4.3 Dark Current, Hot Pixels, and Cosmic Rays

4.3.1 Dark Current

The procedure for creating ACS dark reference files and applying dark subtraction to ACS science data is described in detail in Section 3.4.4. Because it takes around one month to collect enough frames to create the dark reference files, the "best" dark reference file is typically not available in the pipeline for several weeks after the date of observation. Users may verify whether the dark reference file most appropriate to their observations has been delivered for pipeline use by checking the ACS reference file Web pages: https://hst-crds.stsci.edu

Using an old dark reference file will produce a poor dark correction, either leaving too many hot or unstable pixels uncorrected and unflagged, or creating many negative "holes" caused by the correction of hot pixels which were not actually hot in the science data (i.e., if the detectors were annealed in the interim).


The dark current is not constant across the CCDs. Figure 4.7 shows dark current features in the WFC1 (above) and WFC2 (below) chips. These features were observed in pre-flight tests, and have generally remained stable in orbit.

There is a gradient, most noticeable on the WFC1 chip, going from a dark edge in the amplifier A quadrant (upper left) to a bright corner in the amplifier B quadrant (upper right). There are two horizontal bright bands of elevated dark current in the center of the WFC2 chip. Many faint rings are also visible, all concentric with the center of both chips. These features are likely intrinsic to the chips themselves, artifacts embedded in (or on) the silicon during various stages of the CCD manufacturing process. 

Figure 4.7: High S/N Combination of WFC Dark Frames Illustrating Dark Current Structure

The WFC1 and HRC histograms in Figure 4.8 and Figure 4.9 show the growth of hot pixels (for more information, please refer to Section 4.3.2). A less obvious result is that the peak of the normal pixel (Poisson) distribution (i.e., the mean dark current, excluding the hot pixels) has also increased.

The increase in mean dark current for WFC has gone from 6.8 e¯/pixel/hour at launch in March 2002 to 11.1 e¯/pixel/hour (an average of 11.4 for WFC1 and 10.8 for WFC2) in April 2004. Following SM4, a dark current of 20 to 25 e¯/pixel/hour was measured and, as of May 2017, the average dark current measured is 67 e¯/pixel/hour.

For HRC the change in dark current has been from 9.3 e¯/pixel/hour at launch to 13.4 e¯/pixel/hour in April 2004. The dark calibration tracked the mean dark current very closely at 2 week intervals (see Section 3.4.4).

Figure 4.8: WFC Dark Current Histogram for WFC1

Growth of hot pixels over time, as measured for WFC1 in CTE-corrected superdark files. Statistics for WFC2 are nearly identical. Increases in the dark current and number of hot pixels over time were mitigated somewhat by the WFC temperature change from –77°C to –81°C on July 4, 2006. There was a substantial increase in hot and warm pixels starting in January 2015, with the introduction of post-flash to the dark calibration images, which allowed more accurate measurements of the dark frame despite CTE losses.
Figure 4.9: HRC Dark Current Histogram

Data from the first (March 2002) and the last (January 2007) superdarks are shown. Dark current and hot pixels increased over time, as expected. Unlike Figure 4.8, HRC was not affected by the temperature change in WFC.


The SBC, or ACS MAMA detector, intrinsically has no read noise and very low detector noise levels, which are normally negligible compared to statistical fluctuations. Dark frames for SBC are typically taken once per year to monitor their levels. The mean dark rate when the instrument is cooler than 25 °C is 8.11 × 10–6  counts/pixel/second (ACS ISR 2017-04).

Summed SBC dark images are delivered to CRDS from time to time. These are only valid for a period within 2 hours of the SBC turn-on, because the temperature increases with time and causes a subsequent rise in the dark current (see Figure 4.10). Measurements of the dark rate after the instrument has been on for 5 hours show a slight increase in the dark rate over time, at a rate of 2.0 × 10–6 counts/pixel/ second/year (ACS ISR 2017-04). This is driven by an elevated dark rate in the central region of the detector at warmer temperatures. However, dark correction of SBC images remains unnecessary and is not used in the calibration pipeline because the correction is negligible.

Figure 4.10: SBC Dark Rate and Operating Temperature

Top panel shows the dark rates vs. temperature measured from all dark rate monitoring programs. The bottom panel shows how the temperature changed from the time the instrument was turned on until the end of the observations. Dashed lines correspond to observations that were excluded from further analysis. (See ACS ISR 2017-04 for details.)

4.3.2 Hot Pixels

When pixels are damaged by radiation or other causes, they can suffer enhanced dark current. Such pixels are called hot pixels. Although the increase in the mean dark current with proton irradiation is important, of greater consequence is the large increase in dark current non-uniformity.

Field-enhanced pixels have been classified into two categories: warm and hot pixels. The definition of "warm" and "hot" pixel is somewhat arbitrary, and there have been several changes to the definition of warm and hot pixels throughout the lifetime of ACS. In January 2015 the ranges of hot and warm pixels were adjusted as follows. A pixel above 0.14 e¯/pixel/second is considered a "hot" pixel. A pixel below the hot pixel range but above 0.06 e¯/pixel/second is considered a "warm" pixel. This change in definition was made after the addition of post-flash to the dark calibration images, which partially alleviates effects of CTE (ACS ISR 2015-03). The new values were chosen by comparing the hot and warm pixel percentages found in the years following SM4. In this way the hot and warm pixel flag will continue to keep track of the pixels that have the worst levels of artificial signal. Previous post-SM4 limits were set to 0.08 e¯/pixel/second for hot pixels, and 0.04 e¯/pixel/second for warm pixels.

Warm and hot pixels accumulate as a function of time on orbit. Defects responsible for elevated dark rate are created continuously as a result of the ongoing displacement damage on orbit. The number of new pixels with a dark current higher than the mean dark rate increases every day by few to several hundreds depending on the threshold. The reduction of the operating temperature of the WFC CCDs in 2006 dramatically reduced the dark current of the hot pixels, but over time the values have continued to rise. The smoothing effects of CTE have also increased the difficulty of accurately measuring dark current as CTE trail profiles change with time.

Table 4.4: Creation Rate of New Hot Pixels (Pixel/Day)






815 ± 56


125 ± 12


616 ± 22

427 ± 34

96 ± 2


480 ± 13

292 ± 8

66 ± 1


390 ± 9

188 ± 5

48 ± 1


328 ± 8

143 ± 12

35 ± 1


16 ± 1

10 ± 1

1 ± 0.5

Table 4.5: Annual Permanent Hot Pixel Growth (%)





> 0.02




> 0.04




> 0.06




> 0.08




> 0.10




> 1.00




Like other CCDs on HST, the ACS devices undergo a monthly annealing process. (The CCDs and the thermal electric coolers are turned off and the heaters are turned on to warm the CCDs to ~19°C.) Although the annealing mechanism at such low temperatures is not yet understood, after this "thermal cycle" the population of hot pixels is reduced (see Figure 4.11). The anneal rate depends on the dark current rate; very hot pixels are annealed more easily than warmer pixels. For pixels classified as "hot" (those with dark rate > 0.14 e¯/pix/sec.) the anneal heals ~3% for WFC and ~14% for HRC.

Annealing has no effect on the normal pixels that are responsible for the increase in the mean dark current rate. Such behavior was also seen with STIS and WFC3 CCDs during ground radiation testing. Since the anneal cycles do not repair 100% of the hot pixels, there is a growing population of permanent hot pixels (see Figure 4.11).

Figure 4.11: Hot Pixel Growth Rate for HRC and WFC

These figures show hot pixel growth rates (DQ flag 16) in the WFC and HRC. Top plot: The sawtooth patterns correspond to anneal cycles. For HRC, the growth rate increased slightly when the anneal duration was reduced from 12 hours to 6 hours–a slight drop coincided with the switch to Side 2 electronics. Bottom plot: Percent of WFC detector flagged unstable as a function of superdark date. The total percentage of hot pixels, both stable and unstable, is plotted in light pink. The percentages of hot and cold unstable pixels, and their sum, are also plotted. The sharp decline in percentage of hot pixels in 2006 is due to the changing of operating temperature of the instrument. The decline in 2015 is when the darks began to be post-flashed. Only hot unstable pixels will be flagged before 2015, whereas all unstable pixels will be flagged starting in 2015.

The dark current in field-enhanced hot pixels can be dependent on the signal level, so the noise is much higher than the normal shot noise. As a consequence, since the locations of warm and hot pixels are known from dark frames, they are flagged in the data quality array. The hot pixels can be discarded during image combination if multiple exposures have been dithered. The standard "CR-SPLIT" approach allows rejection of cosmic rays, but hot pixels cannot be eliminated in post-observation processing without dithering between exposures.

Given the transient nature of hot pixels, users are reminded that some hot pixels may not be properly flagged in the data quality array (because they spontaneously "healed" or because their status changed in the period spanning the reference file and science frame acquisition), and therefore could create false positive detections in some science programs.

4.3.3 Pixel Stability

In principle, warm and hot pixels should be eliminated by the superdark subtraction. However, some pixels show a dark current that is not stable with time but switches between well-defined levels (see Figure 4.12, see also ACS ISR 2017-05). These fluctuations may have timescales of a few minutes and have the characteristics of random telegraph signal (RTS) noise.

Figure 4.12: ACS Pixel History

Examples of the four classifications of pixel stability in plots of dark current over the lifetime of ACS. Top panel: A pixel whose high average dark current (in electrons) which varies strongly between exposures, now identified as a hot, unstable pixel. Top middle panel: An unstable pixel with normal average dark current, now identified as a cold, unstable pixel. Bottom middle panel: A hot pixel which varies very little between exposures, now identified as a hot, stable pixel. Bottom panel: A cold, stable pixel. Vertical lines are anneal boundaries. The gap in the center is when ACS was inoperable due to electronics failure. (See ACS ISR 2017-05 for details.)

An analysis of the stability of every pixel in the WFC detector was performed using every dark image from the lifetime of ACS (ACS ISR 2017-05). For each anneal cycle, the stability of each pixel was determined by f = (Var(SCI) – Mean(ERR2)) / Mean(SCI) + 1, where SCI and ERR are the values of that pixel in the SCI and ERR extensions, respectively, in all of the dark images taken during the anneal cycle. Pixels with f values above the threshold f threshold= 75*Mean(SCI)–0.75+ 2 are considered unstable. Figure 4.13 shows the f values of each pixel as a function of mean pixel value in electrons as measured from the dark frames from the Nov 18, 2015 anneal. The dotted green line marks the threshold above which pixels are considered unstable.

This analysis has shown that the vast majority of hot pixels are stable over an anneal cycle. In one example anneal from 2015, 213,999 pixels, or 1.27% of the detector, were flagged as hot, but only 257 of them, or 0.002% of the detector, are considered unstable. The total number of pixels considered unstable in this anneal is 20,113, 0.12% of the detector, so there are 19,491 normal unstable pixels.

The superdarks for each anneal will be updated to include flags of value 32 in their DQ extensions at the locations of unstable pixels. calacs will propagate this information into the science images it processes. Therefore, instead of discarding all hot pixels during image combination, as was done previously, stable hot pixels are retained and their noise is propagated into the ERR extensions of each image. An appropriate dither strategy between exposures is nevertheless recommended.

Figure 4.13: Density plot of mean pixel intensity versus stability for the Nov 18, 2015 Anneal

Vertical green line is the hot pixel threshold. Horizontal blue line is a stability of 1. Dotted green line is the stability threshold, everything above will be marked unstable. Note: the axes and colormap are in log space. Mean pixel value is in total electrons in a 1000.5 second dark and includes flash. (See ACS ISR 2017-05 for details.)

4.3.4 Sink Pixels

Sink pixels are certain pixels in a CCD detector that are anomalously low compared to the background. This is likely due to the presence of extra charge traps in the pixel. A study of the sink pixels present in the WFC detector is presented in ACS ISR 2017-01. Charge traps prevent electrons in the sink pixel itself from being read out, and they can also trap electrons from pixels that are transferred through them during the readout process, giving rise to a low-valued trail following the sink pixel. The apparent length of the trail depends on the background level of the image, as shown in Figure 4.14 (ACS ISR 2017-01). In addition, about 30% of the time in WFC, a charge excess is found in the pixel immediately closer to the amplifier.

Sink pixels and the pixels they affect, both above and below the sink pixel, are identified in the average post-flashed 0.5-second WFC dark image from each monthly anneal cycle, which have been available since January 2015. One sink pixel reference image (snk.fits) is produced for each anneal and is used by calacs to flag sink pixels and the pixels they affect with the value 1024 in the DQ extension of WFC science images. Since August 3, 2017, when version 9.2.0 of calacs was implemented, this flagging is performed for all WFC images processed by calacs observed after January 1, 2015. About 0.3 to 0.5% of pixels in the WFC detector are considered sink pixels in a given anneal. Depending on the background level of an image, 1–2% of pixels will be flagged with the value 1024 in its DQ extension (ACS ISR 2017-01).

Figure 4.14: Flash-subtracted Short Dark

A 100 × 100 pixel2 region in the flash-subtracted short dark for the 2016-09-25 anneal cycle centered on a deep SP with a trail extending towards the top of the image. Trails are visible following some of the other SPs, but many appear to be individual pixels.

4.3.5 Cosmic Rays

Like all HST cameras before it, the ACS HRC and WFC images are heavily peppered by cosmic rays in even the shortest of exposures. For full orbit integrations, approximately 5% of the pixels receive significant charge from cosmic rays via direct deposition or from diffusion from nearby pixels. Great care must be taken in planning and analyzing HST ACS observations to minimize the impact of cosmic rays on science images.

Many science observations require a careful consideration of individual cosmic ray events. To either remove cosmic rays or distinguish them from astrophysical sources, users might consider the distributions of observed cosmic ray fluxes, sizes, anisotropies, and the number of attached pixels per event.

Fractional Coverage

For most users of the HRC and WFC, the most important characteristic of cosmic rays is simply the fraction of pixels they impact. This number provides the basis for assessing the risk that the target(s) in any set of exposures will be compromised. For ACS the observed rate of cosmic ray impacts on an individual frame varies by a factor of two depending on the proximity of the spacecraft to the confluence of the Earth's magnetic field lines (e.g., the South Atlantic Anomaly). For a 1000 second exposure, the fraction of pixels affected by cosmic rays (in non-SAA passages) varies between 1.5% and 3%. This fraction is the same for the WFC and HRC despite their factor of two difference in pixel areas because the census of affected pixels is dominated by charge diffusion, not direct impacts. This fraction is also consistent with what was observed for WFPC2.

For most science observations, a single "CR-SPLIT" or dither (i.e., two exposures) is sufficient to ensure that measurements of the targets are not compromised by cosmic rays. Due to detector artifacts like hot pixels, dithered exposures are strongly recommended over a "CR-SPLIT." More consideration is required for survey-type observations with WFC, a bonafide survey instrument. Observers seeking rare or serendipitous objects as well as transients may require that every single WFC pixel in at least one exposure among a set of exposures is free from cosmic ray impacts. For the cosmic ray fractions of 1.5% to 3% in 1000 seconds, a single ~2400 second orbit must be broken into 4 dithered exposures of 500 to 600 seconds each to reduce the number of un-cleanable pixels to 1 or less.

Electron Deposition

The flux deposited on the CCD from an individual cosmic ray does not depend on the energy of the cosmic ray but rather the length it travels in the silicon substrate. As a result, the deposition distribution has a well-defined minimum with few events of less than 500 electrons (where such low-electron events correspond to cosmic rays which pass through the CCD at a normal angle of incidence). As seen in Figure 4.15, the median charge deposited for WFC and HRC is about 1000 electrons, the same as for WFPC2.

A useful characteristic of the deposition distribution is its well-defined minimum; e.g., multi-pixel events which have an apparent magnitude of 25th or fainter, in a 500 second broad-band exposure, are unlikely to be caused by cosmic rays. Such information can help with the removal of false positives from searches for faint transients (e.g., high-redshift SNe).

Figure 4.15: Electron Deposition by Cosmic Rays on HRC (top) and WFC (bottom)

A minimum deposition of ~500 e¯ is seen corresponding to cosmic rays with normal incidence. The median deposition is ~1000 e¯.

Attached Pixels

As seen in Figure 4.16, for HRC and WFC, the salient features of electron deposition are a strong peak in the distribution function at 4 to 5 pixels. On the smaller side there is a sharp decline in events. Although a few events are seen that encompass only one pixel, examination of these events indicate that at least some and maybe all of these sources are actually transient hot pixels or unstable pixels which can appear hot in one exposure (with no charge diffusion) and normal in the next. There is a long tail in the direction towards increasing numbers of attached pixels. Some of these are likely due to two individual events associated by their chance superposition, but more are from oblique incidence cosmic rays that skim the surface of the CCD leaving a long trail (which is wider near the surface). Unfortunately the number of attached pixels is not a very useful characteristic to distinguish cosmic rays from unresolved astrophysical sources.

Figure 4.16: Distribution of the Number of Pixels Associated with a Single Cosmic Ray Event for the HRC (top) and WFC (bottom)

Some bias exists for events > 6 pixels, which may be composed of two events with chance superposition. This distribution does not account for possible charge trails left by the CTE deterioration of WFC.