UV-Optical CCDs Mark Clampin, ABSTRACT We review the performance of large format CCDscurrently planned for instruments in space science missions to be flown the coming decade. With focal planes of up to a billion pixels, we review technical challenges in transferring this technology from ground-based to space imagers and identify areas for future development to facilitate these missions. The current capabilities of CCDs for UV imaging are also summarized and future directions for the development of CCDs suitable for wide field UV imaging are highlighted. 1. INTRODUCTION The last decade saw an exponential increase in the size of CCD focal plane arrays for ground-based astronomy as shown in Figure 1 adapted from Luppino et al. 1998. Focal planes at the start of the decade were typically assembled with 2048x2048 CCDs, but the advent of a three-edge buttable 2048x4096 (2kx4k) format in the mid-nineties led to almost universal adoption of this format for the construction of focal plane mosaics. Current plans for the next generation of astronomical instrumentation require focal plane arrays based on 2kx4k CCDs with up to a billion pixels by the middle of this decade. Luppino et al. (1998) first noted the exponential growth rate in pixel density and predicted that it would flatten out as CCD focal plane arrays fully sampled the working field of view of 8-10 meter telescopes. However, plans for 25 meter (Nelson and Mast 1999) to 100 meter aperture (Gilmozzi 2000) optical telescopes, wide field optimized 8m telescopes (Tyson), and new, high-angular resolution cameras Kaiser et al. 2000) have provided new impetus to drive for even larger focal plane arrays.
Plans for space astronomy missions in the new millennium include a number of missions which require large format imaging capabilities in the UV and optical, and are discussed by Morse in these proceedings. In Table 1 I have identified a selection of missions planned by NASA (Morse 2000), ESA and other missions under development by individual groups which are likely to require CCD imaging. Table 1. A selection of NASA, ESA and University missions, or concepts which are likely to require CCD imagers for UV or visible imaging applications.
Progress in CCDs for UV/Optical applications in space science
has not matched ground-based astronomy, despite the early leadership in this
field provided by the Hubble Space Telescope’s Wide Field Planetary Camera
program (Westphal 1982, Blouke
1981). In Figure 2
we show the evolution of space-based UV/Optical CCD focal
plane arrays, primarily for the Hubble Space Telescope Instruments. It shows
that state of the art for flight arrays lags ground-based astronomy by about
three to five years, probably due to the schedule risk associated with flight
systems, which often precludes selection of new technology. While the current
state of the art in UV/Optical CCDs is the 4096x4096 mosaic scheduled to fly
in the Advanced Camera for Surveys (ACS), a number of future missions plan to
fly focal plane arrays matching those planned for ground-based instruments in
the new decade. Since flight focal plane arrays lag ground-based astronomy we
will address the current status of CCDs for UV/Optical flight applications and
highlight issues which will have to be addressed if the next generations of
flight CCD focal plane arrays are to catch up with ground-based astronomy.
2.
CCD PerfoRmance OVERVIEW
It could be argued that the current performance of CCDs
intended for ground-based astronomy are close to state of the art and unlikely
to improve significantly. Quantum efficiency (QE) is close to unity in the
visible, dark current is negligible, charge transfer efficiencies (CTE)
³
0.999999, full well
³
105K e-,
and read noise ranges from 2 e- to 5 e-
RMS depending upon the readout rate. However, for UV/Optical space
instrumentation CCD performance still has considerable room for improvement,
since the operational environment opens up the UV bandpass for imaging and has
significant impact on CCD performance parameters such as dark current, full
well and CTE, to the extent that unique tradeoffs based on scientific merit
are often required during device design.
The three phase CCD architecture is the standard for
UV/Optical space instrumentation and ground-based astronomy, although some
missions have previously flown alternative device architectures such as the
virtual phase imager on Galileo (Janesick et al. 1981). CCDs for space
applications often feature multi-phased pinned (MPP) or, inverted operation,
which provides significant reduction in dark current. CCD dark current
originates from two sources, Si-SiO interface states at the CCD backside
surface and a lesser contribution from the bulk material. MPP operation
passivates the backside surface states and allows the device to operate at
relatively high temperatures compared to ground-based CCDs. For example at
operating temperatures of
-110oC to –130oC
typical dark currents for non-MPP CCDs are in the range 1 - 4 e-
pixel-1hr-1(Jorden et al 1998; Stover et al. 1998),
compared to 5 e-pixel-1hr-1
for a MPP CCD operating at –83oC (Clampin 2000). In space
instruments CCD array’s are generally cooled by Peltier cooler stacks, so
the use of MPP can provide both margin and design relief on spacecraft power
and thermal budgets. MPP requires additional implants during CCD fabrication
and carries the Figure 1
: Plot showing the growth of focal plane arrays
(total number of pixels), as a function of time. Beyond the year 2000,
the cameras shown are planned for 8-10 meter telescope instruments.
In order to achieve optimum sensitivity CCDs are generally
thinned and backside-illuminated. Frontside illuminated CCDs, such as the Wide
Field and Planetary Camera 2 (WFPC2) CCDs built for HST, carry the penalty of
low QE, especially at shorter wavelengths, due to the polysilicon electrode on
the chip. Backside processing of CCDs to achieve high, stable quantum
efficiency involves thinning the substrate to the epitaxial layer,
backside surface passivation by
charging, ion implantation or molecular beam epitaxy (MBE), and the deposition
of an anti-reflection coating optimized for the science application.
The ACS 4kx2k detectors are manufactured by Scientific
Imaging Technology (SITe) and are processed with a proprietary backside
procedure, and anti-reflection coated with the SITe “VIS-AR” coating.
Typical QE for a SITE “VIS-AR” device is shown in
Figure 3, which
contrasts the reported sensitivities of CCDs summarized in Table1. The ACS
devices are thinned to ~13 mm, which provides a good compromise between long
wavelength QE and modulation transfer function (MTF). SITe CCDs are fabricated
with silicon resistivity r~20-40 W
£
-1
, so charge diffusion in the
field free region becomes a significant problem if the epitaxial layer thickness
exceeds ~15 mm. This is an important consideration for the both the ACS Wide Field
Camera (WFC) and many planned missions, which do not fully sample the point spread
function at all wavelengths. In contrast to SITe’s commercial 2kx4k CCD design the
ACS CCDs are 4kx2k to reduce the number of parallel shifts to 2048.
With the onset of radiation damage CTE degrades and it is advantageous to have the
minimum possible number of parallel shifts. SITe has produced the commercial 2kx4k
devices for several large CCD focal plane arrays, including the NOAO and CTIO
8kx8k imagers (Wolfe et al. 1998). The Massachusetts Institute of Technology Lincoln Laboratory (MIT-LL) is also producing 2kx4k CCDs in large numbers for ground-based astronomy. The MIT-LL devices are built for a consortium of observatories (Burke et al. 1996, Luppino et al. 1998, Wei et al. 1998) including the University of Hawaii and the European Southern Observatory. The CCDs are thick (~40 mm) epitaxial thickness devices yielding enhanced long wavelength QE (see Figure 3).
In order to ensure that significant charge is not lost to
adjacent pixels due to drift in the field free zone, these devices are
fabricated with high resisitivity silicon (r~10^3 W
£-1).
Backside processing is achieved by a process of ion implantation with laser
annealing. The MIT-LL CCDs have low read noises in the range of
2 e- RMS and can be read out at Mhz rates. The CCD design employed in
the ground-based arrays does not have MPP implants since it is typically
operated at –120oC, where dark current is ~2 e-
pixel-1hr-1.
MIT-LL also produces 2kx4k CCDs with a new CCD architecture known as the
orthogonal transfer (OT) CCD (Tonry et al. 1997), which permits interpixel
charge shifting during integration. The OTCCD is primarily intended for
compensation of wavefront tilt (image motion) in real time at large
ground-based observatories, but is also suitable for specialized applications
in spacebased missions.
Marconi Applied Technology (MAT) also manufactures 2kx4k CCDs
for astronomical applications (Pool 1996) and have recently populated the
focal planes of several large telescopes including the CFHT. Marconi offers a
significant degree of customization of their 2kx4k CCDs, including 13.5 mm or
15 mm pixel sizes, and the option of 40 mm thick devices for high long
wavelength QE as shown, or blue optimized devices as shown in
Figure 3.
Marconi have also recently introduced a package that permits four side
abutment of 2x4k CCDs with the minimum loss of active area yielding ~90% focal
plane coverage in a large mosaic.
Lockheed/Fairchild are now manufacturing 2kx2k, 2kx4k and
also a single 4kx4k CCD with four amplifier readout for astronomical imaging.
These devices are currently being thinned and backside processed by Dr Michael
Lesser using his CAT chemisorption process, which yields the typical QE shown
in Figure 3.
Lockheed/Fairchild plan to thin devices in-house using the Lesser
backside process and have recently delivered an NUV optimized CCD for the
Triana mission.
An alternative to CCD thinning is to fabricate a CCD on a
thick substrate of very high resistivity silicon (10,000 W
£
-1
) and to operate the device with the entire 300 mm substrate fully
depleted (Holland et al. 1996). This approach has the benefit that CCD
thinning is not required, thus, improving device yields and reducing
processing costs. "Deep depletion" CCDs can be illuminated from the
backside to preserve blue spectral response and yield excellent
long-wavelength performance, as shown by the measurements of Stover et al.
(1998) in Figure 4.
Long wavelength fringing due to interference found in thinned, 13-25 mm epitaxial
thickness CCDs is also avoided with this approach to CCD architecture.
These devices are now being fabricated in formats as large as 2kx4k and are planned for
use in the SNAP mission. The relative thickness of these devices does result in high
bulk dark current rates of 12 e -pixel-1hr-1 at –130oC,
however, LBL plans to reduce the substrate thickness to 200 mm to help reduce the dark
current.
Clearly, the optical QE performance of CCDs is close to
reaching its limits, however, there is considerable scope for improvement in
their near-UV (200 nm – 400 nm) and far-UV (100 nm – 200 nm) performance.
Photon counters have traditionally dominated UV imaging and spectroscopy,
since the combination of low UV to optical flux ratios in astronomy combined
with their high visible QE places CCDs at a disadvantage. However, photon
counting detectors have their own limitations. Large format, photon-counting
UV detectors are available (Siegmund et al. 1992; Kimble et al. 1997), but can
be easily damaged if their modest local, or global count rate limits are
exceeded. For wide field imaging this can result in significant costs to
safely operate these detectors. Improvements in CCD read noise performance,
combined with significantly improved UV spectral response has made CCDs
competitive for some UV applications. There are two approaches to making UV
sensitive CCDs, downconverting phosphors, and backside processing.
UV sensitive phosphors such as those employed on WFPC and
WFPC2 have achieved UV QEs of 14% at wavelengths from 120-400 nm. Phosphors
are not stable in vacuum, and require passivation by an inert gas which can
provide a conduction path from the CCD to the dewar window, creating a
cold-trap for molecular contaminants (Clampin 1992). UV photons have a
relatively short absorption length in silicon, so detecting UV photons with a
thinned CCD requires careful passivation of the CCD backside by ion
implantation or backside charging techniques such as UV flooding, flashgate or
chemisorption, to achieve stable UV QE. The Space Telescope Imaging
Spectrograph (STIS) successfully developed and flew a 1kx1k CCD with a SITe
near-UV backside process enhanced for 200-400 nm (Kimble et al. 1994), as
shown in the comparison of near-UV CCDs in
Figure 4. SITe improved this
backside process for ACS to achieve the near-UV performance also shown in
Figure 5.
For the ACS near-UV camera, Lesser and Iyer (1998) have combined
their CAT chemisorption charging process with a HfO2
anti-reflection coating to produce a device which offers high QE in the
near-UV and good QE in the optical. The chemisorption process derives from the
passivated platinum flashgate (PPTF) technique first demonstrated by Janesick
et al. (1987), and subsequently adapted by Lesser (1994).
The results of the CAT chemisorption process are presented in
Figure 4
for the ACS flight near-UV detector, a
SITe
1kx1k STIS CCD. MAT also offer a near-UV sensitive CCD backside treatment for
their CCDs which is also shown in
Figure 4.
The primary technical problems in UV imaging with CCDs are
their high QE in the visible, low UV flux levels and low sky backgrounds.
While near-UV filters can be designed with effective red blocking, the
problem for far-UV imaging remains a challenge. Filters with visible light
suppression factors of at least ~106
are required for effective far-UV imaging. Woods filters can
achieve these suppression levels, but at the cost of relatively low far-UV
throughput. More work is required to develop technologies for efficient
visible-light rejection technologies such Wood’s filters, made with Sodium
and also lithium and potassium (see McCandless et al. 1998). Currently, the
most effective backside process for far-UV CCD imaging is “Delta-Doping”
which uses molecular beam epitaxy (MBE) to passivate the backside (McCandliss
1999; Nikzad et al. 1994). Typical quantum efficiencies obtained by Nikzad
with Delta Doped CCDs are shown in
Figure 5. It should be noted that quantum yield which becomes important at
wavelengths less than 350 nm,
with ~2.8 electrons being produced per photon at 122 nm.
Low UV flux levels and sky backgrounds mandate low read noise
and dark current values. Typical read noise values are currently as low as 2 e-
RMS on the MAT and MIT-LL devices. Sub-electron noise performance could be
useful in the far-UV and is possible by employing the Skipper amplifier design
(Janesick 1990). Dark currents on par with photon counting devices require CCD
operating temperatures of <100oC, combined with MPP device
architecture.
3.
Challenges
CCDs are beginning to face competition from technologies that
offer new capabilities such as radiation hardness, low power consumption and
integrated camera systems. These technologies include Active Pixel Sensors
(Pain 2000) and Hybrid CMOS focal plane arrays (Vural 2000) and have yet to
attain the low noise performance and scalability of CCDs. However, they
clearly pose a number of challenges that need to be addressed by development
of CCD technologies with a focus on the needs of future space applications.
There are a number of methods to combat the effects of
radiation damage on CCDs, the simplest of which is to minimize the number of
parallel readouts. An example of this approach is the ACS 4kx2k CCD, where the
standard SITe commercial 2kx4k design was modified to reduce the number of
parallel shifts from 4k to 2k by moving the serial register
and increasing its size to 4k pixels. Mosaicing a larger
number of smaller chips is also an option, especially with the arrival of 4
side buttable packages, such as those offered by MAT. A technology which has
been available for some time is the notch, or supplementary buried channel
(SBC), which restricts small charge packets to a smaller cross section within
the pixel, thus reducing the number of traps seen by the signal.
The use of smaller pixels can also provide a similar benefit. Robbins
(2000) has done a number of simulations which suggests that a 3 mm
mini-channel could be expected to produce a factor of two improvement in
Charge Transfer Inefficiency (CTI) at low signal levels,
where CTI = (1 – CTE) . However,
work is
required to demonstrate the true performance of SBCs for low signal level
imaging, and understand their interaction with other device architectures such
as MPP.
P channel CCDs are another option for increasing the
radiation tolerance of CCDs. The P channel design prevents the formation of
silicon-E centers, the primary cause of trapping (following 1010 10
MeV protons cm2 there will be ~2 101 silicon-E centers
cm-3. Hopkinson (1999) recently demonstrated a reduction in CTI by
at least a factor three in a prototype device fabricated at MAT. Several other
groups are now fabricating P channel CCDs to evaluate their radiation
tolerance, including GSFC, and the LBL who are fabricating fully depleted P
channel devices.
For specialized programs charge can be injected into the
parallel register to improve CTE during readout. The method can be optimized
for different observations, for instance a sequence of charge filled rows can
be placed in front of several targets, so that there these rows are clocked
out ahead of the targets to fill traps without an increase in shot noise for
the targets. The effectiveness is a function of the trap’s emission time
constant. This method has been employed effectively for X-Ray missions, and
should prove especially useful in forthcoming imaging astrometry programs such
as Fame and GAIA. The more traditional way to fill traps prior to an
observation is to preflash the CCD with a calibration lamp, however, this
carries a penalty in lost sensitivity due to increased shot noise in
observations.
Hot pixels are dark current spikes caused by field enhanced
emission from defects in high field regions.
Figure 7 shows the growth of
different hot pixel populations in the Space Telescope Imaging Spectrograph’s
(STIS) 1kx1k CCD. Hot pixels are likely to be a feature of competing
technologies to CCDs as well. Experience with HST’s CCD cameras has shown
that annealing monthly at temperatures of 0oC – 20oC
can significantly curtail the growth
rate of hot pixels. More research is required to determine an optimum practical
temperature for annealing CCDs during flight operations. CCD Controllers
APS and CMOS imagers are particularly attractive because of
their relatively simple and compact controller requirements. The growth in
large format CCD mosaics for ground-based astronomy has been accompanied by
the development of increasingly sophisticated CCD controller systems able to
handle multiple readout channels, such as the ESO Fiera controller (D' Odorico
et al.1998), and the SAO Megacam (Geary and Amato 1998). The extension of
these technologies to large format CCD arrays for future space astronomy
missions will present significant challenges in controlling power consumption,
heat dissipation and instrument cost. This point is illustrated by the case of
ACS where the WFC dissipates up to 27
Watts, and the support electronics
for both CCD cameras
dissipates 197 Watts (Rafal 1998). French et al. (1998) have addressed
this issue by developing an Application Specific Integrated Circuit (ASIC)
which can perform the tasks of waveform generation and sequencing on a single
chip for space-based CCD applications (Waltham et al. 1995). This ASIC
controller is currently being ported to a radiation-hard CMOS process and will
offer a significant gain in reduced mass, power consumption and heat
dissipation. Single chip controllers also offer improved reliability and
protection from single component failures, since they are easily implemented
in redundant configurations for very low cost. Hybrid imaging technologies
(HIT) offer similar benefits with an approach that merges the best features of
CCD and CMOS technologies.
HIT uses bump-bonding to couple CMOS
circuitry to a CCD-based imaging array and is illustrated in
Figure 8 which shows
a device being developed in a program led
by Wadsworth at the Jet Propulsion Laboratory (JPL). A similar concept is being developed
by MIT-LL and is discussed in these proceeding by Bautz. The development of these techniques
is vitally important in order to support the next generation of billion pixel CCD mosaics
in space.
Alternate materials
In the past few years, the
quality of wide band gap materials has
significantly improved and the development of optimized UV sensitive CCDs is
now a possibility. The poor quality of wide band gap materials to date has
resulted in unacceptable trap densities, resulting in poor CTE. In
addition to the possibility of “solar blind” UV imaging, these devices
should also provide excellent radiation tolerance.
CCDs have already been fabricated from SiC
(Sheppard et al. 1996), and Mott (2000) has initiated a program at the
GSFC to develop a GaN CCD. As the commercial market for SiC and GaN
semiconductors develops, opportunities for fabricating wide band gap CCDs
should be pursued.
4.
Summary
Several vendors are currently producing 2kx2k, 2kx4k and 4x4k
CCDs suitable for the next generation of missions featuring large format CCD
arrays. The practical requirements for flight science operations of CCDs
suggest that further development is, however, required to address some of the
issues we have addressed in this review.
Radiation tolerance is still the major hurdle facing
billion pixel imagers for flight applications. Further work is required to
determine the optimum application of SBCs (notches), evaluate the utility of p
channel CCD architectures, and new radiation tolerant materials such as SIC.
Large format CCD focal planes might be better built from larger numbers of
smaller devices (e.g. 2kx2k). New 4 side buttable packaging concepts should be
further developed to facilitate high density arrays of smaller format CCDs. Devices with smaller pixels sizes
will also help address the radiation tolerance issue by reducing pixel cross
section. Given the large number of programs planning to use similar CCDs in
the next generation of missions, the development of a standard flight package
concept for 2kx2k and 2kx4k CCDs would also help reduce mission costs across
programs.
The practical problems of operating large arrays of CCDs for
flight has not yet been addressed. There is an immediate need for
hybrid CCD/CMOS technologies and “single-chip” controllers to be
developed for future missions in order to contain their costs.
Access to vendors able to fabricate flight CCDs is a concern
first raised in the FOSI report (1993). The next generation of billion pixel
missions will need access to vendors who can meet large volume CCD fabrication
schedules, possibly on P type architectures,
at a time when the traditional scientific CCD market may be under attack from alternative
technologies such as CMOS imagers.
5.
Acknowledgements
I am indebted to many people who helped me in the preparation
of this paper, in particular, Morley Blouke, Bill Burmester, Holland Ford,
Steve Holland, Jim Janesick, Murzy Jhabula, Paul Jorden, Mike Jones, Mike
Lesser, Gerry Luppino, Stephan McCandliss, Shouleh Nikzad, Mark Robbins, Scott
Sheppard, Marco Sirianni, Richard Stover,
Paul Vu, Mark Wadsworth and Brad Whitmore. This work was supported by
NASA Grant NAS5-32865.
6.
References |