MITLL2A REVIVAL - TEST SUMMARY ****************************** Revised 24/3/00 Started 20/3/00 J.R.BARTON mitll2a_revival_test_summary.txt MITLL2 has been revived and the CCD is now called MITLL2A so that future data taken using the recovered MITLL2 will be easily distinguished from that taken prior to the mid-January, 2000, failure. This test summary should be read in conjuction with the earlier test summary dated 14/10/97 for MITLL2, as only the changes will be noted in this report. Note that the revived CCD is read out from the opposite CCD amplifier so that new images will now be left/right reversed. In particular, the defective columns will now appear on the left side of images rather than the right side. Images taken will now match those from MITLL3. BACKGROUND ---------- In mid-January, 2000, MITLL2, CCID-20 W20C2, suffered a serious breakdown after a loose nut and a considerable quantity of fine aluminum shavings and powder from a broken LN2 can support found its way onto the dewar electronics printed circuit and, probably, the backs of connectors. Initially, a failure of the CCD temperature monitoring circuits was reported. The dewar was opened and the cause of the problem ascertained. The temperature sensor was found to be operational. The dewar was cleaned and new LN2 can supports fitted. However, the CCD failed to function at all when fitted at UCLES. The CCD output was found to be at only 2 or 3 volts rather than the expected 20V. The dewar and electronics box were taken to Epping for a closer look. On arrival at Epping the dewar was again cleaned as more aluminium shavings had gathered on the window. The wiring, connectors and PCB were inspected but no damage was found. Fortunately, the loose nut appeared not to have physically damaged the imaging surface of the CCD. This time, after cooling, the CCD imaged correctly, its output now up at the expected voltage. However, these first images showed the presence of a very strong light emitting defect (LED) located in the corner containing the amp-b amplifier, i.e. the output then being used for reading out the CCD. Radiation from this corner into the image area was sufficiently intense that remote pixels were badly affected in longer exposures. Even on a short bias frame remote pixels were illuminated as they were shifted up to the readout amp during the course of the readout. The final result was a bright band down the left hand edge of the image which trailed off to a non-zero level on the right hand edge. This rendered the chip useless as there was nowhere on the CCD that was unaffected to some level by the LED. THE SOLUTION ------------ 1. USE OF OPPOSITE READOUT AMPLIFIER Tests showed that there seemed no possibility of quelling the LED while the output of the CCD was being taken from the original amp-b. Amp-a was thus wired up and further tests showed that the LED was associated with the reset drain electrode of the CCD's b-amplifier output. The move to output amp-a freed the reset drain and allowed its voltage to be adjusted to minimise the LED. Lowering its voltage from the original value of 14 volts rapidly reduced the LED intensity but below a threshhold of about 4.4 volts further reduction rendered the CCD imaging completely inoperative. At a value of 4.6 volts the CCD performed well and the LED was deemed sufficiently quelled. Use of the opposite amplifier means that a left to right reversal now applies to images compared with those taken prior to the failure. 2. TOP RIGHT CORNER LED INTENSITY The LED, which initially saturated corner pixels in a bias frame, now produces a hot top-right corner which peaks to about 100e-/pix/2000sec in a few pixels but averages 40e-/pix/2000sec over a block of 15x15 pixels in the corner to less than 10e/pix/2000sec in a 75x75 pixel block. Radiation from this corner LED falls off rapidly to be virtually undetectable beyond 2-300 pixels out from the corner in 2000 sec exposures. Thus, the LED has been quelled by about 6 orders of magnitude. THE PAY-OFF ----------- 3. 1/F NOISE Unfortunately, amp-a was always known to suffer from 1/f noise which randomly caused the bias level to drop by about 12 ADUs for several tens of consecutive pixels (the number affected being quite variable) at random times, The effect was to fill a readout with hundreds of short darkened horizontal lines. Luckily, the 1/f noise could be reduced (but not entirely eliminated) by lowering the operating voltages of amp-a. Now, on occasions, the 1/f noise can be seen as short streaks along a line, with an ADU value slightly lower than its surroundings. 4. REVIVED READOUT NOISE, GAIN AND LINEARITY The price for suppressing the 1/f noise of amp-a by lowering its operating voltages is a slight change in gain (e/ADU), a small increase in the readout noise and a significant reduction in the maximum well size. The readout rates for FAST and NORMAL have been slowed by one microsecond/pixel. This enables the reduced full well of 115Ke- to be fully sampled over 65535 ADUs using the FAST readout speed without hitting either CCD saturation or the ADC limits. At NORMAL readout speeds a worthwhile reduction in the readout noise was attained. SPEED INT GAIN RON ALPHA SAT'N READ RATE # READOUT TIME (sec) (us) e/ADU e-rms Ke- (us/pix) full 2*2 bin 5*5bin 10E6pix NONASTRO 1+1 4.8 5.6 -1.48 125 6.5 60 24 9 40 FAST 2.5+2.5 1.8 3.2 -0.16 115 11.5 105 42 15 67 NORMAL 4.5+4.5 0.9 2.4 0.02 59 19 170 59 19 78 SLOW 12+12 0.37 1.8 0.04 24 34 300 93 24 100 XTRASLOW 48+48 0.091 1.4 0.01 6 106 940 260 56 205 5. BIAS LEVEL JUMPS Another problem associated with amp-a (and which is possibly related to the 1/f noise problem) is that the bias level may jump in level (but by only about 1 ADU in FAST), usually over extended periods, thus affecting many consecutive rows of a readout. Both the image area and the row overscan are equally affected. Fortunately these jumps are quite small, either less than an ADU or less than 1-2 e-/pix depending on the readout speed, and, if considered excessive, should be removable by using the line overscan.