Bright sources may yield faint residual images that persist in subsequent exposures. This persistence is generally low for typical NIRCam observations.
Image persistence, or latency, represents one of the main anomalies of near-IR detectors. Pixels subjected to strong illumination produce a faint residual image in the following integrations. Even if the effect is more visible for pixels exposed to bright sources, any amount of illumination can cause latent images. Persistence may last for several minutes after the end of an exposure and is not restricted to pixels that have reached saturation; nor does it disappear with detector reset.
Persistence is generally low for typical NIRCam exposures and dithers, but it can be noticeable depending on target brightness and exposure setup. Removing or mitigating persistence during data post-processing is among the objectives of the JWST calibration pipeline. This article provides information to estimate the amount of persistence one may expect in an exposure before attempting a software correction.
Update from commissioning
In-flight persistence is measured to be similar to initial estimates in most detectors and higher in short wavelength detectors A3, B2, B3, and B4. We still expect persistence to have a marginal impact on the majority of observations. Detector A3 shows the highest persistence levels and longest recovery time, and also the highest increase (~50%) in levels relative to estimates from ground testing. Effectively, that means a source ~0.5 mag fainter will result in persistence levels previously predicted in detector A3.
The remainder of this article presents predictions and measurements from pre-flight ground testing data.
Smith et al. (2008) have proposed a model explaining the origin of persistence. The detector lattice contains impurities that act as charge traps. When the detector is initially reset, a strong electric field creates a charge-free region within the material; under its influence any photo-generated free charge is immediately swept away and collected, thus no trap capture can occur. As the signal accumulates, the electric field decreases and the charge-free region shrinks, leaving behind an increasingly wider neutral region. Photo-generated free charges will slowly diffuse across the increasingly thicker neutral region to reach the residual charge-free region and be collected. In the process some of them may be captured by the traps, with probability and timescales that depend on the properties of the trap impurities. Eventually, the captured charges will be released back and collected, generating a spurious time-delayed signal (persistence).
The capture and release process can be characterized by a combination of exponential functions with timescales associated to the different trap populations. Leisenring et al. (2016) have analyzed the persistence of NIRCam's flight detectors (or sensor chip assemblies, SCAs) with respect to source flux, rate of discharge of the charge-free region, and over-fill levels, crafting a semi-empirical model of NIRCam persistence.
The data indicate that the average rate of charge releases are consistent within a factor of 2 for the 8 short wavelength (SW) detectors. For the 2 long wavelength (LW) detectors, the average rate of charge release are strongly similar (see Figure 1). The average results for the 2 types of detectors are presented below. Certain detectors show some significant spatial variation; therefore, individual maps to illustrate the regions more affected are also provided. These maps may allow observers to place the brightest offending sources in the less affected areas of the focal plane.
Figures 2 and 3 present the results obtained using the Leisenring et al. (2016) model for the NIRCam short and long wavelength channels, respectively. Each plot shows the accumulated persistence (total released charge, in electrons) after a bright illumination as a function of the fill level of the saturation level ("well fill fraction," vertical axis) and integration time (horizontal axis). The well fill fraction ranges from 0.1 to 5 times the saturation level (about 120,000 e–) whereas the integration times ranges from 0 s to 1,400 s. The results are color-coded according to the color bar at the bottom of each plot. The plots are relative to the following cases:
- Left plot: integration starting immediately after the illumination exposure, without a break between them. This case is representative of the persistence measured in coronagraphic or in time-series observations, and in general, for all cases of exposures containing multiple integrations. The maximum latent signal, about 200 e– at 5× saturation (light blue at the top-right corner), is much smaller than the shot noise of the source, i.e., SQRT(5*120,000) = 774 e–. However, this effect is systematically raising the measured signal in a ramp, so it must be taken into account, for example, to correct for absolute photometry.
- Middle plot: in this case the integration starts 45 s after the previous offending integration. This is about the minimum time needed to restart an exposure after a dither move, excluding overheads (detector setup, wait for next readout cycle, reset all pixels, cleanup after the exposure) that may add about 70 s. The time delay allows trapped charges to be released and swept away by the strong electric field set by the interim reset frames. Since a ~50% fraction of the traps has a short decay timescale, the time delay required by a dither move cuts down significantly the total amount of released charge. Still, the systematic effect may become visible over the sky background (TBD).
- Right plot: in this case, the integration starts after 1,800 s, which is the nominal time (including overheads) charged to slew to a new target. In this case, the release of charges is in practice negligible, reaching 21 e– at the end of a 1,400 s integration following a 5× illumination.
Words in bold italics are also buttons
or parameters in GUI tools. Bold
style represents GUI menus/
panels & data software packages.
The primarily concern is the brightest pixel at the center of the point spread function. To estimate the counts, an extraction aperture, corresponding to the area of a pixel, needs to be set up. Since observations are being done with the F200W filter, the short wavelength channel with a pixel size of 31 mas is being used, an area of about 1/1000 of a square arcsec. The radius of the circular aperture with the same area is 17.5 mas.
In the Strategy tab of the ETC, with Aperture radius = 0.0175" and selection of the noiseless sky background button, a flux of 1,238 e–/s is obtained, having added the small contribution from the background. The total flux is therefore 1,238 e–/s × 236.12 s = 292,317 e–, which is a well fill fraction of about 2.4 (saturation occurs at about 120,000 e–). Using the central panel of Figure 2, the next dithered exposure of the same length and in the same filter will show about 75 e– of persistence at the peak position of the source. A check with the ETC shows that this is comparable to the signal produced, with the same instrument configuration, by a K ~ 25 magnitude source that would be detected with SNR = 4.7.
Figures 4 to 7 show the persistence distribution across each SCA. Figures 4 and 5 refer to the SW and LW channels of module A, while Figures 6 and 7 are relative to module B. These maps were derived from extremely super-saturated data held at saturation for a long time, and therefore, they are representative of the total trap populations. The SW maps tend to show considerable structure, apparently correlated with the dark current distribution. The LW detectors, however, are pretty flat. A value of 1.0 corresponds to the median persistence of all SW or LW detectors. Therefore, one can use this information to tie back to the average contour plots presented in Figures 2 and 3, getting estimates at specific locations. This information may be useful if one has bright sources in the target field, as a guide to avoid certain areas. In particular, grism time-series observations use A5, A1, and A3, and this last SCA is affected by non-uniform persistence. Time-series imaging observations (using primarily B1 and B5) and NIRCam coronagraphy (using A2, A4, and A5) are marginally affected by field dependence of persistence.
Leisenring, J. M., et al. 2016, Proc. SPIE 9915
Characterizing persistence in JWST NIRCam flight detectors
Smith, R. M., et al. 2008, Proc. SPIE 7021
A theory for image persistence in HgCdTephotodiodes