CFH12K: optimizing 12 MIT/LL CCID20 CCDs for a direct imaging application

Jean-Charles Cuillandre (1), Gerard Luppino (2), Barry Starr (1), Sidik Isani (1)
1: Canada-France-Hawaii Telescope Corporation, 65-1238 Mamalahoa Hwy, Kamuel a, HI 96743, USA
2: Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822, USA

In ``Optical Detectors for Astronomy''
Proceedings of the 4th ESO CCD Workshop held in Garching, Germany
September 13-16, 1999
P. Amico and J.W. Beletic Editors
ASSL Series - Kluwer Academic Publishers


Abstract. CFH12K is a 12k by 8k wide field imaging camera for the CFHT prime focus. The mosaic consists of twelve MIT Lincoln laboratories 2k by 4k thinned backside illuminated CCID20 devices. The devices' operating parameters have been optimized to ensure the best data quality for use in broad and narrow band filter imaging mode. Adaptation to the CFHT prime focus environment included modifications to reduce the scattered light seen by the CFH12K. Science data taken by the camera has proven the success of CFHT's new capability for 42 by 28 square arcminute imaging with high resolution subarcsecond seeing.

1. Introduction
Over the years, CCD cameras used at the CFHT prime focus have evolved from Loral 2k x 2k introduced in 1991, to the 4k x 4k mosaic MOCAM in 1994, to the 8k x 8k mosaic UH8K in 1995. Compared to these previous instruments, this latest 12k x 8k thinned CCD mosaic not only increases the sensitivity by a factor of two from 500 nm to 900 nm but it also re-opens access to the B band around 400 nm. This article focuses on describing how the CCDs, the mosaic and the prime focus optics were optimized and arranged to achieve the highest observing efficiency on the sky. This includes getting the highest sensitivity from the combination telescope plus instrument, collecting the data as fast as possible and ensuring the best quality of these data. The first section of this article describes the organization, optimization and characterization of the MIT/LL CCID20 CCDs, which were all tested at once within the CFH12K cryostat. The second section describes the efforts made to improve the instrument environment in the prime focus. Finally a quick outline of the scientific programs covered by CFH12K is given, followed by illustrations of the tremendous scientific capabilities of subarcsecond high resolution wide-field imaging with CFH12K at CFHT atop Mauna Kea.
2. Device selection
The CFH12K mosaic is composed of twelve MIT Lincoln Laboratories CCID20 devices. These 2k by 4k thinned backside illuminated CCDs (Burke et al. 1998), are the result of funding from a consortium of several ground based astronomical facilities (ESO, UH, Keck, CFHT, Subaru, AAO) led by G. Luppino from University of Hawaii (UH). Lick Observatory and UH were responsible for characterizing all the consortium CCDs individually to assert their quality and rate them (Wei & Stover 1998). All members were interested in building mosaics for imaging or spectroscopy and selected in a defined order the CCDs that best suited their application.

This characterization was meant only to qualify the CCDs, not optimize them for a given application. However, the initial performance derived from these simple tests were highly representative of what could ultimately be achieved. The characteristics produced were:
   - functional outputs (CCID20 has two outputs)
   - cosmetic quality at -120 deg. C.
   - charge transfer efficiency (CTE, serial and parallel)
   - quantum efficiency and variations in amplitude across the chip
   - fringing (using a monochromatic source)
   - gain and readout noise (using a SDSU Generation I controller)
   - full well.

As described is the next sections of this article, a direct imaging application sets lower constraints on the cosmetic quality and detector noise. Hence, during the three picks that took place from 1998 to 1999, CFHT and UH focused on getting the CCDs with the best quantum efficiency and the best CTE. The first two drafts permitted going on the sky in January 1999 with a 10k by 8k mosaic, and the third draft brought four CCDs to fill the mosaic and replace the two engineering grade devices used in the initial mosaic. The knowledge built from operations on the sky during the first semester of 1999 provided new information on the initial CCDs to reorganize them within the focal plane based on the criteria described hereafter. This focal plane work was conducted in August 1999 in the newly acquired clean room at the CFHT headquarters in Waimea. CFH12K now has twelve CCDs that qualify for an imaging application (Figure 1).

Figure 1: Left: CFH12K focal plane - twelve 2k x 4k CCID20 MIT/LL CCDs forming
a 12,289 by 8,192 pixel mosaic. Right: A single pre-processed image of an M 31 field
showing the cosmetic quality of the whole mosaic.

3. Mosaic organization
Achieving the highest data quality at the center of the field to benefit from the best optical quality (no optical aberration versus the edges of the 1 square degree field, CFH12K covering 42 by 28 square arcminutes) naturally places the CCDs with the best combination of QE, CTE and cosmetic in the central 8k by 8k mosaic. CFH12K has a central 4k by 8k mosaic nearly perfect in those terms.

There are two types of CCID20 devices: those made out of standard epitaxial silicon (EPI) and those made out of high resistivity bulk silicon (HiRho). The latter have a higher QE (up to 10%) in the red part of the spectrum (Figure 4) and produce less fringing than EPI parts due to their larger thickness. The CFH12K contains both types of devices. They are grouped within the mosaic (all the HiRho parts packed together in the lower right corner) in order to cover large areas on the sky with a similar response (Figure 2).

Figure 2: Left: Organization of the mosaic per CCD type. Right: Organization
of the mosaic per banks (1 bank per controller, one output type per bank).

Each of the two banks of 6 CCDs (12k by 4k) is handled by an SDSU Generation II CCD controller (Leach et al. 1998). This controller, optimized to handle mosaics of identical detectors, constrains the organization of the CCD readout amplifiers (each bank must have the same output per CCD). A few of the CCDs available for CFH12K only have one of the two outputs functional, hence one bank had to be read only from the left outputs and the second one from the right outputs (Figure 2).
4. Mosaic geometry
With a short beam of f/4.2, depth of field is critical on such a large surface (21cm by 14cm). Emphasis was given on designing a focal plane structure that would achieve a flatness of 60 microns, the depth of field at the CFHT prime focus for a 0.4 arcsecond seeing (pixel size is 15 microns, providing a sampling of 0.2 arcsecond per pixel). The CCDs individually are flat within 20 microns. They are mounted on a package standing on three shims which can be machined to within 2 microns after proper measurement of the CCD's height at its four corners. With the twelve CCDs mounted on the molly plate, a flatness of better than 100 microns is measured. This is not enough to degrade images taken under even the best seeing conditions available at CFHT: during first light on the sky in January 1999, image quality was checked on 0.5 arcsecond seeing data and no image degradation larger than 0.04 arcsecond could be measured across the field.

The wide-field corrector induces a radial image distortion, forcing resampling of the data to achieve proper astrometric properties on the observed field when needed. As a consequence and due to an increasing complexity, the relative alignment of the CCDs along the X and Y axis was a low priority. Nonetheless, proper manual mounting of the CCDs led to an amazingly good relative alignment of the devices: the gaps between the CCDs (both in X and Y) range from 30 to 38 pixels (about 500 microns) and the relative angles between devices range from -0.3 degree to +0.3 degree.
5. Device optimization
The CFH12K mosaic was optimized and characterized with all the CCDs integrated in the cryostat. While this speeds up the optimization process by having all the data from the twelve CCDs at once, it implied a software effort to develop all the automatic procedures to reduce the large amount of data produced. Those tools allowed completion of the optimization and calibration phase in the CFHT headquarters CCD lab in less than two weeks. Subsequent knowledge of the detectors and the mosaic was gained through tests and general use by observers on the sky.

5.1. Readout speed
Achieving the highest observing efficiency was a major highlight in the definition of the CFH12K. The SDSU Generation II CCD controller was chosen since it was a logical continuation of CFHT's involvement with this type of controller over the past years and most importantly would result in the camera performance to being limited by the detector, not the controller.

Running the two controllers unsynchronized in parallel during the readout results in high pickup noise. During development of the camera, DSP code was not available from SDSU to facilitate synchronizing the two controllers. Since having this capability would reduce the total readout time by a factor of two, CFHT implemented this code in-house. A DSP routine allowing a synchronization within 6 ns was developed. With the serial CTE being the limiting factor on three CCDs, a readout time of 58 seconds for the whole mosaic is achieved (6.7 microsecond per pixel, or 150 kilopixels per second per CCD or 1.8 Mbytes per second per controller). The limit of the data acquisition system itself (custom interface cards, see Starr et al. in this proceeding) is reached at a 40 second readout.

5.2. CTE
CTE must be high enough to avoid degradation of the image quality (point spread function, PSF) in 0.6 arcsecond seeing conditions along a 2,048 pixels transfer. With a 0.2 arcsecond per pixel sampling, a serial CTE of at least 0.99994 has to be achieved such that the PSF degradation is not larger than 10% after a whole serial register transfer. To get the highest CTE, CCDs are run at -89 deg. C and the serial register timing is slowed down. Three CCDs in CFH12K caused this limitation in readout speed to the whole mosaic (serial CTE is higher than 0.99999 for all the other CCDs). Parallel CTE is at least 0.99999 for all devices.

5.3. QE response
The rather high running temperature of -89 deg. C is not only motivated by the need for the best CTE. It also causes the QE to increase in the red part of the spectrum by up to 10% compared to -110 deg. C. The high temperature is also motivated by the fact that it minimizes the peak-to-peak amplitude of the brick wall pattern (BWP, Wei & Stover 1998), a feature of the early CCID20 devices resulting from the laser annealing process (Figure 3). Typical amplitude of the BWP on the worst CCDs is 10%. See also the "brick wall pattern" paragraph in this article.
QE stability across the chips and in time is crucial to achieve perfect photometric stability. The brick wall pattern being highly dependent on temperature, a fine thermal regulation of +/- 1 degree is required (see accompanying paper by Starr et al.).

5.4. Dark current
In direct imaging application, the sky background noise quickly dominates the signal noise even with the use of narrow band filters. Hence, the dark current was not considered as an important constraint but even the high running temperature of -89 deg. C actually produces only around 1 electron per minute per pixel. This is a very low value that can be neglected in broad band imaging mode and corrected by a simple constant in narrow band imaging mode.

5.5. Linearity
To achieve the photometric accuracy essential to any astrophysical program, a linearity of within 1% over the whole range of the 16 bits analog to digital converter is required. An automatic acquisition and analysis program was developed to map the output drain and reset drain voltages over a 3 volts range with steps of 0.2 volts. The optimal working point was simultaneously computed for all the CCDs by looking for the point where both readout noise and linearity residual tend to be minimal while the gain reaches a plateau. No influence on the linearity was found when changing the reset gate and the output gate voltages in a reasonably low range around the typical values advocated by MIT/LL.

5.6. Gain
The natural gain of the CCID20 devices is very high and to obtain a readout noise sampled on at least two to three analog to digital units (ADU) to allow proper signal analysis, the lowest available gain on the CCD controller had to be selected. The average gain is 1.6 electron per ADU (1.4 to 2.1 over the whole mosaic).

5.7. Detector readout noise
A minor constraint due to the rapidly dominating sky photon noise, the readout noise was derived from the best optimal working point defined during the linearity optimization. Readout noise is very low for such a high readout speed: typically 5 electrons with a correlated double sampling time up and down of 1 microsecond each.

5.8. Other noises
As emphasized by Starr et al. in this proceeding, high attention was given to the general organization of the whole system. As a result of proper grounding within the highly noisy telescope environment, no pickup noise is affecting the camera performance.
Figure 3: Left: A B-band flat-field exhibiting the brick wall pattern. Right: An I-band fringe
frame. HiRho parts are on the lower right corner while the latest EPI parts with a new coating
can be seen on the top bank with no fringing.

5.9. Full well
Full well must be optimized (parallel voltages) to match at least the 16 bits ADC dynamic based on the gain setting. Typical full well of the CCID20 is 150,000 electrons while the gain setting for CFH12K leads to a digital saturation at 100,000 electrons using manufacturer's standards voltages.

Saturated bright stars in such wide field are actually the main cause of pixels lost for science, well before bad cosmetic. Achieving higher full well capacity by increasing the swing between the low and high levels of the parallel clocks is the only solution but comes with a drawback: spurious noise injection that forces a bias frame correction for each image. However, CFH12K runs with the MIT/LL standard parallel voltages since most of the area affected by blooming lies within the obliterated region of the reflection halo caused by the bright object (low level reflection bouncing back from the CCD to the front window then back to the CCD, a one to two percent effect).

5.10. Dithering anti-blooming
As experimented during the early phase of CCD wide field imaging at CFHT (Cuillandre et al. 1996), the dithering anti-blooming mode developed by J. Janesick (1992) reduces the blooming contamination caused by saturated bright objects. It eliminates exceeding charges using the non-perfect nature of silicon (presence of traps at the Si/SiO2 interface) but with the progress made in the past decade, silicon has improved in quality and today it becomes difficult to use this feature without being heavily affected by the drawback of this technique: spurious noise. Also, some CCDs exhibit a large number of pocket pumping sites (Tonry et al. 1997) that can't be corrected by standard techniques during the data reduction (they are basically dead pixels). Dithering anti-blooming was then abandoned on CFH12K since the high full well keeps blooming to a reasonable level.

5.11. Residual image
The CCID20 devices are known to suffer from residual image (Wei & Stover 1998). Inter-image contamination has to be as low as possible since it affects the photometry of any object on top of a previously over exposed area of the detector. The residual image appears only on some CCDs to a various degree and is always associated with a pixel that reached over the full well storage capacity (it is not only located to the bright object site but also the blooming aisles). The residual image is about 20 electrons per pixel on a 5 minutes integration after a saturation state.

The SDSU Generation II controller does not easily allow three clocking states, hence setting the CCDs in full inversion mode is not yet implemented. A full inversion mode would promptly re-populate all the sites at the interface and eliminate all the trapped electrons. In the meantime, the anti-blooming dithering mode proved to be a way of reducing this amount of charges by clocking at very high frequency the parallel clocks during the very short lapse of time available between two exposures (1 second usually).

5.12. Crosstalk
A source of intra-image contamination, crosstalk is a natural concern for multi-readout systems. Crosstalk could not be measured between two independent video boards (see Starr et al. for a description of the CCD controllers), and a negligible value of 8.10E-5 between two channels of a video board was measured.

6. Mosaic characterization

6.1. Cosmetic
While the best CCDs in terms of QE, CTE and cosmetic are at the center of the mosaic, the entire CFH12K exhibits an excellent cosmetic quality. There are approximately 200 bad columns (bad column is defined here as an entire or a fraction of a column), which are actually concentrated in the lower right corner CCD (Figure 1). The number of bad pixel clusters is very low and overall the ratio of bad pixels over the whole mosaic is only 0.4%.

6.2. Brick wall pattern
While reorganizing the CCDs in August 1999, the early phases CCDs with bad brick wall pattern have been pushed out in the corners (Figure 3). These devices exhibit a peak to valley amplitude up to 5 to 10% (worse at shorter wavelengths, i.e. the B band). The most recent CCID20 devices show negligible signature of this effect (1%, seen at the center of the mosaic).

The QE brick wall pattern is a multiplicative effect corrected during the flat-fielding operation. Building a superflat from night sky data is the best solution but is not always possible due to a low number of frames on "empty" fields. Twilight flat-fields flatten very well beyond 600 nm but are a bit marginal below (B and V bands) because the twilight sky spectrum differs quite significantly at those wavelengths from the dark sky spectrum. This is however the only way to collect high signal to noise calibrations, hence special automatic tools have been developed to optimize the number of frames acquired in the short amount of time the sky brightness is within the right range. This is not totally satisfactory though and CFHT is investigating a dome flat-field screen with a set of lamps that would reproduce as closely as possible the shape of the night sky spectrum.

With the dependence of brick wall pattern amplitude versus wavelength, there was a concern on the color effect on astronomical sources. Tests on the sky show that no effect is introduced, proving that the flat-fielding operation corrects for the variations. The only impact is the detection limit that slightly differs from peak to valleys but this can be averaged using a dithering pointing mode during a set of exposures that will move the field around on the focal plane at the BWP geometric scale (60 pixels).

6.3. Fringing
The HiRho parts exhibit a very low fringing of 0.5% in the I band due to the higher thickness of the CCD: 30 microns vs. 15 microns for the EPI parts which suffer from a fringing of 5% at worst. The latest development in coatings (J. Tonry, private communication) actually brings fringing to a negligible level on the EPI parts. Figure 3 shows a CFH12K fringe frame: the three bottom right CCDs are HiRho parts, the top right CCD is a standard EPI part while the CCD to its left (from the latest draft) has the new coating.
The fringing correction relies on obtaining twilight flat-fields to access the fringe frame from the flat-fielded dark sky images. Then, the fringe frame has to be scaled on a per exposure basis and subtracted. The fringing in the Z' band is worse by a factor of two relative to the I band.

Figure 4: Left: CFH12K faces down from the CFHT prime focus onto the primary mirror. Not
shown on this diagram are the power supplies attached to the wall of the prime focus cage.
Rotation of the instrument is not allowed. Right: QE response of the two types of CCDs (EPI
and HiRho) and transmission curves of the filters available for CFH12K.

7. Collecting more photons for CFH12K

7.1. Prime focus
The twenty year old prime focus cage needed some modifications to accommodate this extremely sensitive camera. Experience with previous CFHT wide-field imagers (MOCAM and UH8K, see Cuillandre et al. 1996) showed that scattered light from various sources is a critical limitation for deep imaging (flat-fielding was usually poor). Indeed, the prime focus and the wide-field corrector (WFC, see figure 4) were designed for use with photographic plates and none of the surfaces had been blackened properly, neither were baffling rings installed inside the corrector. The whole unit has been revisited (adding highly low reflectance black velvet everywhere possible and baffling rings) and large scale flat-fielding of CFH12K data is now very efficient (within a percent).

To improve the total efficiency of the instrument, new anti-reflection coatings were installed on the wide field corrector lenses, eventually bringing the total efficiency of the primary mirror plus the three lenses from the WFC to 73%.

Bright stars are an inevitable source of contamination for such wide field instrument. Proper baffling significantly reduces the contamination from most of them but very bright stars on the field (less than the 6th magnitude) will obliterate a large fraction of the data. A special warning tools (Telescope Field Monitoring) has been developed to prepare the observations and also to be used during the instrument operation.

7.2. Filters and zero points
The CFH12K interference filters were produced by Barr Associates (see accompanying paper by Starr et al.) and all have a transmission higher than 85%. The eight filters currently available are shown on figure 4. Note that the Z' filter is open, the high frequency cut down being defined by the drop of the CCD QE. The HiRho parts definitely are a benefit in this band.
With a transmission of 99% for the cryostat front window, the overall efficiency of CFH12K at prime focus at 650 nm is 57% when one includes all the telescope optics plus CCD. This is the highest efficiency ever achieved by an instrument at CFHT. The zero points at zero airmass in electrons per second for the four main photometric bands are: B=26.2, V=26.6, R=26.6, I=26.2.

8. CFH12K on the sky

8.1. Observing efficiency
Not only is the camera efficient in terms of sensitivity and readout time, but also a set of efficient tools have been developed to optimize the use of the observing time. CFH12K can run with scripts that minimize the human input for standard operations (dithering modes, focus sequences, twilight flat-fields sequences,...). With either a graphical interface, a command line interface and a scripting language, any new observer is at ease within the first minutes. With typical exposure times of 5 to 10mn, a global observing efficiency of 90% or more is commonly achieved. This data rate leads to a typical 20 Gbytes of data per night (200 Mbytes per file).
Figure 5: Left: M 31 and NGC 205 in the full field of view of CFH12K. This image
was obtained from a composition of a set of images in the B,V and R bands (resp. 40
mn, 20 mn and 10 mn). Data reduction and composition were conducted using the
FLIPS package developed at CFHT.

8.2. Scientific programs
With a subscription rate of 50% of the whole CFHT observing time, CFH12K is the most popular instrument within the French, Canadian and Hawaiian astrophysical communities. This comes as no surprise when one looks at the global performance of the instrument: covering 42 by 28 arcminutes at 0.2 arcsecond per pixel with a median seeing of 0.7 arcsecond. Image quality as good as 0.45 arcsecond has been often achieved on 10 minute exposures.
Figure 6: This mosaic of images illustrates the tremendous power of CFH12K
at the CFHT atop Mauna Kea. High resolution wide field imaging allows probing
the Universe on many scales. The image quality on these 2h30mn exposures (both
in B and I) is 0.7 arcsecond. The last image (40x40 pixels) of a distant spiral galaxy
beyond the nearby galaxy NGC\,3486 shows the true pixelization from the detector.

The pressure on the instrument is high and emanates from all the observational astronomical communities. Here are a few examples:
   - Solar system: Kuiper belt objects
   - The Galaxy: gravitational tidal effects on stellar clusters
   - Nearby galaxies: dust and stellar populations
   - Cluster of galaxies: luminosity functions of dwarf galaxies in Virgo and Coma clusters.
   - Cosmology: deep wide-field imaging survey & wide-field weak lensing survey

8.3. High resolution wide field imaging
The following images on figures 5 and 6 illustrate the power of wide field imaging associated with high resolution capabilities, the major strengths of CFH12K.

9. Conclusion
The CFHT and its location already promote excellent image quality. CFH12K allows the CFHT to realize the high resolution wide field imaging potential of its prime focus. The performance of the CCD, the camera, and data acquisition environment makes CFH12K the most sensitive and efficient instrument in CFHT's history. In sky coverage and efficiency, it surpasses the previous CFHT CCD mosaics and the competing large mosaic arrays around the world. Despite some troublesome CCDs within the mosaic, they all have been successfully optimized for use in this direct imaging mode application. With proper observing techniques and data reduction tools, the detector signature can be removed and large monolithic 12k by 8k scientific frames can be obtained. As a result, CFH12K is heavily subscribed by the CFH astronomical community, occupying about half of the programmed telescope time.
  • Burke, B., Gregory, J., Mountain, R., Kosicki, B., Savoye, D., Daniels, P., Dolat, V., Loomis, A., Young, D., Luppino, G., Tonry, J., 1998, Proceedings of ESO conference on Optical Detectors for Astronomy, 19
  • Cuillandre, J.-C., Mellier, Y., Dupin, J.-P., Tilloles, P., Murowinski, Crampton, D., Wooff, R., Luppino, G., 1996, PASP, 108, 1120
  • Janesick, J., 1992, Applied Optics, 31, 4890
  • Leach, R., Beale, F., Eriksen, J., 1998, SPIE, 3355, 512
  • Luppino, G., Tonry, J., Stubbs, C, 1998, SPIE, 3355, 469
  • Tonry, J., Burke, B., Schechter, P., 1997, PASP, 109, 1154
  • Wei, M., Stover, R, 1998, SPIE, 3355, 598

Jean-Charles Cuillandre
Canada-France-Hawaii Telescope Corporation, 1999