.e., pixels that the anneals are unable to fix, is growing by 0.03-0.1% per month whether using non-CTE-corrected darks or CTE-corrected darks.
The main difference that we see between the measurement of hot pixels using the non-CTE-corrected darks (Figure 5.12
) vs. the CTE-corrected darks (Figure 5.13
) is a tightening of the spread of the measurements of the percentage of hot pixels over time, especially after the start of post-flashing in 2012. This difference occurs as the CTE correction algorithm helps return electrons that have trailed out of their original pixel back to the pixel they came from, which leads to there being fewer hot pixel false positives. Since we pick out hot pixels as any pixel with a value above a specific cut-off, when electrons trail into pixels they don’t belong in we see a slight increase in the overall number of hot pixels. This leads to a larger spread in the measurements towards higher rates at a given date, as seen in the non-CTE-corrected measurements in Figure 5.12
summarizes the number of hot pixels in each chip over several ranges of dates and the number of bad detector pixels, which is stable. The hot pixel range is the number of hot pixels observed between the sample anneal procedures noted.
Uniform response within each pixel and excellent charge transfer efficiency (CTE) are key to achieving accurate photometric performance. CTE is a measure of how effective the CCD is at moving charge from one pixel location to the next when reading out the chip. A perfect CCD would be able to transfer 100% of the charge as it is shunted across the chip and then out through the serial register. In practice, small traps in the silicon lattice compromise this process by retaining electrons, and then releasing them at a later time. (Depending on the trap type, the release time ranges from a few microseconds to several seconds.) For large charge packets (many thousand electrons), losing a few electrons along the way is not a serious problem, but for smaller signals, it can represent a substantial fraction. The UVIS CCDs are large-format devices, similar in size to those in the ACS WFC, and thus require significantly more charge-shifting steps during readout, with more losses, than smaller devices like the STIS and WFPC2 CCDs. CTE inevitably declines over time as on-orbit radiation damage creates charge traps in the detector. WFC3 was installed during solar minimum, when the cosmic flux is greatest and radiation damage most rapid, so the UVIS detector has experienced a steeper decline in CTE in its early years than the ACS WFC, which was deployed at a more favorable time.
The CTE changes tracked with EPER testing provide a guide to likely evolution in time, but cannot be directly interpreted to predict CTE loss as a function of target and background signal level. Observations of stellar clusters are being used to determine this. Preliminary work in this area is shown in Figure 5.16
, presented by Rajan at the 2010 STScI Calibration Workshop
. This figure illustrates how CTE affects stellar photometry for a stellar cluster as a function of the number of transfers along columns. The normalized ratio of stellar fluxes measured in a long exposure to fluxes measured in a short exposure is shown as a function of Y position on the detector for stars within a limited flux range. The ratio increasingly deviates from 1 at greater distances from the readout amplifiers because CTE losses are relatively greater for the short exposure, where the signal is smaller. Subsequent observations of stellar clusters have shown strong evolution of CTE on the UVIS CCDs, as expected from the commencement of on-orbit operations during the minimum of the solar cycle. See Section 6.9
for further monitoring of CTE using exposures of stellar clusters and advice to observers, and http://www.stsci.edu/hst/wfc3/ins_performance/CTE/
for updated information on CTE and links to relevant documents.
In the UVIS channel, point sources and extended targets generate a low-level, negative mirror images in the quadrant adjoining the target quadrant, on the same chip (i.e., CT does not cross between the chips). The crosstalk effect is linear with flux and is stronger for targets in quadrant A or C. In a full-frame, unbinned, four-amp readout with a target in quadrant A or C, the CT level is ~ -2 x 10–4
that of the source; for a target in quadrant B or D, the CT level is ~ -7 x 10–5
that of the source (WFC3 ISR 2012-02
). To within the errors, the CT due to hot pixels and cosmic rays is the same as that due to point or extended sources. Figure 5.17
, from WFC3 ISR 2009-03
, illustrates the crosstalk effect as observed in an image taken during instrument ground tests.