4.2.3. Charge Transfer Efficiency (CTE)
Accurate photometry depends on stable charge transfer efficiency (CTE), which is the efficiency with which electrons are transferred between pixels during read out. Variable CTE as a function of total charge level, time, or position over the chip results in non-linear sensitivity. Unfortunately, the WFPC2 chips are known to have a "a small parallel CTE problem" (Holtzman et al. 1995b, henceforth H95b; Whitmore, Heyer, & Casertano 1999): the percentage of electrons which are read-out of a pixel depends on the row number of the pixel in question, the number of electrons in that pixel when read-out begins, and the mean charge level in the pixels through which the charge travels during read-out.
Based on in-flight point source photometry and on laboratory tests using a CCD from the same silicon wafer as WFPC2 CCDs, the CTE variability is known to be caused by electron traps which were in the silicon itself before the pixel mask was etched into the wafer (J. Trauger, private communication). Although electrons can be trapped in the silicon at the pixel where they are detected, a greater surface area of silicon is encountered during read-out, providing greater opportunity for trapping if the traps are not already filled before readout begins. When a bright point source is imaged against a faint background, many new traps are encountered during read-out. For example, 4% of a 10,000 e- point source are lost when the charge is transferred over 800 rows if the background level is ~ 10 e-/pixel. If the traps are already filled by a high background level over the whole chip, fewer electrons will be lost from sources during read-out. Surface photometry is less affected by this CTE problem because all pixels are filled to the same charge level and, hence, new traps are not encountered during read-out.
The signal level read out for our images is roughly 80e-/pixel for the F555W and F814W images and 2e-/pixel for the F300W images. We have identified the CTE losses in these data by two methods. First, data from WFPC2 and laboratory tests conducted by J. Trauger of point sources show that ~ 0.35e-/pixel are lost from a point source which is imaged against a background level of roughly 80e-/pixel, while ~ 1.1e-/pixel are lost when the background level is zero (J. Trauger, private communication). The difference between the number of electrons lost at the two different background levels indicates that the number of single-electron traps available in a uniform background of 80e-/pixel is 0.75e-/pixel (~ 1.1-0.35). Second, to confirm that this trapping reflects the CTE losses for uniform sources, we have conducted another set of tests with the help of J. Trauger to measure the non-linear response of the spare WFPC2 CCD to uniform, low-level backgrounds. In this test, we simply expose the CCD to a light source which is stable to better than 0.1% for varying lengths of time and look for non-linearities in the detection rate. The results of these uniform-source test are in excellent agreement with results from point-source tests. We therefore conclude that 0.75 (± 0.25)e-/pixel are lost when a mean level of ~ 80e-/pixel are read-out, and 0.1 (± 0.05) e-/pixel are lost from a mean level of ~ 2e-/pixel. The maximum uncertainty in these corrections is a negligible contribution to our final results (see Table 3). For more detail on the result of the laboratory tests, see Bernstein (1998).