3. DETECTORS AND DEWAR

3.1 Overview of DEIMOS Focal Plane Detectors

3.1.1 Main CCD Mosaic

The CCD mosaic, shown in Figure 1.6, represents an aggregate 8K x 8K 15µ-pixel detector array. No CCD manufacturer can fabricate a single CCD of this size. Therefore it must be built up from a mosaic of smaller devices. Such an array could, in principle, be built from a mosaic of four 4K x 4K CCDs. While this might make the design of the dewar and its CCD adjustments somewhat simpler, it represents a significant risk in our efforts to obtain the CCDs and would drive the CCD cost up significantly. In view of the level of success by several manufacturers in producing 2K x 2K CCDs, our judgment is that they can probably make 2K x 4K CCDs with reasonable yield and at reasonable cost. Therefore, this size was selected for the DEIMOS array. Given this size, a number of other requirements result. First the CCDs must be closely buttable on three sides. This means there will be only one serial register per CCD. With an output amplifier at each end of the serial register, we can read out the entire CCD in 4.2p seconds, where p is the pixel time in microseconds. We hope to achieve a pixel rate of 100K/sec, or 42 seconds for the entire CCD. In addition, we expect to read all output channels from all of the CCDs in parallel, so 42 seconds will also be the time to read the entire mosaic.

3.1.2 Flexure Control CCDs

3.2.2 Detailed Layout

The detailed layout will to some extent be determined by the supplier of the CCDs. Suppliers are free to vary the detailed layout within the constraints defined in Table 3.1. Some parameters, like the 15 µ pixel size, are matched to the optical performance of the spectrograph so we do not want to vary them. Other parameters, like the total number of rows or columns, could be varied a little without seriously affecting the science capabilities of the instrument. We want to give manufacturers as much leeway as possible, since small changes could make a large difference in cost.

3.2.3 Specific Performance Numbers and Justifications

1) Readout noise

Our goal, as stated in Table 3.1 is <5e- readout noise. Note that we have not specified a readout speed. We will read out as fast as possible, within the capabilities of the controller hardware, while keeping within the 5e- limit and maintaining other important readout parameters (such as CTE). As stated in Section 3.1.2, we hope to achieve a readout rate of 100 Kpixels/s. The 5e- readout noise limit is set by the fact that the sky noise in the darkest parts of the spectrum will be 10e- per pixel in a 2000 sec exposure at highest dispersion (see Figure 1.1).

2) RQE and AR coatings

Since the mirrors of Keck II will be silvered, DEIMOS is optimized for the red spectral region. Our RQE specifications are based on demonstrated capabilities for thin devices operating in the 0.5-0.9 µ range using multi-layer antireflection coatings. Even without multilayer coatings we should be able to achieve >80% RQE somewhere in the range 0.5-0.7 µ, and >40% at 0.9 µ. Section 3.4.3 describes investigative work with deep-depletion CCDs which may result in even better red response.

3) Backside passivation

The precise method of backside passivation will depend on the CCD supplier. For instance, Lincoln Labs and SITe utilize a boron implant with anneal. Foundry-produced CCDs on the other hand (such as those from Orbit) may have no built-in passivation and may require UV flooding. Because UV flooding is not permanent, we would like to avoid it if possible. Section 3.4.3 describes work we are doing to address backside passivation for foundry-produced CCDs. On the timescale that we may expect to obtain science-grade CCDs, delta-doping of our Orbit CCDs at JPL may also be possible.

4) Fringing

The calibration and removal of the effects of fringing in thinned CCDs is one of the most difficult data reduction problems today. The problems are made worse if instrument flexure makes the fringe pattern unstable, if the amplitude of the fringes is large, or if the physical peak-to-peak distance between fringes is small. Our specifications define a maximum fringe amplitude since this is an aspect we can hope to control. The peak-to-peak distance between fringes is determined by the detailed characteristics of the thin CCD membrane and is much harder to control. Given sufficient funding and time, we may be able to selectively pick science grade CCDs with particularly large peak-to-peak fringe distances, but it is not obvious at this time that such a selection will be possible.

We will be addressing the fringing problem in three ways. First, a good red-optimized AR coating can help to reduce the amplitude of the interference fringes. This will be done. Second, we are investigating an active Flexure Control system which would help to dynamically remove instrumental flexure and improve fringe pattern stability. Finally, we are doing experimental work with Orbit (see Section 3.4.3) on deep-depletion CCDs which could help to reduce fringe amplitude.

5) Flatness

The flatness requirement is set by the speed and resolution of the cameras and the size of the CCD pixels. A variation of 5 µ causes an image defocus of 2.5 µ, small compared to the nominal image FWHM of 17 µ. Normal silicon wafers are flat to better than 1 µ, and, if the CCD is attached to a solid substrate before thinning, this level of flatness is easy to maintain. We have measured the flatness of Reticon-thinned CCDs and Lesser-thinned CCDs, and both meet our flatness requirements. We know how these CCDs are mounted and thinned, so we are confident that we can achieve similar results with Lick-thinned CCDs. Lincoln Lab personnel have also demonstrated the capability to achieve our flatness requirement.

6) Full Well

The full well specification is a typical charge handling capacity for 15 µ pixels. Given the high amplifier gain typically used during readout (~ 1 e-/DN) and the dynamic range of the readout ADU (16 bits), less than half of the full well is typically used. The additional charge handling capacity helps to insure that non-linear response near pixel full-well does not effect the data.

7) CTE

The CTE specification is based on the desire to minimize smearing of the image during readout. A parallel CTE of 0.999995 would result in about 2% smearing of the charge in the last row of the array, and the same CTE in the serial register would result in about 0.5% smearing of the charge in the last column (using both amplifiers).

8) Dark Current

The dark current specification (< 1 e- in 1000 s) is typical of current CCDs and is low enough that the contribution to the noise from the dark current will still be small, even for a 3-hour exposure.

9) Linearity

The linearity specification is typical of current CCDs and should not be a problem to achieve.

10) Other

The DEIMOS CCDs do not need frame-transfer capability. Whether or not the CCDs selected will have frame transfer capability will depend on the supplier selected. Orbit 2Kx 4K CCDs do not have frame transfer. Lincoln Labs CCDs will. Skipper amplifiers are not planned for the DEIMOS CCDs. In fact, with the improving performance of non-skipper amplifiers the reasons for skipper amplifiers are quickly disappearing. The CCDs will be capable of on-chip binning in both rows and columns. MPP implants may not be used to suppress dark if the devices come from Lincoln. In this case dynamic suppression of the dark will be used.

3.4 Acquisition Plan for Individual CCDs

3.4.1 Overall Plan: SSC and KCCDWG Activities

To address the task of acquiring CCDs for Keck Observatory instrumentation (HIRES, LRIS, and DEIMOS) the Keck Science Steering Committee (SSC) established the Keck CCD Working Group (KCCDWG). The members of the KCCDWG have been:

Chris Stubbs has recently relocated to the University of Washington and has resigned from the KCCDWG. However, all decisions and findings so far by the KCCDWG related to the DEIMOS CCDs have included Chris.

It has been the opinion of the members of the KCCDWG that Keck must take an active role in the development of the CCDs needed by Keck Observatory. This opinion has been backed by the SSC by statements and funding. With regard to the DEIMOS project, the KCCDWG members have undertaken several tasks. First, a set of detailed specifications was developed for the CCDs required for DEIMOS. Then a letter was sent to the following CCD manufacturers asking for budget estimates, development timescales, and fabrication concerns:

Based on the responses received and the current experiences and knowledge of KCCDWG members it was the consensus of the KCCDWG and the SSC that we should pursue further work at Lincoln, Orbit, and Loral. EEV was judged to be too expensive (>$200K per CCD). Although Reticon pricing was attractive, they are more than 2 years behind in their program to develop a thinned 2Kx2K CCD, so they were judged to be too risky. The largest device Sarnoff could make had only 76% of the area of the 2Kx4K CCD we require, and their development costs were high (>$800K). SITe indicated they could not meet our flatness requirements.

At the time the KCCDWG was formed (early 1993), Lick had already been talking to Orbit Semiconductor about a collaborative effort to develop scientific CCDs. With funding from Keck, UCSC, and the University of Hawaii we began our first wafer run at Orbit. This activity was proceeding at the same time the specifications for the DEIMOS CCDs were being developed, so we included a 2Kx4K 3-side buttable CCD in the first wafer run. This first wafer run proved that Orbit could make high quality CCDs.

Following the review of CCD manufacturers by the KCCDWG, the SSC authorized funding for further work at Orbit and Loral, and for additional investigation into Lincoln Labs. The funds for Loral covered a single wafer run at their Newport Beach facility. But since this facility is closing down, it was decided to reserve these funds until the situation at Loral stabilizes.

3.4.2 Lincoln Labs Work

1) Consortium

To investigate Lincoln Labs further, members of the KCCDWG made a site visit to Lincoln. The report on that site visit is included as Appendix D. As a result of the visit it was concluded that further work at Lincoln should be funded. However, the initial phase of development (usually referred to as Non-Recurring Engineering, or NRE) was expected to cost between $0.5M and $0.6M. To lessen the cost to Keck it was decided to try to develop a consortium to fund the NRE phase of the work. The consortium is headed by Gerry Luppino of the University of Hawaii. Consortium members include:

Consortium members will fund equal shares of about $100K. These funds do not come from the DEIMOS budget.

2) NRE phase

The NRE phase of the work at Lincoln will include the following:

It is expected that each consortium member will receive at least one working, thinned CCD for evaluation. The complete NRE phase is expected to take about one year, and is expected to begin in the first quarter of 1995.

The 3-side buttable package (header) for the consortium CCD will not be supplied by Lincoln. This package will be designed by Lick engineers in collaboration with Lincoln personnel and fabricated in the Lick shops.

3) Production phase

If the NRE phase of the work proves successful, then the SSC will have the option to proceed with a production phase at Lincoln. The cost per CCD at this stage may depend on the total number of devices produced by Lincoln, and this will depend on decisions made by the other consortium members. However, the cost per finished, science-grade CCD is expected to be between $50K and $75K. We will need an absolute minimum of 16 devices (for two beams), and 24 would be better. So the total budget for production phase CCDs is between (16 x $50k), or $0.8M and (24 x $75K), or $1.8M. Note that our current budget includes $0.43M for a single beam, so we are just within the minimum needed to secure a full array assuming the lowest cost per device.

3.4.3 Orbit Work

1) NRE phase

The funding provided by the SSC for the NRE phase at Orbit covers fabrication of a new CCD mask set and three additional wafer runs. Lick was able to negotiate a very favorable price for these wafer runs, based on our continued cooperative work with Orbit. We are getting the three runs for a total cost of only $60,000. Three additional runs at the same price are also possible. A new mask set is required because on the original mask set we had one 2Kx4K CCD and two 2Kx2K CCDs. The new mask set will have two 2Kx4K CCDs along with the FC CCDs. The objective of the wafer runs is to test whether or not Orbit can produce good 2Kx4K CCDs with sufficiently high yield to make them economically feasible. Low yield was the one area in which our initial wafer runs were deficient. We hope to achieve at least a 25% yield of working 2Kx4K CCDs, with a science-grade yield of 10%.

To help defray some of the cost of the NRE effort at Orbit we have established Memoranda of Understandings with NRAO and the Observatory of Rome, Italy, in which they will share in the results of these first runs. A similar agreement with a group at Yale is pending. In return for a share of the devices, each group has paid $22K. This doesn't cover our costs for the CCD design, mask fabrication, or testing and packaging, but it does provide us with the funds to continue with additional wafer runs (see below).

The first of the three new wafer runs is now under way. We expect to have devices from this run by mid-January. If we find no problems that indicate a design error or mask fabrication flaw we will start the next two runs as soon as our testing is completed. We should have all three runs completed by June 1995. At the end of the NRE phase Keck should have around six good science-grade CCDs (assuming 10% science grade yield), and a larger number of lower-grade devices. The science-grade devices will have to be thinned (see below).

In addition to Keck SSC funded activities at Orbit, Lick is engaged in a number of additional activities which may have a significant impact on the DEIMOS CCDs. First, we are doing some experimental work with a new kind of silicon wafer which has a double epitaxial layer. The first layer has a very high boron concentration and the second layer has the normal epitaxial properties upon which the CCD structures are fabricated. When thinned, the boron epi-layer acts as a natural etch-stop and, even more importantly, may provide backside passivation properties similar to the boron-implant with laser anneal technique. If this works, it will greatly simplify the fabrication of thinned, high QE CCDs. At least one of our remaining $20K wafer runs, and funds from our MOUs, will be used to test these new wafers. The boron wafers have already been purchased with funds from a Lick NSF grant.

In a separate effort at Orbit, Lick is working with a group at the Lawrence Berkeley Lab to test Orbit's capabilities in fabricating deep-depletion CCDs. These are CCDs made on high-resistivity silicon. The high resistivity allows the potential wells of the pixels to extend deeper into the silicon, thus increasing the effective thickness of the CCD. This means that the CCD would not need to be thinned as much. Thinning would be easier, and the resulting thicker CCD would be more robust. A thicker CCD also means a longer path length for absorption of red photons, thus increasing red RQE while reducing fringing amplitude. We will use another of our $20K wafer runs to test the deep depletion CCDs.

Finally, we have been exploring the possibility of testing Orbit's capabilities in making P-channel CCDs. Most electronic devices are made with N-channel material because the signal carriers are electrons. In P-channel material the signal carriers are holes, which tend to move much more slowly. For this reason most CCDs are made with N-channel material. However, Orbit has some other commercial interests in developing P-channel technology, and if we pay for materials costs (which might include a new mask set at around $22K) they may pay for the wafer run. We are still discussing the details with Orbit. The advantage of P-channel CCDs is that no backside passivation is required since the back surface will naturally repel the signal carriers (holes). Such CCDs should naturally have good blue response when thinned. The lower mobility of holes should not be a problem at CCD operating speeds.

If both deep-depletion and P-channel CCDs work we may be able to combine them to make a CCD with high RQE across the entire spectrum. This would be a very major advance in the state-of-the-art in CCD fabrication. By the end of 1995 all of the Orbit NRE activities should be complete, or near enough to completion, to permit us to judge whether or not Orbit will be a viable choice for further CCD fabrication.

2) Production phase

If the yield of good 2Kx4K CCDs at Orbit is high enough then we can elect to proceed with a production phase. Assuming a 50% loss in thinning, we will probably want about 24 science-grade CCDs (for a single mosaic plus spares). If we get six in the NRE phase then we will need 18 more science-grade CCDs, or 5 runs. At the 100-wafer production level (5 runs) each wafer will cost about $1600 to process. If the double epi-layer wafers are used, that will add an additional $3300 per run. Therefore the budget for production at Orbit would be about $176.5K for a single beam. All of these funds do not have to be committed immediately, and if the thinning losses prove to be less than 50% then the number of runs could be reduced. However, the costs shown here are based on an initial commitment of at least 100 wafers. If we order wafers in smaller lots then the cost per wafer goes up. The production phase would last about six months, with a new run coming off the fab line on average about once every six weeks. Devices would be tested and mounted as they came in. If two mosaics are produced, the cost per wafer may be reduced further so the total cost would not double.

3) Thinning and Packaging

Any foundry-produced CCD (whether at Orbit or Loral) will have to be thinned and mounted in a separate effort. There seem to be three options:

The first option is to contract with Mike Lesser to thin our CCDs. The KCCDWG felt that this approach had several shortcomings. One serious shortcoming is the exceedingly low success level Mike has had for many years. While he now claims a much improved success rate this has yet to be convincingly demonstrated. Equally serious is the high cost of thinning. At $15K per CCD the budget for thinning 12 CCDs (one mosaic plus spares) would be $180K. The three-side buttable package is also a significant departure from his previous thinning efforts, and would require some development effort and additional expense (~$100K). The KCCDWG was also concerned about Mike's ability to deliver the CCDs on time. As the world's only supplier of thinning services for Loral (and now Orbit) CCDs, Mike must have a tremendous backlog of orders, and if he thins CCDs for Gemini, the situation will only get worse. Despite all these concerns, Mike's operation may still prove to be a viable option, and we are not ruling him out at this time.

It has recently come to light that Reticon may be willing to provide thinning services. However, given their poor performance in attempting to produce a thinned 2Kx2K CCD, this does not seem to be a good option at this time. Nevertheless, such situations are highly personnel dependent, and if the situation at Reticon changes, they may become a viable option.

The third option is in-house thinning. For some time there has been interest at Lick in setting up a thinning facility. It has been our opinion that all of the necessary talent is available close to the Santa Cruz campus in Silicon Valley. Furthermore, a detailed estimate of the costs suggests that the actual cost of thinning and packaging a CCD could be as low as $5000 if yields are kept high. Given the interest at Lick and the availability of facilities and skilled personnel, the SSC decided to fund some preliminary thinning work at Lick Observatory. If we are successful at thinning and we can hold costs down, we will have spent no more money than if we had contracted with Mike Lesser, and at the end of the effort Lick and Keck would have a very valuable resource in the thinning lab. The SSC has provided $0.1M for a one-year effort, and $20K for lab remodeling at Lick. To hold costs down we are acquiring as much equipment as possible through corporate dona- tions. We have been very successful with this type of effort so far. Also, to hold costs down and to achieve rapid success we plan to do as little development work as possible, opting instead to purchase expertise and technology that already exists. For instance, we have hired Dr. Charles Hunt from UC Davis as a consultant to help us design our thinning facility and procedures. Dr. Hunt was instrumental in setting up the Tektronix (now SITe) CCD thinning facility. We are also working with packaging specialists to design and fabricate the three-side buttable package (header).

We hope to be able to demonstrate successful thinning and packaging of 2Kx4K CCDs within one year. If we are successful, then in-house thinning of Orbit (or Loral) CCDs becomes a realistic possibility. If we are not successful, then we can still elect to use Mike Lesser (or Reticon) without any impact on the time schedule for DEIMOS. None of these development activities are changed to the DEIMOS budget.

3.4.4 Other Possibilities

Although Keck is investing in Lincoln, Orbit, and possibly Loral, this does not preclude acquisition of the needed CCDs from other sources if they become available. For instance, SITe is said to be working on a 2Kx4K 15m 3-side buttable CCD. SITe CCDs are known to have some serious problems. It is our understanding that they are planning to reduce the 200m p-p curvature seen in their 2Kx2K to around 100m in the 2Kx4K device, but this is still 10 times worse than our specification and is a fatal flaw in any mosaic used behind a fast camera. If SITe can produce a 2Kx4K CCD with acceptable flatness and do it on a reasonable timescale (SITe's record here is not very good but may be improving) then they may become a viable source.

3.4.5 Selection of CCD Path and Costs

All of the NRE efforts at Lincoln and Orbit and the thinning demonstration effort at Lick should be completed by the end of 1995. At that time, the KCCDWG in conjunction with the DEIMOS team will examine the results from those efforts, will re-evaluate other options such as SITe, and will make a recommendation to the Keck SSC. Criteria for selection will include:

The deadline for choice of CCD vendor for DEIMOS is the CCD/Detector CDR in May 1996.

At this time the total cost to produce CCDs for a single mosaic at Lincoln is at least $400K, and the total cost to produce them at Orbit with Lick thinning is projected to be about $300K. These costs do not include the development work at Lincoln, Orbit, or Lick, which is being supported by Keck SSC and other funds. The current budget allocates $433K for CCD acquisition (plus additional funds for making the mosaic). This seems adequate for Orbit-produced CCDs but will cover the Lincoln costs only if we can obtain the lowest projected cost for Lincoln devices. The Keck SSC has indicated in the past that it may be willing to fund further CCD fabrication outside any particular instrument budget. However, further decisions await NRE developments.

3.4.6 Fallback CCD Plans

1) Options

The ideal fallback option would have no impact on the project schedule or budget, and no impact on the science capabilities of the instrument. Such a fallback option does not exist. One obvious fallback is to employ unthinned 2Kx4K CCDs in the mosaic. These should be relatively easy to acquire. They could be produced by either Orbit or Loral. A phosphor coating could be applied to the CCDs to give them some blue response. This fallback plan could be implemented at almost any time. The timescale to produce 16 thick science grade CCDs would be 6 months or less, assuming four wafer runs were started in rapid succession. This option would impact the science capabilities of the instrument because of the reduced RQE in the thick CCDs. However, the complete elimination of fringing provides a slight offsetting advantage. On balance we are not enthusiastic about this fall back position because it would still be costly yet leave us far from our final goals in terms of QE. Also, it does not provide peak performance for single-object observers who require only a 1x2 mosaic.

If we acquire some of the thinned CCDs we need, but not all, two other fallback positions may be possible. First, a single spectrograph beam could be used if we have enough CCDs for only one mosaic. Second we could install fewer CCDs in each mosaic.

In the event that we do not have a full complement of CCDs at first light, it is our highest priority to provide high RQE and full spectrum coverage for single-object users. That means a 1x2 mosaic of thinned CCDs for one beam. This seems well within the scope of the existing budget of $433K for a single-beam spectrograph, given the cost estimates above. As more devices become available at feasible cost, we will extend the mosaic upward along the slit. Since total funds are scarce, we are inclined to bank any unspent dollars rather than build an interim but still costly mosaic of thick CCDs.

2) Decision Time and Criteria