[Skip to Content]


General Links
Professional Links
Australian Human Resources Institute Diversity Awards 2013 Finalist

MITLL3 Deep Depletion 2Kx4K 15um Pixel CCD


  1. The Device
  2. Use with AAO Instruments
    1. UCLES
    2. UHRF
    3. Others (including decommissioned instruments like Taurus & RGO)
  3. A historical Serial Link / NONASTRO speed problem

This page contains information on the AAO's science grade deep depletion 2Kx4K device from the MIT/LL/UH consortium - hereafter MITLL3. It was first commissioned on the AAT in Oct 1998. The information provided is based on tests made at the Epping Lab by John Barton (a copy of his comprehensive report can be found here), observations made at the AAT with LDSS and the RGO spectrograph by C. Tinney and K. Glazebrook, and experience with MITLL3 use on UCLES/UHRF.

1. The Device

Identity & Format

This is device W67c2 (chip 2 from wafer 67) of the CCID-20 format run fabricated at MIT/Lincoln Labs and packed on an AlN substrate by GL Scientific. It is a 2K x 4K edge-buttable device with 15 micron pixels. The wafer used was a high resistivity, or deep depletion, one. The pixels are 40 microns thick.

The pixels on the MITLL3 are read out flipped in X (where X is defined as the X-direction as seen on AAO XMEM images and FITS files) relative to the MITLL2.


Binning in both the vertical or horizontal directions with the MITLL3 is impossible, due to hot columns which bleed charge into adjacent columns when binned.

Quantum Efficiency

The SiO AR coating used on this device is optimised for red performance. This, together with its thickness, produces an enhanced quantum efficiency longwards of 6000 Å. The quantum efficiency results shown for the MITLL3 below were obtained in the Epping lab, with the device being operated in diode mode. The MITLL2 results were mesaured using an earlier set of photodiodes which had gone out of calibration. However, by measuring the MITLL3 QE with both old and new photo-diodes, we have corrected the MITLL2 numbers. These MITLL2 numbers are to be used in preference to previous MITLL2 curves.

A larger version of this figure is also available, as is a colour postscript file. The data used to produce it can be found here.

NB: because the Postscript file was created by an Evil Empire Microsoft product, it will not preview with
Ghostview version 1.5 or earlier. It will preview with Ghostscript version 4 or later, and it will print out.

Longward of 5000 Å the MITLL3 has QE as better than, the Tek 1K. The same wavelength for the MITLL2 is 4500 Å.


Because the detecting layer is so thick this device has generally lower fringing than most thinned CCDs, while retaining excellent red sensitivity. The extent of fringing was explored using the MITLL3 on the 25cm camera of the RGO spectrograph with a 600R grating in October 1998. The following image shows a flatfield in the wavelength range 6400-9500 Å normalised for the flat-field lamp's spectral response. (Click on the image to see a larger version. Note that this image is just the 330x3200 pixel region illuminated by the RGO, and has been rotated and had its x-axis reversed in order to put it in the "conventional" spectral orientation).

Figaro NDF Image (4Mb)
Gzipped FITS Image (3.3Mb)

These results show fringing on two seperate length scales: a large scale effect on scales of hundreds of pixels with 5% p-p amplitude at 9500 Å, and a small-scale effect on scales of 20 pixels (or 20 Å in wavelength) with 5% p-p amplitude at 9500 Å. This can be seen more clearly in the following cut through the above image made at Y=200-210 pixels. We see ±1.5% p-p at 8500 Å, ±4% at 9000 Å and ±5% at 9500 Å. This is about half the fringing seen in the MITLL2, which is itself significantly less than the fringing seen in the TEK.

Postscript version of this plot

The main result is that fringing is (1) negligible below 8000 Å, and (2) even above 8000 Å it is quite small.


The device has a number of hot pixels, and one serious bright defect. The serious bright defect does not affect RGO observing, as it lies right at one edge of the RGO slit. When observing with Taurus or LDSS it will cause the loss of several columns. With UCLES, appropriate set-ups can place it between, rather than on, an order.

The remaining hot pixels cause vertical trails in bias frames. In long dark frames several weaker hot pixels appear and become more prominent. The following images show bias and dark frames acquired approximately 9 hours after the CCD had been powered on.

The device shows a small number of trapping sites. The currently known locations of both bright defects and trapping sites are listed here.

Gzipped Figaro 
BIAS (3.2Mb)
Gzipped Figaro 
DARK (3.2Mb)
Gzipped FITS 
DARK  (3,2Mb)
BIAS Frame in NORMAL Speed created as median of 9
frames. GIF Images are streched between 520 and 530 adu. 
DARK Frame in Normal speed created as median of 5
frames. GIF Images are stretched between 520 and 535adu 

The flat field cosmetics appear to be good with a "brick-wall" considerably reduced over that seen in the MITLL2. Currently we have no instrument which can actually illuminate thr entire CCD. However, an image of the 330x4096 region (X=805-1135, Y=1-4096) illuminated by the RGO can be found here in FITS and FIGARO format. These images are spectrally normalised flat-fields, and should not be trusted in the top and bottom 400 pixels where the RGO is heavily vignetted.

The device does show regular horizontal changes in QE of about 0.5% at a spacing of about 70 pixels. Many of the hot pixels can also be seen as defects in the Si in flat fields, as in the following pair of images from the X=805-1135, Y=1050-1350 region.

Dark frame showing hot pixels.  Flat field showing Si defects 

Cosmic Rays

Because of the thickness of this device it seems to be very susceptible to cosmis rays. The observed hit rate for the entire detector is ~830 per 1000s. A particular point to note is that the cosmic rays are often seen to be quite extended. One hundred pixels is not uncommon. Moreover, cosmic rays are obviously interacting with the Si in this CCD to produce both curved trails and even small showers.


The MITLL3 detector sits in a standard AAO dewar, and will mount on all the AAO instruments - however it has a preferred orientation because it is rectangular. This means, for example, that on the RGO it has to be mounted rotated relative to the TEK.

In general its focus will be different from the MITLL2 and the TEK for each instrument.


NOTE: The readout characteristics presented in this table are appropriate only for the AAO-1 CCD controllers used up until ~August 2004. If your data has a header keyword FIRMVSEQ containing 'Sequencer: AAO2 CCD Controller', then you should refer instead to the CCD Performance with the AAO2 Controllers page.

Modifications have been made to the AAO's existing controllers to take advantage of the superb read-noise performance of these LL devices. This has allowed a device 8 times larger than our TEK 1K to be read in about twice the time with similar or better noise performance. In general the read-noise performance is identical to that of the MITLL2.

FAST  2+2  2.2  2.9  -0.15  148  10.5  108 
NORMAL  4+4  1.1  2.0  0.24  62  18  143 
SLOW  12+12  0.36  1.6  0.15  23  34  285 
XTRASLOW  48+48  0.088  1.3  0.03  106  924 

Note that linearity corrections will need to be applied to all data. Though these are large for an optical detector, they are small compared to those seen in IR devices. After linearity correction, data taken in the NORMAL,FAST and XTRASLOW speed are linear to better than 0.1% over the full dynamic range.

Dark Current

A 160K operating temperature has been adopted. At 160K the dark current is small - a 2000 sec dark shows it was less than 0.4e-/pix/2000sec, once the CCD has been powered on for a day. However, the detector takes several days to reach this low a dark current level, due to a slow CCD cleanout rate. The CCD cleanout rate following power off and on, with a cold CCD, as measured in dark frames are:

Time after Power on
2 hours 
8 hours 
24 hours 
2-3 days 
 0.3 to 0.4 

A further complication is that (at least for the MITLL2) the dark current levels observed when tested in Epping have not been replicated on the telescope. This results in some serious scheduling implications for UHRF observers, and UCLES observers of faint targets for the MITLL2 and MITLL3. UCLES observers of faint targets with MITLL3 are strongly urged to read the following report on MITLL2 dark currents.

Residual Images

There are no signs of residuals in 300s dark frames taken directly after pinhole exposures of 350 times chip saturation, i.e. at 50Me-/pix.

Clock-induced Dark Currents

Are too week to be detected without binning, and since binning is not possible, can be ignored.

Saturation Stripes

Saturation stripes in the row direction across bright saturating pinhole images used to simulate bright point sources have not been detected. No bar patterns or regions of depressed or raised bias level were seen when the local spot was overexposed to 50Me-/pix.

2. MITLL3 on AAO Instruments


The MITLL3 device has been commissioned with UCLES, and gives smaller pixels (15um) than the TEK, improving spectral and spatial resolution, a somewhat increased wavelength coverage, and complete sampling (with no inter-order gaps) much further into the red. Users should be aware that MITLL3 images are flipped laterally compared with MITLL2, i.e. lower order numbers are to the right, but wavelength still increases from top to bottom of each order.

By judicious adjustment of the cross disperser grating angle ECH_GAMMA, it is possible for the most part to position the worst of the hot columns in or near an inter-order gap, even with the 31.6 l/mm grating. The inability to bin MITLL3 data on-chip is only a minor concern, but observers are reminded to enter the command "DATA USHORT" after running up the CCD system in order to keep the file sizes manageable. For example, each image taken with the MITLL_PLANET window (2048 data columns + 40 overscan columns x 2496 rows) is 20 Mb in size.

The fringing numbers found with the RGO (amplitude of ±1.5% p-p at 8500A, ±4% at 9000A and ±7% at 10000)  can be expected to produce similar effects with UCLES.

UCLES observers of faint targets are strongly urged to read the following report on MITLL2 dark currents.

An important point to note is that the current UCLES camera optics cannot illuminate the entire area of the LL detector (which is 60 x 30mm in size). In fact the unvignetted region which can be observed is more like 38.5 x 18.8 mm (for less than 10% vignetting). The region covered at 50% vignetting is 60 x 34 mm, which is approximately the entire LL chip, however unless you are working in the very red, the echellogram will not put any light on much of the chip.

The following sample GIF images from ECHWIND the region of the LL chip illuminated and the effects of vignetting.

Suitable windows should therefore enable the read time to be cut by a factor of 2 over the 'full chip' times given above.  (If the fact that we can't actually illuminate the whole chip seems insane to you, then I suggest you contact your ACIAAT representative and start lobbying for the UCLES Camera upgrade as soon as possible!)

Note of course that the 'boxes' shown are for one wavelength set-up. You can move the echellogram anywhere on this field - but the relative locations of the boxes will stay the same. As with Taurus, the main reason for preferring the TEK over the LL device is its superior QE in the blue.

2.2 UHRF

Because it cannot be binned, the MITLL3 is of very little use for UHRF observers. However, in the far red where fringing with the EEV2 device can be a severe problem, then the MITLL3 may still be your best bet, provided the source is bright enough.

A further complication is that the dark current levels observed when tested in Epping have not been replicated on the telescope. This results in some serious scheduling implications for UHRF observers. It has not yet been verified, but it seems extremely likely that the same is true for the MITLL3. See the following report on MITLL2 dark currents.

2.3 Other Instruments

The use of the LL Enginerring CCD on other instruments will be attempted if required by observers on a shared risks basis.


MITLL3 has been used with Taurus in the past. Taurus is now decommissioned for use on the AAT. This information is provided for historical completeness only. Contact Joss Bland-Hawthorn.

Field Sampling with Taurus II 

f/8  f/15 
TEK  0.594"/pix  0.315"/pix 
MIT-LL Eng  0.37"/pix  0.20"/pix 

The TTF at f/8 now gives 0.37"/pix over the full 9.87' field and in 1" seeing, this areal advantage is really paying off for the wide field surveys. There is not much need for f/15 now, and observers who do wish to use f/15 with the LL chips should think carefully about why they want to. Also the detector allows the full 10' diameter field to be charge shuffled between two frequencies, and almost the entire field to be charge shuffled between three.

There is little reason to prefer the TEK over the LL device for Red TTF use. For Taurus use with Blue TTF, however, the TEK device may be preferable.

RGO Spectrograph

MITLL3 has been used with RGO in the past. The RGO spectrograph is now decommissioned for use on the AAT. This information is provided for historical completeness only.

The MITLL3 device has been used on RGO spectrograph's 25cm camera. RGO users get smaller pixels than the TEK (15um vs 24um), together with a much increased wavelength coverage. Observers can either use smaller slits, or get better sampling of sky lines. Once again vignetting stops the whole chip from being used, with only about 3000 pixels being illuminated by the RGO - which is still more than double the wavelength range covered by the TEK1K. You can play with the possible options using the WWW version of RGOANG.

Once again only observers looking at wavelengths longer than 5000 Å should consider using the LL device, while those observing in the UV may prefer the TEK. Observers of single objects with the LL can achieve quite short exposure times by windowing out the largely useless spatial direction.

4. Serial Links and Possible Problems with data 18 Oct 1997 - 14 Feb 1998

During initial testing of the MITLL2 CCD at Epping, it was realised that the limitation to readout speed of this device was the speed of the Serial Data Link between the CCD Controller and the Large External Memory. A modification of the Epping Serial Data Link was undertaken to increase this speed. This included increasing the link clock rate from 20 MHz to 25 MHz, and decreasing the number of bits transmitted from 32 to 16. These modifications permitted use of the MITLL2 "Normal" and "Fast" readout modes. (The link however was still not fast enough to cope with "Non-astro" ).

This link was sent to Site and placed in regular use for the MITLL2 from 16 October 1997. On 20 January 98, the three Serial Data links at Site were modified to allow them to be used with the MITLL2 CCD. During testing of these modifications, it was noted (by John Sullivan) that one link was generating errors (i.e. the link error signal was causing the Error status LED to come on). Further tests revealed that all links (including the Epping link) were generating errors, at a very low rate.

In order to assess how many errors the links were generating, tests have been carried out of the speeded-up links between the AAT Vaccum Lab and the Control Room. A 1100 x 1090 = 1199000 pixel window was used.


Total errors (out of 1199000 pixels)























It should be noted that the type of "errors" occuring have not yet been quantified. The link error indicator can be generated as a result of a number of reasons, including an actual data parity error. Unforunately, to determine the type of error would require significant modifications to the circuitry. It has not been possible so far to determine if any of these errors are actual data errors. There have certainly been no reports of "suspect" data, and it is felt with confidence that there are no data errors occuring which result in values of either 65535 or 0.

A fix implemented on 14 February 1998 by Darren Stafford, resulted in this problem being cured.  It no longer affects any observations (except for use of the NONASTRO mode of the MITLL2 CCD using the old copper cable serial cards, which are no longer supported).


All observations taken with the MITLL2 CCD prior to 14 Feb 1998, and all observations made with all CCDs between 10 Jan 1998 and 14 Feb 1998 are potentially subject to these errors. However, the error rate is small (10 per 1 million pixels) and does not appear to produce any noticeable effect on data (certainly it was never noticed by observers prior to the John Sullivan's discovery of the problem). We do not beleive it to be a serious concern, but wish to make observers aware of what we have found.

Last Updated 24 January 2005 by cgt