next up previous index
Next: Index Up: Future Technology Previous: Design and Status of


2048 x 2048 HgCdTe Arrays for Astronomy at Visible and Infrared Wavelengths

Donald N.B. Hall Institute for Astronomy, University of Hawaii  



In collaboration with the European Southern Observatory and the Subaru project, the University of Hawaii Institute for Astronomy has entered into a partnership with Rockwell International Science Center to develop 2048 x 2048 format HAWAII-2 HgCdTe arrays for low background ground based astronomical applications. Present plans call for production of 2048 x 2048 multiplexers late in 1998 and hybrid arrays utilizing 2.5 $\mu $m PACE HgCdTe in 1999.

However Rockwell has recently achieved a breakthrough in HgCdTe material quality through use of new detector growth technologies which utilize p/n double layer planar heterostructure arrays grown by MBE on lattice matched CdZnTe substrates. Initial tests of this material have demonstrated near theoretical dark current as a function of temperature and cutoff wavelength.

UH IfA and Rockwell plan to evaluate arrays utilizing this new material for use in the Next Generation Space Telescope. The primary goal would be to directly compare the performance of $\lambda $c = 5 $\mu $m cut off wavelength 2048 x 2048 HgCdTe arrays with that of 1024 x 1024 InSb arrays. However HgCdTe offers the advantage that $\lambda $c can be tailored to any value from 1.2 to 17 $\mu $m; a second goal of the investigation would be to fabricate, test and demonstrate 2048 x 2048 HgCdTe arrays with $\lambda $c values of 1.8 and 2.5 $\mu $m Although the investigation would focus primarily on 1 - 5 $\mu $m detectors, the same multiplexers could also be hybridized to Si PIN diode arrays, providing sensitivity down through the visible into the ultraviolet.

These large format arrays offer potential advantages for both space and ground based applications.


Over the last two decades, dramatic improvements in the performance of 1 - 5 $\mu $m infrared detector materials combined with the leap from single detectors to million pixel arrays have led to a billion-fold improvement in our ability to observe astronomical sources at these wavelengths. These advances, along with those in lightweight mirror and adaptive optics technologies, have already revolutionized ground based infrared astronomy and now pave the way for the science program of NGST.

InSb has long been the detector material of choice for ground based astronomical observations spanning the 1 - 5 $\mu $m region and also has space heritage as the detector material chosen for ISO and SIRTF. However HgCdTe is now an attractive alternate technology which offers very significant advantages as a detector in the 1 - 5 $\mu $m region. 2048 x 2048 format arrays - over four million pixels (Mpxl) - are now under development and the first are scheduled for delivery in the last quarter of 1998. New molecular beam epitaxial (MBE) HgCdTe material offers potentially superior detector performance to InSb along with the additional flexibility to tailor the band-gap to set the long wavelength cutoff ($\lambda $c) to any value from 1.2 to 17 $\mu $m. This is particularly valuable under low background conditions as the theoretical dark current is reduced by more than an order of magnitude for each 10% reduction in $\lambda $c; diode capacitance and hence read noise is also reduced with $\lambda $c.

Over the last decade the University of Hawaii (UH) Institute for Astronomy (IfA) and Rockwell International Science Center (Rockwell) have collaborated in a series of programs to develop large format HgCdTe arrays optimized for low background astronomical applications. UH now has three years of experience operating 1024 x 1024 format, 2.5 $\mu $m HAWAII arrays at telescopes on Mauna Kea. Rockwell has already delivered 30 pairs of engineering and science grade HAWAII arrays to other customers and is in the process of filling orders for a similar number. Read noises of 12 e-, reduced to 3 e-with multiple reads, and dark currents below 1 e-per minute have been demonstrated for these HAWAII arrays.

Rockwell has recently achieved a breakthrough in HgCdTe material quality through use of new detector growth technologies which involve p/n double layer planar heterostructure arrays grown by MBE on lattice matched CdZnTe substrates. Initial tests of 256 x 256 arrays of this material have demonstrated near theoretical dark current as a function of temperature and cutoff wavelength (Figure 2). The new material is lattice matched to the CdZnTe substrate, eliminating the lattice defects of earlier PACE material which are thought to be responsible for both its reduced quantum efficiency at short wavelengths and also image persistence effects. In contrast to the low index sapphire substrate used in PACE material, the refractive index of CdZnTe is well matched to the HgCdTe detector material; the exposed surface of the CdZnTe substrate can thus readily be anti reflection coated and Rockwell is now routinely achieving quantum efficiencies in excess of 80% with this technology.

In November 1997, IfA entered into a contract with Rockwell for the development of a 2048 x 2048 format, 18 $\mu $m pitch, low noise HAWAII-2 multiplexer (mux) based on the proven HAWAII technology; the European Southern Observatory and the Subaru Telescope have also joined this effort. The first run of this 4 Mpxl mux will take place late summer in the same Rockwell production foundry where high yields of the HAWAII muxes have been consistently achieved. This contract specifies early 1999 delivery of two science grade arrays to UH; they will consist of the 4 Mpxl muxes hybridized to traditional 2.5 $\mu $m PACE HgCdTe. Rockwell intends to subsequently offer arrays utilizing this mux to the broad astronomical community. The 37 mm physical dimensions of the 4 Mpxl HgCdTe arrays are only fractionally (one third) larger than the corresponding 27mm of the 1024 x 1024 ALADDIN InSb arrays and the 18 $\mu $m pixel size reduces the rate of ionizing events per pixel by better than a factor of two.

The same muxes can also be readily hybridized to both to Si diode arrays and to MBE HgCdTe material with cutoff wavelengths as long as17 $\mu $m, thus providing the option of wavelength coverage from the ultraviolet through much of the MIR. For many applications this offers the flexibility of extended wavelength coverage with the operational simplicity of using the same multiplexers at the same operating temperature..

The planed NGST investigation would consist of the following tasks:

Fabricate 2048 x 2048 muxes for this program by augmenting the silicon foundry runs already planned under the HAWAII-2 subcontract.
Produce $\lambda $c = 1.8, 2.5 and 5 $\mu $m HgCdTe wafers utilizing p/n double layer planar heterostructure MBE material on CdZnTe substrates.
Initially hybridize 256 x 256 HgCdTe sub arrays with these three cutoff wavelengths to high quality sections of engineering grade 2048 x 2048 muxes. Later in the program full 2048 x 2048 hybrid devices would be produced.
Test and demonstrate the performance of these arrays as they relate to NGST requirements down to a temperature of 30 K. The performance of the $\lambda $c = 5 $\mu $m devices will be specifically characterized relative to available 1024 x 1024 InSb devices. Initial measurements at Rockwell will be followed up with detailed evaluation in the laboratory and at the telescope in Hawaii.
Evaluate the visible wavelength sensitivity of thinned HgCdTe at temperatures of 30 - 50 K relative to available data on thinned InSb and demonstrate the performance of 2048 x 2048 HAWAII-2 muxes hybridized to Si PIN diode sub-arrays.
Evaluate, and subject to availability of funding, demonstrate incorporation of these arrays into mosaic focal planes.
Assess the resources needed to deliver significant numbers of these arrays.


The HAWAII-2 mux represents only a modest jump from the 1024 x 1024 HAWAII mux developed under a similar contract in late 1993 and early 1994. The HAWAII devices have been characterized in great detail and continue to provide unparalleled results in astronomical applications worldwide.

Although the 2048 x 2048 device will be, by a factor of four, the largest device yet fabricated in deep submicron CMOS technology, Rockwell extrapolates high yields based on previous experience with their world-class foundry. The yield of the HAWAII muxes has increased 400% over the last two years and, on the basis of ongoing improvements driven by worldwide competition, we anticipate a defect free yield approaching 30% for the larger muxes.

Operation at the 30K temperature of the NGST focal plane represents a significant extrapolation of the 60 - 77K temperatures at which HAWAII muxes have been operated. The lower operating temperature was not anticipated to be a problem and we have confirmed that two critical parameters, gain and read noise, for an engineering grade HAWAII array show no significant change between 78 and 30 K. The results, shown in Figure 1, demonstrate a change of less than 1.5% in mux gain and no measurable change in read noise (23 vs. 22.7 electrons). The performance of both muxes and hybrid arrays at temperatures down to 30k will be fully evaluated during the proposed investigation as described in Section 5..

Figure 1: Noise squared vs. signal shows no discernable change between 78 and 30 K.

1 - 5 $\mu $m HgCdTe Detector Material

Rockwell will utilize molecular beam epitaxy (MBE), the highest performance HgCdTe detector growth technology available, to fabricate the 2048 x 2048 arrays. Based on this proven, flexible and highly reproducible technology, we expect that the focal plane arrays produced will provide excellent performance and operability. The key attributes of MBE HgCdTe detectors are:

Figure: The measured temperature dependence of dark current of MBE HgCdTe with $\lambda $c's near 2.2, 5 and 10 $\mu $m tracks theoretical predictions far better than traditional PACE material.

Rockwell's MBE technology has been shown to produce the best HgCdTe FPA's in the industry spanning cutoff wavelengths from 1.85 to 17.3 $\mu $m.

Tailoring $\lambda $c to shorter wavelengths also improves read noise as the junction capacitance decreases with $\lambda $c. Figure 3 shows the modeled junction capacitance plus input capacitance vs. $\lambda $c for an HgCdTe/CdZnTe MBE double layer planar heterostructure (DLPH) diode. The figure shows that the junction + input capacitance is reduced from 37 to 32 fF as $\lambda $c is reduced from 5 to 2.5 $\mu $m and that there is no change in read noise from 78 to 30K.

Figure: Modeled junction capacitance vs. cut-off wavelength $\lambda $c.

Hybridization of 2048 x 2048 Arrays

The hybrid mating of the 2048 x 2048 NGST FPA will be based on techniques proven in the fabrication of the 1024 x 1024 HAWAII devices. The proposed layout of the 2048 x 2048 hybrid is basically a 4X scale-up of the 1024 x 1024 arrays, with very similar pixel size (18 vs. 18.5 $\mu $m) and the same interconnect dimensions. The use of CdZnTe as the detector substrate should enhance hybridization as the material is easier to polish than the sapphire substrate of the PACE material used in the HAWAII FPAs and produces flatter detector die.

The main challenge in scaling up the hybrid mating method is the need to apply four times the force to the hybrid components to produce a robust indium interconnect bond. Rockwell's current mating machine is capable of mounting and aligning the larger die of the 2048 x 2048 device, but is not designed to apply the required loads. Our plan for hybridization of the HAWAII-2 arrays is to use this mating machine to align and attach the hybrid components at the load currently used for the 1024 x 1024 arrays, then transfer the hybrid to a very rigid air gimbal in a load frame. Such a compressive apparatus can be made much more rigid than a mating machine, since there is no need to accurately control the movement and orientation of the mounting surfaces. The hybrids will then be compressed to the final load, with the air gimbal ensuring that the load is evenly distributed. The air gimbal and other components for this mating method are in house at Rockwell: the first 2048 x 2048 hybrids are scheduled for fabrication under the HAWAII-2 program late in 1998. Mask sets have also been procured for the fabrication of 2048 x 2048 test structures. The 2048 x 2048 hybrid mating technology should be relatively mature and have been fully demonstrated under the HAWAII-2 program well before 2048 x 2048 arrays would be required for the NGST program.

The difference in thermal expansion between the detector and the silicon mux has the potential to create large thermally induced stresses on the indium interconnects and the detector material. With repeated thermal cycling this can result in failure of the interconnects or lead to damage to the detector pixels. Rockwell has developed the Balanced Composite Structure (BCS) design to produce reliable hybrids and applied it to a variety of FPA designs, including the 1024 x 1024 HAWAII FPA. This design relieves the stress at the hybrid interface by compressing the mux to match the thermal expansion of the detector material. The thickness and composition of the compressive layer is adjusted to produce the desired compression of the multiplexer. A balancing layer below the compressing layer prevents bending of the FPA (Figure 4) The BCS design has been successfully applied to a number of FPA's with CdZnTe-substrate detectors and Rockwell does not anticipate the need for any modifications in going to the 2048 x 2048 format hybrids.

Figure 4: The BCS design compresses the mux to match the thermal contraction of the detector without bending the FPA.

Test and Evaluation of Performance Down to 30 K

Both Rockwell and the IfA have existing test facilities capable of characterization of focal plane arrays down to 30 K. We propose to utilize these facilities for initial screening and evaluation of muxes and arrays produced under the proposed investigation. However the verification of full device performance at read noise levels of 2 e and dark currents as low as 4 e / Ksec will require a specialized ultra-low background test facility.

We plan to carry out initial characterization of HAWAII-2 engineering grade muxes hybridized to 256 x 256 sub arrays of 5.0, 2.5 and 1.8 $\mu $m MBE HgCdTe material on lattice matched CdZnTe substrates and to Si PIN diode arrays. Later we intend to demonstrate and characterize full scale 2048 x 2048 array technology. By that time there will exist a substantial body of data on HAWAII-2 muxes hybridized to traditional PACE material.

Options for Extension to Visible Wavelengths

The sensitivity of the proposed 2048 x 2048 devices can be extended through visible wavelengths either by thinning the CdZnTe substrate or by hybridizing an array of Si PIN diodes to the same multiplexer.

Typical HgCdTe FPA's are not sensitive to visible wavelengths due to absorption in the CdZnTe substrate layer short of 0.8 $\mu $m. It is possible to achieve detector response comparable to InSb in the visible by removing the CdZnTe substrate with a selective etch. At least two standard etches are available that remove the CdZnTe substrate but do not affect the layer containing Hg. This process would remove the entire CdZnTe substrate leaving only the very thin active layer of HgCdTe hybridized to the multiplexer. This material would be sensitive to visible wavelengths.

Several issues would need to be addressed in this approach:

The hybrid gap must be back filled with an epoxy or similar material to support the thin HgCdTe layer, but the backfill material must resist the etch
The multiplexer, including the exposed wirebond pads, must resist the etch
The exposed HgCdTe surface must be passivated to avoid large leakage currents.

Rockwell has established technologies which would address all of these issues.

However, the alternate of hybridizing a back-illuminated Si PIN diode array to the 2048 x 2048 mux appears far more attractive. By utilizing this hybrid focal plane array approach, it is possible to produce a visible imager which will operate at 30K with a combination of performance and other features not attainable with either HgCdTe or InSb in the visible. In this hybrid approach, the detector will be independently optimized for visible quantum efficiency, speed and dark current outside of the standard CMOS process, but will use standard high-volume silicon processing techniques. Due to the maturity of silicon and the available material quality, the cosmetic quality of this device should be very high. The fully-depleted architecture provides low cross-talk and reduces any resistivity variations to improve uniformity of response. This hybrid approach allows for a fill factor of nearly 100%; with an anti-reflection coating of SiO2 tailored to the appropriate thickness, quantum efficiencies exceeding 80% throughout the visible should be achievable. If lower dark current than the thinned wafer approach is required, a bonded-wafer approach in which a thin (10 - 75 $\mu $m) silicon layer is transferred to the multiplexer through bump-bonding with the later removal of the thicker bonded substrate can be employed. In this approach dark currents at the levels of CCD's are attainable, while providing the advantages of operation at 30K, utilization of the same 2048 x 2048 multiplexer and non destructive readout.

Mosaic Focal Planes

Although the primary goal of the proposed investigation is the demonstration of 2048 x 2048 FPAs optimized for NGST core program 1 - 5 $\mu $m requirements, with potential extension through visible wavelengths, the eventual NGST goal is an 8 k x 8 k mosaic imager. This will require the assembly of sixteen 2048 x 2048 FPAs with precision packaging designed to minimize the dead space between the arrays. Both Rockwell and UH have extensive experience with such mosaic focal plane technology.
At UH Gerald Luppino has pioneered the development of several generations of CCD mosaic focal planes for wide field visible imaging. He has extensive experience in the packaging for large CCD focal planes and has developed the first 8192 x 8192 CCD mosaiced from eight 2048 x 4096, three side buttable, devices; similar mosaics are now in routine astronomical use and have been widely duplicated. They utilize an Aluminum Nitride (AlN) ceramic which is an ideal thermal expansion match to silicon and is also a superb thermal conductor. The same techniques are readily applicable to IR array mosaics and involve technology very comparable to the proposed NGST mosaics.
Rockwell has extensive experience mounting large FPAs in the pin grid array commonly used to package large microprocessors. These packages connect to zero insertion-force (ZIF) sockets through an array of pins on the back surface, eliminating the need to clamp the package from the front or sides. This approach lends itself to closer acking of the FPAs in the mosaic.
Figure 5 shows the package designed for the HAWAII-2 2048 x 2048 FPA. This package provides a 46mm well for mounting the FPA and 128 electrical connections that are routed to probe pads on the outer edges and also to the outer two rows of pins on the backside. The central pins in the pin grid array are dedicated to thermal contacts. Rockwell proposes to utilize these packages in the fabrication of initial NGST FPA's.
Figure 5: The proposed package for the HAWAII-2 arrays.

The budget constraints of the current program require us to utilize existing tooling available from the package vendor to minimize cost and is therefore not fully optimized for assembly of a close packed mosaic array. The outer edges of the PGA packages are 57 x 57 mm, resulting in a gap of approximately 20 mm between active areas. The package could be optimized by shrinking the well and outer edges of the package to the minimum dimensions shown in Figure 6, allowing the dead space between arrays to be reduced to 7 mm (most of this gap would be occupied by wirebond pads on both the mux and package). A single custom ZIF socket for all sixteen FPAs would allow exact positioning of the arrays and provide denser packing of the arrays by eliminating the some of the gap due to butting of the socket perimeters.

Figure 6: Conceptual design for a PGA package optimized for assembly of a mosaic array.

Production Approaches

MBE HgCdTe Capability at Rockwell Science Center

The science objectives of the NGST program require near-theoretical performance from very large arrays. This means that only a single array can be processed on each standard 4 cm substrate. This requires detectors that are not only very high performance but are also high yield so as to be produced in relatively large numbers. Rockwell is confident that these requirements can be met utilizing MBE technology.

The ability to fabricate MBE HgCdTe with a variety of $\lambda $c's matched to the science requirements is a key advantage of HgCdTE over fixed band-gap materials. An important attribute of Rockwell's process is the ability to precisely control the band-gap for any run and to rapidly modify it from one run to another. This is illustrated in Figure 7 which documents the growth of a series of LWIR, MWIR and SWIR layers which were grown sequentially demanding the high composition control flexibility inherent to this technology.

Rockwell is currently using MBE processes to produce many arrays. These include all LWIR FPA's ($\lambda $c up to 17.3 $\mu $m), key MWIR deliverables for operation at elevated temperatures and all NIR (1.8 to 2.3 $\mu $m) prototypes. The performance benefits of the DLPH MBE HgCdTe detector technology are illustrated by the dark current density data shown in Figure 2 for recent devices with $\lambda $c from 2.2 to 16 $\mu $m. The MBE dark current is far superior to traditional PACE material and is in good agreement with the theoretical limit for conventional (unthinned) p/n HgCdTe detectors.

Figure 7: An important attribute of Rockwell's MBE growth technology is the ability to rapidly modify the composition of the Hg1-x Cdx Te material. The figure illustrates this capability where a series of LWIR, MWIR and SWIR layers were grown sequentially.


In the next several years we expect to develop and evaluate visible and infrared 2048 x 2048 hybrid focal plane arrays optimized for extremely low background astronomical applications. They will be specifically optimized for the Next Generation Space Telescope. However Rockwell intends to make these arrays available to the broad astronomical community. They offer major opportunities for large ground based telescopes such as the VLT, particularly in spectroscopic applications.


The author wishes to acknowledge the substantial contributions of K. W. Hodapp and G. Luppino at the University of Hawaii Institute for Astronomy and K.Vural, J.T. Montroy and L. J. Kozlowski at Rockwell International Science Center to the planning of this project.

next up previous index
Next: Index Up: Future Technology Previous: Design and Status of
Norbert Pirzkal