We have no detailed description of the physical make-up of the Thomson CCD. The following is a description of the chip as gleaned from several weeks of testing.
The pixels within the 1024x 1024 image area are very uniformly responsive to illumination. Outside this area there are:
- the column adjacent to the image area receives about 60% the image area
- the next outside column gets about 1.5-2% of image area illumination
- the third column gets 0.1% of image area illumination
- the response peaks again at the 5th column with 0.3% of image area responsivity
- the 10th column has 0.1% of image area responsivity
- power-on cleanout charge and normal dark current is generated over these ten columns
- the ten columns on either side are responsive to cosmic rays
- unexpected things are likely to happen in these ten columns
- we certainly cannot trust these ten columns for bias level estimation
- saturation from flat fields does not necessarily produce dark current in these ten columns except for the first column
- ``columns'' beyond the ten are real overscan columns which do not exist adjacent to the image area.
- the row adjacent to image area gets about 85% of image area illumination
- the second row gets about 6% of image area and can show hot top row effects in long dark frames
- both rows are cosmic ray sensitive and show dark current
- the number of overhanging pixels between the image area and the
readout amplifier appears to be 18
- the readout register is cosmic ray sensitive
After powering on the CCD, it takes some time for the chip to clean itself out and for the dark current to become sufficiently low for astronomical applications. Power-on dark current is flat and featureless but takes several hours to fall to negligible levels, as shown in the following table. This table is based on NO strong light being applied to the CCD as it cleans itself (see the section on residuals).
Power-on dark current extends into the ten columns either side of the 1024 square image area and the two rows above and below this area.
Poor horizontal charge transfer efficiency (HCTE) was found using the fast H clock rates recommended by the CCD manufacturer. Major improvements were made by providing longer (slower) H clocks - the longer the clock time the better the result. A total H shift time of 50 us looked pretty good, but the readout noise does suffer and readout times are increased by about 60 seconds.
Observers should use the readout speeds SLOW_BESTCTE or NORMAL_BESTCTE for the best HCTE or the FAST_GOODCTE for rapid readout with not quite the best HCTE. NONASTRO speeds provide fairly poor HCTE and will show low level charge trails in the row direction. This speed is intended for use only on images with high backgrounds of several hundred e/pixel.
In July 1989, the AAO CCD controllers were fitted with a fast cleanout facility. Before this feature was added, the cleanout phase that takes place just after a CCD readout was achieved by the repetition (1000 times for small CCDs, 2000 times for the Thomson) of the sequence one vertical shift followed by a shift out the full row (plus a few more pixels) of the readout register. The cleanout of a Thomson CCD would take about 120 seconds in BESTCTE mode or 18 seconds otherwise.
Fast cleanout is a repetition (1000 or 2000 as before) of the sequence one vertical shift followed by only one horizontal shift of the readout register. This cleans out the Thomson CCD in just over 0.1 sec in the BESTCTE mode of operation.
By the same technique, fast cleanout is used to reject the unwanted CCD rows ahead of and following an astronomer window. This avoids the otherwise long delays that would result if these unwanted rows were fully shifted out in the customary fashion. In fact, the minimum BESTCTE readout time of the Thomson CCD would be 72 seconds for a very small window if the fast cleanout mode was not available.
HOWEVER, it should be recognised that the fast cleanout process in effect bins all of the unwanted rows into the readout register and if these rows have a high intensity background then there is a risk that the readout register will saturate (note that the readout register has a much higher charge handling capacity than an image area pixel) and saturating charge will spread down the columns from the readout register. The extent of downward spread from the unwanted charge existing above the window will generally not reach the astronomer window . Also, the readout register is thoroughly cleaned out by extra rows added above and read out just prior to the astronomer window.
But if there are a large number of unwanted rows below the astronomer window to the last row of the CCD, and these contain extended areas of high intensity, the resulting saturation when these rows are binned in the readout register may spread down from the readout register into the astronomer window area (but not into his data as that has already been read out). The result, however, will be the generation of higher dark currents emanating from the residuals of this saturated charge spread.
NOTE that this is a problem only when windowing, and then only when windowing in the vertical direction (ie full ROWS are being discarded), and then only if extended intense images are attempted such as high level flat fields with, say, 20,000e-/pix. The resultant dark current effects on the next image are too difficult to quantify as they depend on the window size, position, intensity of previous bright images or flat fields, time since these images, etc.
In the past, variations in the bias levels with time have been noticed. Generally, the bias level may fluctuate slowly plus or minus one or two ADUs but on rare occasions the bias level was seen to jump by several ADU between two readouts (not during a readout).
Considerable effort has been spent to remedy this but although some changes have been made and suspect components replaced, some drifting is still apparent. The situation is made worse by the fast readout time of the Thomson CCD requiring an extra gain (when compared with the GEC and RCA CCD systems) of 2.5 to be placed after the DCS where, it appears, some of the drifts take place. The setting of a ``gain balance'' facility in the DCS circuit has been found to be readout rate dependant and the reason for this is not clear. The setting has been optimised for the NORMAL and the SLOW readout rates - FAST and NONASTRO are thus poorly compensated for general system drifts and will show a worse bias level stability.
Astronomers should note that the bias level will be affected by all of the following:
Bias frames and zero background images show faint traces of striations attributed to mains injected noise. These are not so apparent on NORMAL or SLOW readouts but can be seen on FAST and NONASTRO bias frames where the readout noise is digitisation limited. The effect is at a very low level and made more noticeable by the high contrast LEM images which span only a few ADUs. This noise is made worse (compared with the GEC and RCA systems) by the 2.5 times gain after the DCS and the use of long cable lengths at the focal reducer and at the CES.
The following are the ``final'' readout noise (RON), Gain and Alpha values (where Alpha is the non-linearity coefficient such that
N(measured) = N(actual) x [1 - Alpha * N(actual)]
and readout times are approximate only):
N.B. For all practical purposes the CCD is linear.
Measurements in the lab put the cosmic ray event rate at about 220 per 2000 seconds.
On 28/6/89, at 170K in final dewar and including window:
The quantum efficiency of the Red Thomson CCD is compared with that of the other AAO CCD's in Figure 6.1 of the AAO Observer's Guide. PUT LINKS HERE!
Readouts at SLOW, NORMAL and FAST rates are all ADC limited and even at the 65K ADU level the CCD is nowhere near saturation. CCD saturation occurs somewhere about 400,000e-/pixel.
Readouts using NONASTRO are limited to a maximum of about 30K ADU because of the saturation of the electronics amplifiers ahead of the DCS stage and NOT the CCD.
Residual images after bright illumination appear to be of two types.
A dark frame taken 20 minutes after an approximately 250Ke-/pix bright vignetted flat field showed the dark current to be 32e-/pix/2000sec with the same bowl-shaped uneven illumination of the flat field. Twenty minutes min later, the dark current was 20e-/pix/2000sec.
Above and below saturation area refers to the trailing appearing above the residual and the trails below the saturation area. O'scan refers to that trailing into the V overscan. Gen DK refers to the general background well away from the spot image.
Higher CCD operating temperatures do not affect the strength or decay rate of the residuals.
Some problems of data conversion and data corruption in the serial link to the LEM were experienced in the first couple of months of use of the Thomson CCD at the telescope. These have been cleared up. However, it seems that cutting down the analog delay times in the Thomson system to facilitate fast readout times has led to the possibility that the analog to digital conversion may be less than perfect. Astronomers should keep an eye open for recurring odd data values or histograms of data that show the presence of missing ADC codes.
These occur fairly frequently at random locations and may appear as single pixel events or in horizontal or vertical pairs with amplitudes up to several thousand electrons. They look very much like cosmic ray events except they are distinguished by not showing the low-level cloud of electrons found in adjacent pixels of a cosmic ray event. In two Thomson frames, during lab tests, 9 single events and 2 pairs occurred. Magnitudes were up to several 1000e-. These ``events'' are definitely on the CCD as they show trailing when the HCTE is deliberately made poor.
Examples of non-recurring hot pixels are:
The column and row numbers are based on a 1024 square image. The intensity is a measurement of the trailed charge level on the third overscan row, (ie the third row below the last row of the 1024 square image) with the CCD illuminated with a flat field of about 70,000e-/pix. The trap length is an estimate of the number of rows beyond the trapping site that the dark column extends (the dark column being produced by the trapping site on the low-level flat field). These measurements were done at flat-field intensities of approximately 100 e-/pix and 20 e-/pix.