We report here on the behavior of saturated data taken in the upper part of the STIS CCD detector, closer to the default readout register. Saturation in this region of the detector can lead to image artifacts and deviations from linearity larger than those previously reported.
The STIS Instrument Handbook (IHB) states that when an individual pixel of the STIS CCD detector reaches its full well of about 144,000 electrons, the excess electrons bleed along the CCD column, (i.e., in the direction of parallel transfer), and it also states that when used with gain=4, none of these electrons get lost, and all are fully recovered during the readout. Since this bleeding of electrons is perpendicular to the dispersion direction of the STIS CCD gratings, this means that after integrating over the cross-dispersion direction in each column, much of the linearity, photometric precision, and spectral resolution of a spectrum extracted from a saturated image should be preserved. Only some information about the spatial profile is irrecoverably smeared. Bohlin & Gilliland (2004, AJ, 127, 3508) exploited these behaviors to show that linearity can be preserved at the 0.2% level even for exposures at about 80 times the full well level.
However, these previous tests were done near the central rows of the STIS CCD detector. It appears that saturated data taken near the top of the detector can show significantly larger departures from linearity than were found by these earlier studies and can also show substantial image artifacts in the serial transfer direction that take the form of long tails emanating from the saturated pixels.
A good example of this can be seen in a deep image of a star field in the open cluster M67 taken in 1997 a few months after STIS was installed into HST. The serial readout register used for this observation is located at the top of the image, and the "D" amplifier used for the readout is at the upper right corner. After the exposure the accumulated charge was shifted up one row at a time ,and, once in the serial register, each row was shifted out towards the right one pixel at a time to be measured by the amplifier. An image of star near row 845 of the detector that was exposed to a level of about 6X over full well in its central pixel shows a strong serial tail that continues through the serial overscan and wraps back around onto the other side of the detector. However, other stellar images lower on the detector at similar or much larger saturation levels do not show such artifacts.
The circumstances under which such artifacts can appear have been clarified by some new data recently obtained as part of CAL/STIS program 13142. This program used internal calibration lamps and specialized apertures to illuminate narrow stripes on the detector at various illumination levels near both row 515 at the center of the detector, and also in separate exposures near row 907, closer to the serial readout. It appears that the full well depth of the STIS CCD increases towards the top of the detector. Near row 515, individual pixels begin to depart from linear behavior and bleed along the detector columns after collecting about 130,000 e-, while near row 907 this does not start to happen until about 155,000 e- have been collected in an individual pixel. This allows rows saturated near the top of the detector to reach the upper serial readout register with larger amounts of charge in each pixel than do saturated rows from near the center of the detector even for cases when the incident illumination was identical. Near the center of the detector the charge in a saturated column can bleed over more rows allowing the peak level to remain lower. Normally, one would not expect this to cause any additional issues; however, when individual pixels with more than approximately 150,000 e- reach the serial register, the serial transfer artifacts discussed above then begin to appear.
The detector locations at which these artifacts begin to appear has not been completely defined and may depend in part on the degree of saturation. Near the E1 position at row 900, which was defined close to the readout to minimize losses due to charge transfer inefficiency effects, they may begin to appear at a very low level at or near the full well level. Near the center of the detector, overexposures of more than 50 times may be required before any hint of these features appear. For modest saturation levels near the E1 position at row 900, the fraction of signal in the serial tails is still very small. For the case shown in the figure, we estimate that about 0.5% of the total light from that star went into the serial artifact.
The behavior discussed here will primarily be of concern for users who either are very heavily over-saturating the detector or who require very high contrast ratios in the presence of moderately saturated data. Users requiring very large over-saturation factors should avoid use of the E1 position near row 900 and instead put their target close the center of the detector. However, users of targets that are well below the CCD saturation level should have no concerns about continuing to use the E1 position. Coronagraphic users putting their target at positions near the top of the detector should carefully consider how this issue may affect their science goals and analysis and may instead prefer target positions in the lower half of the detector.